lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/fs_struct.h>
  43 #include <linux/kallsyms.h>
  44 #include <linux/module.h>
  45 #include <linux/user_namespace.h>
  46 #include <linux/uidgid.h>
  47
  48 /* prerequisite for linux/xattr.h */
  49 #include <linux/types.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/fs.h>
  52 /* XATTR_{REPLACE,CREATE} */
  53 #include <linux/xattr.h>
  54
  55 #include <ldiskfs/ldiskfs.h>
  56 #include <ldiskfs/xattr.h>
  57 #include <ldiskfs/ldiskfs_extents.h>
  58 #undef ENTRY
  59 /*
  60  * struct OBD_{ALLOC,FREE}*()
  61  * OBD_FAIL_CHECK
  62  */
  63 #include <obd_support.h>
  64 /* struct ptlrpc_thread */
  65 #include <lustre_net.h>
  66 #include <lustre_fid.h>
  67 /* process_config */
  68 #include <uapi/linux/lustre/lustre_param.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_dynlocks.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75 #include <lustre_quota.h>
  76
  77 #include <lustre_linkea.h>
  78
  79 /* Maximum EA size is limited by LNET_MTU for remote objects */
  80 #define OSD_MAX_EA_SIZE 1048364
  81
  82 int ldiskfs_pdo = 1;
  83 module_param(ldiskfs_pdo, int, 0644);
  84 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  85
  86 int ldiskfs_track_declares_assert;
  87 module_param(ldiskfs_track_declares_assert, int, 0644);
  88 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  89
  90 /* Slab to allocate dynlocks */
  91 struct kmem_cache *dynlock_cachep;
  92
  93 /* Slab to allocate osd_it_ea */
  94 struct kmem_cache *osd_itea_cachep;
  95
  96 static struct lu_kmem_descr ldiskfs_caches[] = {
  97         {
  98                 .ckd_cache = &dynlock_cachep,
  99                 .ckd_name  = "dynlock_cache",
 100                 .ckd_size  = sizeof(struct dynlock_handle)
 101         },
 102         {
 103                 .ckd_cache = &osd_itea_cachep,
 104                 .ckd_name  = "osd_itea_cache",
 105                 .ckd_size  = sizeof(struct osd_it_ea)
 106         },
 107         {
 108                 .ckd_cache = NULL
 109         }
 110 };
 111
 112 static const char dot[] = ".";
 113 static const char dotdot[] = "..";
 114
 115 static const struct lu_object_operations      osd_lu_obj_ops;
 116 static const struct dt_object_operations      osd_obj_ops;
 117 static const struct dt_object_operations      osd_obj_otable_it_ops;
 118 static const struct dt_index_operations       osd_index_iam_ops;
 119 static const struct dt_index_operations       osd_index_ea_ops;
 120
 121 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 122                           const struct lu_fid *fid);
 123 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 124                                                 struct osd_device *osd);
 125
 126 int osd_trans_declare_op2rb[] = {
 127         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 128         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 129         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 130         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 131         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 132         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 133         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 134         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 135         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 136         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 137         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 138 };
 139
 140 static int osd_has_index(const struct osd_object *obj)
 141 {
 142         return obj->oo_dt.do_index_ops != NULL;
 143 }
 144
 145 static int osd_object_invariant(const struct lu_object *l)
 146 {
 147         return osd_invariant(osd_obj(l));
 148 }
 149
 150 /*
 151  * Concurrency: doesn't matter
 152  */
 153 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 154 {
 155         struct osd_thread_info *oti = osd_oti_get(env);
 156
 157         return oti->oti_w_locks > 0 && o->oo_owner == env;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't access mutable data
 162  */
 163 static int osd_root_get(const struct lu_env *env,
 164                         struct dt_device *dev, struct lu_fid *f)
 165 {
 166         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 167         return 0;
 168 }
 169
 170 /*
 171  * the following set of functions are used to maintain per-thread
 172  * cache of FID->ino mapping. this mechanism is needed to resolve
 173  * FID to inode at dt_insert() which in turn stores ino in the
 174  * directory entries to keep ldiskfs compatible with ext[34].
 175  * due to locking-originated restrictions we can't lookup ino
 176  * using LU cache (deadlock is possible). lookup using OI is quite
 177  * expensive. so instead we maintain this cache and methods like
 178  * dt_create() fill it. so in the majority of cases dt_insert() is
 179  * able to find needed mapping in lockless manner.
 180  */
 181 static struct osd_idmap_cache *
 182 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 183              const struct lu_fid *fid)
 184 {
 185         struct osd_thread_info *oti = osd_oti_get(env);
 186         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 187         int i;
 188
 189         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 190                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 191                         continue;
 192                 if (idc[i].oic_dev != osd)
 193                         continue;
 194
 195                 return idc + i;
 196         }
 197
 198         return NULL;
 199 }
 200
 201 static struct osd_idmap_cache *
 202 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 203             const struct lu_fid *fid)
 204 {
 205         struct osd_thread_info *oti   = osd_oti_get(env);
 206         struct osd_idmap_cache *idc;
 207         int i;
 208
 209         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 210                 i = oti->oti_ins_cache_size * 2;
 211                 if (i == 0)
 212                         i = OSD_INS_CACHE_SIZE;
 213                 OBD_ALLOC_PTR_ARRAY(idc, i);
 214                 if (idc == NULL)
 215                         return ERR_PTR(-ENOMEM);
 216                 if (oti->oti_ins_cache != NULL) {
 217                         memcpy(idc, oti->oti_ins_cache,
 218                                oti->oti_ins_cache_used * sizeof(*idc));
 219                         OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
 220                                            oti->oti_ins_cache_used);
 221                 }
 222                 oti->oti_ins_cache = idc;
 223                 oti->oti_ins_cache_size = i;
 224         }
 225
 226         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 227         idc->oic_fid = *fid;
 228         idc->oic_dev = osd;
 229         idc->oic_lid.oii_ino = 0;
 230         idc->oic_lid.oii_gen = 0;
 231         idc->oic_remote = 0;
 232
 233         return idc;
 234 }
 235
 236 /*
 237  * lookup mapping for the given fid in the cache, initialize a
 238  * new one if not found. the initialization checks whether the
 239  * object is local or remote. for local objects, OI is used to
 240  * learn ino/generation. the function is used when the caller
 241  * has no information about the object, e.g. at dt_insert().
 242  */
 243 static struct osd_idmap_cache *
 244 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 245                      const struct lu_fid *fid)
 246 {
 247         struct osd_idmap_cache *idc;
 248         int rc;
 249
 250         idc = osd_idc_find(env, osd, fid);
 251         LASSERT(!IS_ERR(idc));
 252         if (idc != NULL)
 253                 return idc;
 254
 255         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 256                osd->od_svname, PFID(fid));
 257
 258         /* new mapping is needed */
 259         idc = osd_idc_add(env, osd, fid);
 260         if (IS_ERR(idc)) {
 261                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 262                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 263                 return idc;
 264         }
 265
 266         /* initialize it */
 267         rc = osd_remote_fid(env, osd, fid);
 268         if (unlikely(rc < 0))
 269                 return ERR_PTR(rc);
 270
 271         if (rc == 0) {
 272                 /* the object is local, lookup in OI */
 273                 /* XXX: probably cheaper to lookup in LU first? */
 274                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 275                                    &idc->oic_lid, 0);
 276                 if (unlikely(rc < 0)) {
 277                         CERROR("can't lookup: rc = %d\n", rc);
 278                         return ERR_PTR(rc);
 279                 }
 280         } else {
 281                 /* the object is remote */
 282                 idc->oic_remote = 1;
 283         }
 284
 285         return idc;
 286 }
 287
 288 static void osd_idc_dump_lma(const struct lu_env *env,
 289                                 struct osd_device *osd,
 290                                 unsigned long ino,
 291                                 bool check_in_oi)
 292 {
 293         struct osd_thread_info *info = osd_oti_get(env);
 294         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 295         const struct lu_fid *fid;
 296         struct osd_inode_id lid;
 297         struct inode *inode;
 298         int rc;
 299
 300         inode = osd_ldiskfs_iget(osd_sb(osd), ino);
 301         if (IS_ERR(inode)) {
 302                 CERROR("%s: can't get inode %lu: rc = %d\n",
 303                        osd->od_svname, ino, (int)PTR_ERR(inode));
 304                 return;
 305         }
 306         if (is_bad_inode(inode)) {
 307                 CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
 308                 goto put;
 309         }
 310         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 311         if (rc) {
 312                 CERROR("%s: can't get LMA for %lu: rc = %d\n",
 313                        osd->od_svname, ino, rc);
 314                 goto put;
 315         }
 316         fid = &loa->loa_lma.lma_self_fid;
 317         LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
 318                       PFID(fid), ino, (unsigned)inode->i_generation);
 319         if (!check_in_oi)
 320                 goto put;
 321         rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
 322         if (rc) {
 323                 CERROR("%s: can't lookup "DFID": rc = %d\n",
 324                        osd->od_svname, PFID(fid), rc);
 325                 goto put;
 326         }
 327         LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
 328                       PFID(fid), lid.oii_ino, lid.oii_gen);
 329 put:
 330         iput(inode);
 331 }
 332
 333 static void osd_idc_dump_debug(const struct lu_env *env,
 334                                 struct osd_device *osd,
 335                                 const struct lu_fid *fid,
 336                                 unsigned long ino1,
 337                                 unsigned long ino2)
 338 {
 339         struct osd_inode_id lid;
 340
 341         int rc;
 342
 343         rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
 344         if (!rc) {
 345                 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
 346                         osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
 347                 osd_idc_dump_lma(env, osd, lid.oii_ino, false);
 348         } else {
 349                 CERROR("%s: can't lookup "DFID": rc = %d\n",
 350                        osd->od_svname, PFID(fid), rc);
 351         }
 352         if (ino1)
 353                 osd_idc_dump_lma(env, osd, ino1, true);
 354         if (ino2)
 355                 osd_idc_dump_lma(env, osd, ino2, true);
 356 }
 357
 358 /*
 359  * lookup mapping for given FID and fill it from the given object.
 360  * the object is lolcal by definition.
 361  */
 362 static int osd_idc_find_and_init(const struct lu_env *env,
 363                                  struct osd_device *osd,
 364                                  struct osd_object *obj)
 365 {
 366         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 367         struct osd_idmap_cache *idc;
 368
 369         idc = osd_idc_find(env, osd, fid);
 370         LASSERT(!IS_ERR(idc));
 371         if (idc != NULL) {
 372                 if (obj->oo_inode == NULL)
 373                         return 0;
 374                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 375                         if (idc->oic_lid.oii_ino) {
 376                                 osd_idc_dump_debug(env, osd, fid,
 377                                                    idc->oic_lid.oii_ino,
 378                                                    obj->oo_inode->i_ino);
 379                                 return -EINVAL;
 380                         }
 381                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 382                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 383                 }
 384                 return 0;
 385         }
 386
 387         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 388                osd->od_svname, PFID(fid));
 389
 390         /* new mapping is needed */
 391         idc = osd_idc_add(env, osd, fid);
 392         if (IS_ERR(idc)) {
 393                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 394                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 395                 return PTR_ERR(idc);
 396         }
 397
 398         if (obj->oo_inode != NULL) {
 399                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 400                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 401         }
 402         return 0;
 403 }
 404
 405 /*
 406  * OSD object methods.
 407  */
 408
 409 /*
 410  * Concurrency: no concurrent access is possible that early in object
 411  * life-cycle.
 412  */
 413 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 414                                           const struct lu_object_header *hdr,
 415                                           struct lu_device *d)
 416 {
 417         struct osd_object *mo;
 418
 419         OBD_ALLOC_PTR(mo);
 420         if (mo != NULL) {
 421                 struct lu_object *l;
 422                 struct lu_object_header *h;
 423                 struct osd_device *o = osd_dev(d);
 424
 425                 l = &mo->oo_dt.do_lu;
 426                 if (unlikely(o->od_in_init)) {
 427                         OBD_ALLOC_PTR(h);
 428                         if (!h) {
 429                                 OBD_FREE_PTR(mo);
 430                                 return NULL;
 431                         }
 432
 433                         lu_object_header_init(h);
 434                         lu_object_init(l, h, d);
 435                         lu_object_add_top(h, l);
 436                         mo->oo_header = h;
 437                 } else {
 438                         dt_object_init(&mo->oo_dt, NULL, d);
 439                         mo->oo_header = NULL;
 440                 }
 441
 442                 mo->oo_dt.do_ops = &osd_obj_ops;
 443                 l->lo_ops = &osd_lu_obj_ops;
 444                 init_rwsem(&mo->oo_sem);
 445                 init_rwsem(&mo->oo_ext_idx_sem);
 446                 spin_lock_init(&mo->oo_guard);
 447                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 448                 return l;
 449         }
 450         return NULL;
 451 }
 452
 453 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 454                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 455 {
 456         int rc;
 457
 458         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 459                              (void *)loa, sizeof(*loa));
 460         if (rc > 0) {
 461                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 462
 463                 if (rc < sizeof(*lma))
 464                         return -EINVAL;
 465
 466                 rc = 0;
 467                 lustre_loa_swab(loa, true);
 468                 /* Check LMA compatibility */
 469                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 470                         rc = -EOPNOTSUPP;
 471                         CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
 472                               osd_ino2name(inode),
 473                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 474                               PFID(&lma->lma_self_fid), inode->i_ino, rc);
 475                 }
 476         } else if (rc == 0) {
 477                 rc = -ENODATA;
 478         }
 479
 480         return rc;
 481 }
 482
 483 /*
 484  * retrieve object from backend ext fs.
 485  **/
 486 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 487                        struct osd_inode_id *id)
 488 {
 489         int rc;
 490         struct inode *inode = NULL;
 491
 492         /*
 493          * if we look for an inode withing a running
 494          * transaction, then we risk to deadlock
 495          * osd_dirent_check_repair() breaks this
 496          */
 497          /* LASSERT(current->journal_info == NULL); */
 498
 499         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 500         if (IS_ERR(inode)) {
 501                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 502                        id->oii_ino, PTR_ERR(inode));
 503         } else if (id->oii_gen != OSD_OII_NOGEN &&
 504                    inode->i_generation != id->oii_gen) {
 505                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 506                        "i_generation = %u\n",
 507                        id->oii_ino, id->oii_gen, inode->i_generation);
 508                 iput(inode);
 509                 inode = ERR_PTR(-ESTALE);
 510         } else if (inode->i_nlink == 0) {
 511                 /*
 512                  * due to parallel readdir and unlink,
 513                  * we can have dead inode here.
 514                  */
 515                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 516                 iput(inode);
 517                 inode = ERR_PTR(-ESTALE);
 518         } else if (is_bad_inode(inode)) {
 519                 rc = -ENOENT;
 520                 CWARN("%s: bad inode: ino = %u: rc = %d\n",
 521                       osd_dev2name(dev), id->oii_ino, rc);
 522                 iput(inode);
 523                 inode = ERR_PTR(rc);
 524         } else if ((rc = osd_attach_jinode(inode))) {
 525                 iput(inode);
 526                 inode = ERR_PTR(rc);
 527         } else {
 528                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 529                 if (id->oii_gen == OSD_OII_NOGEN)
 530                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 531
 532                 /*
 533                  * Do not update file c/mtime in ldiskfs.
 534                  * NB: we don't have any lock to protect this because we don't
 535                  * have reference on osd_object now, but contention with
 536                  * another lookup + attr_set can't happen in the tiny window
 537                  * between if (...) and set S_NOCMTIME.
 538                  */
 539                 if (!(inode->i_flags & S_NOCMTIME))
 540                         inode->i_flags |= S_NOCMTIME;
 541         }
 542         return inode;
 543 }
 544
 545 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 546                           handle_t *handle, struct dentry *child,
 547                           struct inode *inode, struct htree_lock *hlock)
 548 {
 549         int rc, rc2;
 550
 551         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 552         if (rc == -ENOBUFS || rc == -ENOSPC) {
 553                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 554                 struct inode *parent = child->d_parent->d_inode;
 555                 struct lu_fid *fid = NULL;
 556                 char fidstr[FID_LEN + 1] = "unknown";
 557
 558                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 559                 if (!rc2) {
 560                         fid = &loa->loa_lma.lma_self_fid;
 561                 } else if (rc2 == -ENODATA) {
 562                         if (unlikely(is_root_inode(parent))) {
 563                                 fid = &info->oti_fid3;
 564                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 565                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 566                                 fid = &info->oti_fid3;
 567                                 lu_igif_build(fid, parent->i_ino,
 568                                               parent->i_generation);
 569                         }
 570                 }
 571
 572                 if (fid != NULL)
 573                         snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
 574
 575                 /* below message is checked in sanity.sh test_129 */
 576                 if (rc == -ENOSPC) {
 577                         CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
 578                               osd_name(osd), parent->i_ino, fidstr);
 579                 } else {
 580                         rc = 0; /* ignore such error now */
 581                         CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
 582                               osd_name(osd), parent->i_ino, fidstr);
 583                 }
 584
 585         }
 586
 587         return rc;
 588 }
 589
 590
 591 struct inode *
 592 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 593              struct osd_inode_id *id, struct lu_fid *fid)
 594 {
 595         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 596         struct inode *inode;
 597         int rc;
 598
 599         inode = osd_iget(info, dev, id);
 600         if (IS_ERR(inode))
 601                 return inode;
 602
 603         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 604         if (!rc) {
 605                 *fid = loa->loa_lma.lma_self_fid;
 606         } else if (rc == -ENODATA) {
 607                 if (unlikely(is_root_inode(inode)))
 608                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 609                 else
 610                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 611         } else {
 612                 iput(inode);
 613                 inode = ERR_PTR(rc);
 614         }
 615         return inode;
 616 }
 617
 618 static struct inode *osd_iget_check(struct osd_thread_info *info,
 619                                     struct osd_device *dev,
 620                                     const struct lu_fid *fid,
 621                                     struct osd_inode_id *id,
 622                                     bool trusted)
 623 {
 624         struct inode *inode;
 625         int rc = 0;
 626
 627         ENTRY;
 628
 629         /*
 630          * The cached OI mapping is trustable. If we cannot locate the inode
 631          * via the cached OI mapping, then return the failure to the caller
 632          * directly without further OI checking.
 633          */
 634
 635 again:
 636         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 637         if (IS_ERR(inode)) {
 638                 rc = PTR_ERR(inode);
 639                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 640                         goto check_oi;
 641
 642                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 643                        PFID(fid), id->oii_ino, rc);
 644                 GOTO(put, rc);
 645         }
 646
 647         if (is_bad_inode(inode)) {
 648                 rc = -ENOENT;
 649                 if (!trusted)
 650                         goto check_oi;
 651
 652                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 653                        PFID(fid), id->oii_ino);
 654                 GOTO(put, rc);
 655         }
 656
 657         if (id->oii_gen != OSD_OII_NOGEN &&
 658             inode->i_generation != id->oii_gen) {
 659                 rc = -ESTALE;
 660                 if (!trusted)
 661                         goto check_oi;
 662
 663                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 664                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 665                        id->oii_ino, id->oii_gen, inode->i_generation);
 666                 GOTO(put, rc);
 667         }
 668
 669         if (inode->i_nlink == 0) {
 670                 rc = -ENOENT;
 671                 if (!trusted)
 672                         goto check_oi;
 673
 674                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 675                        PFID(fid), id->oii_ino);
 676                 GOTO(put, rc);
 677         }
 678
 679         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 680
 681 check_oi:
 682         if (rc != 0) {
 683                 __u32 saved_ino = id->oii_ino;
 684                 __u32 saved_gen = id->oii_gen;
 685
 686                 LASSERT(!trusted);
 687                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 688
 689                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 690                 /*
 691                  * XXX: There are four possible cases:
 692                  *      1. rc = 0.
 693                  *         Backup/restore caused the OI invalid.
 694                  *      2. rc = 0.
 695                  *         Someone unlinked the object but NOT removed
 696                  *         the OI mapping, such as mount target device
 697                  *         as ldiskfs, and modify something directly.
 698                  *      3. rc = -ENOENT.
 699                  *         Someone just removed the object between the
 700                  *         former oi_lookup and the iget. It is normal.
 701                  *      4. Other failure cases.
 702                  *
 703                  *      Generally, when the device is mounted, it will
 704                  *      auto check whether the system is restored from
 705                  *      file-level backup or not. We trust such detect
 706                  *      to distinguish the 1st case from the 2nd case:
 707                  *      if the OI files are consistent but may contain
 708                  *      stale OI mappings because of case 2, if iget()
 709                  *      returns -ENOENT or -ESTALE, then it should be
 710                  *      the case 2.
 711                  */
 712                 if (rc != 0)
 713                         /*
 714                          * If the OI mapping was in OI file before the
 715                          * osd_iget_check(), but now, it is disappear,
 716                          * then it must be removed by race. That is a
 717                          * normal race case.
 718                          */
 719                         GOTO(put, rc);
 720
 721                 /*
 722                  * It is the OI scrub updated the OI mapping by race.
 723                  * The new OI mapping must be valid.
 724                  */
 725                 if (saved_ino != id->oii_ino ||
 726                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 727                         if (!IS_ERR(inode))
 728                                 iput(inode);
 729
 730                         trusted = true;
 731                         goto again;
 732                 }
 733
 734                 if (IS_ERR(inode)) {
 735                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 736                             SF_INCONSISTENT)
 737                                 /*
 738                                  * It still can be the case 2, but we cannot
 739                                  * distinguish it from the case 1. So return
 740                                  * -EREMCHG to block current operation until
 741                                  *  OI scrub rebuilt the OI mappings.
 742                                  */
 743                                 rc = -EREMCHG;
 744                         else
 745                                 rc = -ENOENT;
 746
 747                         GOTO(put, rc);
 748                 }
 749
 750                 if (inode->i_generation == id->oii_gen)
 751                         rc = -ENOENT;
 752                 else
 753                         rc = -EREMCHG;
 754         } else {
 755                 if (id->oii_gen == OSD_OII_NOGEN)
 756                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 757
 758                 /*
 759                  * Do not update file c/mtime in ldiskfs.
 760                  * NB: we don't have any lock to protect this because we don't
 761                  * have reference on osd_object now, but contention with
 762                  * another lookup + attr_set can't happen in the tiny window
 763                  * between if (...) and set S_NOCMTIME.
 764                  */
 765                 if (!(inode->i_flags & S_NOCMTIME))
 766                         inode->i_flags |= S_NOCMTIME;
 767         }
 768
 769         GOTO(put, rc);
 770
 771 put:
 772         if (rc != 0) {
 773                 if (!IS_ERR(inode))
 774                         iput(inode);
 775
 776                 inode = ERR_PTR(rc);
 777         }
 778
 779         return inode;
 780 }
 781
 782 /**
 783  * \retval +v: new filter_fid does not contain self-fid
 784  * \retval 0:  filter_fid_18_23, contains self-fid
 785  * \retval -v: other failure cases
 786  */
 787 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 788                  struct dentry *dentry, struct lu_fid *fid)
 789 {
 790         struct filter_fid *ff = &info->oti_ff;
 791         struct ost_id *ostid = &info->oti_ostid;
 792         int rc;
 793
 794         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 795         if (rc == sizeof(struct filter_fid_18_23)) {
 796                 struct filter_fid_18_23 *ff_old = (void *)ff;
 797
 798                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 799                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 800                 /*
 801                  * XXX: use 0 as the index for compatibility, the caller will
 802                  * handle index related issues when necessary.
 803                  */
 804                 if (!rc)
 805                         ostid_to_fid(fid, ostid, 0);
 806         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 807                 rc = 1;
 808         } else if (rc >= 0) {
 809                 rc = -EINVAL;
 810         }
 811
 812         return rc;
 813 }
 814
 815 static int osd_lma_self_repair(struct osd_thread_info *info,
 816                                struct osd_device *osd, struct inode *inode,
 817                                const struct lu_fid *fid, __u32 compat)
 818 {
 819         handle_t *jh;
 820         int rc;
 821
 822         LASSERT(current->journal_info == NULL);
 823
 824         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 825                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 826         if (IS_ERR(jh)) {
 827                 rc = PTR_ERR(jh);
 828                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 829                       osd_name(osd), rc);
 830                 return rc;
 831         }
 832
 833         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 834         if (rc != 0)
 835                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 836                       osd_name(osd), rc);
 837         ldiskfs_journal_stop(jh);
 838         return rc;
 839 }
 840
 841 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 842 {
 843         struct osd_thread_info *info = osd_oti_get(env);
 844         struct osd_device *osd = osd_obj2dev(obj);
 845         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 846         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 847         struct inode *inode = obj->oo_inode;
 848         struct dentry *dentry = &info->oti_obj_dentry;
 849         struct lu_fid *fid = NULL;
 850         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 851         int rc;
 852
 853         ENTRY;
 854
 855         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 856                              (void *)loa, sizeof(*loa));
 857         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 858                 fid = &lma->lma_self_fid;
 859                 rc = osd_get_idif(info, inode, dentry, fid);
 860                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 861                         /*
 862                          * For the given OST-object, if it has neither LMA nor
 863                          * FID in XATTR_NAME_FID, then the given FID (which is
 864                          * contained in the @obj, from client RPC for locating
 865                          * the OST-object) is trusted. We use it to generate
 866                          * the LMA.
 867                          */
 868                         osd_lma_self_repair(info, osd, inode, rfid,
 869                                             LMAC_FID_ON_OST);
 870                         RETURN(0);
 871                 }
 872         }
 873
 874         if (rc < 0)
 875                 RETURN(rc);
 876
 877         if (rc > 0) {
 878                 rc = 0;
 879                 lustre_lma_swab(lma);
 880                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 881                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 882                               S_ISREG(inode->i_mode)))) {
 883                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 884                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 885                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 886                               PFID(rfid), inode->i_ino);
 887                         rc = -EOPNOTSUPP;
 888                 } else {
 889                         fid = &lma->lma_self_fid;
 890                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 891                             osd->od_is_ost)
 892                                 obj->oo_pfid_in_lma = 1;
 893                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 894                             !osd->od_is_ost)
 895                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 896                 }
 897         }
 898
 899         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 900                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 901                         struct ost_id   *oi   = &info->oti_ostid;
 902                         struct lu_fid   *fid1 = &info->oti_fid3;
 903                         __u32            idx  = fid_idif_ost_idx(rfid);
 904
 905                         /*
 906                          * For old IDIF, the OST index is not part of the IDIF,
 907                          * Means that different OSTs may have the same IDIFs.
 908                          * Under such case, we need to make some compatible
 909                          * check to make sure to trigger OI scrub properly.
 910                          */
 911                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 912                                 /* Given @rfid is new, LMA is old. */
 913                                 fid_to_ostid(fid, oi);
 914                                 ostid_to_fid(fid1, oi, idx);
 915                                 if (lu_fid_eq(fid1, rfid)) {
 916                                         if (osd->od_index_in_idif)
 917                                                 osd_lma_self_repair(info, osd,
 918                                                         inode, rfid,
 919                                                         LMAC_FID_ON_OST);
 920                                         RETURN(0);
 921                                 }
 922                         }
 923                 }
 924
 925                 rc = -EREMCHG;
 926         }
 927
 928         RETURN(rc);
 929 }
 930
 931 struct osd_check_lmv_buf {
 932         /* please keep it as first member */
 933         struct dir_context ctx;
 934         struct osd_thread_info *oclb_info;
 935         struct osd_device *oclb_dev;
 936         struct osd_idmap_cache *oclb_oic;
 937         int oclb_items;
 938         bool oclb_found;
 939 };
 940
 941 /**
 942  * It is called internally by ->iterate*() to filter out the
 943  * local slave object's FID of the striped directory.
 944  *
 945  * \retval      1 found the local slave's FID
 946  * \retval      0 continue to check next item
 947  * \retval      -ve for failure
 948  */
 949 #ifdef HAVE_FILLDIR_USE_CTX
 950 static int osd_stripe_dir_filldir(struct dir_context *buf,
 951 #else
 952 static int osd_stripe_dir_filldir(void *buf,
 953 #endif
 954                                   const char *name, int namelen,
 955                                   loff_t offset, __u64 ino, unsigned int d_type)
 956 {
 957         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 958         struct osd_thread_info *oti = oclb->oclb_info;
 959         struct lu_fid *fid = &oti->oti_fid3;
 960         struct osd_inode_id *id = &oti->oti_id3;
 961         struct osd_device *dev = oclb->oclb_dev;
 962         struct osd_idmap_cache *oic = oclb->oclb_oic;
 963         struct inode *inode;
 964
 965         oclb->oclb_items++;
 966
 967         if (name[0] == '.')
 968                 return 0;
 969
 970         fid_zero(fid);
 971         sscanf(name + 1, SFID, RFID(fid));
 972         if (!fid_is_sane(fid))
 973                 return 0;
 974
 975         if (osd_remote_fid(oti->oti_env, dev, fid))
 976                 return 0;
 977
 978         osd_id_gen(id, ino, OSD_OII_NOGEN);
 979         inode = osd_iget(oti, dev, id);
 980         if (IS_ERR(inode))
 981                 return PTR_ERR(inode);
 982
 983         iput(inode);
 984         osd_add_oi_cache(oti, dev, id, fid);
 985         oic->oic_fid = *fid;
 986         oic->oic_lid = *id;
 987         oic->oic_dev = dev;
 988         osd_oii_insert(dev, oic, true);
 989         oclb->oclb_found = true;
 990
 991         return 1;
 992 }
 993
 994 /*
 995  * When lookup item under striped directory, we need to locate the master
 996  * MDT-object of the striped directory firstly, then the client will send
 997  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 998  * and the item's name. If the system is restored from MDT file level backup,
 999  * then before the OI scrub completely built the OI files, the OI mappings of
1000  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
1001  * not a problem for the master MDT-object. Because when locate the master
1002  * MDT-object, we will do name based lookup (for the striped directory itself)
1003  * firstly, during such process we can setup the correct OI mapping for the
1004  * master MDT-object. But it will be trouble for the slave MDT-object. Because
1005  * the client will not trigger name based lookup on the MDT to locate the slave
1006  * MDT-object before locating item under the striped directory, then when
1007  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
1008  * is invalid and does not know what the right OI mapping is, then the MDT has
1009  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
1010  * the OI file, related OI mapping is unknown yet, please try again later. And
1011  * then client will re-try the RPC again and again until related OI mapping has
1012  * been updated. That is quite inefficient.
1013  *
1014  * To resolve above trouble, we will handle it as the following two cases:
1015  *
1016  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
1017  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
1018  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
1019  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
1020  *    directly. Please check osd_fid_lookup().
1021  *
1022  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
1023  *    Under such case, during lookup the master MDT-object, we will lookup the
1024  *    slave MDT-object via readdir against the master MDT-object, because the
1025  *    slave MDT-objects information are stored as sub-directories with the name
1026  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
1027  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
1028  *    the correct OI mapping for the slave MDT-object.
1029  */
1030 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
1031                          struct inode *inode, struct osd_idmap_cache *oic)
1032 {
1033         struct lu_buf *buf = &oti->oti_big_buf;
1034         struct dentry *dentry = &oti->oti_obj_dentry;
1035         struct file *filp;
1036         struct lmv_mds_md_v1 *lmv1;
1037         struct osd_check_lmv_buf oclb = {
1038                 .ctx.actor = osd_stripe_dir_filldir,
1039                 .oclb_info = oti,
1040                 .oclb_dev = dev,
1041                 .oclb_oic = oic,
1042                 .oclb_found = false,
1043         };
1044         int rc = 0;
1045
1046         ENTRY;
1047
1048 again:
1049         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
1050                              buf->lb_len);
1051         if (rc == -ERANGE) {
1052                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
1053                 if (rc > 0) {
1054                         lu_buf_realloc(buf, rc);
1055                         if (buf->lb_buf == NULL)
1056                                 GOTO(out, rc = -ENOMEM);
1057
1058                         goto again;
1059                 }
1060         }
1061
1062         if (unlikely(rc == 0 || rc == -ENODATA))
1063                 GOTO(out, rc = 0);
1064
1065         if (rc < 0)
1066                 GOTO(out, rc);
1067
1068         if (unlikely(buf->lb_buf == NULL)) {
1069                 lu_buf_realloc(buf, rc);
1070                 if (buf->lb_buf == NULL)
1071                         GOTO(out, rc = -ENOMEM);
1072
1073                 goto again;
1074         }
1075
1076         lmv1 = buf->lb_buf;
1077         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1078                 GOTO(out, rc = 0);
1079
1080         filp = osd_quasi_file(oti->oti_env, inode);
1081         rc = osd_security_file_alloc(filp);
1082         if (rc)
1083                 goto out;
1084
1085         do {
1086                 oclb.oclb_items = 0;
1087                 rc = iterate_dir(filp, &oclb.ctx);
1088         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1089                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1090         inode->i_fop->release(inode, filp);
1091
1092 out:
1093         if (rc < 0)
1094                 CDEBUG(D_LFSCK,
1095                        "%s: cannot check LMV, ino = %lu/%u "DFID": rc = %d\n",
1096                        osd_ino2name(inode), inode->i_ino, inode->i_generation,
1097                        PFID(&oic->oic_fid), rc);
1098         else
1099                 rc = 0;
1100
1101         RETURN(rc);
1102 }
1103
1104 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1105                           const struct lu_fid *fid,
1106                           const struct lu_object_conf *conf)
1107 {
1108         struct osd_thread_info *info;
1109         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1110         struct osd_device *dev;
1111         struct osd_idmap_cache *oic;
1112         struct osd_inode_id *id;
1113         struct inode *inode = NULL;
1114         struct lustre_scrub *scrub;
1115         struct scrub_file *sf;
1116         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1117         __u32 saved_ino;
1118         __u32 saved_gen;
1119         int result = 0;
1120         int rc1 = 0;
1121         bool remote = false;
1122         bool trusted = true;
1123         bool updated = false;
1124         bool checked = false;
1125
1126         ENTRY;
1127
1128         LINVRNT(osd_invariant(obj));
1129         LASSERT(obj->oo_inode == NULL);
1130
1131         if (fid_is_sane(fid) == 0) {
1132                 CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
1133                        PFID(fid));
1134                 dump_stack();
1135                 RETURN(-EINVAL);
1136         }
1137
1138         dev = osd_dev(ldev);
1139         scrub = &dev->od_scrub.os_scrub;
1140         sf = &scrub->os_file;
1141         info = osd_oti_get(env);
1142         LASSERT(info);
1143         oic = &info->oti_cache;
1144
1145         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1146                 RETURN(-ENOENT);
1147
1148         /*
1149          * For the object is created as locking anchor, or for the object to
1150          * be created on disk. No need to osd_oi_lookup() at here because FID
1151          * shouldn't never be re-used, if it's really a duplicate FID from
1152          * unexpected reason, we should be able to detect it later by calling
1153          * do_create->osd_oi_insert().
1154          */
1155         if (conf && conf->loc_flags & LOC_F_NEW)
1156                 GOTO(out, result = 0);
1157
1158         /* Search order: 1. per-thread cache. */
1159         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1160                 id = &oic->oic_lid;
1161                 goto iget;
1162         }
1163
1164         id = &info->oti_id;
1165         if (!list_empty(&scrub->os_inconsistent_items)) {
1166                 /* Search order: 2. OI scrub pending list. */
1167                 result = osd_oii_lookup(dev, fid, id);
1168                 if (!result)
1169                         goto iget;
1170         }
1171
1172         /*
1173          * The OI mapping in the OI file can be updated by the OI scrub
1174          * when we locate the inode via FID. So it may be not trustable.
1175          */
1176         trusted = false;
1177
1178         /* Search order: 3. OI files. */
1179         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1180         if (result == -ENOENT) {
1181                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1182                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1183                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1184                                       sf->sf_oi_bitmap))
1185                         GOTO(out, result = 0);
1186
1187                 goto trigger;
1188         }
1189
1190         /* -ESTALE is returned if inode of OST object doesn't exist */
1191         if (result == -ESTALE &&
1192             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1193                 GOTO(out, result = 0);
1194         }
1195
1196         if (result)
1197                 GOTO(out, result);
1198
1199 iget:
1200         obj->oo_inode = NULL;
1201         /* for later passes through checks, not true on first pass */
1202         if (!IS_ERR_OR_NULL(inode))
1203                 iput(inode);
1204
1205         inode = osd_iget_check(info, dev, fid, id, trusted);
1206         if (!IS_ERR(inode)) {
1207                 obj->oo_inode = inode;
1208                 result = 0;
1209                 if (remote)
1210                         goto trigger;
1211
1212                 goto check_lma;
1213         }
1214
1215         result = PTR_ERR(inode);
1216         if (result == -ENOENT || result == -ESTALE)
1217                 GOTO(out, result = 0);
1218
1219         if (result != -EREMCHG)
1220                 GOTO(out, result);
1221
1222 trigger:
1223         /*
1224          * We still have chance to get the valid inode: for the
1225          * object which is referenced by remote name entry, the
1226          * object on the local MDT will be linked under the dir
1227          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1228          *
1229          * We do not know whether the object for the given FID
1230          * is referenced by some remote name entry or not, and
1231          * especially for DNE II, a multiple-linked object may
1232          * have many name entries reside on many MDTs.
1233          *
1234          * To simplify the operation, OSD will not distinguish
1235          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1236          * only happened for the RPC from other MDT during the
1237          * OI scrub, or for the client side RPC with FID only,
1238          * such as FID to path, or from old connected client.
1239          */
1240         if (!remote) {
1241                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1242                 if (!rc1) {
1243                         remote = true;
1244                         trusted = true;
1245                         flags |= SS_AUTO_PARTIAL;
1246                         flags &= ~SS_AUTO_FULL;
1247                         goto iget;
1248                 }
1249         }
1250
1251         if (scrub->os_running) {
1252                 if (scrub->os_partial_scan && !scrub->os_in_join)
1253                         goto join;
1254
1255                 osd_add_oi_cache(info, dev, id, fid);
1256                 if (IS_ERR_OR_NULL(inode) || result) {
1257                         osd_oii_insert(dev, oic, result == -ENOENT);
1258                         GOTO(out, result = -EINPROGRESS);
1259                 }
1260
1261                 LASSERT(remote);
1262                 LASSERT(obj->oo_inode == inode);
1263
1264                 osd_oii_insert(dev, oic, true);
1265                 goto found;
1266         }
1267
1268         if (dev->od_auto_scrub_interval == AS_NEVER) {
1269                 if (!remote)
1270                         GOTO(out, result = -EREMCHG);
1271
1272                 LASSERT(!result);
1273                 LASSERT(obj->oo_inode == inode);
1274
1275                 osd_add_oi_cache(info, dev, id, fid);
1276                 goto found;
1277         }
1278
1279 join:
1280         rc1 = osd_scrub_start(env, dev, flags);
1281         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1282                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1283         if (rc1 && rc1 != -EALREADY)
1284                 GOTO(out, result = -EREMCHG);
1285
1286         osd_add_oi_cache(info, dev, id, fid);
1287         if (IS_ERR_OR_NULL(inode) || result) {
1288                 osd_oii_insert(dev, oic, result == -ENOENT);
1289                 GOTO(out, result = -EINPROGRESS);
1290         }
1291
1292         LASSERT(remote);
1293         LASSERT(obj->oo_inode == inode);
1294
1295         osd_oii_insert(dev, oic, true);
1296         goto found;
1297
1298 check_lma:
1299         checked = true;
1300         if (unlikely(obj->oo_header))
1301                 goto found;
1302
1303         result = osd_check_lma(env, obj);
1304         if (!result)
1305                 goto found;
1306
1307         LASSERTF(id->oii_ino == inode->i_ino &&
1308                  id->oii_gen == inode->i_generation,
1309                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1310                  PFID(fid), id->oii_ino, id->oii_gen,
1311                  inode->i_ino, inode->i_generation);
1312
1313         saved_ino = inode->i_ino;
1314         saved_gen = inode->i_generation;
1315
1316         if (unlikely(result == -ENODATA)) {
1317                 /*
1318                  * If the OI scrub updated the OI mapping by race, it
1319                  * must be valid. Trust the inode that has no LMA EA.
1320                  */
1321                 if (updated)
1322                         goto found;
1323
1324                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1325                 if (!result) {
1326                         /*
1327                          * The OI mapping is still there, the inode is still
1328                          * valid. It is just becaues the inode has no LMA EA.
1329                          */
1330                         if (saved_ino == id->oii_ino &&
1331                             saved_gen == id->oii_gen)
1332                                 goto found;
1333
1334                         /*
1335                          * It is the OI scrub updated the OI mapping by race.
1336                          * The new OI mapping must be valid.
1337                          */
1338                         trusted = true;
1339                         updated = true;
1340                         goto iget;
1341                 }
1342
1343                 /*
1344                  * "result == -ENOENT" means that the OI mappinghas been
1345                  * removed by race, so the inode belongs to other object.
1346                  *
1347                  * Others error can be returned  directly.
1348                  */
1349                 if (result == -ENOENT) {
1350                         LASSERT(trusted);
1351
1352                         obj->oo_inode = NULL;
1353                         result = 0;
1354                 }
1355         }
1356
1357         if (result != -EREMCHG)
1358                 GOTO(out, result);
1359
1360         LASSERT(!updated);
1361
1362         /*
1363          * if two OST objects map to the same inode, and inode mode is
1364          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1365          * reserved by precreate, and not written yet, in this case, don't
1366          * set inode for the object whose FID mismatch, so that it can create
1367          * inode and not block precreate.
1368          */
1369         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1370             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1371                 obj->oo_inode = NULL;
1372                 GOTO(out, result = 0);
1373         }
1374
1375         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1376         /*
1377          * "result == -ENOENT" means the cached OI mapping has been removed
1378          * from the OI file by race, above inode belongs to other object.
1379          */
1380         if (result == -ENOENT) {
1381                 LASSERT(trusted);
1382
1383                 obj->oo_inode = NULL;
1384                 GOTO(out, result = 0);
1385         }
1386
1387         if (result)
1388                 GOTO(out, result);
1389
1390         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1391                 result = -EREMCHG;
1392                 goto trigger;
1393         }
1394
1395         /*
1396          * It is the OI scrub updated the OI mapping by race.
1397          * The new OI mapping must be valid.
1398          */
1399         trusted = true;
1400         updated = true;
1401         goto iget;
1402
1403 found:
1404         if (!checked) {
1405                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1406                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1407
1408                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1409                 if (!result) {
1410                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1411                             dev->od_is_ost)
1412                                 obj->oo_pfid_in_lma = 1;
1413                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1414                             !dev->od_is_ost)
1415                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1416                 } else if (result != -ENODATA) {
1417                         GOTO(out, result);
1418                 }
1419         }
1420
1421         obj->oo_compat_dot_created = 1;
1422         obj->oo_compat_dotdot_created = 1;
1423
1424         if (S_ISDIR(inode->i_mode) &&
1425             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1426                 osd_check_lmv(info, dev, inode, oic);
1427
1428         result = osd_attach_jinode(inode);
1429         if (result)
1430                 GOTO(out, result);
1431
1432         if (!ldiskfs_pdo)
1433                 GOTO(out, result = 0);
1434
1435         LASSERT(!obj->oo_hl_head);
1436         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1437
1438         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1439
1440 out:
1441         if (result || !obj->oo_inode) {
1442                 if (!IS_ERR_OR_NULL(inode))
1443                         iput(inode);
1444
1445                 obj->oo_inode = NULL;
1446                 if (trusted)
1447                         fid_zero(&oic->oic_fid);
1448         }
1449
1450         LINVRNT(osd_invariant(obj));
1451         return result;
1452 }
1453
1454 /*
1455  * Concurrency: shouldn't matter.
1456  */
1457 static void osd_object_init0(struct osd_object *obj)
1458 {
1459         LASSERT(obj->oo_inode != NULL);
1460         obj->oo_dt.do_body_ops = &osd_body_ops;
1461         obj->oo_dt.do_lu.lo_header->loh_attr |=
1462                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1463 }
1464
1465 /*
1466  * Concurrency: no concurrent access is possible that early in object
1467  * life-cycle.
1468  */
1469 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1470                            const struct lu_object_conf *conf)
1471 {
1472         struct osd_object *obj = osd_obj(l);
1473         int result;
1474
1475         LINVRNT(osd_invariant(obj));
1476
1477         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1478             cfs_fail_val == 2) {
1479                 struct osd_thread_info *info = osd_oti_get(env);
1480                 struct osd_idmap_cache *oic = &info->oti_cache;
1481                 /* invalidate thread cache */
1482                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1483         }
1484         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1485                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1486                 l->lo_header->loh_attr |= LOHA_EXISTS;
1487                 return 0;
1488         }
1489
1490         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1491         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1492         if (result == 0 && obj->oo_inode != NULL) {
1493                 struct osd_thread_info *oti = osd_oti_get(env);
1494                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1495
1496                 osd_object_init0(obj);
1497                 if (unlikely(obj->oo_header))
1498                         return 0;
1499
1500                 result = osd_get_lma(oti, obj->oo_inode,
1501                                      &oti->oti_obj_dentry, loa);
1502                 if (!result) {
1503                         /*
1504                          * Convert LMAI flags to lustre LMA flags
1505                          * and cache it to oo_lma_flags
1506                          */
1507                         obj->oo_lma_flags =
1508                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1509                 } else if (result == -ENODATA) {
1510                         result = 0;
1511                 }
1512         }
1513         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1514
1515         LINVRNT(osd_invariant(obj));
1516         return result;
1517 }
1518
1519 /*
1520  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1521  * The left part is for value, binary mode.
1522  */
1523 struct osd_xattr_entry {
1524         struct list_head        oxe_list;
1525         size_t                  oxe_len;
1526         size_t                  oxe_namelen;
1527         bool                    oxe_exist;
1528         struct rcu_head         oxe_rcu;
1529         char                    oxe_buf[0];
1530 };
1531
1532 static int osd_oxc_get(struct osd_object *obj, const char *name,
1533                        struct lu_buf *buf)
1534 {
1535         struct osd_xattr_entry *tmp;
1536         struct osd_xattr_entry *oxe = NULL;
1537         size_t namelen = strlen(name);
1538         int rc;
1539
1540         rcu_read_lock();
1541         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1542                 if (namelen == tmp->oxe_namelen &&
1543                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1544                         oxe = tmp;
1545                         break;
1546                 }
1547         }
1548
1549         if (oxe == NULL)
1550                 GOTO(out, rc = -ENOENT);
1551
1552         if (!oxe->oxe_exist)
1553                 GOTO(out, rc = -ENODATA);
1554
1555         /* vallen */
1556         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1557         LASSERT(rc > 0);
1558
1559         if (buf->lb_buf == NULL)
1560                 GOTO(out, rc);
1561
1562         if (buf->lb_len < rc)
1563                 GOTO(out, rc = -ERANGE);
1564
1565         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1566 out:
1567         rcu_read_unlock();
1568
1569         return rc;
1570 }
1571
1572 static void osd_oxc_free(struct rcu_head *head)
1573 {
1574         struct osd_xattr_entry *oxe;
1575
1576         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1577         OBD_FREE(oxe, oxe->oxe_len);
1578 }
1579
1580 static void osd_oxc_add(struct osd_object *obj, const char *name,
1581                         const char *buf, int buflen)
1582 {
1583         struct osd_xattr_entry *oxe;
1584         struct osd_xattr_entry *old = NULL;
1585         struct osd_xattr_entry *tmp;
1586         size_t namelen = strlen(name);
1587         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1588
1589         OBD_ALLOC(oxe, len);
1590         if (oxe == NULL)
1591                 return;
1592
1593         INIT_LIST_HEAD(&oxe->oxe_list);
1594         oxe->oxe_len = len;
1595         oxe->oxe_namelen = namelen;
1596         memcpy(oxe->oxe_buf, name, namelen);
1597         if (buflen > 0) {
1598                 LASSERT(buf != NULL);
1599                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1600                 oxe->oxe_exist = true;
1601         } else {
1602                 oxe->oxe_exist = false;
1603         }
1604
1605         /* this should be rarely called, just remove old and add new */
1606         spin_lock(&obj->oo_guard);
1607         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1608                 if (namelen == tmp->oxe_namelen &&
1609                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1610                         old = tmp;
1611                         break;
1612                 }
1613         }
1614         if (old != NULL) {
1615                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1616                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1617         } else {
1618                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1619         }
1620         spin_unlock(&obj->oo_guard);
1621 }
1622
1623 static void osd_oxc_del(struct osd_object *obj, const char *name)
1624 {
1625         struct osd_xattr_entry *oxe;
1626         size_t namelen = strlen(name);
1627
1628         spin_lock(&obj->oo_guard);
1629         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1630                 if (namelen == oxe->oxe_namelen &&
1631                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1632                         list_del_rcu(&oxe->oxe_list);
1633                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1634                         break;
1635                 }
1636         }
1637         spin_unlock(&obj->oo_guard);
1638 }
1639
1640 static void osd_oxc_fini(struct osd_object *obj)
1641 {
1642         struct osd_xattr_entry *oxe, *next;
1643
1644         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1645                 list_del(&oxe->oxe_list);
1646                 OBD_FREE(oxe, oxe->oxe_len);
1647         }
1648 }
1649
1650 /*
1651  * Concurrency: no concurrent access is possible that late in object
1652  * life-cycle.
1653  */
1654 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1655 {
1656         struct osd_object *obj = osd_obj(l);
1657         struct lu_object_header *h = obj->oo_header;
1658
1659         LINVRNT(osd_invariant(obj));
1660
1661         osd_oxc_fini(obj);
1662         dt_object_fini(&obj->oo_dt);
1663         if (obj->oo_hl_head != NULL)
1664                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1665         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1666         OBD_FREE_PTR(obj);
1667         if (unlikely(h))
1668                 lu_object_header_free(h);
1669 }
1670
1671 /*
1672  * Concurrency: no concurrent access is possible that late in object
1673  * life-cycle.
1674  */
1675 static void osd_index_fini(struct osd_object *o)
1676 {
1677         struct iam_container *bag;
1678
1679         if (o->oo_dir != NULL) {
1680                 bag = &o->oo_dir->od_container;
1681                 if (o->oo_inode != NULL) {
1682                         if (bag->ic_object == o->oo_inode)
1683                                 iam_container_fini(bag);
1684                 }
1685                 OBD_FREE_PTR(o->oo_dir);
1686                 o->oo_dir = NULL;
1687         }
1688 }
1689
1690 enum {
1691         OSD_TXN_OI_DELETE_CREDITS    = 20,
1692         OSD_TXN_INODE_DELETE_CREDITS = 20
1693 };
1694
1695 /*
1696  * Journal
1697  */
1698
1699 #if OSD_THANDLE_STATS
1700 /**
1701  * Set time when the handle is allocated
1702  */
1703 static void osd_th_alloced(struct osd_thandle *oth)
1704 {
1705         oth->oth_alloced = ktime_get();
1706 }
1707
1708 /**
1709  * Set time when the handle started
1710  */
1711 static void osd_th_started(struct osd_thandle *oth)
1712 {
1713         oth->oth_started = ktime_get();
1714 }
1715
1716 /**
1717  * Check whether the we deal with this handle for too long.
1718  */
1719 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1720                                 ktime_t alloced, ktime_t started,
1721                                 ktime_t closed)
1722 {
1723         ktime_t now = ktime_get();
1724
1725         LASSERT(dev != NULL);
1726
1727         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1728                             ktime_us_delta(started, alloced));
1729         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1730                             ktime_us_delta(closed, started));
1731         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1732                             ktime_us_delta(now, closed));
1733
1734         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1735                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1736                                 oth, now, alloced, started, closed);
1737                 libcfs_debug_dumpstack(NULL);
1738         }
1739 }
1740
1741 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1742 {                                                                       \
1743         ktime_t __closed = ktime_get();                                 \
1744         ktime_t __alloced = oth->oth_alloced;                           \
1745         ktime_t __started = oth->oth_started;                           \
1746                                                                         \
1747         expr;                                                           \
1748         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1749 }
1750
1751 #else /* OSD_THANDLE_STATS */
1752
1753 #define osd_th_alloced(h)                  do {} while(0)
1754 #define osd_th_started(h)                  do {} while(0)
1755 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1756
1757 #endif /* OSD_THANDLE_STATS */
1758
1759 /*
1760  * Concurrency: doesn't access mutable data.
1761  */
1762 static int osd_param_is_not_sane(const struct osd_device *dev,
1763                                  const struct thandle *th)
1764 {
1765         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1766
1767         return oh->ot_credits > osd_transaction_size(dev);
1768 }
1769
1770 /*
1771  * Concurrency: shouldn't matter.
1772  */
1773 static void osd_trans_commit_cb(struct super_block *sb,
1774                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1775 {
1776         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1777         struct thandle *th = &oh->ot_super;
1778         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1779         struct osd_device *osd = osd_dev(lud);
1780         struct dt_txn_commit_cb *dcb, *tmp;
1781
1782         LASSERT(oh->ot_handle == NULL);
1783
1784         if (error)
1785                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1786
1787         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1788         /* call per-transaction callbacks if any */
1789         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1790                                  dcb_linkage) {
1791                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1792                          "commit callback entry: magic=%x name='%s'\n",
1793                          dcb->dcb_magic, dcb->dcb_name);
1794                 list_del_init(&dcb->dcb_linkage);
1795                 dcb->dcb_func(NULL, th, dcb, error);
1796         }
1797
1798         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1799         if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
1800                 wake_up(&osd->od_commit_cb_done);
1801         th->th_dev = NULL;
1802
1803         OBD_FREE_PTR(oh);
1804 }
1805
1806 static struct thandle *osd_trans_create(const struct lu_env *env,
1807                                         struct dt_device *d)
1808 {
1809         struct osd_thread_info *oti = osd_oti_get(env);
1810         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1811         struct osd_thandle *oh;
1812         struct thandle *th;
1813
1814         ENTRY;
1815
1816         if (d->dd_rdonly) {
1817                 CERROR("%s: someone try to start transaction under "
1818                        "readonly mode, should be disabled.\n",
1819                        osd_name(osd_dt_dev(d)));
1820                 dump_stack();
1821                 RETURN(ERR_PTR(-EROFS));
1822         }
1823
1824         /* on pending IO in this thread should left from prev. request */
1825         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1826
1827         sb_start_write(osd_sb(osd_dt_dev(d)));
1828
1829         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1830         if (!oh) {
1831                 sb_end_write(osd_sb(osd_dt_dev(d)));
1832                 RETURN(ERR_PTR(-ENOMEM));
1833         }
1834
1835         oh->ot_quota_trans = &oti->oti_quota_trans;
1836         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1837         th = &oh->ot_super;
1838         th->th_dev = d;
1839         th->th_result = 0;
1840         oh->ot_credits = 0;
1841         oh->oh_declared_ext = 0;
1842         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1843         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1844         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1845         osd_th_alloced(oh);
1846
1847         memset(oti->oti_declare_ops, 0,
1848                sizeof(oti->oti_declare_ops));
1849         memset(oti->oti_declare_ops_cred, 0,
1850                sizeof(oti->oti_declare_ops_cred));
1851         memset(oti->oti_declare_ops_used, 0,
1852                sizeof(oti->oti_declare_ops_used));
1853
1854         oti->oti_ins_cache_depth++;
1855
1856         RETURN(th);
1857 }
1858
1859 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1860 {
1861         struct osd_thread_info *oti = osd_oti_get(env);
1862         struct osd_thandle *oh;
1863
1864         oh = container_of(th, struct osd_thandle, ot_super);
1865         LASSERT(oh != NULL);
1866
1867         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1868               oti->oti_declare_ops[OSD_OT_CREATE],
1869               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1870               oti->oti_declare_ops_used[OSD_OT_CREATE],
1871               oti->oti_declare_ops[OSD_OT_DESTROY],
1872               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1873               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1874         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1875               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1876               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1877               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1878               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1879               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1880               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1881         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1882               oti->oti_declare_ops[OSD_OT_WRITE],
1883               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1884               oti->oti_declare_ops_used[OSD_OT_WRITE],
1885               oti->oti_declare_ops[OSD_OT_PUNCH],
1886               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1887               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1888               oti->oti_declare_ops[OSD_OT_QUOTA],
1889               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1890               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1891         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1892               oti->oti_declare_ops[OSD_OT_INSERT],
1893               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1894               oti->oti_declare_ops_used[OSD_OT_INSERT],
1895               oti->oti_declare_ops[OSD_OT_DELETE],
1896               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1897               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1898         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1899               oti->oti_declare_ops[OSD_OT_REF_ADD],
1900               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1901               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1902               oti->oti_declare_ops[OSD_OT_REF_DEL],
1903               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1904               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1905 }
1906
1907 /*
1908  * Concurrency: shouldn't matter.
1909  */
1910 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1911                            struct thandle *th)
1912 {
1913         struct osd_thread_info *oti = osd_oti_get(env);
1914         struct osd_device *dev = osd_dt_dev(d);
1915         handle_t *jh;
1916         struct osd_thandle *oh;
1917         int rc;
1918
1919         ENTRY;
1920
1921         LASSERT(current->journal_info == NULL);
1922
1923         oh = container_of(th, struct osd_thandle, ot_super);
1924         LASSERT(oh != NULL);
1925         LASSERT(oh->ot_handle == NULL);
1926
1927         rc = dt_txn_hook_start(env, d, th);
1928         if (rc != 0)
1929                 GOTO(out, rc);
1930
1931         if (unlikely(osd_param_is_not_sane(dev, th))) {
1932                 static unsigned long last_printed;
1933                 static int last_credits;
1934
1935                 /*
1936                  * don't make noise on a tiny testing systems
1937                  * actual credits misuse will be caught anyway
1938                  */
1939                 if (last_credits != oh->ot_credits &&
1940                     time_after(jiffies, last_printed +
1941                                cfs_time_seconds(60)) &&
1942                     osd_transaction_size(dev) > 512) {
1943                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1944                               oh->ot_credits, osd_transaction_size(dev));
1945                         osd_trans_dump_creds(env, th);
1946                         libcfs_debug_dumpstack(NULL);
1947                         last_credits = oh->ot_credits;
1948                         last_printed = jiffies;
1949                 }
1950                 /*
1951                  * XXX Limit the credits to 'max_transaction_buffers', and
1952                  *     let the underlying filesystem to catch the error if
1953                  *     we really need so many credits.
1954                  *
1955                  *     This should be removed when we can calculate the
1956                  *     credits precisely.
1957                  */
1958                 oh->ot_credits = osd_transaction_size(dev);
1959         } else if (ldiskfs_track_declares_assert != 0) {
1960                 /*
1961                  * reserve few credits to prevent an assertion in JBD
1962                  * our debugging mechanism will be able to detected
1963                  * overuse. this can help to debug single-update
1964                  * transactions
1965                  */
1966                 oh->ot_credits += 10;
1967                 if (unlikely(osd_param_is_not_sane(dev, th)))
1968                         oh->ot_credits = osd_transaction_size(dev);
1969         }
1970
1971         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1972                 GOTO(out, rc = -EIO);
1973
1974         /*
1975          * XXX temporary stuff. Some abstraction layer should
1976          * be used.
1977          */
1978         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1979         osd_th_started(oh);
1980         if (!IS_ERR(jh)) {
1981                 oh->ot_handle = jh;
1982                 LASSERT(oti->oti_txns == 0);
1983
1984                 atomic_inc(&dev->od_commit_cb_in_flight);
1985                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1986                               "osd-tx", th);
1987                 oti->oti_txns++;
1988                 rc = 0;
1989         } else {
1990                 rc = PTR_ERR(jh);
1991         }
1992 out:
1993         RETURN(rc);
1994 }
1995
1996 static int osd_seq_exists(const struct lu_env *env,
1997                           struct osd_device *osd, u64 seq)
1998 {
1999         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
2000         struct seq_server_site *ss = osd_seq_site(osd);
2001         int rc;
2002
2003         ENTRY;
2004
2005         LASSERT(ss != NULL);
2006         LASSERT(ss->ss_server_fld != NULL);
2007
2008         rc = osd_fld_lookup(env, osd, seq, range);
2009         if (rc != 0) {
2010                 if (rc != -ENOENT)
2011                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
2012                                osd_name(osd), seq, rc);
2013                 RETURN(0);
2014         }
2015
2016         RETURN(ss->ss_node_id == range->lsr_index);
2017 }
2018
2019 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
2020 {
2021         struct dt_txn_commit_cb *dcb;
2022         struct dt_txn_commit_cb *tmp;
2023
2024         /* call per-transaction stop callbacks if any */
2025         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
2026                                  dcb_linkage) {
2027                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
2028                          "commit callback entry: magic=%x name='%s'\n",
2029                          dcb->dcb_magic, dcb->dcb_name);
2030                 list_del_init(&dcb->dcb_linkage);
2031                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
2032         }
2033 }
2034
2035 /*
2036  * Concurrency: shouldn't matter.
2037  */
2038 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
2039                           struct thandle *th)
2040 {
2041         struct osd_thread_info *oti = osd_oti_get(env);
2042         struct osd_thandle *oh;
2043         struct osd_iobuf *iobuf = &oti->oti_iobuf;
2044         struct osd_device *osd = osd_dt_dev(th->th_dev);
2045         struct qsd_instance *qsd = osd_def_qsd(osd);
2046         struct lquota_trans *qtrans;
2047         LIST_HEAD(truncates);
2048         int rc = 0, remove_agents = 0;
2049
2050         ENTRY;
2051
2052         oh = container_of(th, struct osd_thandle, ot_super);
2053
2054         remove_agents = oh->ot_remove_agents;
2055
2056         qtrans = oh->ot_quota_trans;
2057         oh->ot_quota_trans = NULL;
2058
2059         /* move locks to local list, stop tx, execute truncates */
2060         list_splice(&oh->ot_trunc_locks, &truncates);
2061
2062         if (oh->ot_handle != NULL) {
2063                 int rc2;
2064
2065                 handle_t *hdl = oh->ot_handle;
2066
2067                 /*
2068                  * add commit callback
2069                  * notice we don't do this in osd_trans_start()
2070                  * as underlying transaction can change during truncate
2071                  */
2072                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2073                                              &oh->ot_jcb);
2074
2075                 LASSERT(oti->oti_txns == 1);
2076                 oti->oti_txns--;
2077
2078                 rc = dt_txn_hook_stop(env, th);
2079                 if (rc != 0)
2080                         CERROR("%s: failed in transaction hook: rc = %d\n",
2081                                osd_name(osd), rc);
2082
2083                 osd_trans_stop_cb(oh, rc);
2084                 /* hook functions might modify th_sync */
2085                 hdl->h_sync = th->th_sync;
2086
2087                 oh->ot_handle = NULL;
2088                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2089                 if (rc2 != 0)
2090                         CERROR("%s: failed to stop transaction: rc = %d\n",
2091                                osd_name(osd), rc2);
2092                 if (!rc)
2093                         rc = rc2;
2094
2095                 osd_process_truncates(&truncates);
2096         } else {
2097                 osd_trans_stop_cb(oh, th->th_result);
2098                 OBD_FREE_PTR(oh);
2099         }
2100
2101         osd_trunc_unlock_all(env, &truncates);
2102
2103         /* inform the quota slave device that the transaction is stopping */
2104         qsd_op_end(env, qsd, qtrans);
2105
2106         /*
2107          * as we want IO to journal and data IO be concurrent, we don't block
2108          * awaiting data IO completion in osd_do_bio(), instead we wait here
2109          * once transaction is submitted to the journal. all reqular requests
2110          * don't do direct IO (except read/write), thus this wait_event becomes
2111          * no-op for them.
2112          *
2113          * IMPORTANT: we have to wait till any IO submited by the thread is
2114          * completed otherwise iobuf may be corrupted by different request
2115          */
2116         wait_event(iobuf->dr_wait,
2117                        atomic_read(&iobuf->dr_numreqs) == 0);
2118         osd_fini_iobuf(osd, iobuf);
2119         if (!rc)
2120                 rc = iobuf->dr_error;
2121
2122         if (unlikely(remove_agents != 0))
2123                 osd_process_scheduled_agent_removals(env, osd);
2124
2125         LASSERT(oti->oti_ins_cache_depth > 0);
2126         oti->oti_ins_cache_depth--;
2127         /* reset OI cache for safety */
2128         if (oti->oti_ins_cache_depth == 0)
2129                 oti->oti_ins_cache_used = 0;
2130
2131         sb_end_write(osd_sb(osd));
2132
2133         RETURN(rc);
2134 }
2135
2136 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2137 {
2138         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2139                                               ot_super);
2140
2141         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2142         LASSERT(&dcb->dcb_func != NULL);
2143         if (dcb->dcb_flags & DCB_TRANS_STOP)
2144                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2145         else
2146                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2147
2148         return 0;
2149 }
2150
2151 /*
2152  * Called just before object is freed. Releases all resources except for
2153  * object itself (that is released by osd_object_free()).
2154  *
2155  * Concurrency: no concurrent access is possible that late in object
2156  * life-cycle.
2157  */
2158 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2159 {
2160         struct osd_object *obj = osd_obj(l);
2161         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2162         struct inode *inode = obj->oo_inode;
2163         __u64 projid;
2164         qid_t uid;
2165         qid_t gid;
2166
2167         LINVRNT(osd_invariant(obj));
2168
2169         /*
2170          * If object is unlinked remove fid->ino mapping from object index.
2171          */
2172
2173         osd_index_fini(obj);
2174
2175         if (!inode)
2176                 return;
2177
2178         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2179                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2180
2181         uid = i_uid_read(inode);
2182         gid = i_gid_read(inode);
2183         projid = i_projid_read(inode);
2184
2185         obj->oo_inode = NULL;
2186         iput(inode);
2187
2188         /* do not rebalance quota if the caller needs to release memory
2189          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2190          * transaction and risk to deadlock - LU-12178 */
2191         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2192                 return;
2193
2194         if (!obj->oo_header && qsd) {
2195                 struct osd_thread_info *info = osd_oti_get(env);
2196                 struct lquota_id_info *qi = &info->oti_qi;
2197
2198                 /* Release granted quota to master if necessary */
2199                 qi->lqi_id.qid_uid = uid;
2200                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2201
2202                 qi->lqi_id.qid_uid = gid;
2203                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2204
2205                 qi->lqi_id.qid_uid = projid;
2206                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2207         }
2208 }
2209
2210 /*
2211  * Concurrency: ->loo_object_release() is called under site spin-lock.
2212  */
2213 static void osd_object_release(const struct lu_env *env,
2214                                struct lu_object *l)
2215 {
2216         struct osd_object *o = osd_obj(l);
2217
2218         /*
2219          * nobody should be releasing a non-destroyed object with nlink=0
2220          * the API allows this, but ldiskfs doesn't like and then report
2221          * this inode as deleted
2222          */
2223         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2224                   o->oo_inode->i_nlink == 0));
2225 }
2226
2227 /*
2228  * Concurrency: shouldn't matter.
2229  */
2230 static int osd_object_print(const struct lu_env *env, void *cookie,
2231                             lu_printer_t p, const struct lu_object *l)
2232 {
2233         struct osd_object *o = osd_obj(l);
2234         struct iam_descr *d;
2235
2236         if (o->oo_dir != NULL)
2237                 d = o->oo_dir->od_container.ic_descr;
2238         else
2239                 d = NULL;
2240         return (*p)(env, cookie,
2241                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2242                     o, o->oo_inode,
2243                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2244                     o->oo_inode ? o->oo_inode->i_generation : 0,
2245                     d ? d->id_ops->id_name : "plain");
2246 }
2247
2248 /*
2249  * Concurrency: shouldn't matter.
2250  */
2251 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2252                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2253 {
2254         struct osd_device *osd = osd_dt_dev(d);
2255         struct super_block *sb = osd_sb(osd);
2256         struct kstatfs *ksfs;
2257         __u64 reserved;
2258         int result = 0;
2259
2260         if (unlikely(osd->od_mnt == NULL))
2261                 return -EINPROGRESS;
2262
2263         /* osd_lproc.c call this without env, allocate ksfs for that case */
2264         if (unlikely(env == NULL)) {
2265                 OBD_ALLOC_PTR(ksfs);
2266                 if (ksfs == NULL)
2267                         return -ENOMEM;
2268         } else {
2269                 ksfs = &osd_oti_get(env)->oti_ksfs;
2270         }
2271
2272         result = sb->s_op->statfs(sb->s_root, ksfs);
2273         if (result)
2274                 goto out;
2275
2276         statfs_pack(sfs, ksfs);
2277         if (unlikely(sb->s_flags & SB_RDONLY))
2278                 sfs->os_state |= OS_STATFS_READONLY;
2279
2280         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2281
2282         if (ldiskfs_has_feature_extents(sb))
2283                 sfs->os_maxbytes = sb->s_maxbytes;
2284         else
2285                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2286
2287         /*
2288          * Reserve some space so to avoid fragmenting the filesystem too much.
2289          * Fragmentation not only impacts performance, but can also increase
2290          * metadata overhead significantly, causing grant calculation to be
2291          * wrong.
2292          *
2293          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2294          */
2295         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2296         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2297         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2298                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2299
2300         sfs->os_blocks -= reserved;
2301         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2302         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2303
2304 out:
2305         if (unlikely(env == NULL))
2306                 OBD_FREE_PTR(ksfs);
2307         return result;
2308 }
2309
2310 /**
2311  * Estimate space needed for file creations. We assume the largest filename
2312  * which is 2^64 - 1, hence a filename of 20 chars.
2313  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2314  */
2315 #ifdef __LDISKFS_DIR_REC_LEN
2316 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2317 #else
2318 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2319 #endif
2320
2321 /*
2322  * Concurrency: doesn't access mutable data.
2323  */
2324 static void osd_conf_get(const struct lu_env *env,
2325                          const struct dt_device *dev,
2326                          struct dt_device_param *param)
2327 {
2328         struct osd_device *d = osd_dt_dev(dev);
2329         struct super_block *sb = osd_sb(d);
2330         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2331         const char *name;
2332         int ea_overhead;
2333
2334         /*
2335          * XXX should be taken from not-yet-existing fs abstraction layer.
2336          */
2337         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2338         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2339         param->ddp_symlink_max  = sb->s_blocksize;
2340         param->ddp_mount_type   = LDD_MT_LDISKFS;
2341         if (ldiskfs_has_feature_extents(sb))
2342                 param->ddp_maxbytes = sb->s_maxbytes;
2343         else
2344                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2345         /*
2346          * inode are statically allocated, so per-inode space consumption
2347          * is the space consumed by the directory entry
2348          */
2349         param->ddp_inodespace     = PER_OBJ_USAGE;
2350         /*
2351          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2352          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2353          * maximum length to not under-count the actual number of extents
2354          * needed for writing a file if there are sub-optimal block allocations.
2355          */
2356         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2357         /* worst-case extent insertion metadata overhead */
2358         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2359         param->ddp_mntopts = 0;
2360         if (test_opt(sb, XATTR_USER))
2361                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2362         if (test_opt(sb, POSIX_ACL))
2363                 param->ddp_mntopts |= MNTOPT_ACL;
2364
2365         /*
2366          * LOD might calculate the max stripe count based on max_ea_size,
2367          * so we need take account in the overhead as well,
2368          * xattr_header + magic + xattr_entry_head
2369          */
2370         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2371                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2372
2373 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2374         if (ldiskfs_has_feature_ea_inode(sb))
2375                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2376                                                                 ea_overhead;
2377         else
2378 #endif
2379                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2380
2381         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2382                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2383
2384         /*
2385          * Preferred RPC size for efficient disk IO.  4MB shows good
2386          * all-around performance for ldiskfs, but use bigalloc chunk size
2387          * by default if larger.
2388          */
2389 #if defined(LDISKFS_CLUSTER_SIZE)
2390         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2391                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2392         else
2393 #endif
2394                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2395
2396         param->ddp_t10_cksum_type = 0;
2397         if (bi) {
2398                 unsigned short interval = blk_integrity_interval(bi);
2399                 name = blk_integrity_name(bi);
2400                 /*
2401                  * Expected values:
2402                  * T10-DIF-TYPE1-CRC
2403                  * T10-DIF-TYPE3-CRC
2404                  * T10-DIF-TYPE1-IP
2405                  * T10-DIF-TYPE3-IP
2406                  */
2407                 if (strncmp(name, "T10-DIF-TYPE",
2408                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2409                         /* also skip "1/3-" at end */
2410                         const int type_off = sizeof("T10-DIF-TYPE.");
2411                         char type_number = name[type_off - 2];
2412
2413                         if (interval != 512 && interval != 4096) {
2414                                 CERROR("%s: unsupported T10PI sector size %u\n",
2415                                        d->od_svname, interval);
2416                         } else if (type_number != '1' && type_number != '3') {
2417                                 CERROR("%s: unsupported T10PI type %s\n",
2418                                        d->od_svname, name);
2419                         } else if (strcmp(name + type_off, "CRC") == 0) {
2420                                 d->od_t10_type = type_number == '1' ?
2421                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2422                                 param->ddp_t10_cksum_type = interval == 512 ?
2423                                         OBD_CKSUM_T10CRC512 :
2424                                         OBD_CKSUM_T10CRC4K;
2425                         } else if (strcmp(name + type_off, "IP") == 0) {
2426                                 d->od_t10_type = type_number == '1' ?
2427                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2428                                 param->ddp_t10_cksum_type = interval == 512 ?
2429                                         OBD_CKSUM_T10IP512 :
2430                                         OBD_CKSUM_T10IP4K;
2431                         } else {
2432                                 CERROR("%s: unsupported checksum type of "
2433                                        "T10PI type '%s'",
2434                                        d->od_svname, name);
2435                         }
2436
2437                 } else {
2438                         CERROR("%s: unsupported T10PI type '%s'",
2439                                d->od_svname, name);
2440                 }
2441         }
2442
2443         param->ddp_has_lseek_data_hole = true;
2444 }
2445
2446 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2447 {
2448         return osd_sb(osd_dt_dev(d));
2449 }
2450
2451 /*
2452  * Concurrency: shouldn't matter.
2453  */
2454 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2455 {
2456         int rc;
2457         struct super_block *s = osd_sb(osd_dt_dev(d));
2458         ENTRY;
2459
2460         down_read(&s->s_umount);
2461         rc = s->s_op->sync_fs(s, 1);
2462         up_read(&s->s_umount);
2463
2464         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2465                rc);
2466
2467         return rc;
2468 }
2469
2470 /**
2471  * Start commit for OSD device.
2472  *
2473  * An implementation of dt_commit_async method for OSD device.
2474  * Asychronously starts underlayng fs sync and thereby a transaction
2475  * commit.
2476  *
2477  * \param env environment
2478  * \param d dt device
2479  *
2480  * \see dt_device_operations
2481  */
2482 static int osd_commit_async(const struct lu_env *env,
2483                             struct dt_device *d)
2484 {
2485         struct super_block *s = osd_sb(osd_dt_dev(d));
2486         int rc;
2487
2488         ENTRY;
2489
2490         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2491         down_read(&s->s_umount);
2492         rc = s->s_op->sync_fs(s, 0);
2493         up_read(&s->s_umount);
2494
2495         RETURN(rc);
2496 }
2497
2498 static int (*priv_security_file_alloc)(struct file *file);
2499
2500 int osd_security_file_alloc(struct file *file)
2501 {
2502         if (priv_security_file_alloc)
2503                 return priv_security_file_alloc(file);
2504         return 0;
2505 }
2506
2507 /*
2508  * Concurrency: shouldn't matter.
2509  */
2510 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2511 {
2512         struct super_block *sb = osd_sb(osd_dt_dev(d));
2513         struct block_device *dev = sb->s_bdev;
2514         int rc = -EOPNOTSUPP;
2515
2516         ENTRY;
2517
2518         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2519                osd_dt_dev(d)->od_svname, (long)dev, rc);
2520
2521         RETURN(rc);
2522 }
2523
2524 /**
2525  * Note: we do not count into QUOTA here.
2526  * If we mount with --data_journal we may need more.
2527  */
2528 const int osd_dto_credits_noquota[DTO_NR] = {
2529         /**
2530          * Insert.
2531          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2532          * SINGLEDATA_TRANS_BLOCKS(8)
2533          * XXX Note: maybe iam need more, since iam have more level than
2534          *           EXT3 htree.
2535          */
2536         [DTO_INDEX_INSERT]  = 16,
2537         /**
2538          * Delete
2539          * just modify a single entry, probably merge few within a block
2540          */
2541         [DTO_INDEX_DELETE]  = 1,
2542         /**
2543          * Used for OI scrub
2544          */
2545         [DTO_INDEX_UPDATE]  = 16,
2546         /**
2547          * 4(inode, inode bits, groups, GDT)
2548          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2549          */
2550         [DTO_OBJECT_CREATE] = 4,
2551         /**
2552          * 4(inode, inode bits, groups, GDT)
2553          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2554          */
2555         [DTO_OBJECT_DELETE] = 4,
2556         /**
2557          * Attr set credits (inode)
2558          */
2559         [DTO_ATTR_SET_BASE] = 1,
2560         /**
2561          * Xattr set. The same as xattr of EXT3.
2562          * DATA_TRANS_BLOCKS(14)
2563          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2564          * are also counted in. Do not know why?
2565          */
2566         [DTO_XATTR_SET]     = 14,
2567         /**
2568          * credits for inode change during write.
2569          */
2570         [DTO_WRITE_BASE]    = 3,
2571         /**
2572          * credits for single block write.
2573          */
2574         [DTO_WRITE_BLOCK]   = 14,
2575         /**
2576          * Attr set credits for chown.
2577          * This is extra credits for setattr, and it is null without quota
2578          */
2579         [DTO_ATTR_SET_CHOWN] = 0
2580 };
2581
2582 static const struct dt_device_operations osd_dt_ops = {
2583         .dt_root_get       = osd_root_get,
2584         .dt_statfs         = osd_statfs,
2585         .dt_trans_create   = osd_trans_create,
2586         .dt_trans_start    = osd_trans_start,
2587         .dt_trans_stop     = osd_trans_stop,
2588         .dt_trans_cb_add   = osd_trans_cb_add,
2589         .dt_conf_get       = osd_conf_get,
2590         .dt_mnt_sb_get     = osd_mnt_sb_get,
2591         .dt_sync           = osd_sync,
2592         .dt_ro             = osd_ro,
2593         .dt_commit_async   = osd_commit_async,
2594 };
2595
2596 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2597                           unsigned int role)
2598 {
2599         struct osd_object *obj = osd_dt_obj(dt);
2600         struct osd_thread_info *oti = osd_oti_get(env);
2601
2602         LINVRNT(osd_invariant(obj));
2603
2604         LASSERT(obj->oo_owner != env);
2605         down_read_nested(&obj->oo_sem, role);
2606
2607         LASSERT(obj->oo_owner == NULL);
2608         oti->oti_r_locks++;
2609 }
2610
2611 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2612                            unsigned int role)
2613 {
2614         struct osd_object *obj = osd_dt_obj(dt);
2615         struct osd_thread_info *oti = osd_oti_get(env);
2616
2617         LINVRNT(osd_invariant(obj));
2618
2619         LASSERT(obj->oo_owner != env);
2620         down_write_nested(&obj->oo_sem, role);
2621
2622         LASSERT(obj->oo_owner == NULL);
2623         obj->oo_owner = env;
2624         oti->oti_w_locks++;
2625 }
2626
2627 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2628 {
2629         struct osd_object *obj = osd_dt_obj(dt);
2630         struct osd_thread_info *oti = osd_oti_get(env);
2631
2632         LINVRNT(osd_invariant(obj));
2633
2634         LASSERT(oti->oti_r_locks > 0);
2635         oti->oti_r_locks--;
2636         up_read(&obj->oo_sem);
2637 }
2638
2639 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2640 {
2641         struct osd_object *obj = osd_dt_obj(dt);
2642         struct osd_thread_info *oti = osd_oti_get(env);
2643
2644         LINVRNT(osd_invariant(obj));
2645
2646         LASSERT(obj->oo_owner == env);
2647         LASSERT(oti->oti_w_locks > 0);
2648         oti->oti_w_locks--;
2649         obj->oo_owner = NULL;
2650         up_write(&obj->oo_sem);
2651 }
2652
2653 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2654 {
2655         struct osd_object *obj = osd_dt_obj(dt);
2656
2657         LINVRNT(osd_invariant(obj));
2658
2659         return obj->oo_owner == env;
2660 }
2661
2662 static void osd_inode_getattr(const struct lu_env *env,
2663                               struct inode *inode, struct lu_attr *attr)
2664 {
2665         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2666                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2667                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2668                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2669
2670         attr->la_atime = inode->i_atime.tv_sec;
2671         attr->la_mtime = inode->i_mtime.tv_sec;
2672         attr->la_ctime = inode->i_ctime.tv_sec;
2673         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2674         attr->la_mode    = inode->i_mode;
2675         attr->la_size    = i_size_read(inode);
2676         attr->la_blocks  = inode->i_blocks;
2677         attr->la_uid     = i_uid_read(inode);
2678         attr->la_gid     = i_gid_read(inode);
2679         attr->la_projid  = i_projid_read(inode);
2680         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2681         attr->la_nlink   = inode->i_nlink;
2682         attr->la_rdev    = inode->i_rdev;
2683         attr->la_blksize = 1 << inode->i_blkbits;
2684         attr->la_blkbits = inode->i_blkbits;
2685         /*
2686          * Ext4 did not transfer inherit flags from raw inode
2687          * to inode flags, and ext4 internally test raw inode
2688          * @i_flags directly. Instead of patching ext4, we do it here.
2689          */
2690         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2691                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2692 }
2693
2694 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2695                             u64 *count)
2696 {
2697         struct osd_object *obj = osd_dt_obj(dt);
2698         const struct dt_it_ops *iops;
2699         struct dt_it *it;
2700         int rc;
2701
2702         ENTRY;
2703
2704         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2705         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2706
2707         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2708                 *count = obj->oo_dirent_count;
2709                 RETURN(0);
2710         }
2711
2712         /* directory not initialized yet */
2713         if (!dt->do_index_ops) {
2714                 *count = 0;
2715                 RETURN(0);
2716         }
2717
2718         iops = &dt->do_index_ops->dio_it;
2719         it = iops->init(env, dt, LUDA_64BITHASH);
2720         if (IS_ERR(it))
2721                 RETURN(PTR_ERR(it));
2722
2723         rc = iops->load(env, it, 0);
2724         if (rc < 0) {
2725                 if (rc == -ENODATA) {
2726                         rc = 0;
2727                         *count = 0;
2728                 }
2729                 GOTO(out, rc);
2730         }
2731         if (rc > 0)
2732                 rc = iops->next(env, it);
2733
2734         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2735                 if (rc == -ESTALE)
2736                         continue;
2737
2738                 if (iops->key_size(env, it) == 0)
2739                         continue;
2740
2741                 (*count)++;
2742         }
2743         if (rc == 1) {
2744                 obj->oo_dirent_count = *count;
2745                 rc = 0;
2746         }
2747 out:
2748         iops->put(env, it);
2749         iops->fini(env, it);
2750
2751         RETURN(rc);
2752 }
2753
2754 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2755                         struct lu_attr *attr)
2756 {
2757         struct osd_object *obj = osd_dt_obj(dt);
2758         int rc = 0;
2759
2760         if (unlikely(!dt_object_exists(dt)))
2761                 return -ENOENT;
2762         if (unlikely(obj->oo_destroyed))
2763                 return -ENOENT;
2764
2765         LASSERT(!dt_object_remote(dt));
2766         LINVRNT(osd_invariant(obj));
2767
2768         spin_lock(&obj->oo_guard);
2769         osd_inode_getattr(env, obj->oo_inode, attr);
2770         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2771                 attr->la_valid |= LA_FLAGS;
2772                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2773         }
2774         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2775                 attr->la_valid |= LA_FLAGS;
2776                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2777         }
2778         spin_unlock(&obj->oo_guard);
2779
2780         if (S_ISDIR(obj->oo_inode->i_mode) &&
2781             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2782                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2783
2784         return rc;
2785 }
2786
2787 static int osd_declare_attr_qid(const struct lu_env *env,
2788                                 struct osd_object *obj,
2789                                 struct osd_thandle *oh, long long bspace,
2790                                 qid_t old_id, qid_t new_id, bool enforce,
2791                                 unsigned int type, bool ignore_edquot)
2792 {
2793         int rc;
2794         struct osd_thread_info *info = osd_oti_get(env);
2795         struct lquota_id_info  *qi = &info->oti_qi;
2796
2797         qi->lqi_type = type;
2798         /* inode accounting */
2799         qi->lqi_is_blk = false;
2800
2801         /* one more inode for the new id ... */
2802         qi->lqi_id.qid_uid = new_id;
2803         qi->lqi_space      = 1;
2804         /* Reserve credits for the new id */
2805         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2806         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2807                 rc = 0;
2808         if (rc)
2809                 RETURN(rc);
2810
2811         /* and one less inode for the current id */
2812         qi->lqi_id.qid_uid = old_id;
2813         qi->lqi_space = -1;
2814         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2815         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2816                 rc = 0;
2817         if (rc)
2818                 RETURN(rc);
2819
2820         /* block accounting */
2821         qi->lqi_is_blk = true;
2822
2823         /* more blocks for the new id ... */
2824         qi->lqi_id.qid_uid = new_id;
2825         qi->lqi_space      = bspace;
2826         /*
2827          * Credits for the new uid has been reserved, re-use "obj"
2828          * to save credit reservation.
2829          */
2830         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2831         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2832                 rc = 0;
2833         if (rc)
2834                 RETURN(rc);
2835
2836         /* and finally less blocks for the current uid */
2837         qi->lqi_id.qid_uid = old_id;
2838         qi->lqi_space      = -bspace;
2839         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2840         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2841                 rc = 0;
2842
2843         RETURN(rc);
2844 }
2845
2846 static int osd_declare_attr_set(const struct lu_env *env,
2847                                 struct dt_object *dt,
2848                                 const struct lu_attr *attr,
2849                                 struct thandle *handle)
2850 {
2851         struct osd_thandle *oh;
2852         struct osd_object *obj;
2853         qid_t uid;
2854         qid_t gid;
2855         long long bspace;
2856         int rc = 0;
2857         bool enforce;
2858
2859         ENTRY;
2860
2861         LASSERT(dt != NULL);
2862         LASSERT(handle != NULL);
2863
2864         obj = osd_dt_obj(dt);
2865         LASSERT(osd_invariant(obj));
2866
2867         oh = container_of(handle, struct osd_thandle, ot_super);
2868         LASSERT(oh->ot_handle == NULL);
2869
2870         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2871                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2872
2873         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2874                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2875
2876         if (attr == NULL || obj->oo_inode == NULL)
2877                 RETURN(rc);
2878
2879         bspace   = obj->oo_inode->i_blocks << 9;
2880         bspace   = toqb(bspace);
2881
2882         /*
2883          * Changing ownership is always preformed by super user, it should not
2884          * fail with EDQUOT unless required explicitly.
2885          *
2886          * We still need to call the osd_declare_qid() to calculate the journal
2887          * credits for updating quota accounting files and to trigger quota
2888          * space adjustment once the operation is completed.
2889          */
2890         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2891                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2892
2893                 if (!ignore_edquot)
2894                         CDEBUG(D_QUOTA,
2895                                "%s: enforce quota on UID %u, GID %u (quota space is %lld)\n",
2896                                osd_ino2name(obj->oo_inode), attr->la_uid,
2897                                attr->la_gid, bspace);
2898
2899                 /* USERQUOTA */
2900                 uid = i_uid_read(obj->oo_inode);
2901                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2902                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2903                                           attr->la_uid, enforce, USRQUOTA,
2904                                           true);
2905                 if (rc)
2906                         RETURN(rc);
2907
2908                 gid = i_gid_read(obj->oo_inode);
2909                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2910                        attr->la_uid, gid, attr->la_gid);
2911                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2912                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2913                                           attr->la_gid, enforce, GRPQUOTA,
2914                                           ignore_edquot);
2915                 if (rc)
2916                         RETURN(rc);
2917
2918         }
2919 #ifdef HAVE_PROJECT_QUOTA
2920         if (attr->la_valid & LA_PROJID) {
2921                 __u32 projid = i_projid_read(obj->oo_inode);
2922
2923                 enforce = (attr->la_valid & LA_PROJID) &&
2924                                         (attr->la_projid != projid);
2925                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2926                                           (qid_t)projid, (qid_t)attr->la_projid,
2927                                           enforce, PRJQUOTA, true);
2928                 if (rc)
2929                         RETURN(rc);
2930         }
2931 #endif
2932         RETURN(rc);
2933 }
2934
2935 static int osd_inode_setattr(const struct lu_env *env,
2936                              struct inode *inode, const struct lu_attr *attr)
2937 {
2938         __u64 bits = attr->la_valid;
2939
2940         /* Only allow set size for regular file */
2941         if (!S_ISREG(inode->i_mode))
2942                 bits &= ~(LA_SIZE | LA_BLOCKS);
2943
2944         if (bits == 0)
2945                 return 0;
2946
2947         if (bits & LA_ATIME)
2948                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2949         if (bits & LA_CTIME)
2950                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2951         if (bits & LA_MTIME)
2952                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2953         if (bits & LA_SIZE) {
2954                 spin_lock(&inode->i_lock);
2955                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2956                 i_size_write(inode, attr->la_size);
2957                 spin_unlock(&inode->i_lock);
2958         }
2959
2960         /*
2961          * OSD should not change "i_blocks" which is used by quota.
2962          * "i_blocks" should be changed by ldiskfs only.
2963          */
2964         if (bits & LA_MODE)
2965                 inode->i_mode = (inode->i_mode & S_IFMT) |
2966                                 (attr->la_mode & ~S_IFMT);
2967         if (bits & LA_UID)
2968                 i_uid_write(inode, attr->la_uid);
2969         if (bits & LA_GID)
2970                 i_gid_write(inode, attr->la_gid);
2971         if (bits & LA_PROJID)
2972                 i_projid_write(inode, attr->la_projid);
2973         if (bits & LA_NLINK)
2974                 set_nlink(inode, attr->la_nlink);
2975         if (bits & LA_RDEV)
2976                 inode->i_rdev = attr->la_rdev;
2977
2978         if (bits & LA_FLAGS) {
2979                 /* always keep S_NOCMTIME */
2980                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2981                                  S_NOCMTIME;
2982 #if defined(S_ENCRYPTED)
2983                 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2984                  * aware of encryption status. It is just stored into LMA
2985                  * so that it can be forwared to client side.
2986                  */
2987                 inode->i_flags &= ~S_ENCRYPTED;
2988 #endif
2989                 /*
2990                  * Ext4 did not transfer inherit flags from
2991                  * @inode->i_flags to raw inode i_flags when writing
2992                  * flags, we do it explictly here.
2993                  */
2994                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2995                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2996                 else
2997                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2998         }
2999         return 0;
3000 }
3001
3002 #ifdef HAVE_PROJECT_QUOTA
3003 static int osd_transfer_project(struct inode *inode, __u32 projid,
3004                                 struct thandle *handle)
3005 {
3006         struct super_block *sb = inode->i_sb;
3007         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
3008         int err;
3009         kprojid_t kprojid;
3010         struct ldiskfs_iloc iloc;
3011         struct ldiskfs_inode *raw_inode;
3012         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
3013
3014         if (!ldiskfs_has_feature_project(sb)) {
3015                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
3016                         == LDISKFS_DEF_PROJID);
3017                 if (projid != LDISKFS_DEF_PROJID)
3018                         return -EOPNOTSUPP;
3019                 else
3020                         return 0;
3021         }
3022
3023         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
3024                 return -EOPNOTSUPP;
3025
3026         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
3027         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
3028                 return 0;
3029
3030         err = ldiskfs_get_inode_loc(inode, &iloc);
3031         if (err)
3032                 return err;
3033
3034         raw_inode = ldiskfs_raw_inode(&iloc);
3035         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3036                 struct osd_thandle *oh = container_of(handle,
3037                                                       struct osd_thandle,
3038                                                       ot_super);
3039                 /**
3040                  * try to expand inode size automatically.
3041                  */
3042                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3043                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3044                         err = -EOVERFLOW;
3045                         brelse(iloc.bh);
3046                         return err;
3047                 }
3048         }
3049         brelse(iloc.bh);
3050
3051         dquot_initialize(inode);
3052         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3053         if (transfer_to[PRJQUOTA]) {
3054                 err = __dquot_transfer(inode, transfer_to);
3055                 dqput(transfer_to[PRJQUOTA]);
3056                 if (err)
3057                         return err;
3058         }
3059
3060         return err;
3061 }
3062 #endif
3063
3064 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3065                               struct thandle *handle)
3066 {
3067         int rc;
3068
3069         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3070             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3071                 struct iattr iattr;
3072
3073                 CDEBUG(D_QUOTA,
3074                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3075                        inode->i_ino, i_uid_read(inode), attr->la_uid,
3076                        i_gid_read(inode), attr->la_gid);
3077
3078                 dquot_initialize(inode);
3079                 iattr.ia_valid = 0;
3080                 if (attr->la_valid & LA_UID)
3081                         iattr.ia_valid |= ATTR_UID;
3082                 if (attr->la_valid & LA_GID)
3083                         iattr.ia_valid |= ATTR_GID;
3084                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3085                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3086
3087                 rc = dquot_transfer(inode, &iattr);
3088                 if (rc) {
3089                         CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3090                                osd_ino2name(inode), rc);
3091                         return rc;
3092                 }
3093         }
3094
3095         /* Handle project id transfer here properly */
3096         if (attr->la_valid & LA_PROJID &&
3097             attr->la_projid != i_projid_read(inode)) {
3098 #ifdef HAVE_PROJECT_QUOTA
3099                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3100 #else
3101                 rc = -ENOTSUPP;
3102 #endif
3103                 if (rc) {
3104                         CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3105                                osd_ino2name(inode), rc);
3106                         return rc;
3107                 }
3108         }
3109         return 0;
3110 }
3111
3112 static int osd_attr_set(const struct lu_env *env,
3113                         struct dt_object *dt,
3114                         const struct lu_attr *attr,
3115                         struct thandle *handle)
3116 {
3117         struct osd_object *obj = osd_dt_obj(dt);
3118         struct inode *inode;
3119         int rc;
3120
3121         if (!dt_object_exists(dt))
3122                 return -ENOENT;
3123
3124         LASSERT(handle != NULL);
3125         LASSERT(!dt_object_remote(dt));
3126         LASSERT(osd_invariant(obj));
3127
3128         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3129
3130         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3131             !osd_obj2dev(obj)->od_is_ost) {
3132                 struct osd_thread_info *oti = osd_oti_get(env);
3133                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3134                 struct lu_fid *fid1 = &oti->oti_fid;
3135                 struct osd_inode_id *id = &oti->oti_id;
3136                 struct iam_path_descr *ipd;
3137                 struct iam_container *bag;
3138                 struct osd_thandle *oh;
3139                 int rc;
3140
3141                 fid_cpu_to_be(fid1, fid0);
3142                 memset(id, 1, sizeof(*id));
3143                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3144                                   fid0)->oi_dir.od_container;
3145                 ipd = osd_idx_ipd_get(env, bag);
3146                 if (unlikely(ipd == NULL))
3147                         RETURN(-ENOMEM);
3148
3149                 oh = container_of(handle, struct osd_thandle, ot_super);
3150                 rc = iam_update(oh->ot_handle, bag,
3151                                 (const struct iam_key *)fid1,
3152                                 (const struct iam_rec *)id, ipd);
3153                 osd_ipd_put(env, bag, ipd);
3154                 return(rc > 0 ? 0 : rc);
3155         }
3156
3157         inode = obj->oo_inode;
3158
3159         rc = osd_quota_transfer(inode, attr, handle);
3160         if (rc)
3161                 return rc;
3162
3163         spin_lock(&obj->oo_guard);
3164         rc = osd_inode_setattr(env, inode, attr);
3165         spin_unlock(&obj->oo_guard);
3166         if (rc != 0)
3167                 GOTO(out, rc);
3168
3169         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3170
3171         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3172
3173         if (!(attr->la_valid & LA_FLAGS))
3174                 GOTO(out, rc);
3175
3176         /* Let's check if there are extra flags need to be set into LMA */
3177         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3178                 struct osd_thread_info *info = osd_oti_get(env);
3179                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3180
3181                 LASSERT(!obj->oo_pfid_in_lma);
3182
3183                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3184                                  &info->oti_ost_attrs);
3185                 if (rc)
3186                         GOTO(out, rc);
3187
3188                 lma->lma_incompat |=
3189                         lustre_to_lma_flags(attr->la_flags);
3190                 lustre_lma_swab(lma);
3191
3192                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3193
3194                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3195                                      lma, sizeof(*lma), XATTR_REPLACE);
3196                 if (rc != 0) {
3197                         struct osd_device *osd = osd_obj2dev(obj);
3198
3199                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3200                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3201                               lma->lma_incompat, rc);
3202                 } else {
3203                         obj->oo_lma_flags =
3204                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3205                 }
3206                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3207         }
3208 out:
3209
3210         return rc;
3211 }
3212
3213 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3214                                            struct osd_object *obj,
3215                                            const char *name, const int namelen)
3216 {
3217         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3218 }
3219
3220 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3221                       umode_t mode, struct dt_allocation_hint *hint,
3222                       struct thandle *th, struct lu_attr *attr)
3223 {
3224         int result;
3225         struct osd_device *osd = osd_obj2dev(obj);
3226         struct osd_thandle *oth;
3227         struct dt_object *parent = NULL;
3228         struct inode *inode;
3229         struct iattr iattr = {
3230                 .ia_valid = ATTR_UID | ATTR_GID |
3231                             ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3232                 .ia_ctime.tv_sec = attr->la_ctime,
3233                 .ia_mtime.tv_sec = attr->la_mtime,
3234                 .ia_atime.tv_sec = attr->la_atime,
3235                 .ia_uid = GLOBAL_ROOT_UID,
3236                 .ia_gid = GLOBAL_ROOT_GID,
3237         };
3238         const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
3239
3240         if (attr->la_valid & LA_UID)
3241                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3242         if (attr->la_valid & LA_GID)
3243                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3244
3245         LINVRNT(osd_invariant(obj));
3246         LASSERT(obj->oo_inode == NULL);
3247         LASSERT(obj->oo_hl_head == NULL);
3248
3249         if (S_ISDIR(mode) && ldiskfs_pdo) {
3250                 obj->oo_hl_head =
3251                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3252                 if (obj->oo_hl_head == NULL)
3253                         return -ENOMEM;
3254         }
3255
3256         oth = container_of(th, struct osd_thandle, ot_super);
3257         LASSERT(oth->ot_handle->h_transaction != NULL);
3258
3259         if (hint != NULL && hint->dah_parent != NULL &&
3260             !dt_object_remote(hint->dah_parent))
3261                 parent = hint->dah_parent;
3262
3263         /* if a time component is not valid set it to UTIME_OMIT */
3264         if (!(attr->la_valid & LA_CTIME))
3265                 iattr.ia_ctime = omit;
3266         if (!(attr->la_valid & LA_MTIME))
3267                 iattr.ia_mtime = omit;
3268         if (!(attr->la_valid & LA_ATIME))
3269                 iattr.ia_atime = omit;
3270
3271         inode = ldiskfs_create_inode(oth->ot_handle,
3272                                      parent ? osd_dt_obj(parent)->oo_inode :
3273                                               osd_sb(osd)->s_root->d_inode,
3274                                      mode, &iattr);
3275         if (!IS_ERR(inode)) {
3276                 /* Do not update file c/mtime in ldiskfs. */
3277                 inode->i_flags |= S_NOCMTIME;
3278
3279                 /*
3280                  * For new created object, it must be consistent,
3281                  * and it is unnecessary to scrub against it.
3282                  */
3283                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3284
3285                 obj->oo_inode = inode;
3286                 result = 0;
3287         } else {
3288                 if (obj->oo_hl_head != NULL) {
3289                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3290                         obj->oo_hl_head = NULL;
3291                 }
3292                 result = PTR_ERR(inode);
3293         }
3294         LINVRNT(osd_invariant(obj));
3295         return result;
3296 }
3297
3298 enum {
3299         OSD_NAME_LEN = 255
3300 };
3301
3302 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3303                      struct lu_attr *attr,
3304                      struct dt_allocation_hint *hint,
3305                      struct dt_object_format *dof,
3306                      struct thandle *th)
3307 {
3308         int result;
3309         struct osd_thandle *oth;
3310         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3311
3312         LASSERT(S_ISDIR(attr->la_mode));
3313
3314         oth = container_of(th, struct osd_thandle, ot_super);
3315         LASSERT(oth->ot_handle->h_transaction != NULL);
3316         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3317                 obj->oo_dirent_count = 0;
3318         result = osd_mkfile(info, obj, mode, hint, th, attr);
3319
3320         return result;
3321 }
3322
3323 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3324                         struct lu_attr *attr,
3325                         struct dt_allocation_hint *hint,
3326                         struct dt_object_format *dof,
3327                         struct thandle *th)
3328 {
3329         int result;
3330         struct osd_thandle *oth;
3331         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3332
3333         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3334
3335         LASSERT(S_ISREG(attr->la_mode));
3336
3337         oth = container_of(th, struct osd_thandle, ot_super);
3338         LASSERT(oth->ot_handle->h_transaction != NULL);
3339
3340         result = osd_mkfile(info, obj, mode, hint, th, attr);
3341         if (result == 0) {
3342                 LASSERT(obj->oo_inode != NULL);
3343                 if (feat->dif_flags & DT_IND_VARKEY)
3344                         result = iam_lvar_create(obj->oo_inode,
3345                                                  feat->dif_keysize_max,
3346                                                  feat->dif_ptrsize,
3347                                                  feat->dif_recsize_max,
3348                                                  oth->ot_handle);
3349                 else
3350                         result = iam_lfix_create(obj->oo_inode,
3351                                                  feat->dif_keysize_max,
3352                                                  feat->dif_ptrsize,
3353                                                  feat->dif_recsize_max,
3354                                                  oth->ot_handle);
3355         }
3356         return result;
3357 }
3358
3359 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3360                      struct lu_attr *attr,
3361                      struct dt_allocation_hint *hint,
3362                      struct dt_object_format *dof,
3363                      struct thandle *th)
3364 {
3365         LASSERT(S_ISREG(attr->la_mode));
3366         return osd_mkfile(info, obj, (attr->la_mode &
3367                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3368                           attr);
3369 }
3370
3371 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3372                      struct lu_attr *attr,
3373                      struct dt_allocation_hint *hint,
3374                      struct dt_object_format *dof,
3375                      struct thandle *th)
3376 {
3377         LASSERT(S_ISLNK(attr->la_mode));
3378         return osd_mkfile(info, obj, (attr->la_mode &
3379                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3380                           attr);
3381 }
3382
3383 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3384                      struct lu_attr *attr,
3385                      struct dt_allocation_hint *hint,
3386                      struct dt_object_format *dof,
3387                      struct thandle *th)
3388 {
3389         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3390         int result;
3391
3392         LINVRNT(osd_invariant(obj));
3393         LASSERT(obj->oo_inode == NULL);
3394         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3395                 S_ISFIFO(mode) || S_ISSOCK(mode));
3396
3397         result = osd_mkfile(info, obj, mode, hint, th, attr);
3398         if (result == 0) {
3399                 LASSERT(obj->oo_inode != NULL);
3400                 /*
3401                  * This inode should be marked dirty for i_rdev.  Currently
3402                  * that is done in the osd_attr_init().
3403                  */
3404                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3405                                    attr->la_rdev);
3406         }
3407         LINVRNT(osd_invariant(obj));
3408         return result;
3409 }
3410
3411 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3412                               struct lu_attr *,
3413                               struct dt_allocation_hint *hint,
3414                               struct dt_object_format *dof,
3415                               struct thandle *);
3416
3417 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3418 {
3419         osd_obj_type_f result;
3420
3421         switch (type) {
3422         case DFT_DIR:
3423                 result = osd_mkdir;
3424                 break;
3425         case DFT_REGULAR:
3426                 result = osd_mkreg;
3427                 break;
3428         case DFT_SYM:
3429                 result = osd_mksym;
3430                 break;
3431         case DFT_NODE:
3432                 result = osd_mknod;
3433                 break;
3434         case DFT_INDEX:
3435                 result = osd_mk_index;
3436                 break;
3437
3438         default:
3439                 LBUG();
3440                 break;
3441         }
3442         return result;
3443 }
3444
3445
3446 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3447                         struct dt_object *parent, struct dt_object *child,
3448                         umode_t child_mode)
3449 {
3450         LASSERT(ah);
3451
3452         ah->dah_parent = parent;
3453         ah->dah_mode = child_mode;
3454
3455         if (parent != NULL && !dt_object_remote(parent)) {
3456                 /* will help to find FID->ino at dt_insert("..") */
3457                 struct osd_object *pobj = osd_dt_obj(parent);
3458
3459                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3460         }
3461 }
3462
3463 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3464                           struct lu_attr *attr, struct dt_object_format *dof,
3465                           struct thandle *handle)
3466 {
3467         struct inode *inode = obj->oo_inode;
3468         __u64 valid = attr->la_valid;
3469         int result;
3470
3471         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3472
3473         if (dof->dof_type != DFT_NODE)
3474                 attr->la_valid &= ~LA_RDEV;
3475         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3476                 attr->la_valid &= ~LA_ATIME;
3477         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3478                 attr->la_valid &= ~LA_CTIME;
3479         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3480                 attr->la_valid &= ~LA_MTIME;
3481
3482         result = osd_quota_transfer(inode, attr, handle);
3483         if (result)
3484                 return;
3485
3486         if (attr->la_valid != 0) {
3487                 result = osd_inode_setattr(info->oti_env, inode, attr);
3488                 /*
3489                  * The osd_inode_setattr() should always succeed here.  The
3490                  * only error that could be returned is EDQUOT when we are
3491                  * trying to change the UID or GID of the inode. However, this
3492                  * should not happen since quota enforcement is no longer
3493                  * enabled on ldiskfs (lquota takes care of it).
3494                  */
3495                 LASSERTF(result == 0, "%d\n", result);
3496                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3497         }
3498
3499         attr->la_valid = valid;
3500 }
3501
3502 /**
3503  * Helper function for osd_create()
3504  *
3505  * \retval 0, on success
3506  */
3507 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3508                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3509                         struct dt_object_format *dof, struct thandle *th)
3510 {
3511         int result;
3512         __u32 umask;
3513
3514         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3515
3516         /* we drop umask so that permissions we pass are not affected */
3517         umask = current->fs->umask;
3518         current->fs->umask = 0;
3519
3520         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3521                                                   th);
3522         if (likely(obj->oo_inode != NULL)) {
3523                 LASSERT(obj->oo_inode->i_state & I_NEW);
3524
3525                 /*
3526                  * Unlock the inode before attr initialization to avoid
3527                  * unnecessary dqget operations. LU-6378
3528                  */
3529                 unlock_new_inode(obj->oo_inode);
3530         }
3531
3532         if (likely(result == 0)) {
3533                 osd_attr_init(info, obj, attr, dof, th);
3534                 osd_object_init0(obj);
3535         }
3536
3537         /* restore previous umask value */
3538         current->fs->umask = umask;
3539
3540         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3541
3542         return result;
3543 }
3544
3545 /**
3546  * Helper function for osd_create()
3547  *
3548  * \retval 0, on success
3549  */
3550 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3551                            const struct lu_fid *fid, struct thandle *th)
3552 {
3553         struct osd_thread_info *info = osd_oti_get(env);
3554         struct osd_inode_id    *id   = &info->oti_id;
3555         struct osd_device      *osd  = osd_obj2dev(obj);
3556         struct osd_thandle     *oh;
3557         int rc;
3558
3559         LASSERT(obj->oo_inode != NULL);
3560
3561         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
3562                 return -ENOSPC;
3563
3564         oh = container_of(th, struct osd_thandle, ot_super);
3565         LASSERT(oh->ot_handle);
3566         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3567
3568         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3569         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3570                            OI_CHECK_FLD, NULL);
3571         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3572                 struct lu_fid next_fid = *fid;
3573
3574                 /* insert next object in advance, and map to the same inode */
3575                 next_fid.f_oid++;
3576                 if (next_fid.f_oid != 0) {
3577                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3578                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3579                                       OI_CHECK_FLD, NULL);
3580                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3581                 }
3582         }
3583
3584         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3585
3586         return rc;
3587 }
3588
3589 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3590                    u64 seq, struct lu_seq_range *range)
3591 {
3592         struct seq_server_site *ss = osd_seq_site(osd);
3593
3594         if (fid_seq_is_idif(seq)) {
3595                 fld_range_set_ost(range);
3596                 range->lsr_index = idif_ost_idx(seq);
3597                 return 0;
3598         }
3599
3600         if (!fid_seq_in_fldb(seq)) {
3601                 fld_range_set_mdt(range);
3602                 if (ss != NULL)
3603                         /*
3604                          * FIXME: If ss is NULL, it suppose not get lsr_index
3605                          * at all
3606                          */
3607                         range->lsr_index = ss->ss_node_id;
3608                 return 0;
3609         }
3610
3611         LASSERT(ss != NULL);
3612         fld_range_set_any(range);
3613         /* OSD will only do local fld lookup */
3614         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3615 }
3616
3617 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3618                               struct lu_attr *attr,
3619                               struct dt_allocation_hint *hint,
3620                               struct dt_object_format *dof,
3621                               struct thandle *handle)
3622 {
3623         struct osd_thandle *oh;
3624         int rc;
3625
3626         ENTRY;
3627
3628         LASSERT(handle != NULL);
3629
3630         oh = container_of(handle, struct osd_thandle, ot_super);
3631         LASSERT(oh->ot_handle == NULL);
3632
3633         /*
3634          * EA object consumes more credits than regular object: osd_mk_index
3635          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3636          * leaf_node, could involves the block, block bitmap, groups, GDT
3637          * change for each block, so add 4 * 2 credits in that case.
3638          */
3639         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3640                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3641                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3642         /*
3643          * Reuse idle OI block may cause additional one OI block
3644          * to be changed.
3645          */
3646         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3647                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3648         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3649                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3650                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3651
3652         /* will help to find FID->ino mapping at dt_insert() */
3653         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3654                                    osd_dt_obj(dt));
3655         if (rc != 0)
3656                 RETURN(rc);
3657
3658         if (!attr)
3659                 RETURN(0);
3660
3661         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3662                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3663                                    NULL, OSD_QID_INODE);
3664         if (rc != 0)
3665                 RETURN(rc);
3666
3667         RETURN(rc);
3668 }
3669
3670 /**
3671  * Called to destroy on-disk representation of the object
3672  *
3673  * Concurrency: must be locked
3674  */
3675 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3676                                struct thandle *th)
3677 {
3678         struct osd_object *obj = osd_dt_obj(dt);
3679         struct inode *inode = obj->oo_inode;
3680         struct osd_thandle *oh;
3681         int rc;
3682
3683         ENTRY;
3684
3685         if (inode == NULL)
3686                 RETURN(-ENOENT);
3687
3688         oh = container_of(th, struct osd_thandle, ot_super);
3689         LASSERT(oh->ot_handle == NULL);
3690
3691         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3692                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3693
3694         /* For removing agent entry */
3695         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3696                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3697
3698         /*
3699          * Recycle idle OI leaf may cause additional three OI blocks
3700          * to be changed.
3701          */
3702         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3703                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3704                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3705         /* one less inode */
3706         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3707                                    i_projid_read(inode), -1, oh, obj, NULL,
3708                                    OSD_QID_INODE);
3709         if (rc)
3710                 RETURN(rc);
3711         /* data to be truncated */
3712         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3713                                    i_projid_read(inode), 0, oh, obj, NULL,
3714                                    OSD_QID_BLK);
3715         if (rc)
3716                 RETURN(rc);
3717
3718         /*
3719          * will help to find FID->ino when this object is being
3720          * added to PENDING
3721          */
3722         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3723
3724         RETURN(rc);
3725 }
3726
3727 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3728                        struct thandle *th)
3729 {
3730         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3731         struct osd_object *obj = osd_dt_obj(dt);
3732         struct inode *inode = obj->oo_inode;
3733         struct osd_device *osd = osd_obj2dev(obj);
3734         struct osd_thandle *oh;
3735         int result;
3736
3737         ENTRY;
3738
3739         oh = container_of(th, struct osd_thandle, ot_super);
3740         LASSERT(oh->ot_handle);
3741         LASSERT(inode);
3742         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3743
3744         if (unlikely(fid_is_acct(fid)))
3745                 RETURN(-EPERM);
3746
3747         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3748                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3749                 if (result != 0)
3750                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3751                                osd_name(osd), PFID(fid), result);
3752         }
3753
3754         if (S_ISDIR(inode->i_mode)) {
3755                 if (inode->i_nlink > 2)
3756                         CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3757                                osd_name(osd), PFID(fid), inode->i_ino,
3758                                inode->i_nlink);
3759
3760                 spin_lock(&obj->oo_guard);
3761                 clear_nlink(inode);
3762                 spin_unlock(&obj->oo_guard);
3763                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3764         }
3765
3766         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3767
3768         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3769
3770         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3771                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3772                                        oh->ot_handle, OI_CHECK_FLD);
3773
3774         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3775         /* XXX: add to ext3 orphan list */
3776         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3777
3778         /* not needed in the cache anymore */
3779         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3780         obj->oo_destroyed = 1;
3781
3782         RETURN(0);
3783 }
3784
3785 /**
3786  * Put the fid into lustre_mdt_attrs, and then place the structure
3787  * inode's ea. This fid should not be altered during the life time
3788  * of the inode.
3789  *
3790  * \retval +ve, on success
3791  * \retval -ve, on error
3792  *
3793  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3794  */
3795 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3796                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3797 {
3798         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3799         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3800         int rc;
3801
3802         ENTRY;
3803
3804         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3805                 RETURN(0);
3806
3807         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3808                 rc = -ENOMEM;
3809
3810         lustre_loa_init(loa, fid, compat, incompat);
3811         lustre_loa_swab(loa, false);
3812
3813         /*
3814          * For the OST device with 256 bytes inode size by default,
3815          * the PFID EA will be stored together with LMA EA to avoid
3816          * performance trouble. Otherwise the PFID EA can be stored
3817          * independently. LU-8998
3818          */
3819         if ((compat & LMAC_FID_ON_OST) &&
3820             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3821                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3822                                      sizeof(*loa), XATTR_CREATE);
3823         else
3824                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3825                                      sizeof(*lma), XATTR_CREATE);
3826         /*
3827          * LMA may already exist, but we need to check that all the
3828          * desired compat/incompat flags have been added.
3829          */
3830         if (unlikely(rc == -EEXIST)) {
3831                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3832                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3833                 if (rc < 0)
3834                         RETURN(rc);
3835
3836                 if (rc < sizeof(*lma))
3837                         RETURN(-EINVAL);
3838
3839                 lustre_loa_swab(loa, true);
3840                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3841                     ((compat == 0 && incompat == 0) ||
3842                      (!(~lma->lma_compat & compat) &&
3843                       !(~lma->lma_incompat & incompat))))
3844                         RETURN(0);
3845
3846                 lma->lma_self_fid = *fid;
3847                 lma->lma_compat |= compat;
3848                 lma->lma_incompat |= incompat;
3849                 if (rc == sizeof(*lma)) {
3850                         lustre_lma_swab(lma);
3851                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3852                                              sizeof(*lma), XATTR_REPLACE);
3853                 } else {
3854                         lustre_loa_swab(loa, false);
3855                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3856                                              sizeof(*loa), XATTR_REPLACE);
3857                 }
3858         }
3859
3860         RETURN(rc);
3861 }
3862
3863 /**
3864  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3865  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3866  * To have compatilibility with 1.8 ldiskfs driver we need to have
3867  * magic number at start of fid data.
3868  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3869  * its inmemory API.
3870  */
3871 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3872                                          const struct lu_fid *fid)
3873 {
3874         if (!fid_is_namespace_visible(fid) ||
3875             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3876                 param->edp_magic = 0;
3877                 return;
3878         }
3879
3880         param->edp_magic = LDISKFS_LUFID_MAGIC;
3881         param->edp_len =  sizeof(struct lu_fid) + 1;
3882         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3883 }
3884
3885 /**
3886  * Try to read the fid from inode ea into dt_rec.
3887  *
3888  * \param fid object fid.
3889  *
3890  * \retval 0 on success
3891  */
3892 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3893                           __u32 ino, struct lu_fid *fid,
3894                           struct osd_inode_id *id)
3895 {
3896         struct osd_thread_info *info  = osd_oti_get(env);
3897         struct inode *inode;
3898
3899         ENTRY;
3900
3901         osd_id_gen(id, ino, OSD_OII_NOGEN);
3902         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3903         if (IS_ERR(inode))
3904                 RETURN(PTR_ERR(inode));
3905
3906         iput(inode);
3907         RETURN(0);
3908 }
3909
3910 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3911                                         struct inode *dir,
3912                                         struct inode *parent_dir,
3913                                         const struct lu_fid *dot_fid,
3914                                         const struct lu_fid *dot_dot_fid,
3915                                         struct osd_thandle *oth)
3916 {
3917         struct ldiskfs_dentry_param *dot_ldp;
3918         struct ldiskfs_dentry_param *dot_dot_ldp;
3919         __u32 saved_nlink = dir->i_nlink;
3920         int rc;
3921
3922         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
3923                 return -ENOSPC;
3924
3925         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3926         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3927
3928         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3929         dot_ldp->edp_magic = 0;
3930
3931         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3932                                     dir, dot_ldp, dot_dot_ldp);
3933         /*
3934          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3935          * the subseqent ref_add() will increase the dir->i_nlink
3936          * as 3. That is incorrect for new created directory.
3937          *
3938          * It looks like hack, because we want to make the OSD API
3939          * to be order-independent for new created directory object
3940          * between dt_insert(..) and ref_add() operations.
3941          *
3942          * Here, we only restore the in-RAM dir-inode's nlink attr,
3943          * becuase if the nlink attr is not 2, then there will be
3944          * ref_add() called following the dt_insert(..), such call
3945          * will make both the in-RAM and on-disk dir-inode's nlink
3946          * attr to be set as 2. LU-7447
3947          */
3948         set_nlink(dir, saved_nlink);
3949         return rc;
3950 }
3951
3952 /**
3953  * Create an local agent inode for remote entry
3954  */
3955 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3956                                                   struct osd_device *osd,
3957                                                   struct osd_object *pobj,
3958                                                   const struct lu_fid *fid,
3959                                                   __u32 type,
3960                                                   struct thandle *th)
3961 {
3962         struct osd_thread_info *info = osd_oti_get(env);
3963         struct inode *local;
3964         struct osd_thandle *oh;
3965         struct iattr iattr = {
3966                 .ia_valid = ATTR_UID | ATTR_GID |
3967                             ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3968                 .ia_ctime.tv_nsec = UTIME_OMIT,
3969                 .ia_mtime.tv_nsec = UTIME_OMIT,
3970                 .ia_atime.tv_nsec = UTIME_OMIT,
3971                 .ia_uid = GLOBAL_ROOT_UID,
3972                 .ia_gid = GLOBAL_ROOT_GID,
3973         };
3974         int rc;
3975
3976         ENTRY;
3977
3978         LASSERT(th);
3979         oh = container_of(th, struct osd_thandle, ot_super);
3980         LASSERT(oh->ot_handle->h_transaction != NULL);
3981
3982         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
3983                                      type, &iattr);
3984         if (IS_ERR(local)) {
3985                 CERROR("%s: create local error %d\n", osd_name(osd),
3986                        (int)PTR_ERR(local));
3987                 RETURN(local);
3988         }
3989
3990         /*
3991          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3992          * correct gid on remote file, not agent here
3993          */
3994         local->i_gid = current_fsgid();
3995         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3996
3997         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3998          * That gives e2fsck something to look at and be happy, and allows
3999          * debugging if we need to determine where this symlink came from.
4000          */
4001         if (S_ISLNK(type)) {
4002                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
4003                 rc = scnprintf((char *)LDISKFS_I(local)->i_data,
4004                                LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
4005
4006                 i_size_write(local, rc);
4007                 LDISKFS_I(local)->i_disksize = rc;
4008         }
4009         unlock_new_inode(local);
4010
4011         /* Agent inode should not have project ID */
4012 #ifdef  HAVE_PROJECT_QUOTA
4013         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
4014             i_projid_read(pobj->oo_inode) != 0) {
4015                 rc = osd_transfer_project(local, 0, th);
4016                 if (rc) {
4017                         CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
4018                                osd_ino2name(local), rc);
4019                         RETURN(ERR_PTR(rc));
4020                 }
4021         }
4022 #endif
4023         /* Set special LMA flag for local agent inode */
4024         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
4025         if (rc != 0) {
4026                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
4027                        osd_name(osd), PFID(fid), rc);
4028                 RETURN(ERR_PTR(rc));
4029         }
4030
4031         if (!S_ISDIR(type))
4032                 RETURN(local);
4033
4034         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
4035                                          lu_object_fid(&pobj->oo_dt.do_lu),
4036                                          fid, oh);
4037         if (rc != 0) {
4038                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
4039                         osd_name(osd), PFID(fid), rc);
4040                 RETURN(ERR_PTR(rc));
4041         }
4042
4043         RETURN(local);
4044 }
4045
4046 /**
4047  * when direntry is deleted, we have to take care of possible agent inode
4048  * referenced by that. unfortunately we can't do this at that point:
4049  * iget() within a running transaction leads to deadlock and we better do
4050  * not call that every delete declaration to save performance. so we put
4051  * a potention agent inode on a list and process that once the transaction
4052  * is over. Notice it's not any worse in terms of real orphans as regular
4053  * object destroy doesn't put inodes on the on-disk orphan list. this should
4054  * be addressed separately
4055  */
4056 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
4057                                             struct osd_thandle *oh,
4058                                             __u32 ino)
4059 {
4060         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
4061         struct osd_obj_orphan *oor;
4062
4063         OBD_ALLOC_PTR(oor);
4064         if (oor == NULL)
4065                 return -ENOMEM;
4066
4067         oor->oor_ino = ino;
4068         oor->oor_env = (struct lu_env *)env;
4069         spin_lock(&osd->od_osfs_lock);
4070         list_add(&oor->oor_list, &osd->od_orphan_list);
4071         spin_unlock(&osd->od_osfs_lock);
4072
4073         oh->ot_remove_agents = 1;
4074
4075         return 0;
4076
4077 }
4078
4079 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
4080                                                 struct osd_device *osd)
4081 {
4082         struct osd_thread_info *info = osd_oti_get(env);
4083         struct osd_obj_orphan *oor, *tmp;
4084         struct osd_inode_id id;
4085         LIST_HEAD(list);
4086         struct inode *inode;
4087         struct lu_fid fid;
4088         handle_t *jh;
4089         __u32 ino;
4090
4091         spin_lock(&osd->od_osfs_lock);
4092         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4093                 if (oor->oor_env == env)
4094                         list_move(&oor->oor_list, &list);
4095         }
4096         spin_unlock(&osd->od_osfs_lock);
4097
4098         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4099
4100                 ino = oor->oor_ino;
4101
4102                 list_del(&oor->oor_list);
4103                 OBD_FREE_PTR(oor);
4104
4105                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4106                 inode = osd_iget_fid(info, osd, &id, &fid);
4107                 if (IS_ERR(inode))
4108                         continue;
4109
4110                 if (!osd_remote_fid(env, osd, &fid)) {
4111                         iput(inode);
4112                         continue;
4113                 }
4114
4115                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4116                 clear_nlink(inode);
4117                 mark_inode_dirty(inode);
4118                 ldiskfs_journal_stop(jh);
4119                 iput(inode);
4120         }
4121
4122         return 0;
4123 }
4124
4125 /**
4126  * OSD layer object create function for OST objects (b=11826).
4127  *
4128  * The FID is inserted into inode xattr here.
4129  *
4130  * \retval   0, on success
4131  * \retval -ve, on error
4132  */
4133 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4134                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4135                       struct dt_object_format *dof, struct thandle *th)
4136 {
4137         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4138         struct osd_object *obj = osd_dt_obj(dt);
4139         struct osd_thread_info *info = osd_oti_get(env);
4140         int result, on_ost = 0;
4141
4142         ENTRY;
4143
4144         if (dt_object_exists(dt))
4145                 RETURN(-EEXIST);
4146
4147         LINVRNT(osd_invariant(obj));
4148         LASSERT(!dt_object_remote(dt));
4149         LASSERT(osd_is_write_locked(env, obj));
4150         LASSERT(th != NULL);
4151
4152         if (unlikely(fid_is_acct(fid)))
4153                 /*
4154                  * Quota files can't be created from the kernel any more,
4155                  * 'tune2fs -O quota' will take care of creating them
4156                  */
4157                 RETURN(-EPERM);
4158
4159         result = __osd_create(info, obj, attr, hint, dof, th);
4160         if (result == 0) {
4161                 if (fid_is_idif(fid) &&
4162                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4163                         struct lu_fid *tfid = &info->oti_fid;
4164                         struct ost_id *oi   = &info->oti_ostid;
4165
4166                         fid_to_ostid(fid, oi);
4167                         ostid_to_fid(tfid, oi, 0);
4168                         on_ost = 1;
4169                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4170                                                 LMAC_FID_ON_OST, 0);
4171                 } else {
4172                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4173                                                fid, OI_CHECK_FLD);
4174                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4175                                                 on_ost ? LMAC_FID_ON_OST : 0,
4176                                                 0);
4177                 }
4178                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4179                         obj->oo_dt.do_body_ops = &osd_body_ops;
4180         }
4181
4182         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
4183                 struct inode *inode = obj->oo_inode;
4184
4185                 result = __osd_oi_insert(env, obj, fid, th);
4186                 if (result && inode) {
4187                         spin_lock(&obj->oo_guard);
4188                         clear_nlink(inode);
4189                         spin_unlock(&obj->oo_guard);
4190                         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4191                         ldiskfs_set_inode_state(inode,
4192                                                 LDISKFS_STATE_LUSTRE_DESTROY);
4193                         iput(inode);
4194                         obj->oo_inode = NULL;
4195                 }
4196         }
4197
4198         /*
4199          * a small optimization - dt_insert() isn't usually applied
4200          * to OST objects, so we don't need to cache OI mapping for
4201          * OST objects
4202          */
4203         if (result == 0 && on_ost == 0) {
4204                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4205
4206                 result = osd_idc_find_and_init(env, osd, obj);
4207                 LASSERT(result == 0);
4208         }
4209
4210         LASSERT(ergo(result == 0,
4211                      dt_object_exists(dt) && !dt_object_remote(dt)));
4212         LINVRNT(osd_invariant(obj));
4213         RETURN(result);
4214 }
4215
4216 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4217                                struct thandle *handle)
4218 {
4219         struct osd_thandle *oh;
4220         int rc;
4221
4222         /* it's possible that object doesn't exist yet */
4223         LASSERT(handle != NULL);
4224
4225         oh = container_of(handle, struct osd_thandle, ot_super);
4226         LASSERT(oh->ot_handle == NULL);
4227
4228         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4229                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4230
4231         rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
4232                                    osd_dt_obj(dt));
4233
4234         return rc;
4235 }
4236
4237 /*
4238  * Concurrency: @dt is write locked.
4239  */
4240 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4241                        struct thandle *th)
4242 {
4243         struct osd_object *obj = osd_dt_obj(dt);
4244         struct inode *inode = obj->oo_inode;
4245         struct osd_thandle *oh;
4246         int rc = 0;
4247
4248         if (!dt_object_exists(dt) || obj->oo_destroyed)
4249                 return -ENOENT;
4250
4251         LINVRNT(osd_invariant(obj));
4252         LASSERT(!dt_object_remote(dt));
4253         LASSERT(osd_is_write_locked(env, obj));
4254         LASSERT(th != NULL);
4255
4256         oh = container_of(th, struct osd_thandle, ot_super);
4257         LASSERT(oh->ot_handle != NULL);
4258
4259         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4260
4261         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4262                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4263         /*
4264          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4265          * (65000) subdirectories by storing "1" in i_nlink if the link count
4266          * would otherwise overflow. Directory tranversal tools understand
4267          * that (st_nlink == 1) indicates that the filesystem dose not track
4268          * hard links count on the directory, and will not abort subdirectory
4269          * scanning early once (st_nlink - 2) subdirs have been found.
4270          *
4271          * This also has to properly handle the case of inodes with nlink == 0
4272          * in case they are being linked into the PENDING directory
4273          */
4274         spin_lock(&obj->oo_guard);
4275         if (unlikely(inode->i_nlink == 0))
4276                 /* inc_nlink from 0 may cause WARN_ON */
4277                 set_nlink(inode, 1);
4278         else {
4279                 ldiskfs_inc_count(oh->ot_handle, inode);
4280                 if (!S_ISDIR(inode->i_mode))
4281                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4282         }
4283         spin_unlock(&obj->oo_guard);
4284
4285         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4286         LINVRNT(osd_invariant(obj));
4287
4288         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4289
4290         return rc;
4291 }
4292
4293 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4294                                struct thandle *handle)
4295 {
4296         struct osd_thandle *oh;
4297
4298         if (!dt_object_exists(dt))
4299                 return -ENOENT;
4300
4301         LASSERT(!dt_object_remote(dt));
4302         LASSERT(handle != NULL);
4303
4304         oh = container_of(handle, struct osd_thandle, ot_super);
4305         LASSERT(oh->ot_handle == NULL);
4306
4307         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4308                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4309
4310         return 0;
4311 }
4312
4313 /*
4314  * Concurrency: @dt is write locked.
4315  */
4316 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4317                        struct thandle *th)
4318 {
4319         struct osd_object *obj = osd_dt_obj(dt);
4320         struct inode *inode = obj->oo_inode;
4321         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4322         struct osd_thandle *oh;
4323
4324         if (!dt_object_exists(dt))
4325                 return -ENOENT;
4326
4327         LINVRNT(osd_invariant(obj));
4328         LASSERT(!dt_object_remote(dt));
4329         LASSERT(osd_is_write_locked(env, obj));
4330         LASSERT(th != NULL);
4331
4332         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4333                 return -EIO;
4334
4335         oh = container_of(th, struct osd_thandle, ot_super);
4336         LASSERT(oh->ot_handle != NULL);
4337
4338         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4339
4340         spin_lock(&obj->oo_guard);
4341         /*
4342          * That can be result of upgrade from old Lustre version and
4343          * applied only to local files.  Just skip this ref_del call.
4344          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4345          */
4346         if (inode->i_nlink == 0) {
4347                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4348                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4349                              ", maybe an upgraded file? (LU-3915)\n",
4350                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4351                 spin_unlock(&obj->oo_guard);
4352                 return 0;
4353         }
4354
4355         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4356                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4357
4358         ldiskfs_dec_count(oh->ot_handle, inode);
4359         spin_unlock(&obj->oo_guard);
4360
4361         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4362         LINVRNT(osd_invariant(obj));
4363
4364         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4365
4366         return 0;
4367 }
4368
4369 /*
4370  * Concurrency: @dt is read locked.
4371  */
4372 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4373                          struct lu_buf *buf, const char *name)
4374 {
4375         struct osd_object      *obj    = osd_dt_obj(dt);
4376         struct inode           *inode  = obj->oo_inode;
4377         struct osd_thread_info *info   = osd_oti_get(env);
4378         struct dentry          *dentry = &info->oti_obj_dentry;
4379         bool cache_xattr = false;
4380         int rc;
4381
4382         LASSERT(buf);
4383
4384         /* version get is not real XATTR but uses xattr API */
4385         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4386                 dt_obj_version_t *ver = buf->lb_buf;
4387
4388                 /*
4389                  * for version we are just using xattr API but change inode
4390                  * field instead
4391                  */
4392                 if (buf->lb_len == 0)
4393                         return sizeof(dt_obj_version_t);
4394
4395                 if (buf->lb_len < sizeof(dt_obj_version_t))
4396                         return -ERANGE;
4397
4398                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4399                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4400
4401                 *ver = LDISKFS_I(inode)->i_fs_version;
4402
4403                 return sizeof(dt_obj_version_t);
4404         }
4405
4406         if (!dt_object_exists(dt))
4407                 return -ENOENT;
4408
4409         LASSERT(!dt_object_remote(dt));
4410         LASSERT(inode->i_op != NULL);
4411 #ifdef HAVE_IOP_XATTR
4412         LASSERT(inode->i_op->getxattr != NULL);
4413 #endif
4414
4415         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4416             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4417                 cache_xattr = true;
4418
4419         if (cache_xattr) {
4420                 rc = osd_oxc_get(obj, name, buf);
4421                 if (rc != -ENOENT)
4422                         return rc;
4423         }
4424
4425         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4426                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4427                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4428                 struct filter_fid *ff;
4429                 struct ost_layout *ol;
4430
4431                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4432                 if (rc)
4433                         return rc;
4434
4435                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4436
4437                 rc = sizeof(*ff);
4438                 if (buf->lb_len == 0 || !buf->lb_buf)
4439                         return rc;
4440
4441                 if (buf->lb_len < rc)
4442                         return -ERANGE;
4443
4444                 ff = buf->lb_buf;
4445                 ol = &ff->ff_layout;
4446                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4447                                                   PFID_STRIPE_IDX_BITS);
4448                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4449                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4450                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4451                 if (lma->lma_compat & LMAC_COMP_INFO) {
4452                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4453                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4454                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4455                 } else {
4456                         ol->ol_comp_start = 0;
4457                         ol->ol_comp_end = 0;
4458                         ol->ol_comp_id = 0;
4459                 }
4460         } else {
4461                 rc = __osd_xattr_get(inode, dentry, name,
4462                                      buf->lb_buf, buf->lb_len);
4463         }
4464
4465         if (cache_xattr) {
4466                 if (rc == -ENOENT || rc == -ENODATA)
4467                         osd_oxc_add(obj, name, NULL, 0);
4468                 else if (rc > 0 && buf->lb_buf != NULL)
4469                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4470         }
4471
4472         return rc;
4473 }
4474
4475 static int osd_declare_xattr_set(const struct lu_env *env,
4476                                  struct dt_object *dt,
4477                                  const struct lu_buf *buf, const char *name,
4478                                  int fl, struct thandle *handle)
4479 {
4480         struct osd_thandle *oh;
4481         int credits = 0;
4482         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4483
4484         LASSERT(handle != NULL);
4485
4486         oh = container_of(handle, struct osd_thandle, ot_super);
4487         LASSERT(oh->ot_handle == NULL);
4488
4489         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4490                 /*
4491                  * For non-upgrading case, the LMA is set first and
4492                  * usually fit inode. But for upgrade case, the LMA
4493                  * may be in another separated EA block.
4494                  */
4495                 if (dt_object_exists(dt)) {
4496                         if (fl == LU_XATTR_REPLACE)
4497                                 credits = 1;
4498                         else
4499                                 goto upgrade;
4500                 }
4501         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4502                 credits = 1;
4503         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4504                 /* We may need to delete the old PFID EA. */
4505                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4506                 if (fl == LU_XATTR_REPLACE)
4507                         credits += 1;
4508                 else
4509                         goto upgrade;
4510         } else {
4511                 /*
4512                  * If some name entry resides on remote MDT, then will create
4513                  * agent entry under remote parent. On the other hand, if the
4514                  * remote entry will be removed, then related agent entry may
4515                  * need to be removed from the remote parent. So there may be
4516                  * kinds of cases, let's declare enough credits. The credits
4517                  * for create agent entry is enough for remove case.
4518                  */
4519                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4520                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4521                         if (dt_object_exists(dt))
4522                                 credits += 1; /* For updating LMA */
4523                 }
4524
4525 upgrade:
4526                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4527
4528                 if (buf != NULL) {
4529                         ssize_t buflen;
4530
4531                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4532                                 /*
4533                                  * learn xattr size from osd_xattr_get if
4534                                  * attribute has not been read yet
4535                                  */
4536                                 buflen = __osd_xattr_get(
4537                                     osd_dt_obj(dt)->oo_inode,
4538                                     &osd_oti_get(env)->oti_obj_dentry,
4539                                     name, NULL, 0);
4540                                 if (buflen < 0)
4541                                         buflen = 0;
4542                         } else {
4543                                 buflen = buf->lb_len;
4544                         }
4545
4546                         if (buflen > sb->s_blocksize) {
4547                                 credits += osd_calc_bkmap_credits(
4548                                     sb, NULL, 0, -1,
4549                                     (buflen + sb->s_blocksize - 1) >>
4550                                     sb->s_blocksize_bits);
4551                         }
4552                 }
4553                 /*
4554                  * xattr set may involve inode quota change, reserve credits for
4555                  * dquot_initialize()
4556                  */
4557                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4558         }
4559
4560         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4561
4562         return 0;
4563 }
4564
4565 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4566                               const struct lu_buf *buf, int fl,
4567                               struct thandle *handle)
4568 {
4569         struct osd_thread_info *info = osd_oti_get(env);
4570         struct dentry *dentry = &info->oti_obj_dentry;
4571         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4572         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4573         struct inode *inode = obj->oo_inode;
4574         struct filter_fid *ff = buf->lb_buf;
4575         struct ost_layout *ol = &ff->ff_layout;
4576         int flags = XATTR_REPLACE;
4577         int rc;
4578
4579         ENTRY;
4580
4581         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4582                 RETURN(-EINVAL);
4583
4584         rc = osd_get_lma(info, inode, dentry, loa);
4585         if (rc == -ENODATA) {
4586                 /* Usually for upgarding from old device */
4587                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4588                                 LMAC_FID_ON_OST, 0);
4589                 flags = XATTR_CREATE;
4590         } else if (rc) {
4591                 RETURN(rc);
4592         }
4593
4594         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4595                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4596                         RETURN(-EEXIST);
4597
4598                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4599                         /* Separate PFID EA from LMA */
4600                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4601                         lustre_lma_swab(lma);
4602                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4603                                              sizeof(*lma), XATTR_REPLACE);
4604                         if (!rc) {
4605                                 obj->oo_pfid_in_lma = 0;
4606                                 rc = LU_XATTR_CREATE;
4607                         }
4608
4609                         RETURN(rc);
4610                 }
4611         } else {
4612                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4613                         RETURN(fl);
4614
4615                 /*
4616                  * Old client does not send stripe information,
4617                  * then store the PFID EA on disk separatedly.
4618                  */
4619                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4620                              ol->ol_stripe_size == 0))
4621                         RETURN(fl);
4622
4623                 /* Remove old PFID EA entry firstly. */
4624                 dquot_initialize(inode);
4625                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4626                 if (rc == -ENODATA) {
4627                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4628                                 RETURN(rc);
4629                 } else if (rc) {
4630                         RETURN(rc);
4631                 }
4632         }
4633
4634         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4635         if (likely(ol->ol_stripe_size != 0)) {
4636                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4637                                              PFID_STRIPE_IDX_BITS;
4638                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4639                 lma->lma_compat |= LMAC_STRIPE_INFO;
4640                 if (ol->ol_comp_id != 0) {
4641                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4642                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4643                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4644                         lma->lma_compat |= LMAC_COMP_INFO;
4645                 }
4646         }
4647
4648         lustre_loa_swab(loa, false);
4649
4650         /* Store the PFID EA inside LMA. */
4651         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4652                              flags);
4653         if (!rc)
4654                 obj->oo_pfid_in_lma = 1;
4655
4656         RETURN(rc);
4657 }
4658
4659 /*
4660  * In DNE environment, the object (in spite of regular file or directory)
4661  * and its name entry may reside on different MDTs. Under such case, we will
4662  * create an agent entry on the MDT where the object resides. The agent entry
4663  * references the object locally, that makes the object to be visible to the
4664  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4665  * such as 'tar' can handle the object properly.
4666  *
4667  * We handle the agent entry during set linkEA that is the common interface
4668  * for both regular file and directroy, can handle kinds of cases, such as
4669  * create/link/unlink/rename, and so on.
4670  *
4671  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4672  *
4673  * XXX: There are two known issues:
4674  * 1. For one object, we will create at most one agent entry even if there
4675  *    may be more than one cross-MDTs hard links on the object. So the local
4676  *    e2fsck may claim that the object's nlink is larger than the name entries
4677  *    that reference such inode. And in further, the e2fsck will fix the nlink
4678  *    attribute to match the local references. Then it will cause the object's
4679  *    nlink attribute to be inconsistent with the global references. it is bad
4680  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4681  *    other hand, the global namespace LFSCK can repair the object's attribute
4682  *    according to the linkEA.
4683  * 2. There may be too many hard links on the object as to its linkEA overflow,
4684  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4685  *    case happened, then at this point, we do not know whether there are some
4686  *    cross-MDTs reference. But there are local references, it guarantees that
4687  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4688  */
4689 static int osd_xattr_handle_linkea(const struct lu_env *env,
4690                                    struct osd_device *osd,
4691                                    struct osd_object *obj,
4692                                    const struct lu_buf *buf,
4693                                    struct thandle *handle)
4694 {
4695         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4696         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4697         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4698         struct lu_name tmpname;
4699         struct osd_thandle *oh;
4700         int rc;
4701         bool remote = false;
4702
4703         ENTRY;
4704
4705         oh = container_of(handle, struct osd_thandle, ot_super);
4706         LASSERT(oh->ot_handle != NULL);
4707
4708         rc = linkea_init_with_rec(&ldata);
4709         if (!rc) {
4710                 linkea_first_entry(&ldata);
4711                 while (ldata.ld_lee != NULL && !remote) {
4712                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4713                                             &tmpname, tfid);
4714                         if (osd_remote_fid(env, osd, tfid) > 0)
4715                                 remote = true;
4716                         else
4717                                 linkea_next_entry(&ldata);
4718                 }
4719         } else if (rc == -ENODATA) {
4720                 rc = 0;
4721         } else {
4722                 RETURN(rc);
4723         }
4724
4725         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4726                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4727                 if (rc)
4728                         CERROR("%s: failed to remove agent entry for "DFID
4729                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4730         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4731                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4732                 if (rc)
4733                         CERROR("%s: failed to create agent entry for "DFID
4734                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4735         }
4736
4737         RETURN(rc);
4738 }
4739
4740 /*
4741  * Concurrency: @dt is write locked.
4742  */
4743 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4744                          const struct lu_buf *buf, const char *name, int fl,
4745                          struct thandle *handle)
4746 {
4747         struct osd_object *obj = osd_dt_obj(dt);
4748         struct osd_device *osd = osd_obj2dev(obj);
4749         struct inode *inode = obj->oo_inode;
4750         struct osd_thread_info *info = osd_oti_get(env);
4751         int fs_flags = 0;
4752         int len;
4753         int rc;
4754
4755         ENTRY;
4756
4757         LASSERT(handle);
4758         LASSERT(buf);
4759
4760         /* version set is not real XATTR */
4761         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4762                 dt_obj_version_t *version = buf->lb_buf;
4763
4764                 /*
4765                  * for version we are just using xattr API but change inode
4766                  * field instead
4767                  */
4768                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4769
4770                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4771                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4772
4773                 LDISKFS_I(inode)->i_fs_version = *version;
4774                 /*
4775                  * Version is set after all inode operations are finished,
4776                  * so we should mark it dirty here
4777                  */
4778                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4779
4780                 RETURN(0);
4781         }
4782
4783         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4784                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4785
4786         len = buf->lb_len;
4787         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4788
4789         /*
4790          * For the OST device with 256 bytes inode size by default,
4791          * the PFID EA will be stored together with LMA EA to avoid
4792          * performance trouble. Otherwise the PFID EA can be stored
4793          * independently. LU-8998
4794          */
4795         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4796             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4797                 LASSERT(buf->lb_buf);
4798
4799                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4800                 if (fl <= 0)
4801                         RETURN(fl);
4802         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4803                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4804                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4805
4806                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4807                 if (rc)
4808                         RETURN(rc);
4809
4810                 lma->lma_incompat |= LMAI_STRIPED;
4811                 lustre_lma_swab(lma);
4812                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4813                                      sizeof(*lma), XATTR_REPLACE);
4814                 if (rc != 0)
4815                         RETURN(rc);
4816         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4817                 LASSERT(!osd->od_is_ost);
4818
4819                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4820                 if (rc)
4821                         RETURN(rc);
4822         }
4823
4824         if (fl & LU_XATTR_REPLACE)
4825                 fs_flags |= XATTR_REPLACE;
4826
4827         if (fl & LU_XATTR_CREATE)
4828                 fs_flags |= XATTR_CREATE;
4829
4830         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4831         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4832
4833         if (rc == 0 &&
4834             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4835              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4836                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4837
4838         return rc;
4839 }
4840
4841 /*
4842  * Concurrency: @dt is read locked.
4843  */
4844 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4845                           const struct lu_buf *buf)
4846 {
4847         struct osd_object *obj = osd_dt_obj(dt);
4848         struct inode *inode = obj->oo_inode;
4849         struct osd_thread_info *info = osd_oti_get(env);
4850         struct dentry *dentry = &info->oti_obj_dentry;
4851
4852         if (!dt_object_exists(dt))
4853                 return -ENOENT;
4854
4855         LASSERT(!dt_object_remote(dt));
4856         LASSERT(inode->i_op != NULL);
4857         LASSERT(inode->i_op->listxattr != NULL);
4858
4859         dentry->d_inode = inode;
4860         dentry->d_sb = inode->i_sb;
4861         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4862 }
4863
4864 static int osd_declare_xattr_del(const struct lu_env *env,
4865                                  struct dt_object *dt, const char *name,
4866                                  struct thandle *handle)
4867 {
4868         struct osd_thandle *oh;
4869         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4870
4871         LASSERT(!dt_object_remote(dt));
4872         LASSERT(handle != NULL);
4873
4874         oh = container_of(handle, struct osd_thandle, ot_super);
4875         LASSERT(oh->ot_handle == NULL);
4876
4877         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4878                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4879         /*
4880          * xattr del may involve inode quota change, reserve credits for
4881          * dquot_initialize()
4882          */
4883         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4884
4885         return 0;
4886 }
4887
4888 /*
4889  * Concurrency: @dt is write locked.
4890  */
4891 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4892                          const char *name, struct thandle *handle)
4893 {
4894         struct osd_object *obj = osd_dt_obj(dt);
4895         struct inode *inode = obj->oo_inode;
4896         struct osd_thread_info *info = osd_oti_get(env);
4897         struct dentry *dentry = &info->oti_obj_dentry;
4898         int rc;
4899
4900         if (!dt_object_exists(dt))
4901                 return -ENOENT;
4902
4903         LASSERT(!dt_object_remote(dt));
4904         LASSERT(inode->i_op != NULL);
4905         LASSERT(handle != NULL);
4906 #ifdef HAVE_IOP_XATTR
4907         LASSERT(inode->i_op->removexattr != NULL);
4908 #endif
4909
4910         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4911
4912         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4913                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4914
4915                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4916                                  &info->oti_ost_attrs);
4917                 if (!rc) {
4918                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4919
4920                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4921                         lustre_lma_swab(lma);
4922                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4923                                              sizeof(*lma), XATTR_REPLACE);
4924                         if (!rc)
4925                                 obj->oo_pfid_in_lma = 0;
4926                 }
4927         } else {
4928                 dquot_initialize(inode);
4929                 dentry->d_inode = inode;
4930                 dentry->d_sb = inode->i_sb;
4931                 rc = ll_vfs_removexattr(dentry, inode, name);
4932         }
4933
4934         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4935
4936         if (rc == 0 &&
4937             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4938              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4939                 osd_oxc_del(obj, name);
4940
4941         return rc;
4942 }
4943
4944 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4945                            __u64 start, __u64 end)
4946 {
4947         struct osd_object *obj = osd_dt_obj(dt);
4948         struct inode *inode = obj->oo_inode;
4949         struct file *file = osd_quasi_file(env, inode);
4950         int rc;
4951
4952         ENTRY;
4953
4954         rc = vfs_fsync_range(file, start, end, 0);
4955
4956         RETURN(rc);
4957 }
4958
4959 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4960 {
4961         return 0;
4962 }
4963
4964 static bool osd_check_stale(struct dt_object *dt)
4965 {
4966         return false;
4967 }
4968 /*
4969  * Index operations.
4970  */
4971
4972 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4973                                const struct dt_index_features *feat)
4974 {
4975         struct iam_descr *descr;
4976
4977         if (osd_object_is_root(o))
4978                 return feat == &dt_directory_features;
4979
4980         LASSERT(o->oo_dir != NULL);
4981
4982         descr = o->oo_dir->od_container.ic_descr;
4983         if (feat == &dt_directory_features) {
4984                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4985                         return 1;
4986                 else
4987                         return 0;
4988         } else {
4989                 return feat->dif_keysize_min <= descr->id_key_size &&
4990                        descr->id_key_size <= feat->dif_keysize_max &&
4991                        feat->dif_recsize_min <= descr->id_rec_size &&
4992                        descr->id_rec_size <= feat->dif_recsize_max &&
4993                        !(feat->dif_flags & (DT_IND_VARKEY |
4994                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4995                        ergo(feat->dif_flags & DT_IND_UPDATE,
4996                             1 /* XXX check that object (and fs) is writable */);
4997         }
4998 }
4999
5000 static int osd_iam_container_init(const struct lu_env *env,
5001                                   struct osd_object *obj,
5002                                   struct osd_directory *dir)
5003 {
5004         struct iam_container *bag = &dir->od_container;
5005         int result;
5006
5007         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
5008         if (result != 0)
5009                 return result;
5010
5011         result = iam_container_setup(bag);
5012         if (result == 0)
5013                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
5014         else
5015                 iam_container_fini(bag);
5016
5017         return result;
5018 }
5019
5020
5021 /*
5022  * Concurrency: no external locking is necessary.
5023  */
5024 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
5025                          const struct dt_index_features *feat)
5026 {
5027         int result;
5028         int skip_iam = 0;
5029         struct osd_object *obj = osd_dt_obj(dt);
5030
5031         LINVRNT(osd_invariant(obj));
5032
5033         if (osd_object_is_root(obj)) {
5034                 dt->do_index_ops = &osd_index_ea_ops;
5035                 result = 0;
5036         } else if (feat == &dt_directory_features) {
5037                 dt->do_index_ops = &osd_index_ea_ops;
5038                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
5039                         result = 0;
5040                 else
5041                         result = -ENOTDIR;
5042                 skip_iam = 1;
5043         } else if (unlikely(feat == &dt_otable_features)) {
5044                 dt->do_index_ops = &osd_otable_ops;
5045                 return 0;
5046         } else if (unlikely(feat == &dt_acct_features)) {
5047                 dt->do_index_ops = &osd_acct_index_ops;
5048                 result = 0;
5049                 skip_iam = 1;
5050         } else if (!osd_has_index(obj)) {
5051                 struct osd_directory *dir;
5052                 struct osd_device *osd = osd_obj2dev(obj);
5053                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
5054
5055                 OBD_ALLOC_PTR(dir);
5056                 if (dir) {
5057
5058                         spin_lock(&obj->oo_guard);
5059                         if (obj->oo_dir == NULL)
5060                                 obj->oo_dir = dir;
5061                         else
5062                                 /*
5063                                  * Concurrent thread allocated container data.
5064                                  */
5065                                 OBD_FREE_PTR(dir);
5066                         spin_unlock(&obj->oo_guard);
5067                         /*
5068                          * Now, that we have container data, serialize its
5069                          * initialization.
5070                          */
5071                         down_write(&obj->oo_ext_idx_sem);
5072                         /*
5073                          * recheck under lock.
5074                          */
5075
5076                         if (osd_has_index(obj)) {
5077                                 result = 0;
5078                                 goto unlock;
5079                         }
5080
5081                         result = osd_iam_container_init(env, obj, obj->oo_dir);
5082                         if (result || feat == &dt_lfsck_namespace_features ||
5083                             feat == &dt_lfsck_layout_orphan_features ||
5084                             feat == &dt_lfsck_layout_dangling_features)
5085                                 goto unlock;
5086
5087                         result = osd_index_register(osd, fid,
5088                                                     feat->dif_keysize_max,
5089                                                     feat->dif_recsize_max);
5090                         if (result < 0)
5091                                 CWARN("%s: failed to register index "
5092                                       DFID": rc = %d\n",
5093                                       osd_name(osd), PFID(fid), result);
5094                         else if (result > 0)
5095                                 result = 0;
5096                         else
5097                                 CDEBUG(D_LFSCK, "%s: index object "DFID
5098                                        " (%d/%d) registered\n",
5099                                        osd_name(osd), PFID(fid),
5100                                        (int)feat->dif_keysize_max,
5101                                        (int)feat->dif_recsize_max);
5102
5103 unlock:
5104                         up_write(&obj->oo_ext_idx_sem);
5105                 } else {
5106                         result = -ENOMEM;
5107                 }
5108         } else {
5109                 result = 0;
5110         }
5111
5112         if (result == 0 && skip_iam == 0) {
5113                 if (!osd_iam_index_probe(env, obj, feat))
5114                         result = -ENOTDIR;
5115         }
5116         LINVRNT(osd_invariant(obj));
5117
5118         return result;
5119 }
5120
5121 static int osd_otable_it_attr_get(const struct lu_env *env,
5122                                  struct dt_object *dt,
5123                                  struct lu_attr *attr)
5124 {
5125         attr->la_valid = 0;
5126         return 0;
5127 }
5128
5129 static const struct dt_object_operations osd_obj_ops = {
5130         .do_read_lock           = osd_read_lock,
5131         .do_write_lock          = osd_write_lock,
5132         .do_read_unlock         = osd_read_unlock,
5133         .do_write_unlock        = osd_write_unlock,
5134         .do_write_locked        = osd_write_locked,
5135         .do_attr_get            = osd_attr_get,
5136         .do_declare_attr_set    = osd_declare_attr_set,
5137         .do_attr_set            = osd_attr_set,
5138         .do_ah_init             = osd_ah_init,
5139         .do_declare_create      = osd_declare_create,
5140         .do_create              = osd_create,
5141         .do_declare_destroy     = osd_declare_destroy,
5142         .do_destroy             = osd_destroy,
5143         .do_index_try           = osd_index_try,
5144         .do_declare_ref_add     = osd_declare_ref_add,
5145         .do_ref_add             = osd_ref_add,
5146         .do_declare_ref_del     = osd_declare_ref_del,
5147         .do_ref_del             = osd_ref_del,
5148         .do_xattr_get           = osd_xattr_get,
5149         .do_declare_xattr_set   = osd_declare_xattr_set,
5150         .do_xattr_set           = osd_xattr_set,
5151         .do_declare_xattr_del   = osd_declare_xattr_del,
5152         .do_xattr_del           = osd_xattr_del,
5153         .do_xattr_list          = osd_xattr_list,
5154         .do_object_sync         = osd_object_sync,
5155         .do_invalidate          = osd_invalidate,
5156         .do_check_stale         = osd_check_stale,
5157 };
5158
5159 static const struct dt_object_operations osd_obj_otable_it_ops = {
5160         .do_attr_get    = osd_otable_it_attr_get,
5161         .do_index_try   = osd_index_try,
5162 };
5163
5164 static int osd_index_declare_iam_delete(const struct lu_env *env,
5165                                         struct dt_object *dt,
5166                                         const struct dt_key *key,
5167                                         struct thandle *handle)
5168 {
5169         struct osd_thandle *oh;
5170
5171         oh = container_of(handle, struct osd_thandle, ot_super);
5172         LASSERT(oh->ot_handle == NULL);
5173
5174         /* Recycle  may cause additional three blocks to be changed. */
5175         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5176                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5177
5178         return 0;
5179 }
5180
5181 /**
5182  *      delete a (key, value) pair from index \a dt specified by \a key
5183  *
5184  *      \param  dt      osd index object
5185  *      \param  key     key for index
5186  *      \param  rec     record reference
5187  *      \param  handle  transaction handler
5188  *
5189  *      \retval  0  success
5190  *      \retval -ve   failure
5191  */
5192 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5193                                 const struct dt_key *key,
5194                                 struct thandle *handle)
5195 {
5196         struct osd_thread_info *oti = osd_oti_get(env);
5197         struct osd_object *obj = osd_dt_obj(dt);
5198         struct osd_thandle *oh;
5199         struct iam_path_descr *ipd;
5200         struct iam_container *bag = &obj->oo_dir->od_container;
5201         int rc;
5202
5203         ENTRY;
5204
5205         if (!dt_object_exists(dt))
5206                 RETURN(-ENOENT);
5207
5208         LINVRNT(osd_invariant(obj));
5209         LASSERT(!dt_object_remote(dt));
5210         LASSERT(bag->ic_object == obj->oo_inode);
5211         LASSERT(handle != NULL);
5212
5213         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5214
5215         ipd = osd_idx_ipd_get(env, bag);
5216         if (unlikely(ipd == NULL))
5217                 RETURN(-ENOMEM);
5218
5219         oh = container_of(handle, struct osd_thandle, ot_super);
5220         LASSERT(oh->ot_handle != NULL);
5221         LASSERT(oh->ot_handle->h_transaction != NULL);
5222
5223         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5224                 /* swab quota uid/gid provided by caller */
5225                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5226                 key = (const struct dt_key *)&oti->oti_quota_id;
5227         }
5228
5229         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5230         osd_ipd_put(env, bag, ipd);
5231         LINVRNT(osd_invariant(obj));
5232         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5233         RETURN(rc);
5234 }
5235
5236 static int osd_index_declare_ea_delete(const struct lu_env *env,
5237                                        struct dt_object *dt,
5238                                        const struct dt_key *key,
5239                                        struct thandle *handle)
5240 {
5241         struct osd_thandle *oh;
5242         struct inode *inode;
5243         int rc, credits;
5244
5245         ENTRY;
5246
5247         LASSERT(!dt_object_remote(dt));
5248         LASSERT(handle != NULL);
5249
5250         oh = container_of(handle, struct osd_thandle, ot_super);
5251         LASSERT(oh->ot_handle == NULL);
5252
5253         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5254         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5255
5256         inode = osd_dt_obj(dt)->oo_inode;
5257         if (inode == NULL)
5258                 RETURN(-ENOENT);
5259
5260         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5261                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5262                                    NULL, OSD_QID_BLK);
5263         RETURN(rc);
5264 }
5265
5266 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5267                                           struct dt_rec *fid)
5268 {
5269         struct osd_fid_pack *rec;
5270         int rc = -ENODATA;
5271
5272         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5273                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5274                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5275                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5276                         rc = -EINVAL;
5277         }
5278         return rc;
5279 }
5280
5281 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5282                           const struct lu_fid *fid)
5283 {
5284         struct seq_server_site *ss = osd_seq_site(osd);
5285
5286         ENTRY;
5287
5288         /* FID seqs not in FLDB, must be local seq */
5289         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5290                 RETURN(0);
5291
5292         /*
5293          * If FLD is not being initialized yet, it only happens during the
5294          * initialization, likely during mgs initialization, and we assume
5295          * this is local FID.
5296          */
5297         if (ss == NULL || ss->ss_server_fld == NULL)
5298                 RETURN(0);
5299
5300         /* Only check the local FLDB here */
5301         if (osd_seq_exists(env, osd, fid_seq(fid)))
5302                 RETURN(0);
5303
5304         RETURN(1);
5305 }
5306
5307 static void osd_take_care_of_agent(const struct lu_env *env,
5308                                    struct osd_device *osd,
5309                                    struct osd_thandle *oh,
5310                                    struct ldiskfs_dir_entry_2 *de)
5311 {
5312         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5313         struct osd_idmap_cache *idc;
5314         int rc, schedule = 0;
5315
5316         LASSERT(de != NULL);
5317
5318         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5319         if (likely(rc == 0)) {
5320                 idc = osd_idc_find_or_init(env, osd, fid);
5321                 if (IS_ERR(idc) || idc->oic_remote)
5322                         schedule = 1;
5323         } else if (rc == -ENODATA) {
5324                 /*
5325                  * can't get FID, postpone to the end of the
5326                  * transaction when iget() is safe
5327                  */
5328                 schedule = 1;
5329         } else {
5330                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5331         }
5332         if (schedule)
5333                 osd_schedule_agent_inode_removal(env, oh,
5334                                                  le32_to_cpu(de->inode));
5335 }
5336
5337 /**
5338  * Index delete function for interoperability mode (b11826).
5339  * It will remove the directory entry added by osd_index_ea_insert().
5340  * This entry is needed to maintain name->fid mapping.
5341  *
5342  * \param key,  key i.e. file entry to be deleted
5343  *
5344  * \retval   0, on success
5345  * \retval -ve, on error
5346  */
5347 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5348                                const struct dt_key *key, struct thandle *handle)
5349 {
5350         struct osd_object *obj = osd_dt_obj(dt);
5351         struct inode *dir = obj->oo_inode;
5352         struct dentry *dentry;
5353         struct osd_thandle *oh;
5354         struct ldiskfs_dir_entry_2 *de = NULL;
5355         struct buffer_head *bh;
5356         struct htree_lock *hlock = NULL;
5357         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5358         int rc;
5359
5360         ENTRY;
5361
5362         if (!dt_object_exists(dt))
5363                 RETURN(-ENOENT);
5364
5365         LINVRNT(osd_invariant(obj));
5366         LASSERT(!dt_object_remote(dt));
5367         LASSERT(handle != NULL);
5368
5369         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5370
5371         oh = container_of(handle, struct osd_thandle, ot_super);
5372         LASSERT(oh->ot_handle != NULL);
5373         LASSERT(oh->ot_handle->h_transaction != NULL);
5374
5375         dquot_initialize(dir);
5376         dentry = osd_child_dentry_get(env, obj,
5377                                       (char *)key, strlen((char *)key));
5378
5379         if (obj->oo_hl_head != NULL) {
5380                 hlock = osd_oti_get(env)->oti_hlock;
5381                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5382                                    dir, LDISKFS_HLOCK_DEL);
5383         } else {
5384                 down_write(&obj->oo_ext_idx_sem);
5385         }
5386
5387         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5388         if (!IS_ERR(bh)) {
5389                 /*
5390                  * If this is not the ".." entry, it might be a remote DNE
5391                  * entry and  we need to check if the FID is for a remote
5392                  * MDT.  If the FID is  not in the directory entry (e.g.
5393                  * upgraded 1.8 filesystem without dirdata enabled) then
5394                  * we need to get the FID from the LMA. For a remote directory
5395                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5396                  * case.
5397                  *
5398                  * Delete the entry before the agent inode in order to
5399                  * simplify error handling.  At worst an error after deleting
5400                  * the entry first might leak the agent inode afterward. The
5401                  * reverse would need filesystem abort in case of error deleting
5402                  * the entry after the agent had been removed, or leave a
5403                  * dangling entry pointing at a random inode.
5404                  */
5405                 if (strcmp((char *)key, dotdot) != 0)
5406                         osd_take_care_of_agent(env, osd, oh, de);
5407                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5408                 brelse(bh);
5409         } else {
5410                 rc = PTR_ERR(bh);
5411         }
5412
5413         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5414             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5415                 /* NB, dirent count may not be accurate, because it's counted
5416                  * without lock.
5417                  */
5418                 if (obj->oo_dirent_count)
5419                         obj->oo_dirent_count--;
5420                 else
5421                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5422         }
5423         if (hlock != NULL)
5424                 ldiskfs_htree_unlock(hlock);
5425         else
5426                 up_write(&obj->oo_ext_idx_sem);
5427         GOTO(out, rc);
5428 out:
5429         LASSERT(osd_invariant(obj));
5430         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5431         RETURN(rc);
5432 }
5433
5434 /**
5435  *      Lookup index for \a key and copy record to \a rec.
5436  *
5437  *      \param  dt      osd index object
5438  *      \param  key     key for index
5439  *      \param  rec     record reference
5440  *
5441  *      \retval  +ve  success : exact mach
5442  *      \retval  0    return record with key not greater than \a key
5443  *      \retval -ve   failure
5444  */
5445 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5446                                 struct dt_rec *rec, const struct dt_key *key)
5447 {
5448         struct osd_object *obj = osd_dt_obj(dt);
5449         struct iam_path_descr *ipd;
5450         struct iam_container *bag = &obj->oo_dir->od_container;
5451         struct osd_thread_info *oti = osd_oti_get(env);
5452         struct iam_iterator *it = &oti->oti_idx_it;
5453         struct iam_rec *iam_rec;
5454         int rc;
5455
5456         ENTRY;
5457
5458         if (!dt_object_exists(dt))
5459                 RETURN(-ENOENT);
5460
5461         LASSERT(osd_invariant(obj));
5462         LASSERT(!dt_object_remote(dt));
5463         LASSERT(bag->ic_object == obj->oo_inode);
5464
5465         ipd = osd_idx_ipd_get(env, bag);
5466         if (IS_ERR(ipd))
5467                 RETURN(-ENOMEM);
5468
5469         /* got ipd now we can start iterator. */
5470         iam_it_init(it, bag, 0, ipd);
5471
5472         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5473                 /* swab quota uid/gid provided by caller */
5474                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5475                 key = (const struct dt_key *)&oti->oti_quota_id;
5476         }
5477
5478         rc = iam_it_get(it, (struct iam_key *)key);
5479         if (rc >= 0) {
5480                 if (S_ISDIR(obj->oo_inode->i_mode))
5481                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5482                 else
5483                         iam_rec = (struct iam_rec *)rec;
5484
5485                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5486
5487                 if (S_ISDIR(obj->oo_inode->i_mode))
5488                         osd_fid_unpack((struct lu_fid *)rec,
5489                                        (struct osd_fid_pack *)iam_rec);
5490                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5491                         osd_quota_unpack(obj, rec);
5492         }
5493
5494         iam_it_put(it);
5495         iam_it_fini(it);
5496         osd_ipd_put(env, bag, ipd);
5497
5498         LINVRNT(osd_invariant(obj));
5499
5500         RETURN(rc);
5501 }
5502
5503 static int osd_index_declare_iam_insert(const struct lu_env *env,
5504                                         struct dt_object *dt,
5505                                         const struct dt_rec *rec,
5506                                         const struct dt_key *key,
5507                                         struct thandle *handle)
5508 {
5509         struct osd_thandle *oh;
5510
5511         LASSERT(handle != NULL);
5512
5513         oh = container_of(handle, struct osd_thandle, ot_super);
5514         LASSERT(oh->ot_handle == NULL);
5515
5516         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5517                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5518
5519         return 0;
5520 }
5521
5522 /**
5523  *      Inserts (key, value) pair in \a dt index object.
5524  *
5525  *      \param  dt      osd index object
5526  *      \param  key     key for index
5527  *      \param  rec     record reference
5528  *      \param  th      transaction handler
5529  *
5530  *      \retval  0  success
5531  *      \retval -ve failure
5532  */
5533 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5534                                 const struct dt_rec *rec,
5535                                 const struct dt_key *key, struct thandle *th)
5536 {
5537         struct osd_object *obj = osd_dt_obj(dt);
5538         struct iam_path_descr *ipd;
5539         struct osd_thandle *oh;
5540         struct iam_container *bag;
5541         struct osd_thread_info *oti = osd_oti_get(env);
5542         struct iam_rec *iam_rec;
5543         int rc;
5544
5545         ENTRY;
5546
5547         if (!dt_object_exists(dt))
5548                 RETURN(-ENOENT);
5549
5550         LINVRNT(osd_invariant(obj));
5551         LASSERT(!dt_object_remote(dt));
5552
5553         bag = &obj->oo_dir->od_container;
5554         LASSERT(bag->ic_object == obj->oo_inode);
5555         LASSERT(th != NULL);
5556
5557         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5558
5559         ipd = osd_idx_ipd_get(env, bag);
5560         if (unlikely(ipd == NULL))
5561                 RETURN(-ENOMEM);
5562
5563         oh = container_of(th, struct osd_thandle, ot_super);
5564         LASSERT(oh->ot_handle != NULL);
5565         LASSERT(oh->ot_handle->h_transaction != NULL);
5566         if (S_ISDIR(obj->oo_inode->i_mode)) {
5567                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5568                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5569                              &oti->oti_fid);
5570         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5571                 /* pack quota uid/gid */
5572                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5573                 key = (const struct dt_key *)&oti->oti_quota_id;
5574                 /* pack quota record */
5575                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5576                 iam_rec = (struct iam_rec *)rec;
5577         } else {
5578                 iam_rec = (struct iam_rec *)rec;
5579         }
5580
5581         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5582                         iam_rec, ipd);
5583         osd_ipd_put(env, bag, ipd);
5584         LINVRNT(osd_invariant(obj));
5585         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5586         RETURN(rc);
5587 }
5588
5589 /**
5590  * Calls ldiskfs_add_entry() to add directory entry
5591  * into the directory. This is required for
5592  * interoperability mode (b11826)
5593  *
5594  * \retval   0, on success
5595  * \retval -ve, on error
5596  */
5597 static int __osd_ea_add_rec(struct osd_thread_info *info,
5598                             struct osd_object *pobj, struct inode  *cinode,
5599                             const char *name, const struct lu_fid *fid,
5600                             struct htree_lock *hlock, struct thandle *th)
5601 {
5602         struct ldiskfs_dentry_param *ldp;
5603         struct dentry *child;
5604         struct osd_thandle *oth;
5605         int rc;
5606
5607         oth = container_of(th, struct osd_thandle, ot_super);
5608         LASSERT(oth->ot_handle != NULL);
5609         LASSERT(oth->ot_handle->h_transaction != NULL);
5610         LASSERT(pobj->oo_inode);
5611
5612         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5613         if (unlikely(osd_object_is_root(pobj)))
5614                 ldp->edp_magic = 0;
5615         else
5616                 osd_get_ldiskfs_dirent_param(ldp, fid);
5617         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5618         child->d_fsdata = (void *)ldp;
5619         dquot_initialize(pobj->oo_inode);
5620         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5621                                    child, cinode, hlock);
5622         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5623                 struct ldiskfs_dir_entry_2      *de;
5624                 struct buffer_head              *bh;
5625                 int                              rc1;
5626
5627                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5628                                             NULL, hlock);
5629                 if (!IS_ERR(bh)) {
5630                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5631                                                                bh);
5632                         if (rc1 == 0) {
5633                                 if (S_ISDIR(cinode->i_mode))
5634                                         de->file_type = LDISKFS_DIRENT_LUFID |
5635                                                         LDISKFS_FT_REG_FILE;
5636                                 else
5637                                         de->file_type = LDISKFS_DIRENT_LUFID |
5638                                                         LDISKFS_FT_DIR;
5639                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5640                                                               NULL, bh);
5641                         }
5642                         brelse(bh);
5643                 }
5644         }
5645
5646         RETURN(rc);
5647 }
5648
5649 /**
5650  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5651  * into the directory.Also sets flags into osd object to
5652  * indicate dot and dotdot are created. This is required for
5653  * interoperability mode (b11826)
5654  *
5655  * \param dir   directory for dot and dotdot fixup.
5656  * \param obj   child object for linking
5657  *
5658  * \retval   0, on success
5659  * \retval -ve, on error
5660  */
5661 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5662                               struct osd_object *dir,
5663                               struct inode *parent_dir, const char *name,
5664                               const struct lu_fid *dot_fid,
5665                               const struct lu_fid *dot_dot_fid,
5666                               struct thandle *th)
5667 {
5668         struct inode *inode = dir->oo_inode;
5669         struct osd_thandle *oth;
5670         int result = 0;
5671
5672         oth = container_of(th, struct osd_thandle, ot_super);
5673         LASSERT(oth->ot_handle->h_transaction != NULL);
5674         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5675
5676         if (strcmp(name, dot) == 0) {
5677                 if (dir->oo_compat_dot_created) {
5678                         result = -EEXIST;
5679                 } else {
5680                         LASSERT(inode->i_ino == parent_dir->i_ino);
5681                         dir->oo_compat_dot_created = 1;
5682                         result = 0;
5683                 }
5684         } else if (strcmp(name, dotdot) == 0) {
5685                 if (!dir->oo_compat_dot_created)
5686                         return -EINVAL;
5687                 /* in case of rename, dotdot is already created */
5688                 if (dir->oo_compat_dotdot_created) {
5689                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5690                                                 dot_dot_fid, NULL, th);
5691                 }
5692
5693                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5694                         struct lu_fid tfid = *dot_dot_fid;
5695
5696                         tfid.f_oid--;
5697                         result = osd_add_dot_dotdot_internal(info,
5698                                         dir->oo_inode, parent_dir, dot_fid,
5699                                         &tfid, oth);
5700                 } else {
5701                         result = osd_add_dot_dotdot_internal(info,
5702                                         dir->oo_inode, parent_dir, dot_fid,
5703                                         dot_dot_fid, oth);
5704                 }
5705
5706                 if (result == 0)
5707                         dir->oo_compat_dotdot_created = 1;
5708         }
5709
5710         return result;
5711 }
5712
5713
5714 /**
5715  * It will call the appropriate osd_add* function and return the
5716  * value, return by respective functions.
5717  */
5718 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5719                           struct inode *cinode, const char *name,
5720                           const struct lu_fid *fid, struct thandle *th)
5721 {
5722         struct osd_thread_info *info = osd_oti_get(env);
5723         struct htree_lock *hlock;
5724         int rc;
5725
5726         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5727
5728         if (name[0] == '.' && (name[1] == '\0' ||
5729                                (name[1] == '.' && name[2] == '\0'))) {
5730                 if (hlock != NULL) {
5731                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5732                                            pobj->oo_inode, 0);
5733                 } else {
5734                         down_write(&pobj->oo_ext_idx_sem);
5735                 }
5736
5737                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5738                                         lu_object_fid(&pobj->oo_dt.do_lu),
5739                                         fid, th);
5740         } else {
5741                 if (hlock != NULL) {
5742                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5743                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5744                 } else {
5745                         down_write(&pobj->oo_ext_idx_sem);
5746                 }
5747
5748                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5749                         struct lu_fid *tfid = &info->oti_fid;
5750
5751                         *tfid = *fid;
5752                         tfid->f_ver = ~0;
5753                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5754                                               tfid, hlock, th);
5755                 } else {
5756                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5757                                               hlock, th);
5758                 }
5759         }
5760         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5761             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5762                 pobj->oo_dirent_count++;
5763
5764         if (hlock != NULL)
5765                 ldiskfs_htree_unlock(hlock);
5766         else
5767                 up_write(&pobj->oo_ext_idx_sem);
5768
5769         return rc;
5770 }
5771
5772 static int
5773 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5774                       struct osd_idmap_cache *oic)
5775 {
5776         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5777         struct lu_fid *fid = &oic->oic_fid;
5778         struct osd_inode_id *id = &oic->oic_lid;
5779         struct inode *inode = NULL;
5780         int once = 0;
5781         bool insert;
5782         int rc;
5783
5784         ENTRY;
5785
5786         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5787                 RETURN(0);
5788
5789         if (scrub->os_running && scrub->os_pos_current > id->oii_ino)
5790                 RETURN(0);
5791
5792         if (dev->od_auto_scrub_interval == AS_NEVER ||
5793             ktime_get_real_seconds() <
5794             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5795                 RETURN(0);
5796
5797 again:
5798         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5799         if (rc == -ENOENT) {
5800                 __u32 gen = id->oii_gen;
5801
5802                 insert = true;
5803                 if (inode != NULL)
5804                         goto trigger;
5805
5806                 inode = osd_iget(oti, dev, id);
5807                 /* The inode has been removed (by race maybe). */
5808                 if (IS_ERR(inode)) {
5809                         rc = PTR_ERR(inode);
5810
5811                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5812                 }
5813
5814                 /* The OI mapping is lost. */
5815                 if (gen != OSD_OII_NOGEN)
5816                         goto trigger;
5817
5818                 /*
5819                  * The inode may has been reused by others, we do not know,
5820                  * leave it to be handled by subsequent osd_fid_lookup().
5821                  */
5822                 GOTO(out, rc = 0);
5823         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5824                 GOTO(out, rc);
5825         }
5826
5827         insert = false;
5828
5829 trigger:
5830         if (scrub->os_running) {
5831                 if (inode == NULL) {
5832                         inode = osd_iget(oti, dev, id);
5833                         /* The inode has been removed (by race maybe). */
5834                         if (IS_ERR(inode)) {
5835                                 rc = PTR_ERR(inode);
5836
5837                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5838                         }
5839                 }
5840
5841                 rc = osd_oii_insert(dev, oic, insert);
5842                 /*
5843                  * There is race condition between osd_oi_lookup and OI scrub.
5844                  * The OI scrub finished just after osd_oi_lookup() failure.
5845                  * Under such case, it is unnecessary to trigger OI scrub again,
5846                  * but try to call osd_oi_lookup() again.
5847                  */
5848                 if (unlikely(rc == -EAGAIN))
5849                         goto again;
5850
5851                 if (!S_ISDIR(inode->i_mode))
5852                         rc = 0;
5853                 else
5854                         rc = osd_check_lmv(oti, dev, inode, oic);
5855
5856                 GOTO(out, rc);
5857         }
5858
5859         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5860                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5861                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5862                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5863                        "%s: trigger partial OI scrub for RPC inconsistency checking FID "DFID": rc = %d\n",
5864                        osd_dev2name(dev), PFID(fid), rc);
5865                 if (rc == 0 || rc == -EALREADY)
5866                         goto again;
5867         }
5868
5869         GOTO(out, rc);
5870
5871 out:
5872         iput(inode);
5873
5874         RETURN(rc);
5875 }
5876
5877 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5878                                struct osd_device *dev,
5879                                struct osd_idmap_cache *oic,
5880                                struct lu_fid *fid, __u32 ino)
5881 {
5882         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5883         struct inode *inode;
5884         int rc;
5885
5886         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5887         inode = osd_iget(oti, dev, &oic->oic_lid);
5888         if (IS_ERR(inode)) {
5889                 fid_zero(&oic->oic_fid);
5890                 return PTR_ERR(inode);
5891         }
5892
5893         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5894         iput(inode);
5895         if (rc != 0)
5896                 fid_zero(&oic->oic_fid);
5897         else
5898                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5899         return rc;
5900 }
5901
5902 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5903                       struct osd_inode_id *id, const struct lu_fid *fid)
5904 {
5905         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5906                id->oii_ino, id->oii_gen, info);
5907         info->oti_cache.oic_lid = *id;
5908         info->oti_cache.oic_fid = *fid;
5909         info->oti_cache.oic_dev = osd;
5910 }
5911
5912 /**
5913  * Get parent FID from the linkEA.
5914  *
5915  * For a directory which parent resides on remote MDT, to satisfy the
5916  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5917  * the other hand, to make the lookup(..) on the directory can return
5918  * the real parent FID, we append the real parent FID after its ".."
5919  * name entry in the /REMOTE_PARENT_DIR.
5920  *
5921  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5922  * backup. So after the restore, we cannot get the right parent FID from
5923  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5924  * we have stored the real parent FID in the directory object's linkEA,
5925  * we can parse the linkEA for the real parent FID.
5926  *
5927  * \param[in] env       pointer to the thread context
5928  * \param[in] obj       pointer to the object to be handled
5929  * \param[out]fid       pointer to the buffer to hold the parent FID
5930  *
5931  * \retval              0 for getting the real parent FID successfully
5932  * \retval              negative error number on failure
5933  */
5934 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5935                                     struct osd_object *obj,
5936                                     struct lu_fid *fid)
5937 {
5938         struct osd_thread_info *oti = osd_oti_get(env);
5939         struct lu_buf *buf = &oti->oti_big_buf;
5940         struct dentry *dentry = &oti->oti_obj_dentry;
5941         struct inode *inode = obj->oo_inode;
5942         struct linkea_data ldata = { NULL };
5943         int rc;
5944
5945         ENTRY;
5946
5947         fid_zero(fid);
5948         if (!S_ISDIR(inode->i_mode))
5949                 RETURN(-EIO);
5950
5951 again:
5952         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5953                              buf->lb_buf, buf->lb_len);
5954         if (rc == -ERANGE) {
5955                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5956                                      NULL, 0);
5957                 if (rc > 0) {
5958                         lu_buf_realloc(buf, rc);
5959                         if (buf->lb_buf == NULL)
5960                                 RETURN(-ENOMEM);
5961
5962                         goto again;
5963                 }
5964         }
5965
5966         if (unlikely(rc == 0))
5967                 RETURN(-ENODATA);
5968
5969         if (rc < 0)
5970                 RETURN(rc);
5971
5972         if (unlikely(buf->lb_buf == NULL)) {
5973                 lu_buf_realloc(buf, rc);
5974                 if (buf->lb_buf == NULL)
5975                         RETURN(-ENOMEM);
5976
5977                 goto again;
5978         }
5979
5980         ldata.ld_buf = buf;
5981         rc = linkea_init_with_rec(&ldata);
5982         if (!rc) {
5983                 linkea_first_entry(&ldata);
5984                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5985         }
5986
5987         RETURN(rc);
5988 }
5989
5990 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5991                                     struct inode *inode,
5992                                     const struct lu_fid *pfid,
5993                                     const char *name, const int namelen)
5994 {
5995         struct osd_thread_info *oti = osd_oti_get(env);
5996         struct lu_buf *buf = &oti->oti_big_buf;
5997         struct dentry *dentry = &oti->oti_obj_dentry;
5998         struct linkea_data ldata = { NULL };
5999         struct lu_name cname = { .ln_name = name,
6000                                  .ln_namelen = namelen };
6001         int rc;
6002
6003         ENTRY;
6004
6005 again:
6006         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6007                              buf->lb_buf, buf->lb_len);
6008         if (rc == -ERANGE)
6009                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
6010
6011         if (rc < 0)
6012                 RETURN(rc);
6013
6014         if (unlikely(rc == 0))
6015                 RETURN(-ENODATA);
6016
6017         if (buf->lb_len < rc) {
6018                 lu_buf_realloc(buf, rc);
6019                 if (buf->lb_buf == NULL)
6020                         RETURN(-ENOMEM);
6021
6022                 goto again;
6023         }
6024
6025         ldata.ld_buf = buf;
6026         rc = linkea_init_with_rec(&ldata);
6027         if (!rc)
6028                 rc = linkea_links_find(&ldata, &cname, pfid);
6029
6030         RETURN(rc);
6031 }
6032
6033 /**
6034  * Calls ->lookup() to find dentry. From dentry get inode and
6035  * read inode's ea to get fid. This is required for  interoperability
6036  * mode (b11826)
6037  *
6038  * \retval   0, on success
6039  * \retval -ve, on error
6040  */
6041 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
6042                              struct dt_rec *rec, const struct dt_key *key)
6043 {
6044         struct inode *dir = obj->oo_inode;
6045         struct dentry *dentry;
6046         struct ldiskfs_dir_entry_2 *de;
6047         struct buffer_head *bh;
6048         struct lu_fid *fid = (struct lu_fid *)rec;
6049         struct htree_lock *hlock = NULL;
6050         int ino;
6051         int rc;
6052
6053         ENTRY;
6054
6055         LASSERT(dir->i_op != NULL);
6056         LASSERT(dir->i_op->lookup != NULL);
6057
6058         dentry = osd_child_dentry_get(env, obj,
6059                                       (char *)key, strlen((char *)key));
6060
6061         if (obj->oo_hl_head != NULL) {
6062                 hlock = osd_oti_get(env)->oti_hlock;
6063                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6064                                    dir, LDISKFS_HLOCK_LOOKUP);
6065         } else {
6066                 down_read(&obj->oo_ext_idx_sem);
6067         }
6068
6069         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6070         if (!IS_ERR(bh)) {
6071                 struct osd_thread_info *oti = osd_oti_get(env);
6072                 struct osd_inode_id *id = &oti->oti_id;
6073                 struct osd_idmap_cache *oic = &oti->oti_cache;
6074                 struct osd_device *dev = osd_obj2dev(obj);
6075
6076                 ino = le32_to_cpu(de->inode);
6077                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
6078                         brelse(bh);
6079                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
6080                         GOTO(out, rc);
6081                 }
6082
6083                 rc = osd_get_fid_from_dentry(de, rec);
6084
6085                 /* done with de, release bh */
6086                 brelse(bh);
6087                 if (rc != 0) {
6088                         if (unlikely(is_remote_parent_ino(dev, ino))) {
6089                                 const char *name = (const char *)key;
6090
6091                                 /*
6092                                  * If the parent is on remote MDT, and there
6093                                  * is no FID-in-dirent, then we have to get
6094                                  * the parent FID from the linkEA.
6095                                  */
6096                                 if (likely(strlen(name) == 2 &&
6097                                            name[0] == '.' && name[1] == '.'))
6098                                         rc = osd_get_pfid_from_linkea(env, obj,
6099                                                                       fid);
6100                         } else {
6101                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6102                         }
6103                 } else {
6104                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6105                 }
6106
6107                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
6108                         fid_zero(&oic->oic_fid);
6109
6110                         GOTO(out, rc);
6111                 }
6112
6113                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
6114                 rc = osd_consistency_check(oti, dev, oic);
6115                 if (rc == -ENOENT)
6116                         fid_zero(&oic->oic_fid);
6117                 else
6118                         /* Other error should not affect lookup result. */
6119                         rc = 0;
6120         } else {
6121                 rc = PTR_ERR(bh);
6122         }
6123
6124         GOTO(out, rc);
6125
6126 out:
6127         if (hlock != NULL)
6128                 ldiskfs_htree_unlock(hlock);
6129         else
6130                 up_read(&obj->oo_ext_idx_sem);
6131         return rc;
6132 }
6133
6134 static int osd_index_declare_ea_insert(const struct lu_env *env,
6135                                        struct dt_object *dt,
6136                                        const struct dt_rec *rec,
6137                                        const struct dt_key *key,
6138                                        struct thandle *handle)
6139 {
6140         struct osd_thandle *oh;
6141         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6142         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6143         const struct lu_fid *fid = rec1->rec_fid;
6144         int credits, rc = 0;
6145         struct osd_idmap_cache *idc;
6146
6147         ENTRY;
6148
6149         LASSERT(!dt_object_remote(dt));
6150         LASSERT(handle != NULL);
6151         LASSERT(fid != NULL);
6152         LASSERT(rec1->rec_type != 0);
6153
6154         oh = container_of(handle, struct osd_thandle, ot_super);
6155         LASSERT(oh->ot_handle == NULL);
6156
6157         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6158
6159         /*
6160          * we can't call iget() while a transactions is running
6161          * (this can lead to a deadlock), but we need to know
6162          * inum and object type. so we find this information at
6163          * declaration and cache in per-thread info
6164          */
6165         idc = osd_idc_find_or_init(env, osd, fid);
6166         if (IS_ERR(idc))
6167                 RETURN(PTR_ERR(idc));
6168         if (idc->oic_remote) {
6169                 /*
6170                  * a reference to remote inode is represented by an
6171                  * agent inode which we have to create
6172                  */
6173                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6174                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6175         }
6176
6177         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6178
6179         if (osd_dt_obj(dt)->oo_inode != NULL) {
6180                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6181
6182                 /*
6183                  * We ignore block quota on meta pool (MDTs), so needn't
6184                  * calculate how many blocks will be consumed by this index
6185                  * insert
6186                  */
6187                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6188                                            i_gid_read(inode),
6189                                            i_projid_read(inode), 0,
6190                                            oh, osd_dt_obj(dt), NULL,
6191                                            OSD_QID_BLK);
6192                 if (rc)
6193                         RETURN(rc);
6194
6195 #ifdef HAVE_PROJECT_QUOTA
6196                 /*
6197                  * Reserve credits for local agent inode to transfer
6198                  * to 0, quota enforcement is ignored in this case.
6199                  */
6200                 if (idc->oic_remote &&
6201                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6202                     i_projid_read(inode) != 0)
6203                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6204                                                   0, i_projid_read(inode),
6205                                                   0, false, PRJQUOTA, true);
6206 #endif
6207         }
6208
6209         RETURN(rc);
6210 }
6211
6212 /**
6213  * Index add function for interoperability mode (b11826).
6214  * It will add the directory entry.This entry is needed to
6215  * maintain name->fid mapping.
6216  *
6217  * \param key it is key i.e. file entry to be inserted
6218  * \param rec it is value of given key i.e. fid
6219  *
6220  * \retval   0, on success
6221  * \retval -ve, on error
6222  */
6223 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6224                                const struct dt_rec *rec,
6225                                const struct dt_key *key, struct thandle *th)
6226 {
6227         struct osd_object *obj = osd_dt_obj(dt);
6228         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6229         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6230         const struct lu_fid *fid = rec1->rec_fid;
6231         const char *name = (const char *)key;
6232         struct osd_thread_info *oti = osd_oti_get(env);
6233         struct inode *child_inode = NULL;
6234         struct osd_idmap_cache *idc;
6235         int rc;
6236
6237         ENTRY;
6238
6239         if (!dt_object_exists(dt))
6240                 RETURN(-ENOENT);
6241
6242         LASSERT(osd_invariant(obj));
6243         LASSERT(!dt_object_remote(dt));
6244         LASSERT(th != NULL);
6245
6246         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6247
6248         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6249
6250         idc = osd_idc_find(env, osd, fid);
6251         if (unlikely(idc == NULL)) {
6252                 idc = osd_idc_find_or_init(env, osd, fid);
6253                 if (IS_ERR(idc)) {
6254                         /*
6255                          * this dt_insert() wasn't declared properly, so
6256                          * FID is missing in OI cache. we better do not
6257                          * lookup FID in FLDB/OI and don't risk to deadlock,
6258                          * but in some special cases (lfsck testing, etc)
6259                          * it's much simpler than fixing a caller.
6260                          *
6261                          * normally this error should be placed after the first
6262                          * find, but migrate may attach source stripes to
6263                          * target, which doesn't create stripes.
6264                          */
6265                         CERROR("%s: "DFID" wasn't declared for insert\n",
6266                                osd_name(osd), PFID(fid));
6267                         dump_stack();
6268                         RETURN(PTR_ERR(idc));
6269                 }
6270         }
6271
6272         if (idc->oic_remote) {
6273                 /* Insert remote entry */
6274                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6275                         child_inode =
6276                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6277                 } else {
6278                         child_inode = osd_create_local_agent_inode(env, osd,
6279                                         obj, fid, rec1->rec_type & S_IFMT, th);
6280                         if (IS_ERR(child_inode))
6281                                 RETURN(PTR_ERR(child_inode));
6282                 }
6283         } else {
6284                 /* Insert local entry */
6285                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6286                         /* for a reason OI cache wasn't filled properly */
6287                         CERROR("%s: OIC for "DFID" isn't filled\n",
6288                                osd_name(osd), PFID(fid));
6289                         RETURN(-EINVAL);
6290                 }
6291                 child_inode = oti->oti_inode;
6292                 if (unlikely(child_inode == NULL)) {
6293                         struct ldiskfs_inode_info *lii;
6294
6295                         OBD_ALLOC_PTR(lii);
6296                         if (lii == NULL)
6297                                 RETURN(-ENOMEM);
6298                         child_inode = oti->oti_inode = &lii->vfs_inode;
6299                 }
6300                 child_inode->i_sb = osd_sb(osd);
6301                 child_inode->i_ino = idc->oic_lid.oii_ino;
6302                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6303         }
6304
6305         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6306
6307         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6308                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6309
6310         if (child_inode && child_inode != oti->oti_inode)
6311                 iput(child_inode);
6312         LASSERT(osd_invariant(obj));
6313         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6314
6315         RETURN(rc);
6316 }
6317
6318 /**
6319  *  Initialize osd Iterator for given osd index object.
6320  *
6321  *  \param  dt      osd index object
6322  */
6323
6324 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6325                                      struct dt_object *dt,
6326                                      __u32 unused)
6327 {
6328         struct osd_it_iam *it;
6329         struct osd_object *obj = osd_dt_obj(dt);
6330         struct lu_object *lo = &dt->do_lu;
6331         struct iam_path_descr *ipd;
6332         struct iam_container *bag = &obj->oo_dir->od_container;
6333
6334         if (!dt_object_exists(dt))
6335                 return ERR_PTR(-ENOENT);
6336
6337         OBD_ALLOC_PTR(it);
6338         if (it == NULL)
6339                 return ERR_PTR(-ENOMEM);
6340
6341         ipd = osd_it_ipd_get(env, bag);
6342         if (likely(ipd != NULL)) {
6343                 it->oi_obj = obj;
6344                 it->oi_ipd = ipd;
6345                 lu_object_get(lo);
6346                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6347                 return (struct dt_it *)it;
6348         } else {
6349                 OBD_FREE_PTR(it);
6350                 return ERR_PTR(-ENOMEM);
6351         }
6352 }
6353
6354 /**
6355  * free given Iterator.
6356  */
6357 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6358 {
6359         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6360         struct osd_object *obj = it->oi_obj;
6361
6362         iam_it_fini(&it->oi_it);
6363         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6364         osd_object_put(env, obj);
6365         OBD_FREE_PTR(it);
6366 }
6367
6368 /**
6369  *  Move Iterator to record specified by \a key
6370  *
6371  *  \param  di      osd iterator
6372  *  \param  key     key for index
6373  *
6374  *  \retval +ve  di points to record with least key not larger than key
6375  *  \retval  0   di points to exact matched key
6376  *  \retval -ve  failure
6377  */
6378
6379 static int osd_it_iam_get(const struct lu_env *env,
6380                           struct dt_it *di, const struct dt_key *key)
6381 {
6382         struct osd_thread_info *oti = osd_oti_get(env);
6383         struct osd_it_iam *it = (struct osd_it_iam *)di;
6384
6385         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6386                 /* swab quota uid/gid */
6387                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6388                 key = (struct dt_key *)&oti->oti_quota_id;
6389         }
6390
6391         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6392 }
6393
6394 /**
6395  *  Release Iterator
6396  *
6397  *  \param  di      osd iterator
6398  */
6399 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6400 {
6401         struct osd_it_iam *it = (struct osd_it_iam *)di;
6402
6403         iam_it_put(&it->oi_it);
6404 }
6405
6406 /**
6407  *  Move iterator by one record
6408  *
6409  *  \param  di      osd iterator
6410  *
6411  *  \retval +1   end of container reached
6412  *  \retval  0   success
6413  *  \retval -ve  failure
6414  */
6415
6416 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6417 {
6418         struct osd_it_iam *it = (struct osd_it_iam *)di;
6419
6420         return iam_it_next(&it->oi_it);
6421 }
6422
6423 /**
6424  * Return pointer to the key under iterator.
6425  */
6426
6427 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6428                                      const struct dt_it *di)
6429 {
6430         struct osd_thread_info *oti = osd_oti_get(env);
6431         struct osd_it_iam *it = (struct osd_it_iam *)di;
6432         struct osd_object *obj = it->oi_obj;
6433         struct dt_key *key;
6434
6435         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6436
6437         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6438                 /* swab quota uid/gid */
6439                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6440                 key = (struct dt_key *)&oti->oti_quota_id;
6441         }
6442
6443         return key;
6444 }
6445
6446 /**
6447  * Return size of key under iterator (in bytes)
6448  */
6449
6450 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6451 {
6452         struct osd_it_iam *it = (struct osd_it_iam *)di;
6453
6454         return iam_it_key_size(&it->oi_it);
6455 }
6456
6457 static inline void
6458 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6459 {
6460         /* check if file type is required */
6461         if (ent->lde_attrs & LUDA_TYPE) {
6462                 struct luda_type *lt;
6463                 int align = sizeof(*lt) - 1;
6464
6465                 len = (len + align) & ~align;
6466                 lt = (struct luda_type *)(ent->lde_name + len);
6467                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6468         }
6469
6470         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6471 }
6472
6473 /**
6474  * build lu direct from backend fs dirent.
6475  */
6476
6477 static inline void
6478 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6479                    char *name, __u16 namelen, __u16 type, __u32 attr)
6480 {
6481         ent->lde_attrs = attr | LUDA_FID;
6482         fid_cpu_to_le(&ent->lde_fid, fid);
6483
6484         ent->lde_hash = cpu_to_le64(offset);
6485         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6486
6487         strncpy(ent->lde_name, name, namelen);
6488         ent->lde_name[namelen] = '\0';
6489         ent->lde_namelen = cpu_to_le16(namelen);
6490
6491         /* append lustre attributes */
6492         osd_it_append_attrs(ent, namelen, type);
6493 }
6494
6495 /**
6496  * Return pointer to the record under iterator.
6497  */
6498 static int osd_it_iam_rec(const struct lu_env *env,
6499                           const struct dt_it *di,
6500                           struct dt_rec *dtrec, __u32 attr)
6501 {
6502         struct osd_it_iam *it = (struct osd_it_iam *)di;
6503         struct osd_thread_info *info = osd_oti_get(env);
6504
6505         ENTRY;
6506
6507         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6508                 const struct osd_fid_pack *rec;
6509                 struct lu_fid *fid = &info->oti_fid;
6510                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6511                 char *name;
6512                 int namelen;
6513                 __u64 hash;
6514                 int rc;
6515
6516                 name = (char *)iam_it_key_get(&it->oi_it);
6517                 if (IS_ERR(name))
6518                         RETURN(PTR_ERR(name));
6519
6520                 namelen = iam_it_key_size(&it->oi_it);
6521
6522                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6523                 if (IS_ERR(rec))
6524                         RETURN(PTR_ERR(rec));
6525
6526                 rc = osd_fid_unpack(fid, rec);
6527                 if (rc)
6528                         RETURN(rc);
6529
6530                 hash = iam_it_store(&it->oi_it);
6531
6532                 /* IAM does not store object type in IAM index (dir) */
6533                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6534                                    0, LUDA_FID);
6535         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6536                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6537                            (struct iam_rec *)dtrec);
6538                 osd_quota_unpack(it->oi_obj, dtrec);
6539         } else {
6540                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6541                            (struct iam_rec *)dtrec);
6542         }
6543
6544         RETURN(0);
6545 }
6546
6547 /**
6548  * Returns cookie for current Iterator position.
6549  */
6550 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6551 {
6552         struct osd_it_iam *it = (struct osd_it_iam *)di;
6553
6554         return iam_it_store(&it->oi_it);
6555 }
6556
6557 /**
6558  * Restore iterator from cookie.
6559  *
6560  * \param  di      osd iterator
6561  * \param  hash    Iterator location cookie
6562  *
6563  * \retval +ve  di points to record with least key not larger than key.
6564  * \retval  0   di points to exact matched key
6565  * \retval -ve  failure
6566  */
6567
6568 static int osd_it_iam_load(const struct lu_env *env,
6569                            const struct dt_it *di, __u64 hash)
6570 {
6571         struct osd_it_iam *it = (struct osd_it_iam *)di;
6572
6573         return iam_it_load(&it->oi_it, hash);
6574 }
6575
6576 static const struct dt_index_operations osd_index_iam_ops = {
6577         .dio_lookup         = osd_index_iam_lookup,
6578         .dio_declare_insert = osd_index_declare_iam_insert,
6579         .dio_insert         = osd_index_iam_insert,
6580         .dio_declare_delete = osd_index_declare_iam_delete,
6581         .dio_delete         = osd_index_iam_delete,
6582         .dio_it     = {
6583                 .init     = osd_it_iam_init,
6584                 .fini     = osd_it_iam_fini,
6585                 .get      = osd_it_iam_get,
6586                 .put      = osd_it_iam_put,
6587                 .next     = osd_it_iam_next,
6588                 .key      = osd_it_iam_key,
6589                 .key_size = osd_it_iam_key_size,
6590                 .rec      = osd_it_iam_rec,
6591                 .store    = osd_it_iam_store,
6592                 .load     = osd_it_iam_load
6593         }
6594 };
6595
6596
6597 /**
6598  * Creates or initializes iterator context.
6599  *
6600  * \retval struct osd_it_ea, iterator structure on success
6601  *
6602  */
6603 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6604                                     struct dt_object *dt,
6605                                     __u32 attr)
6606 {
6607         struct osd_object *obj = osd_dt_obj(dt);
6608         struct osd_thread_info *info = osd_oti_get(env);
6609         struct osd_it_ea *oie;
6610         struct file *file;
6611         struct lu_object *lo = &dt->do_lu;
6612         struct dentry *obj_dentry;
6613
6614         ENTRY;
6615
6616         if (!dt_object_exists(dt) || obj->oo_destroyed)
6617                 RETURN(ERR_PTR(-ENOENT));
6618
6619         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6620         if (oie == NULL)
6621                 RETURN(ERR_PTR(-ENOMEM));
6622         obj_dentry = &oie->oie_dentry;
6623
6624         obj_dentry->d_inode = obj->oo_inode;
6625         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6626         obj_dentry->d_name.hash = 0;
6627
6628         oie->oie_rd_dirent       = 0;
6629         oie->oie_it_dirent       = 0;
6630         oie->oie_dirent          = NULL;
6631         if (unlikely(!info->oti_it_ea_buf_used)) {
6632                 oie->oie_buf = info->oti_it_ea_buf;
6633                 info->oti_it_ea_buf_used = 1;
6634         } else {
6635                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6636                 if (oie->oie_buf == NULL)
6637                         RETURN(ERR_PTR(-ENOMEM));
6638         }
6639         oie->oie_obj = obj;
6640
6641         file = &oie->oie_file;
6642
6643         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6644         if (attr & LUDA_64BITHASH)
6645                 file->f_mode    = FMODE_64BITHASH;
6646         else
6647                 file->f_mode    = FMODE_32BITHASH;
6648         file->f_path.dentry     = obj_dentry;
6649         file->f_mapping         = obj->oo_inode->i_mapping;
6650         file->f_op              = obj->oo_inode->i_fop;
6651         file->f_inode = obj->oo_inode;
6652
6653         lu_object_get(lo);
6654         RETURN((struct dt_it *)oie);
6655 }
6656
6657 /**
6658  * Destroy or finishes iterator context.
6659  *
6660  * \param di iterator structure to be destroyed
6661  */
6662 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6663 {
6664         struct osd_thread_info *info = osd_oti_get(env);
6665         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6666         struct osd_object *obj = oie->oie_obj;
6667         struct inode *inode = obj->oo_inode;
6668
6669         ENTRY;
6670         oie->oie_file.f_op->release(inode, &oie->oie_file);
6671         osd_object_put(env, obj);
6672         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6673                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6674         else
6675                 info->oti_it_ea_buf_used = 0;
6676         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6677         EXIT;
6678 }
6679
6680 /**
6681  * It position the iterator at given key, so that next lookup continues from
6682  * that key Or it is similar to dio_it->load() but based on a key,
6683  * rather than file position.
6684  *
6685  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6686  * to the beginning.
6687  *
6688  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6689  */
6690 static int osd_it_ea_get(const struct lu_env *env,
6691                          struct dt_it *di, const struct dt_key *key)
6692 {
6693         struct osd_it_ea *it = (struct osd_it_ea *)di;
6694
6695         ENTRY;
6696         LASSERT(((const char *)key)[0] == '\0');
6697         it->oie_file.f_pos = 0;
6698         it->oie_rd_dirent = 0;
6699         it->oie_it_dirent = 0;
6700         it->oie_dirent = NULL;
6701
6702         RETURN(+1);
6703 }
6704
6705 /**
6706  * Does nothing
6707  */
6708 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6709 {
6710 }
6711
6712 struct osd_filldir_cbs {
6713         struct dir_context ctx;
6714         struct osd_it_ea  *it;
6715 };
6716 /**
6717  * It is called internally by ->iterate*(). It fills the
6718  * iterator's in-memory data structure with required
6719  * information i.e. name, namelen, rec_size etc.
6720  *
6721  * \param buf in which information to be filled in.
6722  * \param name name of the file in given dir
6723  *
6724  * \retval 0 on success
6725  * \retval 1 on buffer full
6726  */
6727 #ifdef HAVE_FILLDIR_USE_CTX
6728 static int osd_ldiskfs_filldir(struct dir_context *buf,
6729 #else
6730 static int osd_ldiskfs_filldir(void *buf,
6731 #endif
6732                                const char *name, int namelen,
6733                                loff_t offset, __u64 ino, unsigned int d_type)
6734 {
6735         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6736         struct osd_object *obj = it->oie_obj;
6737         struct osd_it_ea_dirent *ent = it->oie_dirent;
6738         struct lu_fid *fid = &ent->oied_fid;
6739         struct osd_fid_pack *rec;
6740
6741         ENTRY;
6742
6743 /* this should never happen */
6744         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6745                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6746                 RETURN(-EIO);
6747         }
6748
6749         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6750             OSD_IT_EA_BUFSIZE)
6751                 RETURN(1);
6752
6753         /* "." is just the object itself. */
6754         if (namelen == 1 && name[0] == '.') {
6755                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6756         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6757                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6758                 if (osd_fid_unpack(fid, rec) != 0)
6759                         fid_zero(fid);
6760         } else {
6761                 fid_zero(fid);
6762         }
6763         d_type &= ~LDISKFS_DIRENT_LUFID;
6764
6765         /* NOT export local root. */
6766         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6767                 ino = obj->oo_inode->i_ino;
6768                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6769         }
6770
6771         ent->oied_ino     = ino;
6772         ent->oied_off     = offset;
6773         ent->oied_namelen = namelen;
6774         ent->oied_type    = d_type;
6775
6776         memcpy(ent->oied_name, name, namelen);
6777
6778         it->oie_rd_dirent++;
6779         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6780         RETURN(0);
6781 }
6782
6783 /**
6784  * Calls ->iterate*() to load a directory entry at a time
6785  * and stored it in iterator's in-memory data structure.
6786  *
6787  * \param di iterator's in memory structure
6788  *
6789  * \retval   0 on success
6790  * \retval -ve on error
6791  * \retval +1 reach the end of entry
6792  */
6793 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6794                                const struct dt_it *di)
6795 {
6796         struct osd_it_ea *it = (struct osd_it_ea *)di;
6797         struct osd_object *obj = it->oie_obj;
6798         struct inode *inode = obj->oo_inode;
6799         struct htree_lock *hlock = NULL;
6800         struct file *filp = &it->oie_file;
6801         int rc = 0;
6802         struct osd_filldir_cbs buf = {
6803                 .ctx.actor = osd_ldiskfs_filldir,
6804                 .it = it
6805         };
6806
6807         ENTRY;
6808         it->oie_dirent = it->oie_buf;
6809         it->oie_rd_dirent = 0;
6810
6811         if (obj->oo_hl_head != NULL) {
6812                 hlock = osd_oti_get(env)->oti_hlock;
6813                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6814                                    inode, LDISKFS_HLOCK_READDIR);
6815         } else {
6816                 down_read(&obj->oo_ext_idx_sem);
6817         }
6818
6819         filp->f_cred = current_cred();
6820         rc = osd_security_file_alloc(filp);
6821         if (rc)
6822                 GOTO(unlock, rc);
6823
6824         filp->f_flags |= O_NOATIME;
6825         filp->f_mode |= FMODE_NONOTIFY;
6826         rc = iterate_dir(filp, &buf.ctx);
6827         if (rc)
6828                 GOTO(unlock, rc);
6829
6830         if (it->oie_rd_dirent == 0) {
6831                 /*
6832                  * If it does not get any dirent, it means it has been reached
6833                  * to the end of the dir
6834                  */
6835                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6836                 if (rc == 0)
6837                         rc = 1;
6838         } else {
6839                 it->oie_dirent = it->oie_buf;
6840                 it->oie_it_dirent = 1;
6841         }
6842 unlock:
6843         if (hlock != NULL)
6844                 ldiskfs_htree_unlock(hlock);
6845         else
6846                 up_read(&obj->oo_ext_idx_sem);
6847
6848         RETURN(rc);
6849 }
6850
6851 /**
6852  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6853  * to load a directory entry at a time and stored it in
6854  * iterator's in-memory data structure.
6855  *
6856  * \param di iterator's in memory structure
6857  *
6858  * \retval +ve iterator reached to end
6859  * \retval   0 iterator not reached to end
6860  * \retval -ve on error
6861  */
6862 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6863 {
6864         struct osd_it_ea *it = (struct osd_it_ea *)di;
6865         int rc;
6866
6867         ENTRY;
6868
6869         if (it->oie_it_dirent < it->oie_rd_dirent) {
6870                 it->oie_dirent =
6871                         (void *)it->oie_dirent +
6872                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6873                                        it->oie_dirent->oied_namelen);
6874                 it->oie_it_dirent++;
6875                 rc = 0;
6876         } else {
6877                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6878                         rc = 1;
6879                 else
6880                         rc = osd_ldiskfs_it_fill(env, di);
6881         }
6882
6883         RETURN(rc);
6884 }
6885
6886 /**
6887  * Returns the key at current position from iterator's in memory structure.
6888  *
6889  * \param di iterator's in memory structure
6890  *
6891  * \retval key i.e. struct dt_key on success
6892  */
6893 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6894                                     const struct dt_it *di)
6895 {
6896         struct osd_it_ea *it = (struct osd_it_ea *)di;
6897
6898         return (struct dt_key *)it->oie_dirent->oied_name;
6899 }
6900
6901 /**
6902  * Returns key's size at current position from iterator's in memory structure.
6903  *
6904  * \param di iterator's in memory structure
6905  *
6906  * \retval key_size i.e. struct dt_key on success
6907  */
6908 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6909 {
6910         struct osd_it_ea *it = (struct osd_it_ea *)di;
6911
6912         return it->oie_dirent->oied_namelen;
6913 }
6914
6915 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6916 #undef LDISKFS_DIR_REC_LEN
6917 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6918 #endif
6919
6920 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6921 {
6922         if (LDISKFS_DIR_REC_LEN(de) >=
6923             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6924                 return true;
6925
6926         return false;
6927 }
6928
6929 static inline bool
6930 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6931                      unsigned int blocksize, bool dotdot)
6932 {
6933         if (dotdot)
6934                 return osd_dotdot_has_space(de);
6935
6936         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6937             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6938                 return true;
6939
6940         return false;
6941 }
6942
6943 static int
6944 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6945                     handle_t *jh, struct dentry *dentry,
6946                     const struct lu_fid *fid, struct buffer_head *bh,
6947                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6948                     bool dotdot)
6949 {
6950         struct inode *dir = dentry->d_parent->d_inode;
6951         struct inode *inode = dentry->d_inode;
6952         struct osd_fid_pack *rec;
6953         struct ldiskfs_dentry_param *ldp;
6954         int namelen = dentry->d_name.len;
6955         int rc;
6956         struct osd_thread_info *info = osd_oti_get(env);
6957
6958         ENTRY;
6959
6960         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6961                 RETURN(0);
6962
6963         /* There is enough space to hold the FID-in-dirent. */
6964         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6965                 rc = ldiskfs_journal_get_write_access(jh, bh);
6966                 if (rc != 0)
6967                         RETURN(rc);
6968
6969                 de->name[namelen] = 0;
6970                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6971                 rec->fp_len = sizeof(struct lu_fid) + 1;
6972                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6973                 de->file_type |= LDISKFS_DIRENT_LUFID;
6974                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6975
6976                 RETURN(rc);
6977         }
6978
6979         LASSERT(!dotdot);
6980
6981         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6982         if (rc != 0)
6983                 RETURN(rc);
6984
6985         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6986         osd_get_ldiskfs_dirent_param(ldp, fid);
6987         dentry->d_fsdata = (void *)ldp;
6988         dquot_initialize(dir);
6989         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6990         /*
6991          * It is too bad, we cannot reinsert the name entry back.
6992          * That means we lose it!
6993          */
6994         if (rc != 0)
6995                 CDEBUG(D_LFSCK,
6996                        "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6997                        osd_ino2name(inode), dir->i_ino, dir->i_generation,
6998                        namelen, dentry->d_name.name, PFID(fid), rc);
6999
7000         RETURN(rc);
7001 }
7002
7003 static int
7004 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
7005                         struct osd_it_ea *it, struct lu_fid *fid,
7006                         struct osd_inode_id *id, __u32 *attr)
7007 {
7008         struct osd_thread_info *info = osd_oti_get(env);
7009         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
7010         struct osd_device *dev = osd_obj2dev(obj);
7011         struct super_block *sb = osd_sb(dev);
7012         const char *devname = osd_name(dev);
7013         struct osd_it_ea_dirent *ent = it->oie_dirent;
7014         struct inode *dir = obj->oo_inode;
7015         struct htree_lock *hlock = NULL;
7016         struct buffer_head *bh = NULL;
7017         handle_t *jh = NULL;
7018         struct ldiskfs_dir_entry_2 *de;
7019         struct dentry *dentry;
7020         struct inode *inode;
7021         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
7022         int credits;
7023         int rc;
7024         bool dotdot = false;
7025         bool dirty = false;
7026
7027         ENTRY;
7028
7029         if (ent->oied_name[0] == '.') {
7030                 if (ent->oied_namelen == 1)
7031                         RETURN(0);
7032
7033                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
7034                         dotdot = true;
7035         }
7036
7037         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
7038         inode = osd_iget(info, dev, id);
7039         if (IS_ERR(inode)) {
7040                 rc = PTR_ERR(inode);
7041                 if (rc == -ENOENT || rc == -ESTALE) {
7042                         /*
7043                          * Maybe dangling name entry, or
7044                          * corrupted directory entry.
7045                          */
7046                         *attr |= LUDA_UNKNOWN;
7047                         rc = 0;
7048                 } else {
7049                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
7050                                "check_repair, dir = %lu/%u, name = %.*s, "
7051                                "ino = %llu, rc = %d\n",
7052                                devname, dir->i_ino, dir->i_generation,
7053                                ent->oied_namelen, ent->oied_name,
7054                                ent->oied_ino, rc);
7055                 }
7056
7057                 RETURN(rc);
7058         }
7059
7060         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
7061                                            ent->oied_namelen);
7062         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
7063         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
7064                 lma = NULL;
7065         else if (rc != 0)
7066                 GOTO(out, rc);
7067
7068         /*
7069          * We need to ensure that the name entry is still valid.
7070          * Because it may be removed or renamed by other already.
7071          *
7072          * The unlink or rename operation will start journal before PDO lock,
7073          * so to avoid deadlock, here we need to start journal handle before
7074          * related PDO lock also. But because we do not know whether there
7075          * will be something to be repaired before PDO lock, we just start
7076          * journal without conditions.
7077          *
7078          * We may need to remove the name entry firstly, then insert back.
7079          * One credit is for user quota file update.
7080          * One credit is for group quota file update.
7081          * Two credits are for dirty inode.
7082          */
7083         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
7084                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
7085
7086         if (dev->od_dirent_journal != 0) {
7087
7088 again:
7089                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
7090                 if (IS_ERR(jh)) {
7091                         rc = PTR_ERR(jh);
7092                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
7093                                "check_repair, dir = %lu/%u, credits = %d, "
7094                                "name = %.*s, ino = %llu: rc = %d\n",
7095                                devname, dir->i_ino, dir->i_generation, credits,
7096                                ent->oied_namelen, ent->oied_name,
7097                                ent->oied_ino, rc);
7098
7099                         GOTO(out_inode, rc);
7100                 }
7101
7102                 if (obj->oo_hl_head != NULL) {
7103                         hlock = osd_oti_get(env)->oti_hlock;
7104                         /*
7105                          * "0" means exclusive lock for the whole directory.
7106                          * We need to prevent others access such name entry
7107                          * during the delete + insert. Neither HLOCK_ADD nor
7108                          * HLOCK_DEL cannot guarantee the atomicity.
7109                          */
7110                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7111                 } else {
7112                         down_write(&obj->oo_ext_idx_sem);
7113                 }
7114         } else {
7115                 if (obj->oo_hl_head != NULL) {
7116                         hlock = osd_oti_get(env)->oti_hlock;
7117                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7118                                            LDISKFS_HLOCK_LOOKUP);
7119                 } else {
7120                         down_read(&obj->oo_ext_idx_sem);
7121                 }
7122         }
7123
7124         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7125         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7126                 *attr |= LUDA_IGNORE;
7127
7128                 GOTO(out, rc = 0);
7129         }
7130
7131         /*
7132          * For dotdot entry, if there is not enough space to hold the
7133          * FID-in-dirent, just keep them there. It only happens when the
7134          * device upgraded from 1.8 or restored from MDT file-level backup.
7135          * For the whole directory, only dotdot entry have no FID-in-dirent
7136          * and needs to get FID from LMA when readdir, it will not affect the
7137          * performance much.
7138          */
7139         if (dotdot && !osd_dotdot_has_space(de)) {
7140                 *attr |= LUDA_UNKNOWN;
7141
7142                 GOTO(out, rc = 0);
7143         }
7144
7145         if (lma != NULL) {
7146                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7147                         GOTO(out, rc = 0);
7148
7149                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7150                         struct lu_fid *tfid = &lma->lma_self_fid;
7151
7152                         if (likely(dotdot &&
7153                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7154                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7155                                 /*
7156                                  * It must be REMOTE_PARENT_DIR and as the
7157                                  * 'dotdot' entry of remote directory
7158                                  */
7159                                 *attr |= LUDA_IGNORE;
7160                         } else {
7161                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7162                                        "parent directory, but got %.*s under "
7163                                        "dir = %lu/%u with the FID "DFID"\n",
7164                                        devname, ent->oied_namelen,
7165                                        ent->oied_name, dir->i_ino,
7166                                        dir->i_generation, PFID(tfid));
7167
7168                                 *attr |= LUDA_UNKNOWN;
7169                         }
7170
7171                         GOTO(out, rc = 0);
7172                 }
7173         }
7174
7175         if (!fid_is_zero(fid)) {
7176                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7177                                               ent->oied_namelen);
7178                 if (rc == -ENOENT ||
7179                     (rc == -ENODATA &&
7180                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7181                         /*
7182                          * linkEA does not recognize the dirent entry,
7183                          * it may because the dirent entry corruption
7184                          * and points to other's inode.
7185                          */
7186                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7187                                "recognize the dirent, dir = %lu/%u, "
7188                                " name = %.*s, ino = %llu, "
7189                                DFID": rc = %d\n", devname, dir->i_ino,
7190                                dir->i_generation, ent->oied_namelen,
7191                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7192                         *attr |= LUDA_UNKNOWN;
7193
7194                         GOTO(out, rc = 0);
7195                 }
7196
7197                 if (rc && rc != -ENODATA) {
7198                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7199                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7200                                DFID": rc = %d\n", devname, dir->i_ino,
7201                                dir->i_generation, ent->oied_namelen,
7202                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7203                         *attr |= LUDA_UNKNOWN;
7204
7205                         GOTO(out, rc = 0);
7206                 }
7207         }
7208
7209         if (lma != NULL) {
7210                 /*
7211                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7212                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7213                  */
7214                 if (*attr & LUDA_VERIFY_DRYRUN) {
7215                         *fid = lma->lma_self_fid;
7216                         *attr |= LUDA_REPAIR;
7217
7218                         GOTO(out, rc = 0);
7219                 }
7220
7221                 if (jh == NULL) {
7222                         brelse(bh);
7223                         dev->od_dirent_journal = 1;
7224                         if (hlock != NULL) {
7225                                 ldiskfs_htree_unlock(hlock);
7226                                 hlock = NULL;
7227                         } else {
7228                                 up_read(&obj->oo_ext_idx_sem);
7229                         }
7230
7231                         goto again;
7232                 }
7233
7234                 *fid = lma->lma_self_fid;
7235                 dirty = true;
7236                 /* Update or append the FID-in-dirent. */
7237                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7238                                          bh, de, hlock, dotdot);
7239                 if (rc == 0)
7240                         *attr |= LUDA_REPAIR;
7241                 else
7242                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7243                                "the dirent, dir = %lu/%u, name = %.*s, "
7244                                "ino = %llu, "DFID": rc = %d\n",
7245                                devname, dir->i_ino, dir->i_generation,
7246                                ent->oied_namelen, ent->oied_name,
7247                                ent->oied_ino, PFID(fid), rc);
7248         } else {
7249                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7250                 if (*attr & LUDA_VERIFY_DRYRUN) {
7251                         if (fid_is_sane(fid)) {
7252                                 *attr |= LUDA_REPAIR;
7253                         } else if (dev->od_index == 0) {
7254                                 lu_igif_build(fid, inode->i_ino,
7255                                               inode->i_generation);
7256                                 *attr |= LUDA_UPGRADE;
7257                         }
7258
7259                         GOTO(out, rc = 0);
7260                 }
7261
7262                 if (jh == NULL) {
7263                         brelse(bh);
7264                         dev->od_dirent_journal = 1;
7265                         if (hlock != NULL) {
7266                                 ldiskfs_htree_unlock(hlock);
7267                                 hlock = NULL;
7268                         } else {
7269                                 up_read(&obj->oo_ext_idx_sem);
7270                         }
7271
7272                         goto again;
7273                 }
7274
7275                 dirty = true;
7276                 if (unlikely(fid_is_sane(fid))) {
7277                         /*
7278                          * FID-in-dirent exists, but FID-in-LMA is lost.
7279                          * Trust the FID-in-dirent, and add FID-in-LMA.
7280                          */
7281                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7282                         if (rc == 0)
7283                                 *attr |= LUDA_REPAIR;
7284                         else
7285                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7286                                        "update dirent, dir = %lu/%u, "
7287                                        "name = %.*s, ino = %llu, "
7288                                        DFID": rc = %d\n",
7289                                        devname, dir->i_ino, dir->i_generation,
7290                                        ent->oied_namelen, ent->oied_name,
7291                                        ent->oied_ino, PFID(fid), rc);
7292                 } else if (dev->od_index == 0) {
7293                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7294                         /*
7295                          * It is probably IGIF object. Only aappend the
7296                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7297                          */
7298                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7299                                                  bh, de, hlock, dotdot);
7300                         if (rc == 0)
7301                                 *attr |= LUDA_UPGRADE;
7302                         else
7303                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7304                                        "after the dirent, dir = %lu/%u, "
7305                                        "name = %.*s, ino = %llu, "
7306                                        DFID": rc = %d\n",
7307                                        devname, dir->i_ino, dir->i_generation,
7308                                        ent->oied_namelen, ent->oied_name,
7309                                        ent->oied_ino, PFID(fid), rc);
7310                 }
7311         }
7312
7313         GOTO(out, rc);
7314
7315 out:
7316         if (!IS_ERR(bh))
7317                 brelse(bh);
7318         if (hlock != NULL) {
7319                 ldiskfs_htree_unlock(hlock);
7320         } else {
7321                 if (dev->od_dirent_journal != 0)
7322                         up_write(&obj->oo_ext_idx_sem);
7323                 else
7324                         up_read(&obj->oo_ext_idx_sem);
7325         }
7326
7327         if (jh != NULL)
7328                 ldiskfs_journal_stop(jh);
7329
7330 out_inode:
7331         iput(inode);
7332         if (rc >= 0 && !dirty)
7333                 dev->od_dirent_journal = 0;
7334
7335         return rc;
7336 }
7337
7338 /**
7339  * Returns the value at current position from iterator's in memory structure.
7340  *
7341  * \param di struct osd_it_ea, iterator's in memory structure
7342  * \param attr attr requested for dirent.
7343  * \param lde lustre dirent
7344  *
7345  * \retval   0 no error and \param lde has correct lustre dirent.
7346  * \retval -ve on error
7347  */
7348 static inline int osd_it_ea_rec(const struct lu_env *env,
7349                                 const struct dt_it *di,
7350                                 struct dt_rec *dtrec, __u32 attr)
7351 {
7352         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7353         struct osd_object      *obj   = it->oie_obj;
7354         struct osd_device      *dev   = osd_obj2dev(obj);
7355         struct osd_thread_info *oti   = osd_oti_get(env);
7356         struct osd_inode_id    *id    = &oti->oti_id;
7357         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7358         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7359         __u32 ino = it->oie_dirent->oied_ino;
7360         int rc = 0;
7361
7362         ENTRY;
7363
7364         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7365
7366         if (attr & LUDA_VERIFY) {
7367                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7368                         attr |= LUDA_IGNORE;
7369                         /*
7370                          * If the parent is on remote MDT, and there
7371                          * is no FID-in-dirent, then we have to get
7372                          * the parent FID from the linkEA.
7373                          */
7374                         if (!fid_is_sane(fid) &&
7375                             it->oie_dirent->oied_namelen == 2 &&
7376                             it->oie_dirent->oied_name[0] == '.' &&
7377                             it->oie_dirent->oied_name[1] == '.')
7378                                 osd_get_pfid_from_linkea(env, obj, fid);
7379                 } else {
7380                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7381                                                      &attr);
7382                 }
7383
7384                 if (!fid_is_sane(fid))
7385                         attr |= LUDA_UNKNOWN;
7386         } else {
7387                 attr &= ~LU_DIRENT_ATTRS_MASK;
7388                 if (!fid_is_sane(fid)) {
7389                         bool is_dotdot = false;
7390
7391                         if (it->oie_dirent->oied_namelen == 2 &&
7392                             it->oie_dirent->oied_name[0] == '.' &&
7393                             it->oie_dirent->oied_name[1] == '.')
7394                                 is_dotdot = true;
7395                         /*
7396                          * If the parent is on remote MDT, and there
7397                          * is no FID-in-dirent, then we have to get
7398                          * the parent FID from the linkEA.
7399                          */
7400                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7401                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7402                         } else {
7403                                 if (is_dotdot == false &&
7404                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7405                                         RETURN(-ENOENT);
7406
7407                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7408                         }
7409                 }
7410         }
7411
7412         /* Pack the entry anyway, at least the offset is right. */
7413         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7414                            it->oie_dirent->oied_name,
7415                            it->oie_dirent->oied_namelen,
7416                            it->oie_dirent->oied_type, attr);
7417
7418         RETURN(rc > 0 ? 0 : rc);
7419 }
7420
7421 /**
7422  * Returns the record size size at current position.
7423  *
7424  * This function will return record(lu_dirent) size in bytes.
7425  *
7426  * \param[in] env       execution environment
7427  * \param[in] di        iterator's in memory structure
7428  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7429  *                      calculate the lu_dirent size.
7430  *
7431  * \retval      record size(in bytes & in memory) of the current lu_dirent
7432  *              entry.
7433  */
7434 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7435                               __u32 attr)
7436 {
7437         struct osd_it_ea *it = (struct osd_it_ea *)di;
7438
7439         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7440 }
7441
7442 /**
7443  * Returns a cookie for current position of the iterator head, so that
7444  * user can use this cookie to load/start the iterator next time.
7445  *
7446  * \param di iterator's in memory structure
7447  *
7448  * \retval cookie for current position, on success
7449  */
7450 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7451 {
7452         struct osd_it_ea *it = (struct osd_it_ea *)di;
7453
7454         return it->oie_dirent->oied_off;
7455 }
7456
7457 /**
7458  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7459  * to load a directory entry at a time and stored it i inn,
7460  * in iterator's in-memory data structure.
7461  *
7462  * \param di struct osd_it_ea, iterator's in memory structure
7463  *
7464  * \retval +ve on success
7465  * \retval -ve on error
7466  */
7467 static int osd_it_ea_load(const struct lu_env *env,
7468                           const struct dt_it *di, __u64 hash)
7469 {
7470         struct osd_it_ea *it = (struct osd_it_ea *)di;
7471         int rc;
7472
7473         ENTRY;
7474         it->oie_file.f_pos = hash;
7475
7476         rc =  osd_ldiskfs_it_fill(env, di);
7477         if (rc > 0)
7478                 rc = -ENODATA;
7479
7480         if (rc == 0)
7481                 rc = 1;
7482
7483         RETURN(rc);
7484 }
7485
7486 /**
7487  * Index lookup function for interoperability mode (b11826).
7488  *
7489  * \param key,  key i.e. file name to be searched
7490  *
7491  * \retval +ve, on success
7492  * \retval -ve, on error
7493  */
7494 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7495                                struct dt_rec *rec, const struct dt_key *key)
7496 {
7497         struct osd_object *obj = osd_dt_obj(dt);
7498         int rc = 0;
7499
7500         ENTRY;
7501
7502         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7503         LINVRNT(osd_invariant(obj));
7504
7505         rc = osd_ea_lookup_rec(env, obj, rec, key);
7506         if (rc == 0)
7507                 rc = 1;
7508         RETURN(rc);
7509 }
7510
7511 /**
7512  * Index and Iterator operations for interoperability
7513  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7514  */
7515 static const struct dt_index_operations osd_index_ea_ops = {
7516         .dio_lookup         = osd_index_ea_lookup,
7517         .dio_declare_insert = osd_index_declare_ea_insert,
7518         .dio_insert         = osd_index_ea_insert,
7519         .dio_declare_delete = osd_index_declare_ea_delete,
7520         .dio_delete         = osd_index_ea_delete,
7521         .dio_it     = {
7522                 .init     = osd_it_ea_init,
7523                 .fini     = osd_it_ea_fini,
7524                 .get      = osd_it_ea_get,
7525                 .put      = osd_it_ea_put,
7526                 .next     = osd_it_ea_next,
7527                 .key      = osd_it_ea_key,
7528                 .key_size = osd_it_ea_key_size,
7529                 .rec      = osd_it_ea_rec,
7530                 .rec_size = osd_it_ea_rec_size,
7531                 .store    = osd_it_ea_store,
7532                 .load     = osd_it_ea_load
7533         }
7534 };
7535
7536 static void *osd_key_init(const struct lu_context *ctx,
7537                           struct lu_context_key *key)
7538 {
7539         struct osd_thread_info *info;
7540
7541         OBD_ALLOC_PTR(info);
7542         if (info == NULL)
7543                 return ERR_PTR(-ENOMEM);
7544
7545         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7546         if (info->oti_it_ea_buf == NULL)
7547                 goto out_free_info;
7548
7549         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7550
7551         info->oti_hlock = ldiskfs_htree_lock_alloc();
7552         if (info->oti_hlock == NULL)
7553                 goto out_free_ea;
7554
7555         return info;
7556
7557 out_free_ea:
7558         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7559 out_free_info:
7560         OBD_FREE_PTR(info);
7561         return ERR_PTR(-ENOMEM);
7562 }
7563
7564 static void osd_key_fini(const struct lu_context *ctx,
7565                          struct lu_context_key *key, void *data)
7566 {
7567         struct osd_thread_info *info = data;
7568         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7569         struct osd_idmap_cache *idc = info->oti_ins_cache;
7570
7571         if (info->oti_dio_pages) {
7572                 int i;
7573                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7574                         struct page *page = info->oti_dio_pages[i];
7575                         if (page) {
7576                                 LASSERT(PagePrivate2(page));
7577                                 LASSERT(PageLocked(page));
7578                                 ClearPagePrivate2(page);
7579                                 unlock_page(page);
7580                                 __free_page(page);
7581                         }
7582                 }
7583                 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7584                                          PTLRPC_MAX_BRW_PAGES);
7585         }
7586
7587         if (info->oti_inode != NULL)
7588                 OBD_FREE_PTR(lli);
7589         if (info->oti_hlock != NULL)
7590                 ldiskfs_htree_lock_free(info->oti_hlock);
7591         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7592         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7593         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7594         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7595         lu_buf_free(&info->oti_big_buf);
7596         if (idc != NULL) {
7597                 LASSERT(info->oti_ins_cache_size > 0);
7598                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7599                 info->oti_ins_cache = NULL;
7600                 info->oti_ins_cache_size = 0;
7601         }
7602         OBD_FREE_PTR(info);
7603 }
7604
7605 static void osd_key_exit(const struct lu_context *ctx,
7606                          struct lu_context_key *key, void *data)
7607 {
7608         struct osd_thread_info *info = data;
7609
7610         LASSERT(info->oti_r_locks == 0);
7611         LASSERT(info->oti_w_locks == 0);
7612         LASSERT(info->oti_txns    == 0);
7613 }
7614
7615 /* type constructor/destructor: osd_type_init, osd_type_fini */
7616 LU_TYPE_INIT_FINI(osd, &osd_key);
7617
7618 struct lu_context_key osd_key = {
7619         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7620         .lct_init = osd_key_init,
7621         .lct_fini = osd_key_fini,
7622         .lct_exit = osd_key_exit
7623 };
7624
7625
7626 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7627                            const char *name, struct lu_device *next)
7628 {
7629         struct osd_device *osd = osd_dev(d);
7630
7631         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7632             sizeof(osd->od_svname))
7633                 return -E2BIG;
7634         return osd_procfs_init(osd, name);
7635 }
7636
7637 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7638 {
7639         struct seq_server_site *ss = osd_seq_site(osd);
7640         int rc = 0;
7641
7642         ENTRY;
7643
7644         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7645                 RETURN(0);
7646
7647         if (unlikely(ss == NULL))
7648                 RETURN(-ENODEV);
7649
7650         OBD_ALLOC_PTR(osd->od_cl_seq);
7651         if (osd->od_cl_seq == NULL)
7652                 RETURN(-ENOMEM);
7653
7654         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7655                         osd->od_svname, ss->ss_server_seq);
7656
7657         if (ss->ss_node_id == 0) {
7658                 /*
7659                  * If the OSD on the sequence controller(MDT0), then allocate
7660                  * sequence here, otherwise allocate sequence after connected
7661                  * to MDT0 (see mdt_register_lwp_callback()).
7662                  */
7663                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7664                                    &osd->od_cl_seq->lcs_space, env);
7665         }
7666
7667         RETURN(rc);
7668 }
7669
7670 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7671 {
7672         if (osd->od_cl_seq == NULL)
7673                 return;
7674
7675         seq_client_fini(osd->od_cl_seq);
7676         OBD_FREE_PTR(osd->od_cl_seq);
7677         osd->od_cl_seq = NULL;
7678 }
7679
7680 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7681 {
7682         ENTRY;
7683
7684         /* shutdown quota slave instance associated with the device */
7685         if (o->od_quota_slave_md != NULL) {
7686                 struct qsd_instance *qsd = o->od_quota_slave_md;
7687
7688                 o->od_quota_slave_md = NULL;
7689                 qsd_fini(env, qsd);
7690         }
7691
7692         if (o->od_quota_slave_dt != NULL) {
7693                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7694
7695                 o->od_quota_slave_dt = NULL;
7696                 qsd_fini(env, qsd);
7697         }
7698
7699         osd_fid_fini(env, o);
7700         osd_scrub_cleanup(env, o);
7701
7702         RETURN(0);
7703 }
7704
7705 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7706 {
7707         ENTRY;
7708
7709         if (o->od_mnt != NULL) {
7710                 shrink_dcache_sb(osd_sb(o));
7711                 osd_sync(env, &o->od_dt_dev);
7712                 wait_event(o->od_commit_cb_done,
7713                           !atomic_read(&o->od_commit_cb_in_flight));
7714
7715                 mntput(o->od_mnt);
7716                 o->od_mnt = NULL;
7717         }
7718
7719         EXIT;
7720 }
7721
7722 static int osd_mount(const struct lu_env *env,
7723                      struct osd_device *o, struct lustre_cfg *cfg)
7724 {
7725         const char *name = lustre_cfg_string(cfg, 0);
7726         const char *dev = lustre_cfg_string(cfg, 1);
7727         const char *opts;
7728         unsigned long page, s_flags = 0, lmd_flags = 0;
7729         struct page *__page;
7730         struct file_system_type *type;
7731         char *options = NULL;
7732         const char *str;
7733         struct osd_thread_info *info = osd_oti_get(env);
7734         struct lu_fid *fid = &info->oti_fid;
7735         struct inode *inode;
7736         int rc = 0, force_over_1024tb = 0;
7737
7738         ENTRY;
7739
7740         if (o->od_mnt != NULL)
7741                 RETURN(0);
7742
7743         if (strlen(dev) >= sizeof(o->od_mntdev))
7744                 RETURN(-E2BIG);
7745         strcpy(o->od_mntdev, dev);
7746
7747         str = lustre_cfg_buf(cfg, 2);
7748         sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
7749
7750         opts = lustre_cfg_string(cfg, 3);
7751 #ifdef __BIG_ENDIAN
7752         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7753                 CERROR("%s: device %s extents feature is not guaranteed to "
7754                        "work on big-endian systems. Use \"bigendian_extents\" "
7755                        "mount option to override.\n", name, dev);
7756                 RETURN(-EINVAL);
7757         }
7758 #endif
7759 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7760         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7761                 CWARN("force_over_128tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7762         }
7763 #endif
7764 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7765         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7766                 CWARN("force_over_256tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7767         }
7768 #endif
7769 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7770         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7771                 CWARN("force_over_512tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7772         }
7773 #endif
7774
7775         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7776                 force_over_1024tb = 1;
7777
7778         __page = alloc_page(GFP_KERNEL);
7779         if (__page == NULL)
7780                 GOTO(out, rc = -ENOMEM);
7781         page = (unsigned long)page_address(__page);
7782         options = (char *)page;
7783         *options = '\0';
7784         if (opts != NULL) {
7785                 /* strip out the options for back compatiblity */
7786                 static const char * const sout[] = {
7787                         "mballoc",
7788                         "iopen",
7789                         "noiopen",
7790                         "iopen_nopriv",
7791                         "extents",
7792                         "noextents",
7793                         /* strip out option we processed in osd */
7794                         "bigendian_extents",
7795                         "force_over_128tb",
7796                         "force_over_256tb",
7797                         "force_over_512tb",
7798                         "force_over_1024tb",
7799                         NULL
7800                 };
7801                 strncat(options, opts, PAGE_SIZE);
7802                 for (rc = 0, str = options; sout[rc]; ) {
7803                         char *op = strstr(str, sout[rc]);
7804
7805                         if (op == NULL) {
7806                                 rc++;
7807                                 str = options;
7808                                 continue;
7809                         }
7810                         if (op == options || *(op - 1) == ',') {
7811                                 str = op + strlen(sout[rc]);
7812                                 if (*str == ',' || *str == '\0') {
7813                                         *str == ',' ? str++ : str;
7814                                         memmove(op, str, strlen(str) + 1);
7815                                 }
7816                         }
7817                         for (str = op; *str != ',' && *str != '\0'; str++)
7818                                 ;
7819                 }
7820         } else {
7821                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7822         }
7823
7824         /* Glom up mount options */
7825         if (*options != '\0')
7826                 strncat(options, ",", PAGE_SIZE);
7827         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7828
7829         type = get_fs_type("ldiskfs");
7830         if (!type) {
7831                 CERROR("%s: cannot find ldiskfs module\n", name);
7832                 GOTO(out, rc = -ENODEV);
7833         }
7834
7835         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7836         module_put(type->owner);
7837
7838         if (IS_ERR(o->od_mnt)) {
7839                 rc = PTR_ERR(o->od_mnt);
7840                 o->od_mnt = NULL;
7841                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7842                 GOTO(out, rc);
7843         }
7844
7845         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7846                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7847                                  force_over_1024tb == 0) {
7848                 CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
7849                        name, dev);
7850                 GOTO(out_mnt, rc = -EINVAL);
7851         }
7852
7853         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7854                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7855                               name);
7856
7857                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7858         }
7859
7860         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7861                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7862                 GOTO(out_mnt, rc = -EINVAL);
7863         }
7864
7865 #ifdef LDISKFS_MOUNT_DIRDATA
7866         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7867                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7868         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7869                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7870                       "enabling the dirdata feature. If you do not want to "
7871                       "downgrade to Lustre-1.x again, you can enable it via "
7872                       "'tune2fs -O dirdata device'\n", name, dev);
7873 #endif
7874         inode = osd_sb(o)->s_root->d_inode;
7875         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7876         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7877         if (rc != 0) {
7878                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7879                 GOTO(out_mnt, rc);
7880         }
7881
7882         if (lmd_flags & LMD_FLG_NOSCRUB)
7883                 o->od_auto_scrub_interval = AS_NEVER;
7884
7885         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7886                 /* do not use pagecache with flash-backed storage */
7887                 o->od_writethrough_cache = 0;
7888                 o->od_read_cache = 0;
7889         }
7890
7891         GOTO(out, rc = 0);
7892
7893 out_mnt:
7894         mntput(o->od_mnt);
7895         o->od_mnt = NULL;
7896
7897 out:
7898         if (__page)
7899                 __free_page(__page);
7900
7901         return rc;
7902 }
7903
7904 static struct lu_device *osd_device_fini(const struct lu_env *env,
7905                                          struct lu_device *d)
7906 {
7907         struct osd_device *o = osd_dev(d);
7908
7909         ENTRY;
7910
7911         osd_index_backup(env, o, false);
7912         osd_shutdown(env, o);
7913         osd_procfs_fini(o);
7914         if (o->od_oi_table != NULL)
7915                 osd_oi_fini(osd_oti_get(env), o);
7916         if (o->od_extent_bytes_percpu)
7917                 free_percpu(o->od_extent_bytes_percpu);
7918         osd_obj_map_fini(o);
7919         osd_umount(env, o);
7920
7921         RETURN(NULL);
7922 }
7923
7924 static int osd_device_init0(const struct lu_env *env,
7925                             struct osd_device *o,
7926                             struct lustre_cfg *cfg)
7927 {
7928         struct lu_device *l = osd2lu_dev(o);
7929         struct osd_thread_info *info;
7930         int rc;
7931         int cplen = 0;
7932
7933         /* if the module was re-loaded, env can loose its keys */
7934         rc = lu_env_refill((struct lu_env *)env);
7935         if (rc)
7936                 GOTO(out, rc);
7937         info = osd_oti_get(env);
7938         LASSERT(info);
7939
7940         l->ld_ops = &osd_lu_ops;
7941         o->od_dt_dev.dd_ops = &osd_dt_ops;
7942
7943         spin_lock_init(&o->od_osfs_lock);
7944         mutex_init(&o->od_otable_mutex);
7945         INIT_LIST_HEAD(&o->od_orphan_list);
7946         INIT_LIST_HEAD(&o->od_index_backup_list);
7947         INIT_LIST_HEAD(&o->od_index_restore_list);
7948         spin_lock_init(&o->od_lock);
7949         o->od_index_backup_policy = LIBP_NONE;
7950         o->od_t10_type = 0;
7951         init_waitqueue_head(&o->od_commit_cb_done);
7952
7953         o->od_read_cache = 1;
7954         o->od_writethrough_cache = 1;
7955         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7956         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7957         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7958         o->od_auto_scrub_interval = AS_DEFAULT;
7959         /* default fallocate to unwritten extents: LU-14326/LU-14333 */
7960         o->od_fallocate_zero_blocks = 0;
7961
7962         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7963                         sizeof(o->od_svname));
7964         if (cplen >= sizeof(o->od_svname)) {
7965                 rc = -E2BIG;
7966                 GOTO(out, rc);
7967         }
7968
7969         o->od_index_backup_stop = 0;
7970         o->od_index = -1; /* -1 means index is invalid */
7971         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7972         if (rc == LDD_F_SV_TYPE_OST)
7973                 o->od_is_ost = 1;
7974
7975         o->od_full_scrub_ratio = OFSR_DEFAULT;
7976         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7977         rc = osd_mount(env, o, cfg);
7978         if (rc != 0)
7979                 GOTO(out, rc);
7980
7981         /* Can only check block device after mount */
7982         o->od_nonrotational =
7983                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7984
7985         rc = osd_obj_map_init(env, o);
7986         if (rc != 0)
7987                 GOTO(out_mnt, rc);
7988
7989         rc = lu_site_init(&o->od_site, l);
7990         if (rc != 0)
7991                 GOTO(out_compat, rc);
7992         o->od_site.ls_bottom_dev = l;
7993
7994         rc = lu_site_init_finish(&o->od_site);
7995         if (rc != 0)
7996                 GOTO(out_site, rc);
7997
7998         INIT_LIST_HEAD(&o->od_ios_list);
7999         /* setup scrub, including OI files initialization */
8000         o->od_in_init = 1;
8001         rc = osd_scrub_setup(env, o);
8002         o->od_in_init = 0;
8003         if (rc < 0)
8004                 GOTO(out_site, rc);
8005
8006         rc = osd_procfs_init(o, o->od_svname);
8007         if (rc != 0) {
8008                 CERROR("%s: can't initialize procfs: rc = %d\n",
8009                        o->od_svname, rc);
8010                 GOTO(out_scrub, rc);
8011         }
8012
8013         LASSERT(l->ld_site->ls_linkage.next != NULL);
8014         LASSERT(l->ld_site->ls_linkage.prev != NULL);
8015
8016         /* initialize quota slave instance */
8017         /* currently it's no need to prepare qsd_instance_md for OST */
8018         if (!o->od_is_ost) {
8019                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
8020                                                 &o->od_dt_dev,
8021                                                 o->od_proc_entry, true);
8022                 if (IS_ERR(o->od_quota_slave_md)) {
8023                         rc = PTR_ERR(o->od_quota_slave_md);
8024                         o->od_quota_slave_md = NULL;
8025                         GOTO(out_procfs, rc);
8026                 }
8027         }
8028
8029         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
8030                                         o->od_proc_entry, false);
8031
8032         if (IS_ERR(o->od_quota_slave_dt)) {
8033                 if (o->od_quota_slave_md != NULL) {
8034                         qsd_fini(env, o->od_quota_slave_md);
8035                         o->od_quota_slave_md = NULL;
8036                 }
8037
8038                 rc = PTR_ERR(o->od_quota_slave_dt);
8039                 o->od_quota_slave_dt = NULL;
8040                 GOTO(out_procfs, rc);
8041         }
8042
8043         o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
8044         if (!o->od_extent_bytes_percpu) {
8045                 rc = -ENOMEM;
8046                 GOTO(out_procfs, rc);
8047         }
8048
8049         RETURN(0);
8050
8051 out_procfs:
8052         osd_procfs_fini(o);
8053 out_scrub:
8054         osd_scrub_cleanup(env, o);
8055 out_site:
8056         lu_site_fini(&o->od_site);
8057 out_compat:
8058         osd_obj_map_fini(o);
8059 out_mnt:
8060         osd_umount(env, o);
8061 out:
8062         return rc;
8063 }
8064
8065 static struct lu_device *osd_device_alloc(const struct lu_env *env,
8066                                           struct lu_device_type *t,
8067                                           struct lustre_cfg *cfg)
8068 {
8069         struct osd_device *o;
8070         int rc;
8071
8072         OBD_ALLOC_PTR(o);
8073         if (o == NULL)
8074                 return ERR_PTR(-ENOMEM);
8075
8076         rc = dt_device_init(&o->od_dt_dev, t);
8077         if (rc == 0) {
8078                 /*
8079                  * Because the ctx might be revived in dt_device_init,
8080                  * refill the env here
8081                  */
8082                 lu_env_refill((struct lu_env *)env);
8083                 rc = osd_device_init0(env, o, cfg);
8084                 if (rc)
8085                         dt_device_fini(&o->od_dt_dev);
8086         }
8087
8088         if (unlikely(rc != 0))
8089                 OBD_FREE_PTR(o);
8090
8091         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
8092 }
8093
8094 static struct lu_device *osd_device_free(const struct lu_env *env,
8095                                          struct lu_device *d)
8096 {
8097         struct osd_device *o = osd_dev(d);
8098
8099         ENTRY;
8100
8101         /* XXX: make osd top device in order to release reference */
8102         d->ld_site->ls_top_dev = d;
8103         lu_site_purge(env, d->ld_site, -1);
8104         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8105                       D_ERROR, lu_cdebug_printer);
8106         lu_site_fini(&o->od_site);
8107         dt_device_fini(&o->od_dt_dev);
8108         OBD_FREE_PTR(o);
8109         RETURN(NULL);
8110 }
8111
8112 static int osd_process_config(const struct lu_env *env,
8113                               struct lu_device *d, struct lustre_cfg *cfg)
8114 {
8115         struct osd_device *o = osd_dev(d);
8116         ssize_t count;
8117         int rc;
8118
8119         ENTRY;
8120
8121         switch (cfg->lcfg_command) {
8122         case LCFG_SETUP:
8123                 rc = osd_mount(env, o, cfg);
8124                 break;
8125         case LCFG_CLEANUP:
8126                 /*
8127                  * For the case LCFG_PRE_CLEANUP is not called in advance,
8128                  * that may happend if hit failure during mount process.
8129                  */
8130                 osd_index_backup(env, o, false);
8131                 lu_dev_del_linkage(d->ld_site, d);
8132                 rc = osd_shutdown(env, o);
8133                 break;
8134         case LCFG_PARAM:
8135                 LASSERT(&o->od_dt_dev);
8136                 count  = class_modify_config(cfg, PARAM_OSD,
8137                                              &o->od_dt_dev.dd_kobj);
8138                 if (count < 0)
8139                         count = class_modify_config(cfg, PARAM_OST,
8140                                                     &o->od_dt_dev.dd_kobj);
8141                 rc = count > 0 ? 0 : count;
8142                 break;
8143         case LCFG_PRE_CLEANUP:
8144                 osd_scrub_stop(o);
8145                 osd_index_backup(env, o,
8146                                  o->od_index_backup_policy != LIBP_NONE);
8147                 rc = 0;
8148                 break;
8149         default:
8150                 rc = -ENOSYS;
8151         }
8152
8153         RETURN(rc);
8154 }
8155
8156 static int osd_recovery_complete(const struct lu_env *env,
8157                                  struct lu_device *d)
8158 {
8159         struct osd_device *osd = osd_dev(d);
8160         int rc = 0;
8161
8162         ENTRY;
8163
8164         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8165                 RETURN(0);
8166
8167         /*
8168          * start qsd instance on recovery completion, this notifies the quota
8169          * slave code that we are about to process new requests now
8170          */
8171         rc = qsd_start(env, osd->od_quota_slave_dt);
8172         if (rc == 0 && osd->od_quota_slave_md != NULL)
8173                 rc = qsd_start(env, osd->od_quota_slave_md);
8174
8175         RETURN(rc);
8176 }
8177
8178 /*
8179  * we use exports to track all osd users
8180  */
8181 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8182                            struct obd_device *obd, struct obd_uuid *cluuid,
8183                            struct obd_connect_data *data, void *localdata)
8184 {
8185         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8186         struct lustre_handle conn;
8187         int rc;
8188
8189         ENTRY;
8190
8191         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8192
8193         rc = class_connect(&conn, obd, cluuid);
8194         if (rc)
8195                 RETURN(rc);
8196
8197         *exp = class_conn2export(&conn);
8198
8199         spin_lock(&osd->od_osfs_lock);
8200         osd->od_connects++;
8201         spin_unlock(&osd->od_osfs_lock);
8202
8203         RETURN(0);
8204 }
8205
8206 /*
8207  * once last export (we don't count self-export) disappeared
8208  * osd can be released
8209  */
8210 static int osd_obd_disconnect(struct obd_export *exp)
8211 {
8212         struct obd_device *obd = exp->exp_obd;
8213         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8214         int rc, release = 0;
8215
8216         ENTRY;
8217
8218         /* Only disconnect the underlying layers on the final disconnect. */
8219         spin_lock(&osd->od_osfs_lock);
8220         osd->od_connects--;
8221         if (osd->od_connects == 0)
8222                 release = 1;
8223         spin_unlock(&osd->od_osfs_lock);
8224
8225         rc = class_disconnect(exp); /* bz 9811 */
8226
8227         if (rc == 0 && release)
8228                 class_manual_cleanup(obd);
8229         RETURN(rc);
8230 }
8231
8232 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8233                        struct lu_device *dev)
8234 {
8235         struct osd_device *osd = osd_dev(dev);
8236         struct lr_server_data *lsd =
8237                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8238         int result = 0;
8239
8240         ENTRY;
8241
8242         if (osd->od_quota_slave_md != NULL) {
8243                 /* set up quota slave objects for inode */
8244                 result = qsd_prepare(env, osd->od_quota_slave_md);
8245                 if (result != 0)
8246                         RETURN(result);
8247         }
8248
8249         if (osd->od_quota_slave_dt != NULL) {
8250                 /* set up quota slave objects for block */
8251                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8252                 if (result != 0)
8253                         RETURN(result);
8254         }
8255
8256
8257         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8258 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8259                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8260                         osd->od_index_in_idif = 1;
8261                 } else {
8262                         osd->od_index_in_idif = 0;
8263                         result = osd_register_proc_index_in_idif(osd);
8264                         if (result != 0)
8265                                 RETURN(result);
8266                 }
8267 #else
8268                 osd->od_index_in_idif = 1;
8269 #endif
8270         }
8271
8272         result = osd_fid_init(env, osd);
8273
8274         RETURN(result);
8275 }
8276
8277 /**
8278  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8279  *
8280  * Allocate FID.
8281  *
8282  * see include/lu_object.h for the details.
8283  */
8284 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8285                          struct lu_fid *fid, struct lu_object *parent,
8286                          const struct lu_name *name)
8287 {
8288         struct osd_device *osd = osd_dev(d);
8289
8290         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8291 }
8292
8293 static const struct lu_object_operations osd_lu_obj_ops = {
8294         .loo_object_init      = osd_object_init,
8295         .loo_object_delete    = osd_object_delete,
8296         .loo_object_release   = osd_object_release,
8297         .loo_object_free      = osd_object_free,
8298         .loo_object_print     = osd_object_print,
8299         .loo_object_invariant = osd_object_invariant
8300 };
8301
8302 const struct lu_device_operations osd_lu_ops = {
8303         .ldo_object_alloc      = osd_object_alloc,
8304         .ldo_process_config    = osd_process_config,
8305         .ldo_recovery_complete = osd_recovery_complete,
8306         .ldo_prepare           = osd_prepare,
8307         .ldo_fid_alloc         = osd_fid_alloc,
8308 };
8309
8310 static const struct lu_device_type_operations osd_device_type_ops = {
8311         .ldto_init = osd_type_init,
8312         .ldto_fini = osd_type_fini,
8313
8314         .ldto_start = osd_type_start,
8315         .ldto_stop  = osd_type_stop,
8316
8317         .ldto_device_alloc = osd_device_alloc,
8318         .ldto_device_free  = osd_device_free,
8319
8320         .ldto_device_init = osd_device_init,
8321         .ldto_device_fini = osd_device_fini
8322 };
8323
8324 static struct lu_device_type osd_device_type = {
8325         .ldt_tags     = LU_DEVICE_DT,
8326         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8327         .ldt_ops      = &osd_device_type_ops,
8328         .ldt_ctx_tags = LCT_LOCAL,
8329 };
8330
8331 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8332 {
8333         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8334         struct super_block *sb = osd_sb(osd);
8335
8336         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8337 }
8338
8339 /*
8340  * lprocfs legacy support.
8341  */
8342 static const struct obd_ops osd_obd_device_ops = {
8343         .o_owner = THIS_MODULE,
8344         .o_connect      = osd_obd_connect,
8345         .o_disconnect   = osd_obd_disconnect,
8346         .o_health_check = osd_health_check,
8347 };
8348
8349 static ssize_t track_declares_assert_show(struct kobject *kobj,
8350                                    struct attribute *attr,
8351                                    char *buf)
8352 {
8353         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8354 }
8355
8356 static ssize_t track_declares_assert_store(struct kobject *kobj,
8357                                            struct attribute *attr,
8358                                            const char *buffer, size_t count)
8359 {
8360         bool track_declares_assert;
8361         int rc;
8362
8363         rc = kstrtobool(buffer, &track_declares_assert);
8364         if (rc)
8365                 return rc;
8366
8367         ldiskfs_track_declares_assert = track_declares_assert;
8368
8369         return count;
8370 }
8371 LUSTRE_RW_ATTR(track_declares_assert);
8372
8373 static int __init osd_init(void)
8374 {
8375         struct kobject *kobj;
8376         int rc;
8377
8378         BUILD_BUG_ON(BH_DXLock >=
8379                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8380 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8381         /* please, try to keep osd_thread_info smaller than a page */
8382         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8383 #endif
8384
8385         osd_oi_mod_init();
8386
8387         rc = lu_kmem_init(ldiskfs_caches);
8388         if (rc)
8389                 return rc;
8390
8391 #ifdef CONFIG_KALLSYMS
8392         priv_security_file_alloc =
8393                 (void *)kallsyms_lookup_name("security_file_alloc");
8394 #endif
8395
8396         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8397                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8398         if (rc) {
8399                 lu_kmem_fini(ldiskfs_caches);
8400                 return rc;
8401         }
8402
8403         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8404         if (kobj) {
8405                 rc = sysfs_create_file(kobj,
8406                                        &lustre_attr_track_declares_assert.attr);
8407                 kobject_put(kobj);
8408                 if (rc) {
8409                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8410                         rc = 0;
8411                 }
8412         }
8413         return rc;
8414 }
8415
8416 static void __exit osd_exit(void)
8417 {
8418         struct kobject *kobj;
8419
8420         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8421         if (kobj) {
8422                 sysfs_remove_file(kobj,
8423                                   &lustre_attr_track_declares_assert.attr);
8424                 kobject_put(kobj);
8425         }
8426         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8427         lu_kmem_fini(ldiskfs_caches);
8428 }
8429
8430 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8431 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8432 MODULE_VERSION(LUSTRE_VERSION_STRING);
8433 MODULE_LICENSE("GPL");
8434
8435 module_init(osd_init);
8436 module_exit(osd_exit);