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 T10PI type '%s'\n",
2433                                        d->od_svname, name);
2434                         }
2435
2436                 } else {
2437                         CERROR("%s: unsupported T10PI type '%s'\n",
2438                                d->od_svname, name);
2439                 }
2440         }
2441
2442         param->ddp_has_lseek_data_hole = true;
2443 }
2444
2445 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2446 {
2447         return osd_sb(osd_dt_dev(d));
2448 }
2449
2450 /*
2451  * Concurrency: shouldn't matter.
2452  */
2453 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2454 {
2455         int rc;
2456         struct super_block *s = osd_sb(osd_dt_dev(d));
2457         ENTRY;
2458
2459         down_read(&s->s_umount);
2460         rc = s->s_op->sync_fs(s, 1);
2461         up_read(&s->s_umount);
2462
2463         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2464                rc);
2465
2466         return rc;
2467 }
2468
2469 /**
2470  * Start commit for OSD device.
2471  *
2472  * An implementation of dt_commit_async method for OSD device.
2473  * Asychronously starts underlayng fs sync and thereby a transaction
2474  * commit.
2475  *
2476  * \param env environment
2477  * \param d dt device
2478  *
2479  * \see dt_device_operations
2480  */
2481 static int osd_commit_async(const struct lu_env *env,
2482                             struct dt_device *d)
2483 {
2484         struct super_block *s = osd_sb(osd_dt_dev(d));
2485         int rc;
2486
2487         ENTRY;
2488
2489         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2490         down_read(&s->s_umount);
2491         rc = s->s_op->sync_fs(s, 0);
2492         up_read(&s->s_umount);
2493
2494         RETURN(rc);
2495 }
2496
2497 static int (*priv_security_file_alloc)(struct file *file);
2498
2499 int osd_security_file_alloc(struct file *file)
2500 {
2501         if (priv_security_file_alloc)
2502                 return priv_security_file_alloc(file);
2503         return 0;
2504 }
2505
2506 /*
2507  * Concurrency: shouldn't matter.
2508  */
2509 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2510 {
2511         struct super_block *sb = osd_sb(osd_dt_dev(d));
2512         struct block_device *dev = sb->s_bdev;
2513         int rc = -EOPNOTSUPP;
2514
2515         ENTRY;
2516
2517         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2518                osd_dt_dev(d)->od_svname, (long)dev, rc);
2519
2520         RETURN(rc);
2521 }
2522
2523 /**
2524  * Note: we do not count into QUOTA here.
2525  * If we mount with --data_journal we may need more.
2526  */
2527 const int osd_dto_credits_noquota[DTO_NR] = {
2528         /**
2529          * Insert.
2530          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2531          * SINGLEDATA_TRANS_BLOCKS(8)
2532          * XXX Note: maybe iam need more, since iam have more level than
2533          *           EXT3 htree.
2534          */
2535         [DTO_INDEX_INSERT]  = 16,
2536         /**
2537          * Delete
2538          * just modify a single entry, probably merge few within a block
2539          */
2540         [DTO_INDEX_DELETE]  = 1,
2541         /**
2542          * Used for OI scrub
2543          */
2544         [DTO_INDEX_UPDATE]  = 16,
2545         /**
2546          * 4(inode, inode bits, groups, GDT)
2547          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2548          */
2549         [DTO_OBJECT_CREATE] = 4,
2550         /**
2551          * 4(inode, inode bits, groups, GDT)
2552          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2553          */
2554         [DTO_OBJECT_DELETE] = 4,
2555         /**
2556          * Attr set credits (inode)
2557          */
2558         [DTO_ATTR_SET_BASE] = 1,
2559         /**
2560          * Xattr set. The same as xattr of EXT3.
2561          * DATA_TRANS_BLOCKS(14)
2562          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2563          * are also counted in. Do not know why?
2564          */
2565         [DTO_XATTR_SET]     = 14,
2566         /**
2567          * credits for inode change during write.
2568          */
2569         [DTO_WRITE_BASE]    = 3,
2570         /**
2571          * credits for single block write.
2572          */
2573         [DTO_WRITE_BLOCK]   = 14,
2574         /**
2575          * Attr set credits for chown.
2576          * This is extra credits for setattr, and it is null without quota
2577          */
2578         [DTO_ATTR_SET_CHOWN] = 0
2579 };
2580
2581 static const struct dt_device_operations osd_dt_ops = {
2582         .dt_root_get       = osd_root_get,
2583         .dt_statfs         = osd_statfs,
2584         .dt_trans_create   = osd_trans_create,
2585         .dt_trans_start    = osd_trans_start,
2586         .dt_trans_stop     = osd_trans_stop,
2587         .dt_trans_cb_add   = osd_trans_cb_add,
2588         .dt_conf_get       = osd_conf_get,
2589         .dt_mnt_sb_get     = osd_mnt_sb_get,
2590         .dt_sync           = osd_sync,
2591         .dt_ro             = osd_ro,
2592         .dt_commit_async   = osd_commit_async,
2593 };
2594
2595 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2596                           unsigned int role)
2597 {
2598         struct osd_object *obj = osd_dt_obj(dt);
2599         struct osd_thread_info *oti = osd_oti_get(env);
2600
2601         LINVRNT(osd_invariant(obj));
2602
2603         LASSERT(obj->oo_owner != env);
2604         down_read_nested(&obj->oo_sem, role);
2605
2606         LASSERT(obj->oo_owner == NULL);
2607         oti->oti_r_locks++;
2608 }
2609
2610 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2611                            unsigned int role)
2612 {
2613         struct osd_object *obj = osd_dt_obj(dt);
2614         struct osd_thread_info *oti = osd_oti_get(env);
2615
2616         LINVRNT(osd_invariant(obj));
2617
2618         LASSERT(obj->oo_owner != env);
2619         down_write_nested(&obj->oo_sem, role);
2620
2621         LASSERT(obj->oo_owner == NULL);
2622         obj->oo_owner = env;
2623         oti->oti_w_locks++;
2624 }
2625
2626 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2627 {
2628         struct osd_object *obj = osd_dt_obj(dt);
2629         struct osd_thread_info *oti = osd_oti_get(env);
2630
2631         LINVRNT(osd_invariant(obj));
2632
2633         LASSERT(oti->oti_r_locks > 0);
2634         oti->oti_r_locks--;
2635         up_read(&obj->oo_sem);
2636 }
2637
2638 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2639 {
2640         struct osd_object *obj = osd_dt_obj(dt);
2641         struct osd_thread_info *oti = osd_oti_get(env);
2642
2643         LINVRNT(osd_invariant(obj));
2644
2645         LASSERT(obj->oo_owner == env);
2646         LASSERT(oti->oti_w_locks > 0);
2647         oti->oti_w_locks--;
2648         obj->oo_owner = NULL;
2649         up_write(&obj->oo_sem);
2650 }
2651
2652 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2653 {
2654         struct osd_object *obj = osd_dt_obj(dt);
2655
2656         LINVRNT(osd_invariant(obj));
2657
2658         return obj->oo_owner == env;
2659 }
2660
2661 static void osd_inode_getattr(const struct lu_env *env,
2662                               struct inode *inode, struct lu_attr *attr)
2663 {
2664         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2665                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2666                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2667                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2668
2669         attr->la_atime = inode->i_atime.tv_sec;
2670         attr->la_mtime = inode->i_mtime.tv_sec;
2671         attr->la_ctime = inode->i_ctime.tv_sec;
2672         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2673         attr->la_mode    = inode->i_mode;
2674         attr->la_size    = i_size_read(inode);
2675         attr->la_blocks  = inode->i_blocks;
2676         attr->la_uid     = i_uid_read(inode);
2677         attr->la_gid     = i_gid_read(inode);
2678         attr->la_projid  = i_projid_read(inode);
2679         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2680         attr->la_nlink   = inode->i_nlink;
2681         attr->la_rdev    = inode->i_rdev;
2682         attr->la_blksize = 1 << inode->i_blkbits;
2683         attr->la_blkbits = inode->i_blkbits;
2684         /*
2685          * Ext4 did not transfer inherit flags from raw inode
2686          * to inode flags, and ext4 internally test raw inode
2687          * @i_flags directly. Instead of patching ext4, we do it here.
2688          */
2689         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2690                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2691 }
2692
2693 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2694                             u64 *count)
2695 {
2696         struct osd_object *obj = osd_dt_obj(dt);
2697         const struct dt_it_ops *iops;
2698         struct dt_it *it;
2699         int rc;
2700
2701         ENTRY;
2702
2703         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2704         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2705
2706         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2707                 *count = obj->oo_dirent_count;
2708                 RETURN(0);
2709         }
2710
2711         /* directory not initialized yet */
2712         if (!dt->do_index_ops) {
2713                 *count = 0;
2714                 RETURN(0);
2715         }
2716
2717         iops = &dt->do_index_ops->dio_it;
2718         it = iops->init(env, dt, LUDA_64BITHASH);
2719         if (IS_ERR(it))
2720                 RETURN(PTR_ERR(it));
2721
2722         rc = iops->load(env, it, 0);
2723         if (rc < 0) {
2724                 if (rc == -ENODATA) {
2725                         rc = 0;
2726                         *count = 0;
2727                 }
2728                 GOTO(out, rc);
2729         }
2730         if (rc > 0)
2731                 rc = iops->next(env, it);
2732
2733         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2734                 if (rc == -ESTALE)
2735                         continue;
2736
2737                 if (iops->key_size(env, it) == 0)
2738                         continue;
2739
2740                 (*count)++;
2741         }
2742         if (rc == 1) {
2743                 obj->oo_dirent_count = *count;
2744                 rc = 0;
2745         }
2746 out:
2747         iops->put(env, it);
2748         iops->fini(env, it);
2749
2750         RETURN(rc);
2751 }
2752
2753 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2754                         struct lu_attr *attr)
2755 {
2756         struct osd_object *obj = osd_dt_obj(dt);
2757         int rc = 0;
2758
2759         if (unlikely(!dt_object_exists(dt)))
2760                 return -ENOENT;
2761         if (unlikely(obj->oo_destroyed))
2762                 return -ENOENT;
2763
2764         LASSERT(!dt_object_remote(dt));
2765         LINVRNT(osd_invariant(obj));
2766
2767         spin_lock(&obj->oo_guard);
2768         osd_inode_getattr(env, obj->oo_inode, attr);
2769         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2770                 attr->la_valid |= LA_FLAGS;
2771                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2772         }
2773         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2774                 attr->la_valid |= LA_FLAGS;
2775                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2776         }
2777         spin_unlock(&obj->oo_guard);
2778
2779         if (S_ISDIR(obj->oo_inode->i_mode) &&
2780             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2781                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2782
2783         return rc;
2784 }
2785
2786 static int osd_declare_attr_qid(const struct lu_env *env,
2787                                 struct osd_object *obj,
2788                                 struct osd_thandle *oh, long long bspace,
2789                                 qid_t old_id, qid_t new_id, bool enforce,
2790                                 unsigned int type, bool ignore_edquot)
2791 {
2792         int rc;
2793         struct osd_thread_info *info = osd_oti_get(env);
2794         struct lquota_id_info  *qi = &info->oti_qi;
2795
2796         qi->lqi_type = type;
2797         /* inode accounting */
2798         qi->lqi_is_blk = false;
2799
2800         /* one more inode for the new id ... */
2801         qi->lqi_id.qid_uid = new_id;
2802         qi->lqi_space      = 1;
2803         /* Reserve credits for the new id */
2804         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2805         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2806                 rc = 0;
2807         if (rc)
2808                 RETURN(rc);
2809
2810         /* and one less inode for the current id */
2811         qi->lqi_id.qid_uid = old_id;
2812         qi->lqi_space = -1;
2813         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2814         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2815                 rc = 0;
2816         if (rc)
2817                 RETURN(rc);
2818
2819         /* block accounting */
2820         qi->lqi_is_blk = true;
2821
2822         /* more blocks for the new id ... */
2823         qi->lqi_id.qid_uid = new_id;
2824         qi->lqi_space      = bspace;
2825         /*
2826          * Credits for the new uid has been reserved, re-use "obj"
2827          * to save credit reservation.
2828          */
2829         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2830         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2831                 rc = 0;
2832         if (rc)
2833                 RETURN(rc);
2834
2835         /* and finally less blocks for the current uid */
2836         qi->lqi_id.qid_uid = old_id;
2837         qi->lqi_space      = -bspace;
2838         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2839         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2840                 rc = 0;
2841
2842         RETURN(rc);
2843 }
2844
2845 static int osd_declare_attr_set(const struct lu_env *env,
2846                                 struct dt_object *dt,
2847                                 const struct lu_attr *attr,
2848                                 struct thandle *handle)
2849 {
2850         struct osd_thandle *oh;
2851         struct osd_object *obj;
2852         qid_t uid;
2853         qid_t gid;
2854         long long bspace;
2855         int rc = 0;
2856         bool enforce;
2857
2858         ENTRY;
2859
2860         LASSERT(dt != NULL);
2861         LASSERT(handle != NULL);
2862
2863         obj = osd_dt_obj(dt);
2864         LASSERT(osd_invariant(obj));
2865
2866         oh = container_of(handle, struct osd_thandle, ot_super);
2867         LASSERT(oh->ot_handle == NULL);
2868
2869         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2870                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2871
2872         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2873                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2874
2875         if (attr == NULL || obj->oo_inode == NULL)
2876                 RETURN(rc);
2877
2878         bspace   = obj->oo_inode->i_blocks << 9;
2879         bspace   = toqb(bspace);
2880
2881         /*
2882          * Changing ownership is always preformed by super user, it should not
2883          * fail with EDQUOT unless required explicitly.
2884          *
2885          * We still need to call the osd_declare_qid() to calculate the journal
2886          * credits for updating quota accounting files and to trigger quota
2887          * space adjustment once the operation is completed.
2888          */
2889         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2890                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2891
2892                 if (!ignore_edquot)
2893                         CDEBUG(D_QUOTA,
2894                                "%s: enforce quota on UID %u, GID %u (quota space is %lld)\n",
2895                                osd_ino2name(obj->oo_inode), attr->la_uid,
2896                                attr->la_gid, bspace);
2897
2898                 /* USERQUOTA */
2899                 uid = i_uid_read(obj->oo_inode);
2900                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2901                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2902                                           attr->la_uid, enforce, USRQUOTA,
2903                                           true);
2904                 if (rc)
2905                         RETURN(rc);
2906
2907                 gid = i_gid_read(obj->oo_inode);
2908                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2909                        attr->la_uid, gid, attr->la_gid);
2910                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2911                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2912                                           attr->la_gid, enforce, GRPQUOTA,
2913                                           ignore_edquot);
2914                 if (rc)
2915                         RETURN(rc);
2916
2917         }
2918 #ifdef HAVE_PROJECT_QUOTA
2919         if (attr->la_valid & LA_PROJID) {
2920                 __u32 projid = i_projid_read(obj->oo_inode);
2921
2922                 enforce = (attr->la_valid & LA_PROJID) &&
2923                                         (attr->la_projid != projid);
2924                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2925                                           (qid_t)projid, (qid_t)attr->la_projid,
2926                                           enforce, PRJQUOTA, true);
2927                 if (rc)
2928                         RETURN(rc);
2929         }
2930 #endif
2931         RETURN(rc);
2932 }
2933
2934 static int osd_inode_setattr(const struct lu_env *env,
2935                              struct inode *inode, const struct lu_attr *attr)
2936 {
2937         __u64 bits = attr->la_valid;
2938
2939         /* Only allow set size for regular file */
2940         if (!S_ISREG(inode->i_mode))
2941                 bits &= ~(LA_SIZE | LA_BLOCKS);
2942
2943         if (bits == 0)
2944                 return 0;
2945
2946         if (bits & LA_ATIME)
2947                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2948         if (bits & LA_CTIME)
2949                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2950         if (bits & LA_MTIME)
2951                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2952         if (bits & LA_SIZE) {
2953                 spin_lock(&inode->i_lock);
2954                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2955                 i_size_write(inode, attr->la_size);
2956                 spin_unlock(&inode->i_lock);
2957         }
2958
2959         /*
2960          * OSD should not change "i_blocks" which is used by quota.
2961          * "i_blocks" should be changed by ldiskfs only.
2962          */
2963         if (bits & LA_MODE)
2964                 inode->i_mode = (inode->i_mode & S_IFMT) |
2965                                 (attr->la_mode & ~S_IFMT);
2966         if (bits & LA_UID)
2967                 i_uid_write(inode, attr->la_uid);
2968         if (bits & LA_GID)
2969                 i_gid_write(inode, attr->la_gid);
2970         if (bits & LA_PROJID)
2971                 i_projid_write(inode, attr->la_projid);
2972         if (bits & LA_NLINK)
2973                 set_nlink(inode, attr->la_nlink);
2974         if (bits & LA_RDEV)
2975                 inode->i_rdev = attr->la_rdev;
2976
2977         if (bits & LA_FLAGS) {
2978                 /* always keep S_NOCMTIME */
2979                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2980                                  S_NOCMTIME;
2981 #if defined(S_ENCRYPTED)
2982                 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2983                  * aware of encryption status. It is just stored into LMA
2984                  * so that it can be forwared to client side.
2985                  */
2986                 inode->i_flags &= ~S_ENCRYPTED;
2987 #endif
2988                 /*
2989                  * Ext4 did not transfer inherit flags from
2990                  * @inode->i_flags to raw inode i_flags when writing
2991                  * flags, we do it explictly here.
2992                  */
2993                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2994                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2995                 else
2996                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2997         }
2998         return 0;
2999 }
3000
3001 #ifdef HAVE_PROJECT_QUOTA
3002 static int osd_transfer_project(struct inode *inode, __u32 projid,
3003                                 struct thandle *handle)
3004 {
3005         struct super_block *sb = inode->i_sb;
3006         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
3007         int err;
3008         kprojid_t kprojid;
3009         struct ldiskfs_iloc iloc;
3010         struct ldiskfs_inode *raw_inode;
3011         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
3012
3013         if (!ldiskfs_has_feature_project(sb)) {
3014                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
3015                         == LDISKFS_DEF_PROJID);
3016                 if (projid != LDISKFS_DEF_PROJID)
3017                         return -EOPNOTSUPP;
3018                 else
3019                         return 0;
3020         }
3021
3022         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
3023                 return -EOPNOTSUPP;
3024
3025         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
3026         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
3027                 return 0;
3028
3029         err = ldiskfs_get_inode_loc(inode, &iloc);
3030         if (err)
3031                 return err;
3032
3033         raw_inode = ldiskfs_raw_inode(&iloc);
3034         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3035                 struct osd_thandle *oh = container_of(handle,
3036                                                       struct osd_thandle,
3037                                                       ot_super);
3038                 /**
3039                  * try to expand inode size automatically.
3040                  */
3041                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3042                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3043                         err = -EOVERFLOW;
3044                         brelse(iloc.bh);
3045                         return err;
3046                 }
3047         }
3048         brelse(iloc.bh);
3049
3050         dquot_initialize(inode);
3051         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3052         if (transfer_to[PRJQUOTA]) {
3053                 err = __dquot_transfer(inode, transfer_to);
3054                 dqput(transfer_to[PRJQUOTA]);
3055                 if (err)
3056                         return err;
3057         }
3058
3059         return err;
3060 }
3061 #endif
3062
3063 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3064                               struct thandle *handle)
3065 {
3066         int rc;
3067
3068         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3069             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3070                 struct iattr iattr;
3071
3072                 CDEBUG(D_QUOTA,
3073                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3074                        inode->i_ino, i_uid_read(inode), attr->la_uid,
3075                        i_gid_read(inode), attr->la_gid);
3076
3077                 dquot_initialize(inode);
3078                 iattr.ia_valid = 0;
3079                 if (attr->la_valid & LA_UID)
3080                         iattr.ia_valid |= ATTR_UID;
3081                 if (attr->la_valid & LA_GID)
3082                         iattr.ia_valid |= ATTR_GID;
3083                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3084                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3085
3086                 rc = dquot_transfer(inode, &iattr);
3087                 if (rc) {
3088                         CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3089                                osd_ino2name(inode), rc);
3090                         return rc;
3091                 }
3092         }
3093
3094         /* Handle project id transfer here properly */
3095         if (attr->la_valid & LA_PROJID &&
3096             attr->la_projid != i_projid_read(inode)) {
3097 #ifdef HAVE_PROJECT_QUOTA
3098                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3099 #else
3100                 rc = -ENOTSUPP;
3101 #endif
3102                 if (rc) {
3103                         CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3104                                osd_ino2name(inode), rc);
3105                         return rc;
3106                 }
3107         }
3108         return 0;
3109 }
3110
3111 static int osd_attr_set(const struct lu_env *env,
3112                         struct dt_object *dt,
3113                         const struct lu_attr *attr,
3114                         struct thandle *handle)
3115 {
3116         struct osd_object *obj = osd_dt_obj(dt);
3117         struct inode *inode;
3118         int rc;
3119
3120         if (!dt_object_exists(dt))
3121                 return -ENOENT;
3122
3123         LASSERT(handle != NULL);
3124         LASSERT(!dt_object_remote(dt));
3125         LASSERT(osd_invariant(obj));
3126
3127         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3128
3129         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3130             !osd_obj2dev(obj)->od_is_ost) {
3131                 struct osd_thread_info *oti = osd_oti_get(env);
3132                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3133                 struct lu_fid *fid1 = &oti->oti_fid;
3134                 struct osd_inode_id *id = &oti->oti_id;
3135                 struct iam_path_descr *ipd;
3136                 struct iam_container *bag;
3137                 struct osd_thandle *oh;
3138                 int rc;
3139
3140                 fid_cpu_to_be(fid1, fid0);
3141                 memset(id, 1, sizeof(*id));
3142                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3143                                   fid0)->oi_dir.od_container;
3144                 ipd = osd_idx_ipd_get(env, bag);
3145                 if (unlikely(ipd == NULL))
3146                         RETURN(-ENOMEM);
3147
3148                 oh = container_of(handle, struct osd_thandle, ot_super);
3149                 rc = iam_update(oh->ot_handle, bag,
3150                                 (const struct iam_key *)fid1,
3151                                 (const struct iam_rec *)id, ipd);
3152                 osd_ipd_put(env, bag, ipd);
3153                 return(rc > 0 ? 0 : rc);
3154         }
3155
3156         inode = obj->oo_inode;
3157
3158         rc = osd_quota_transfer(inode, attr, handle);
3159         if (rc)
3160                 return rc;
3161
3162         spin_lock(&obj->oo_guard);
3163         rc = osd_inode_setattr(env, inode, attr);
3164         spin_unlock(&obj->oo_guard);
3165         if (rc != 0)
3166                 GOTO(out, rc);
3167
3168         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3169
3170         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3171
3172         if (!(attr->la_valid & LA_FLAGS))
3173                 GOTO(out, rc);
3174
3175         /* Let's check if there are extra flags need to be set into LMA */
3176         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3177                 struct osd_thread_info *info = osd_oti_get(env);
3178                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3179
3180                 LASSERT(!obj->oo_pfid_in_lma);
3181
3182                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3183                                  &info->oti_ost_attrs);
3184                 if (rc)
3185                         GOTO(out, rc);
3186
3187                 lma->lma_incompat |=
3188                         lustre_to_lma_flags(attr->la_flags);
3189                 lustre_lma_swab(lma);
3190
3191                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3192
3193                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3194                                      lma, sizeof(*lma), XATTR_REPLACE);
3195                 if (rc != 0) {
3196                         struct osd_device *osd = osd_obj2dev(obj);
3197
3198                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3199                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3200                               lma->lma_incompat, rc);
3201                 } else {
3202                         obj->oo_lma_flags =
3203                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3204                 }
3205                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3206         }
3207 out:
3208
3209         return rc;
3210 }
3211
3212 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3213                                            struct osd_object *obj,
3214                                            const char *name, const int namelen)
3215 {
3216         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3217 }
3218
3219 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3220                       umode_t mode, struct dt_allocation_hint *hint,
3221                       struct thandle *th, struct lu_attr *attr)
3222 {
3223         int result;
3224         struct osd_device *osd = osd_obj2dev(obj);
3225         struct osd_thandle *oth;
3226         struct dt_object *parent = NULL;
3227         struct inode *inode;
3228         struct iattr iattr = {
3229                 .ia_valid = ATTR_UID | ATTR_GID |
3230                             ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3231                 .ia_ctime.tv_sec = attr->la_ctime,
3232                 .ia_mtime.tv_sec = attr->la_mtime,
3233                 .ia_atime.tv_sec = attr->la_atime,
3234                 .ia_uid = GLOBAL_ROOT_UID,
3235                 .ia_gid = GLOBAL_ROOT_GID,
3236         };
3237         const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
3238
3239         if (attr->la_valid & LA_UID)
3240                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3241         if (attr->la_valid & LA_GID)
3242                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3243
3244         LINVRNT(osd_invariant(obj));
3245         LASSERT(obj->oo_inode == NULL);
3246         LASSERT(obj->oo_hl_head == NULL);
3247
3248         if (S_ISDIR(mode) && ldiskfs_pdo) {
3249                 obj->oo_hl_head =
3250                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3251                 if (obj->oo_hl_head == NULL)
3252                         return -ENOMEM;
3253         }
3254
3255         oth = container_of(th, struct osd_thandle, ot_super);
3256         LASSERT(oth->ot_handle->h_transaction != NULL);
3257
3258         if (hint != NULL && hint->dah_parent != NULL &&
3259             !dt_object_remote(hint->dah_parent))
3260                 parent = hint->dah_parent;
3261
3262         /* if a time component is not valid set it to UTIME_OMIT */
3263         if (!(attr->la_valid & LA_CTIME))
3264                 iattr.ia_ctime = omit;
3265         if (!(attr->la_valid & LA_MTIME))
3266                 iattr.ia_mtime = omit;
3267         if (!(attr->la_valid & LA_ATIME))
3268                 iattr.ia_atime = omit;
3269
3270         inode = ldiskfs_create_inode(oth->ot_handle,
3271                                      parent ? osd_dt_obj(parent)->oo_inode :
3272                                               osd_sb(osd)->s_root->d_inode,
3273                                      mode, &iattr);
3274         if (!IS_ERR(inode)) {
3275                 /* Do not update file c/mtime in ldiskfs. */
3276                 inode->i_flags |= S_NOCMTIME;
3277
3278                 /*
3279                  * For new created object, it must be consistent,
3280                  * and it is unnecessary to scrub against it.
3281                  */
3282                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3283
3284                 obj->oo_inode = inode;
3285                 result = 0;
3286         } else {
3287                 if (obj->oo_hl_head != NULL) {
3288                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3289                         obj->oo_hl_head = NULL;
3290                 }
3291                 result = PTR_ERR(inode);
3292         }
3293         LINVRNT(osd_invariant(obj));
3294         return result;
3295 }
3296
3297 enum {
3298         OSD_NAME_LEN = 255
3299 };
3300
3301 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3302                      struct lu_attr *attr,
3303                      struct dt_allocation_hint *hint,
3304                      struct dt_object_format *dof,
3305                      struct thandle *th)
3306 {
3307         int result;
3308         struct osd_thandle *oth;
3309         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3310
3311         LASSERT(S_ISDIR(attr->la_mode));
3312
3313         oth = container_of(th, struct osd_thandle, ot_super);
3314         LASSERT(oth->ot_handle->h_transaction != NULL);
3315         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3316                 obj->oo_dirent_count = 0;
3317         result = osd_mkfile(info, obj, mode, hint, th, attr);
3318
3319         return result;
3320 }
3321
3322 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3323                         struct lu_attr *attr,
3324                         struct dt_allocation_hint *hint,
3325                         struct dt_object_format *dof,
3326                         struct thandle *th)
3327 {
3328         int result;
3329         struct osd_thandle *oth;
3330         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3331
3332         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3333
3334         LASSERT(S_ISREG(attr->la_mode));
3335
3336         oth = container_of(th, struct osd_thandle, ot_super);
3337         LASSERT(oth->ot_handle->h_transaction != NULL);
3338
3339         result = osd_mkfile(info, obj, mode, hint, th, attr);
3340         if (result == 0) {
3341                 LASSERT(obj->oo_inode != NULL);
3342                 if (feat->dif_flags & DT_IND_VARKEY)
3343                         result = iam_lvar_create(obj->oo_inode,
3344                                                  feat->dif_keysize_max,
3345                                                  feat->dif_ptrsize,
3346                                                  feat->dif_recsize_max,
3347                                                  oth->ot_handle);
3348                 else
3349                         result = iam_lfix_create(obj->oo_inode,
3350                                                  feat->dif_keysize_max,
3351                                                  feat->dif_ptrsize,
3352                                                  feat->dif_recsize_max,
3353                                                  oth->ot_handle);
3354         }
3355         return result;
3356 }
3357
3358 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3359                      struct lu_attr *attr,
3360                      struct dt_allocation_hint *hint,
3361                      struct dt_object_format *dof,
3362                      struct thandle *th)
3363 {
3364         LASSERT(S_ISREG(attr->la_mode));
3365         return osd_mkfile(info, obj, (attr->la_mode &
3366                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3367                           attr);
3368 }
3369
3370 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3371                      struct lu_attr *attr,
3372                      struct dt_allocation_hint *hint,
3373                      struct dt_object_format *dof,
3374                      struct thandle *th)
3375 {
3376         LASSERT(S_ISLNK(attr->la_mode));
3377         return osd_mkfile(info, obj, (attr->la_mode &
3378                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3379                           attr);
3380 }
3381
3382 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3383                      struct lu_attr *attr,
3384                      struct dt_allocation_hint *hint,
3385                      struct dt_object_format *dof,
3386                      struct thandle *th)
3387 {
3388         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3389         int result;
3390
3391         LINVRNT(osd_invariant(obj));
3392         LASSERT(obj->oo_inode == NULL);
3393         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3394                 S_ISFIFO(mode) || S_ISSOCK(mode));
3395
3396         result = osd_mkfile(info, obj, mode, hint, th, attr);
3397         if (result == 0) {
3398                 LASSERT(obj->oo_inode != NULL);
3399                 /*
3400                  * This inode should be marked dirty for i_rdev.  Currently
3401                  * that is done in the osd_attr_init().
3402                  */
3403                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3404                                    attr->la_rdev);
3405         }
3406         LINVRNT(osd_invariant(obj));
3407         return result;
3408 }
3409
3410 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3411                               struct lu_attr *,
3412                               struct dt_allocation_hint *hint,
3413                               struct dt_object_format *dof,
3414                               struct thandle *);
3415
3416 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3417 {
3418         osd_obj_type_f result;
3419
3420         switch (type) {
3421         case DFT_DIR:
3422                 result = osd_mkdir;
3423                 break;
3424         case DFT_REGULAR:
3425                 result = osd_mkreg;
3426                 break;
3427         case DFT_SYM:
3428                 result = osd_mksym;
3429                 break;
3430         case DFT_NODE:
3431                 result = osd_mknod;
3432                 break;
3433         case DFT_INDEX:
3434                 result = osd_mk_index;
3435                 break;
3436
3437         default:
3438                 LBUG();
3439                 break;
3440         }
3441         return result;
3442 }
3443
3444
3445 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3446                         struct dt_object *parent, struct dt_object *child,
3447                         umode_t child_mode)
3448 {
3449         LASSERT(ah);
3450
3451         ah->dah_parent = parent;
3452         ah->dah_mode = child_mode;
3453
3454         if (parent != NULL && !dt_object_remote(parent)) {
3455                 /* will help to find FID->ino at dt_insert("..") */
3456                 struct osd_object *pobj = osd_dt_obj(parent);
3457
3458                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3459         }
3460 }
3461
3462 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3463                           struct lu_attr *attr, struct dt_object_format *dof,
3464                           struct thandle *handle)
3465 {
3466         struct inode *inode = obj->oo_inode;
3467         __u64 valid = attr->la_valid;
3468         int result;
3469
3470         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3471
3472         if (dof->dof_type != DFT_NODE)
3473                 attr->la_valid &= ~LA_RDEV;
3474         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3475                 attr->la_valid &= ~LA_ATIME;
3476         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3477                 attr->la_valid &= ~LA_CTIME;
3478         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3479                 attr->la_valid &= ~LA_MTIME;
3480
3481         result = osd_quota_transfer(inode, attr, handle);
3482         if (result)
3483                 return;
3484
3485         if (attr->la_valid != 0) {
3486                 result = osd_inode_setattr(info->oti_env, inode, attr);
3487                 /*
3488                  * The osd_inode_setattr() should always succeed here.  The
3489                  * only error that could be returned is EDQUOT when we are
3490                  * trying to change the UID or GID of the inode. However, this
3491                  * should not happen since quota enforcement is no longer
3492                  * enabled on ldiskfs (lquota takes care of it).
3493                  */
3494                 LASSERTF(result == 0, "%d\n", result);
3495                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3496         }
3497
3498         attr->la_valid = valid;
3499 }
3500
3501 /**
3502  * Helper function for osd_create()
3503  *
3504  * \retval 0, on success
3505  */
3506 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3507                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3508                         struct dt_object_format *dof, struct thandle *th)
3509 {
3510         int result;
3511         __u32 umask;
3512
3513         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3514
3515         /* we drop umask so that permissions we pass are not affected */
3516         umask = current->fs->umask;
3517         current->fs->umask = 0;
3518
3519         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3520                                                   th);
3521         if (likely(obj->oo_inode != NULL)) {
3522                 LASSERT(obj->oo_inode->i_state & I_NEW);
3523
3524                 /*
3525                  * Unlock the inode before attr initialization to avoid
3526                  * unnecessary dqget operations. LU-6378
3527                  */
3528                 unlock_new_inode(obj->oo_inode);
3529         }
3530
3531         if (likely(result == 0)) {
3532                 osd_attr_init(info, obj, attr, dof, th);
3533                 osd_object_init0(obj);
3534         }
3535
3536         /* restore previous umask value */
3537         current->fs->umask = umask;
3538
3539         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3540
3541         return result;
3542 }
3543
3544 /**
3545  * Helper function for osd_create()
3546  *
3547  * \retval 0, on success
3548  */
3549 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3550                            const struct lu_fid *fid, struct thandle *th)
3551 {
3552         struct osd_thread_info *info = osd_oti_get(env);
3553         struct osd_inode_id    *id   = &info->oti_id;
3554         struct osd_device      *osd  = osd_obj2dev(obj);
3555         struct osd_thandle     *oh;
3556         int rc;
3557
3558         LASSERT(obj->oo_inode != NULL);
3559
3560         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
3561                 return -ENOSPC;
3562
3563         oh = container_of(th, struct osd_thandle, ot_super);
3564         LASSERT(oh->ot_handle);
3565         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3566
3567         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3568         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3569                            OI_CHECK_FLD, NULL);
3570         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3571                 struct lu_fid next_fid = *fid;
3572
3573                 /* insert next object in advance, and map to the same inode */
3574                 next_fid.f_oid++;
3575                 if (next_fid.f_oid != 0) {
3576                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3577                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3578                                       OI_CHECK_FLD, NULL);
3579                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3580                 }
3581         }
3582
3583         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3584
3585         return rc;
3586 }
3587
3588 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3589                    u64 seq, struct lu_seq_range *range)
3590 {
3591         struct seq_server_site *ss = osd_seq_site(osd);
3592
3593         if (fid_seq_is_idif(seq)) {
3594                 fld_range_set_ost(range);
3595                 range->lsr_index = idif_ost_idx(seq);
3596                 return 0;
3597         }
3598
3599         if (!fid_seq_in_fldb(seq)) {
3600                 fld_range_set_mdt(range);
3601                 if (ss != NULL)
3602                         /*
3603                          * FIXME: If ss is NULL, it suppose not get lsr_index
3604                          * at all
3605                          */
3606                         range->lsr_index = ss->ss_node_id;
3607                 return 0;
3608         }
3609
3610         LASSERT(ss != NULL);
3611         fld_range_set_any(range);
3612         /* OSD will only do local fld lookup */
3613         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3614 }
3615
3616 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3617                               struct lu_attr *attr,
3618                               struct dt_allocation_hint *hint,
3619                               struct dt_object_format *dof,
3620                               struct thandle *handle)
3621 {
3622         struct osd_thandle *oh;
3623         int rc;
3624
3625         ENTRY;
3626
3627         LASSERT(handle != NULL);
3628
3629         oh = container_of(handle, struct osd_thandle, ot_super);
3630         LASSERT(oh->ot_handle == NULL);
3631
3632         /*
3633          * EA object consumes more credits than regular object: osd_mk_index
3634          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3635          * leaf_node, could involves the block, block bitmap, groups, GDT
3636          * change for each block, so add 4 * 2 credits in that case.
3637          */
3638         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3639                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3640                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3641         /*
3642          * Reuse idle OI block may cause additional one OI block
3643          * to be changed.
3644          */
3645         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3646                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3647         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3648                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3649                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3650
3651         /* will help to find FID->ino mapping at dt_insert() */
3652         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3653                                    osd_dt_obj(dt));
3654         if (rc != 0)
3655                 RETURN(rc);
3656
3657         if (!attr)
3658                 RETURN(0);
3659
3660         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3661                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3662                                    NULL, OSD_QID_INODE);
3663         if (rc != 0)
3664                 RETURN(rc);
3665
3666         RETURN(rc);
3667 }
3668
3669 /**
3670  * Called to destroy on-disk representation of the object
3671  *
3672  * Concurrency: must be locked
3673  */
3674 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3675                                struct thandle *th)
3676 {
3677         struct osd_object *obj = osd_dt_obj(dt);
3678         struct inode *inode = obj->oo_inode;
3679         struct osd_thandle *oh;
3680         int rc;
3681
3682         ENTRY;
3683
3684         if (inode == NULL)
3685                 RETURN(-ENOENT);
3686
3687         oh = container_of(th, struct osd_thandle, ot_super);
3688         LASSERT(oh->ot_handle == NULL);
3689
3690         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3691                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3692
3693         /* For removing agent entry */
3694         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3695                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3696
3697         /*
3698          * Recycle idle OI leaf may cause additional three OI blocks
3699          * to be changed.
3700          */
3701         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3702                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3703                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3704         /* one less inode */
3705         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3706                                    i_projid_read(inode), -1, oh, obj, NULL,
3707                                    OSD_QID_INODE);
3708         if (rc)
3709                 RETURN(rc);
3710         /* data to be truncated */
3711         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3712                                    i_projid_read(inode), 0, oh, obj, NULL,
3713                                    OSD_QID_BLK);
3714         if (rc)
3715                 RETURN(rc);
3716
3717         /*
3718          * will help to find FID->ino when this object is being
3719          * added to PENDING
3720          */
3721         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3722
3723         RETURN(rc);
3724 }
3725
3726 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3727                        struct thandle *th)
3728 {
3729         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3730         struct osd_object *obj = osd_dt_obj(dt);
3731         struct inode *inode = obj->oo_inode;
3732         struct osd_device *osd = osd_obj2dev(obj);
3733         struct osd_thandle *oh;
3734         int result;
3735
3736         ENTRY;
3737
3738         oh = container_of(th, struct osd_thandle, ot_super);
3739         LASSERT(oh->ot_handle);
3740         LASSERT(inode);
3741         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3742
3743         if (unlikely(fid_is_acct(fid)))
3744                 RETURN(-EPERM);
3745
3746         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3747                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3748                 if (result != 0)
3749                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3750                                osd_name(osd), PFID(fid), result);
3751         }
3752
3753         if (S_ISDIR(inode->i_mode)) {
3754                 if (inode->i_nlink > 2)
3755                         CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3756                                osd_name(osd), PFID(fid), inode->i_ino,
3757                                inode->i_nlink);
3758
3759                 spin_lock(&obj->oo_guard);
3760                 clear_nlink(inode);
3761                 spin_unlock(&obj->oo_guard);
3762                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3763         }
3764
3765         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3766
3767         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3768
3769         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3770                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3771                                        oh->ot_handle, OI_CHECK_FLD);
3772
3773         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3774         /* XXX: add to ext3 orphan list */
3775         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3776
3777         /* not needed in the cache anymore */
3778         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3779         obj->oo_destroyed = 1;
3780
3781         RETURN(0);
3782 }
3783
3784 /**
3785  * Put the fid into lustre_mdt_attrs, and then place the structure
3786  * inode's ea. This fid should not be altered during the life time
3787  * of the inode.
3788  *
3789  * \retval +ve, on success
3790  * \retval -ve, on error
3791  *
3792  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3793  */
3794 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3795                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3796 {
3797         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3798         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3799         int rc;
3800
3801         ENTRY;
3802
3803         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3804                 RETURN(0);
3805
3806         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3807                 rc = -ENOMEM;
3808
3809         lustre_loa_init(loa, fid, compat, incompat);
3810         lustre_loa_swab(loa, false);
3811
3812         /*
3813          * For the OST device with 256 bytes inode size by default,
3814          * the PFID EA will be stored together with LMA EA to avoid
3815          * performance trouble. Otherwise the PFID EA can be stored
3816          * independently. LU-8998
3817          */
3818         if ((compat & LMAC_FID_ON_OST) &&
3819             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3820                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3821                                      sizeof(*loa), XATTR_CREATE);
3822         else
3823                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3824                                      sizeof(*lma), XATTR_CREATE);
3825         /*
3826          * LMA may already exist, but we need to check that all the
3827          * desired compat/incompat flags have been added.
3828          */
3829         if (unlikely(rc == -EEXIST)) {
3830                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3831                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3832                 if (rc < 0)
3833                         RETURN(rc);
3834
3835                 if (rc < sizeof(*lma))
3836                         RETURN(-EINVAL);
3837
3838                 lustre_loa_swab(loa, true);
3839                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3840                     ((compat == 0 && incompat == 0) ||
3841                      (!(~lma->lma_compat & compat) &&
3842                       !(~lma->lma_incompat & incompat))))
3843                         RETURN(0);
3844
3845                 lma->lma_self_fid = *fid;
3846                 lma->lma_compat |= compat;
3847                 lma->lma_incompat |= incompat;
3848                 if (rc == sizeof(*lma)) {
3849                         lustre_lma_swab(lma);
3850                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3851                                              sizeof(*lma), XATTR_REPLACE);
3852                 } else {
3853                         lustre_loa_swab(loa, false);
3854                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3855                                              sizeof(*loa), XATTR_REPLACE);
3856                 }
3857         }
3858
3859         RETURN(rc);
3860 }
3861
3862 /**
3863  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3864  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3865  * To have compatilibility with 1.8 ldiskfs driver we need to have
3866  * magic number at start of fid data.
3867  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3868  * its inmemory API.
3869  */
3870 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3871                                          const struct lu_fid *fid)
3872 {
3873         if (!fid_is_namespace_visible(fid) ||
3874             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3875                 param->edp_magic = 0;
3876                 return;
3877         }
3878
3879         param->edp_magic = LDISKFS_LUFID_MAGIC;
3880         param->edp_len =  sizeof(struct lu_fid) + 1;
3881         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3882 }
3883
3884 /**
3885  * Try to read the fid from inode ea into dt_rec.
3886  *
3887  * \param fid object fid.
3888  *
3889  * \retval 0 on success
3890  */
3891 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3892                           __u32 ino, struct lu_fid *fid,
3893                           struct osd_inode_id *id)
3894 {
3895         struct osd_thread_info *info  = osd_oti_get(env);
3896         struct inode *inode;
3897
3898         ENTRY;
3899
3900         osd_id_gen(id, ino, OSD_OII_NOGEN);
3901         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3902         if (IS_ERR(inode))
3903                 RETURN(PTR_ERR(inode));
3904
3905         iput(inode);
3906         RETURN(0);
3907 }
3908
3909 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3910                                         struct inode *dir,
3911                                         struct inode *parent_dir,
3912                                         const struct lu_fid *dot_fid,
3913                                         const struct lu_fid *dot_dot_fid,
3914                                         struct osd_thandle *oth)
3915 {
3916         struct ldiskfs_dentry_param *dot_ldp;
3917         struct ldiskfs_dentry_param *dot_dot_ldp;
3918         __u32 saved_nlink = dir->i_nlink;
3919         int rc;
3920
3921         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
3922                 return -ENOSPC;
3923
3924         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3925         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3926
3927         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3928         dot_ldp->edp_magic = 0;
3929
3930         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3931                                     dir, dot_ldp, dot_dot_ldp);
3932         /*
3933          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3934          * the subseqent ref_add() will increase the dir->i_nlink
3935          * as 3. That is incorrect for new created directory.
3936          *
3937          * It looks like hack, because we want to make the OSD API
3938          * to be order-independent for new created directory object
3939          * between dt_insert(..) and ref_add() operations.
3940          *
3941          * Here, we only restore the in-RAM dir-inode's nlink attr,
3942          * becuase if the nlink attr is not 2, then there will be
3943          * ref_add() called following the dt_insert(..), such call
3944          * will make both the in-RAM and on-disk dir-inode's nlink
3945          * attr to be set as 2. LU-7447
3946          */
3947         set_nlink(dir, saved_nlink);
3948         return rc;
3949 }
3950
3951 /**
3952  * Create an local agent inode for remote entry
3953  */
3954 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3955                                                   struct osd_device *osd,
3956                                                   struct osd_object *pobj,
3957                                                   const struct lu_fid *fid,
3958                                                   __u32 type,
3959                                                   struct thandle *th)
3960 {
3961         struct osd_thread_info *info = osd_oti_get(env);
3962         struct inode *local;
3963         struct osd_thandle *oh;
3964         struct iattr iattr = {
3965                 .ia_valid = ATTR_UID | ATTR_GID |
3966                             ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3967                 .ia_ctime.tv_nsec = UTIME_OMIT,
3968                 .ia_mtime.tv_nsec = UTIME_OMIT,
3969                 .ia_atime.tv_nsec = UTIME_OMIT,
3970                 .ia_uid = GLOBAL_ROOT_UID,
3971                 .ia_gid = GLOBAL_ROOT_GID,
3972         };
3973         int rc;
3974
3975         ENTRY;
3976
3977         LASSERT(th);
3978         oh = container_of(th, struct osd_thandle, ot_super);
3979         LASSERT(oh->ot_handle->h_transaction != NULL);
3980
3981         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
3982                                      type, &iattr);
3983         if (IS_ERR(local)) {
3984                 CERROR("%s: create local error %d\n", osd_name(osd),
3985                        (int)PTR_ERR(local));
3986                 RETURN(local);
3987         }
3988
3989         /*
3990          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3991          * correct gid on remote file, not agent here
3992          */
3993         local->i_gid = current_fsgid();
3994         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3995
3996         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3997          * That gives e2fsck something to look at and be happy, and allows
3998          * debugging if we need to determine where this symlink came from.
3999          */
4000         if (S_ISLNK(type)) {
4001                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
4002                 rc = scnprintf((char *)LDISKFS_I(local)->i_data,
4003                                LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
4004
4005                 i_size_write(local, rc);
4006                 LDISKFS_I(local)->i_disksize = rc;
4007         }
4008         unlock_new_inode(local);
4009
4010         /* Agent inode should not have project ID */
4011 #ifdef  HAVE_PROJECT_QUOTA
4012         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
4013             i_projid_read(pobj->oo_inode) != 0) {
4014                 rc = osd_transfer_project(local, 0, th);
4015                 if (rc) {
4016                         CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
4017                                osd_ino2name(local), rc);
4018                         RETURN(ERR_PTR(rc));
4019                 }
4020         }
4021 #endif
4022         /* Set special LMA flag for local agent inode */
4023         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
4024         if (rc != 0) {
4025                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
4026                        osd_name(osd), PFID(fid), rc);
4027                 RETURN(ERR_PTR(rc));
4028         }
4029
4030         if (!S_ISDIR(type))
4031                 RETURN(local);
4032
4033         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
4034                                          lu_object_fid(&pobj->oo_dt.do_lu),
4035                                          fid, oh);
4036         if (rc != 0) {
4037                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
4038                         osd_name(osd), PFID(fid), rc);
4039                 RETURN(ERR_PTR(rc));
4040         }
4041
4042         RETURN(local);
4043 }
4044
4045 /**
4046  * when direntry is deleted, we have to take care of possible agent inode
4047  * referenced by that. unfortunately we can't do this at that point:
4048  * iget() within a running transaction leads to deadlock and we better do
4049  * not call that every delete declaration to save performance. so we put
4050  * a potention agent inode on a list and process that once the transaction
4051  * is over. Notice it's not any worse in terms of real orphans as regular
4052  * object destroy doesn't put inodes on the on-disk orphan list. this should
4053  * be addressed separately
4054  */
4055 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
4056                                             struct osd_thandle *oh,
4057                                             __u32 ino)
4058 {
4059         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
4060         struct osd_obj_orphan *oor;
4061
4062         OBD_ALLOC_PTR(oor);
4063         if (oor == NULL)
4064                 return -ENOMEM;
4065
4066         oor->oor_ino = ino;
4067         oor->oor_env = (struct lu_env *)env;
4068         spin_lock(&osd->od_osfs_lock);
4069         list_add(&oor->oor_list, &osd->od_orphan_list);
4070         spin_unlock(&osd->od_osfs_lock);
4071
4072         oh->ot_remove_agents = 1;
4073
4074         return 0;
4075
4076 }
4077
4078 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
4079                                                 struct osd_device *osd)
4080 {
4081         struct osd_thread_info *info = osd_oti_get(env);
4082         struct osd_obj_orphan *oor, *tmp;
4083         struct osd_inode_id id;
4084         LIST_HEAD(list);
4085         struct inode *inode;
4086         struct lu_fid fid;
4087         handle_t *jh;
4088         __u32 ino;
4089
4090         spin_lock(&osd->od_osfs_lock);
4091         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4092                 if (oor->oor_env == env)
4093                         list_move(&oor->oor_list, &list);
4094         }
4095         spin_unlock(&osd->od_osfs_lock);
4096
4097         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4098
4099                 ino = oor->oor_ino;
4100
4101                 list_del(&oor->oor_list);
4102                 OBD_FREE_PTR(oor);
4103
4104                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4105                 inode = osd_iget_fid(info, osd, &id, &fid);
4106                 if (IS_ERR(inode))
4107                         continue;
4108
4109                 if (!osd_remote_fid(env, osd, &fid)) {
4110                         iput(inode);
4111                         continue;
4112                 }
4113
4114                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4115                 clear_nlink(inode);
4116                 mark_inode_dirty(inode);
4117                 ldiskfs_journal_stop(jh);
4118                 iput(inode);
4119         }
4120
4121         return 0;
4122 }
4123
4124 /**
4125  * OSD layer object create function for OST objects (b=11826).
4126  *
4127  * The FID is inserted into inode xattr here.
4128  *
4129  * \retval   0, on success
4130  * \retval -ve, on error
4131  */
4132 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4133                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4134                       struct dt_object_format *dof, struct thandle *th)
4135 {
4136         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4137         struct osd_object *obj = osd_dt_obj(dt);
4138         struct osd_thread_info *info = osd_oti_get(env);
4139         int result, on_ost = 0;
4140
4141         ENTRY;
4142
4143         if (dt_object_exists(dt))
4144                 RETURN(-EEXIST);
4145
4146         LINVRNT(osd_invariant(obj));
4147         LASSERT(!dt_object_remote(dt));
4148         LASSERT(osd_is_write_locked(env, obj));
4149         LASSERT(th != NULL);
4150
4151         if (unlikely(fid_is_acct(fid)))
4152                 /*
4153                  * Quota files can't be created from the kernel any more,
4154                  * 'tune2fs -O quota' will take care of creating them
4155                  */
4156                 RETURN(-EPERM);
4157
4158         result = __osd_create(info, obj, attr, hint, dof, th);
4159         if (result == 0) {
4160                 if (fid_is_idif(fid) &&
4161                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4162                         struct lu_fid *tfid = &info->oti_fid;
4163                         struct ost_id *oi   = &info->oti_ostid;
4164
4165                         fid_to_ostid(fid, oi);
4166                         ostid_to_fid(tfid, oi, 0);
4167                         on_ost = 1;
4168                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4169                                                 LMAC_FID_ON_OST, 0);
4170                 } else {
4171                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4172                                                fid, OI_CHECK_FLD);
4173                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4174                                                 on_ost ? LMAC_FID_ON_OST : 0,
4175                                                 0);
4176                 }
4177                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4178                         obj->oo_dt.do_body_ops = &osd_body_ops;
4179         }
4180
4181         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
4182                 struct inode *inode = obj->oo_inode;
4183
4184                 result = __osd_oi_insert(env, obj, fid, th);
4185                 if (result && inode) {
4186                         spin_lock(&obj->oo_guard);
4187                         clear_nlink(inode);
4188                         spin_unlock(&obj->oo_guard);
4189                         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4190                         ldiskfs_set_inode_state(inode,
4191                                                 LDISKFS_STATE_LUSTRE_DESTROY);
4192                         iput(inode);
4193                         obj->oo_inode = NULL;
4194                 }
4195         }
4196
4197         /*
4198          * a small optimization - dt_insert() isn't usually applied
4199          * to OST objects, so we don't need to cache OI mapping for
4200          * OST objects
4201          */
4202         if (result == 0 && on_ost == 0) {
4203                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4204
4205                 result = osd_idc_find_and_init(env, osd, obj);
4206                 LASSERT(result == 0);
4207         }
4208
4209         LASSERT(ergo(result == 0,
4210                      dt_object_exists(dt) && !dt_object_remote(dt)));
4211         LINVRNT(osd_invariant(obj));
4212         RETURN(result);
4213 }
4214
4215 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4216                                struct thandle *handle)
4217 {
4218         struct osd_thandle *oh;
4219         int rc;
4220
4221         /* it's possible that object doesn't exist yet */
4222         LASSERT(handle != NULL);
4223
4224         oh = container_of(handle, struct osd_thandle, ot_super);
4225         LASSERT(oh->ot_handle == NULL);
4226
4227         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4228                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4229
4230         rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
4231                                    osd_dt_obj(dt));
4232
4233         return rc;
4234 }
4235
4236 /*
4237  * Concurrency: @dt is write locked.
4238  */
4239 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4240                        struct thandle *th)
4241 {
4242         struct osd_object *obj = osd_dt_obj(dt);
4243         struct inode *inode = obj->oo_inode;
4244         struct osd_thandle *oh;
4245         int rc = 0;
4246
4247         if (!dt_object_exists(dt) || obj->oo_destroyed)
4248                 return -ENOENT;
4249
4250         LINVRNT(osd_invariant(obj));
4251         LASSERT(!dt_object_remote(dt));
4252         LASSERT(osd_is_write_locked(env, obj));
4253         LASSERT(th != NULL);
4254
4255         oh = container_of(th, struct osd_thandle, ot_super);
4256         LASSERT(oh->ot_handle != NULL);
4257
4258         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4259
4260         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4261                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4262         /*
4263          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4264          * (65000) subdirectories by storing "1" in i_nlink if the link count
4265          * would otherwise overflow. Directory tranversal tools understand
4266          * that (st_nlink == 1) indicates that the filesystem dose not track
4267          * hard links count on the directory, and will not abort subdirectory
4268          * scanning early once (st_nlink - 2) subdirs have been found.
4269          *
4270          * This also has to properly handle the case of inodes with nlink == 0
4271          * in case they are being linked into the PENDING directory
4272          */
4273         spin_lock(&obj->oo_guard);
4274         if (unlikely(inode->i_nlink == 0))
4275                 /* inc_nlink from 0 may cause WARN_ON */
4276                 set_nlink(inode, 1);
4277         else {
4278                 ldiskfs_inc_count(oh->ot_handle, inode);
4279                 if (!S_ISDIR(inode->i_mode))
4280                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4281         }
4282         spin_unlock(&obj->oo_guard);
4283
4284         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4285         LINVRNT(osd_invariant(obj));
4286
4287         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4288
4289         return rc;
4290 }
4291
4292 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4293                                struct thandle *handle)
4294 {
4295         struct osd_thandle *oh;
4296
4297         if (!dt_object_exists(dt))
4298                 return -ENOENT;
4299
4300         LASSERT(!dt_object_remote(dt));
4301         LASSERT(handle != NULL);
4302
4303         oh = container_of(handle, struct osd_thandle, ot_super);
4304         LASSERT(oh->ot_handle == NULL);
4305
4306         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4307                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4308
4309         return 0;
4310 }
4311
4312 /*
4313  * Concurrency: @dt is write locked.
4314  */
4315 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4316                        struct thandle *th)
4317 {
4318         struct osd_object *obj = osd_dt_obj(dt);
4319         struct inode *inode = obj->oo_inode;
4320         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4321         struct osd_thandle *oh;
4322
4323         if (!dt_object_exists(dt))
4324                 return -ENOENT;
4325
4326         LINVRNT(osd_invariant(obj));
4327         LASSERT(!dt_object_remote(dt));
4328         LASSERT(osd_is_write_locked(env, obj));
4329         LASSERT(th != NULL);
4330
4331         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4332                 return -EIO;
4333
4334         oh = container_of(th, struct osd_thandle, ot_super);
4335         LASSERT(oh->ot_handle != NULL);
4336
4337         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4338
4339         spin_lock(&obj->oo_guard);
4340         /*
4341          * That can be result of upgrade from old Lustre version and
4342          * applied only to local files.  Just skip this ref_del call.
4343          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4344          */
4345         if (inode->i_nlink == 0) {
4346                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4347                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4348                              ", maybe an upgraded file? (LU-3915)\n",
4349                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4350                 spin_unlock(&obj->oo_guard);
4351                 return 0;
4352         }
4353
4354         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4355                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4356
4357         ldiskfs_dec_count(oh->ot_handle, inode);
4358         spin_unlock(&obj->oo_guard);
4359
4360         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4361         LINVRNT(osd_invariant(obj));
4362
4363         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4364
4365         return 0;
4366 }
4367
4368 /*
4369  * Concurrency: @dt is read locked.
4370  */
4371 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4372                          struct lu_buf *buf, const char *name)
4373 {
4374         struct osd_object      *obj    = osd_dt_obj(dt);
4375         struct inode           *inode  = obj->oo_inode;
4376         struct osd_thread_info *info   = osd_oti_get(env);
4377         struct dentry          *dentry = &info->oti_obj_dentry;
4378         bool cache_xattr = false;
4379         int rc;
4380
4381         LASSERT(buf);
4382
4383         /* version get is not real XATTR but uses xattr API */
4384         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4385                 dt_obj_version_t *ver = buf->lb_buf;
4386
4387                 /*
4388                  * for version we are just using xattr API but change inode
4389                  * field instead
4390                  */
4391                 if (buf->lb_len == 0)
4392                         return sizeof(dt_obj_version_t);
4393
4394                 if (buf->lb_len < sizeof(dt_obj_version_t))
4395                         return -ERANGE;
4396
4397                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4398                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4399
4400                 *ver = LDISKFS_I(inode)->i_fs_version;
4401
4402                 return sizeof(dt_obj_version_t);
4403         }
4404
4405         if (!dt_object_exists(dt))
4406                 return -ENOENT;
4407
4408         LASSERT(!dt_object_remote(dt));
4409         LASSERT(inode->i_op != NULL);
4410 #ifdef HAVE_IOP_XATTR
4411         LASSERT(inode->i_op->getxattr != NULL);
4412 #endif
4413
4414         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4415             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4416                 cache_xattr = true;
4417
4418         if (cache_xattr) {
4419                 rc = osd_oxc_get(obj, name, buf);
4420                 if (rc != -ENOENT)
4421                         return rc;
4422         }
4423
4424         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4425                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4426                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4427                 struct filter_fid *ff;
4428                 struct ost_layout *ol;
4429
4430                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4431                 if (rc)
4432                         return rc;
4433
4434                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4435
4436                 rc = sizeof(*ff);
4437                 if (buf->lb_len == 0 || !buf->lb_buf)
4438                         return rc;
4439
4440                 if (buf->lb_len < rc)
4441                         return -ERANGE;
4442
4443                 ff = buf->lb_buf;
4444                 ol = &ff->ff_layout;
4445                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4446                                                   PFID_STRIPE_IDX_BITS);
4447                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4448                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4449                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4450                 if (lma->lma_compat & LMAC_COMP_INFO) {
4451                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4452                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4453                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4454                 } else {
4455                         ol->ol_comp_start = 0;
4456                         ol->ol_comp_end = 0;
4457                         ol->ol_comp_id = 0;
4458                 }
4459         } else {
4460                 rc = __osd_xattr_get(inode, dentry, name,
4461                                      buf->lb_buf, buf->lb_len);
4462         }
4463
4464         if (cache_xattr) {
4465                 if (rc == -ENOENT || rc == -ENODATA)
4466                         osd_oxc_add(obj, name, NULL, 0);
4467                 else if (rc > 0 && buf->lb_buf != NULL)
4468                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4469         }
4470
4471         return rc;
4472 }
4473
4474 static int osd_declare_xattr_set(const struct lu_env *env,
4475                                  struct dt_object *dt,
4476                                  const struct lu_buf *buf, const char *name,
4477                                  int fl, struct thandle *handle)
4478 {
4479         struct osd_thandle *oh;
4480         int credits = 0;
4481         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4482
4483         LASSERT(handle != NULL);
4484
4485         oh = container_of(handle, struct osd_thandle, ot_super);
4486         LASSERT(oh->ot_handle == NULL);
4487
4488         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4489                 /*
4490                  * For non-upgrading case, the LMA is set first and
4491                  * usually fit inode. But for upgrade case, the LMA
4492                  * may be in another separated EA block.
4493                  */
4494                 if (dt_object_exists(dt)) {
4495                         if (fl == LU_XATTR_REPLACE)
4496                                 credits = 1;
4497                         else
4498                                 goto upgrade;
4499                 }
4500         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4501                 credits = 1;
4502         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4503                 /* We may need to delete the old PFID EA. */
4504                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4505                 if (fl == LU_XATTR_REPLACE)
4506                         credits += 1;
4507                 else
4508                         goto upgrade;
4509         } else {
4510                 /*
4511                  * If some name entry resides on remote MDT, then will create
4512                  * agent entry under remote parent. On the other hand, if the
4513                  * remote entry will be removed, then related agent entry may
4514                  * need to be removed from the remote parent. So there may be
4515                  * kinds of cases, let's declare enough credits. The credits
4516                  * for create agent entry is enough for remove case.
4517                  */
4518                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4519                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4520                         if (dt_object_exists(dt))
4521                                 credits += 1; /* For updating LMA */
4522                 }
4523
4524 upgrade:
4525                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4526
4527                 if (buf != NULL) {
4528                         ssize_t buflen;
4529
4530                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4531                                 /*
4532                                  * learn xattr size from osd_xattr_get if
4533                                  * attribute has not been read yet
4534                                  */
4535                                 buflen = __osd_xattr_get(
4536                                     osd_dt_obj(dt)->oo_inode,
4537                                     &osd_oti_get(env)->oti_obj_dentry,
4538                                     name, NULL, 0);
4539                                 if (buflen < 0)
4540                                         buflen = 0;
4541                         } else {
4542                                 buflen = buf->lb_len;
4543                         }
4544
4545                         if (buflen > sb->s_blocksize) {
4546                                 credits += osd_calc_bkmap_credits(
4547                                     sb, NULL, 0, -1,
4548                                     (buflen + sb->s_blocksize - 1) >>
4549                                     sb->s_blocksize_bits);
4550                         }
4551                 }
4552                 /*
4553                  * xattr set may involve inode quota change, reserve credits for
4554                  * dquot_initialize()
4555                  */
4556                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4557         }
4558
4559         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4560
4561         return 0;
4562 }
4563
4564 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4565                               const struct lu_buf *buf, int fl,
4566                               struct thandle *handle)
4567 {
4568         struct osd_thread_info *info = osd_oti_get(env);
4569         struct dentry *dentry = &info->oti_obj_dentry;
4570         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4571         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4572         struct inode *inode = obj->oo_inode;
4573         struct filter_fid *ff = buf->lb_buf;
4574         struct ost_layout *ol = &ff->ff_layout;
4575         int flags = XATTR_REPLACE;
4576         int rc;
4577
4578         ENTRY;
4579
4580         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4581                 RETURN(-EINVAL);
4582
4583         rc = osd_get_lma(info, inode, dentry, loa);
4584         if (rc == -ENODATA) {
4585                 /* Usually for upgarding from old device */
4586                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4587                                 LMAC_FID_ON_OST, 0);
4588                 flags = XATTR_CREATE;
4589         } else if (rc) {
4590                 RETURN(rc);
4591         }
4592
4593         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4594                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4595                         RETURN(-EEXIST);
4596
4597                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4598                         /* Separate PFID EA from LMA */
4599                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4600                         lustre_lma_swab(lma);
4601                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4602                                              sizeof(*lma), XATTR_REPLACE);
4603                         if (!rc) {
4604                                 obj->oo_pfid_in_lma = 0;
4605                                 rc = LU_XATTR_CREATE;
4606                         }
4607
4608                         RETURN(rc);
4609                 }
4610         } else {
4611                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4612                         RETURN(fl);
4613
4614                 /*
4615                  * Old client does not send stripe information,
4616                  * then store the PFID EA on disk separatedly.
4617                  */
4618                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4619                              ol->ol_stripe_size == 0))
4620                         RETURN(fl);
4621
4622                 /* Remove old PFID EA entry firstly. */
4623                 dquot_initialize(inode);
4624                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4625                 if (rc == -ENODATA) {
4626                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4627                                 RETURN(rc);
4628                 } else if (rc) {
4629                         RETURN(rc);
4630                 }
4631         }
4632
4633         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4634         if (likely(ol->ol_stripe_size != 0)) {
4635                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4636                                              PFID_STRIPE_IDX_BITS;
4637                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4638                 lma->lma_compat |= LMAC_STRIPE_INFO;
4639                 if (ol->ol_comp_id != 0) {
4640                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4641                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4642                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4643                         lma->lma_compat |= LMAC_COMP_INFO;
4644                 }
4645         }
4646
4647         lustre_loa_swab(loa, false);
4648
4649         /* Store the PFID EA inside LMA. */
4650         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4651                              flags);
4652         if (!rc)
4653                 obj->oo_pfid_in_lma = 1;
4654
4655         RETURN(rc);
4656 }
4657
4658 /*
4659  * In DNE environment, the object (in spite of regular file or directory)
4660  * and its name entry may reside on different MDTs. Under such case, we will
4661  * create an agent entry on the MDT where the object resides. The agent entry
4662  * references the object locally, that makes the object to be visible to the
4663  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4664  * such as 'tar' can handle the object properly.
4665  *
4666  * We handle the agent entry during set linkEA that is the common interface
4667  * for both regular file and directroy, can handle kinds of cases, such as
4668  * create/link/unlink/rename, and so on.
4669  *
4670  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4671  *
4672  * XXX: There are two known issues:
4673  * 1. For one object, we will create at most one agent entry even if there
4674  *    may be more than one cross-MDTs hard links on the object. So the local
4675  *    e2fsck may claim that the object's nlink is larger than the name entries
4676  *    that reference such inode. And in further, the e2fsck will fix the nlink
4677  *    attribute to match the local references. Then it will cause the object's
4678  *    nlink attribute to be inconsistent with the global references. it is bad
4679  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4680  *    other hand, the global namespace LFSCK can repair the object's attribute
4681  *    according to the linkEA.
4682  * 2. There may be too many hard links on the object as to its linkEA overflow,
4683  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4684  *    case happened, then at this point, we do not know whether there are some
4685  *    cross-MDTs reference. But there are local references, it guarantees that
4686  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4687  */
4688 static int osd_xattr_handle_linkea(const struct lu_env *env,
4689                                    struct osd_device *osd,
4690                                    struct osd_object *obj,
4691                                    const struct lu_buf *buf,
4692                                    struct thandle *handle)
4693 {
4694         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4695         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4696         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4697         struct lu_name tmpname;
4698         struct osd_thandle *oh;
4699         int rc;
4700         bool remote = false;
4701
4702         ENTRY;
4703
4704         oh = container_of(handle, struct osd_thandle, ot_super);
4705         LASSERT(oh->ot_handle != NULL);
4706
4707         rc = linkea_init_with_rec(&ldata);
4708         if (!rc) {
4709                 linkea_first_entry(&ldata);
4710                 while (ldata.ld_lee != NULL && !remote) {
4711                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4712                                             &tmpname, tfid);
4713                         if (osd_remote_fid(env, osd, tfid) > 0)
4714                                 remote = true;
4715                         else
4716                                 linkea_next_entry(&ldata);
4717                 }
4718         } else if (rc == -ENODATA) {
4719                 rc = 0;
4720         } else {
4721                 RETURN(rc);
4722         }
4723
4724         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4725                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4726                 if (rc)
4727                         CERROR("%s: failed to remove agent entry for "DFID
4728                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4729         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4730                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4731                 if (rc)
4732                         CERROR("%s: failed to create agent entry for "DFID
4733                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4734         }
4735
4736         RETURN(rc);
4737 }
4738
4739 /*
4740  * Concurrency: @dt is write locked.
4741  */
4742 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4743                          const struct lu_buf *buf, const char *name, int fl,
4744                          struct thandle *handle)
4745 {
4746         struct osd_object *obj = osd_dt_obj(dt);
4747         struct osd_device *osd = osd_obj2dev(obj);
4748         struct inode *inode = obj->oo_inode;
4749         struct osd_thread_info *info = osd_oti_get(env);
4750         int fs_flags = 0;
4751         int len;
4752         int rc;
4753
4754         ENTRY;
4755
4756         LASSERT(handle);
4757         LASSERT(buf);
4758
4759         /* version set is not real XATTR */
4760         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4761                 dt_obj_version_t *version = buf->lb_buf;
4762
4763                 /*
4764                  * for version we are just using xattr API but change inode
4765                  * field instead
4766                  */
4767                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4768
4769                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4770                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4771
4772                 LDISKFS_I(inode)->i_fs_version = *version;
4773                 /*
4774                  * Version is set after all inode operations are finished,
4775                  * so we should mark it dirty here
4776                  */
4777                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4778
4779                 RETURN(0);
4780         }
4781
4782         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4783                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4784
4785         len = buf->lb_len;
4786         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4787
4788         /*
4789          * For the OST device with 256 bytes inode size by default,
4790          * the PFID EA will be stored together with LMA EA to avoid
4791          * performance trouble. Otherwise the PFID EA can be stored
4792          * independently. LU-8998
4793          */
4794         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4795             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4796                 LASSERT(buf->lb_buf);
4797
4798                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4799                 if (fl <= 0)
4800                         RETURN(fl);
4801         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4802                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4803                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4804
4805                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4806                 if (rc)
4807                         RETURN(rc);
4808
4809                 lma->lma_incompat |= LMAI_STRIPED;
4810                 lustre_lma_swab(lma);
4811                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4812                                      sizeof(*lma), XATTR_REPLACE);
4813                 if (rc != 0)
4814                         RETURN(rc);
4815         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4816                 LASSERT(!osd->od_is_ost);
4817
4818                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4819                 if (rc)
4820                         RETURN(rc);
4821         }
4822
4823         if (fl & LU_XATTR_REPLACE)
4824                 fs_flags |= XATTR_REPLACE;
4825
4826         if (fl & LU_XATTR_CREATE)
4827                 fs_flags |= XATTR_CREATE;
4828
4829         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4830         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4831
4832         if (rc == 0 &&
4833             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4834              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4835                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4836
4837         return rc;
4838 }
4839
4840 /*
4841  * Concurrency: @dt is read locked.
4842  */
4843 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4844                           const struct lu_buf *buf)
4845 {
4846         struct osd_object *obj = osd_dt_obj(dt);
4847         struct inode *inode = obj->oo_inode;
4848         struct osd_thread_info *info = osd_oti_get(env);
4849         struct dentry *dentry = &info->oti_obj_dentry;
4850
4851         if (!dt_object_exists(dt))
4852                 return -ENOENT;
4853
4854         LASSERT(!dt_object_remote(dt));
4855         LASSERT(inode->i_op != NULL);
4856         LASSERT(inode->i_op->listxattr != NULL);
4857
4858         dentry->d_inode = inode;
4859         dentry->d_sb = inode->i_sb;
4860         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4861 }
4862
4863 static int osd_declare_xattr_del(const struct lu_env *env,
4864                                  struct dt_object *dt, const char *name,
4865                                  struct thandle *handle)
4866 {
4867         struct osd_thandle *oh;
4868         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4869
4870         LASSERT(!dt_object_remote(dt));
4871         LASSERT(handle != NULL);
4872
4873         oh = container_of(handle, struct osd_thandle, ot_super);
4874         LASSERT(oh->ot_handle == NULL);
4875
4876         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4877                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4878         /*
4879          * xattr del may involve inode quota change, reserve credits for
4880          * dquot_initialize()
4881          */
4882         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4883
4884         return 0;
4885 }
4886
4887 /*
4888  * Concurrency: @dt is write locked.
4889  */
4890 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4891                          const char *name, struct thandle *handle)
4892 {
4893         struct osd_object *obj = osd_dt_obj(dt);
4894         struct inode *inode = obj->oo_inode;
4895         struct osd_thread_info *info = osd_oti_get(env);
4896         struct dentry *dentry = &info->oti_obj_dentry;
4897         int rc;
4898
4899         if (!dt_object_exists(dt))
4900                 return -ENOENT;
4901
4902         LASSERT(!dt_object_remote(dt));
4903         LASSERT(inode->i_op != NULL);
4904         LASSERT(handle != NULL);
4905 #ifdef HAVE_IOP_XATTR
4906         LASSERT(inode->i_op->removexattr != NULL);
4907 #endif
4908
4909         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4910
4911         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4912                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4913
4914                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4915                                  &info->oti_ost_attrs);
4916                 if (!rc) {
4917                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4918
4919                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4920                         lustre_lma_swab(lma);
4921                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4922                                              sizeof(*lma), XATTR_REPLACE);
4923                         if (!rc)
4924                                 obj->oo_pfid_in_lma = 0;
4925                 }
4926         } else {
4927                 dquot_initialize(inode);
4928                 dentry->d_inode = inode;
4929                 dentry->d_sb = inode->i_sb;
4930                 rc = ll_vfs_removexattr(dentry, inode, name);
4931         }
4932
4933         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4934
4935         if (rc == 0 &&
4936             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4937              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4938                 osd_oxc_del(obj, name);
4939
4940         return rc;
4941 }
4942
4943 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4944                            __u64 start, __u64 end)
4945 {
4946         struct osd_object *obj = osd_dt_obj(dt);
4947         struct inode *inode = obj->oo_inode;
4948         struct file *file = osd_quasi_file(env, inode);
4949         int rc;
4950
4951         ENTRY;
4952
4953         rc = vfs_fsync_range(file, start, end, 0);
4954
4955         RETURN(rc);
4956 }
4957
4958 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4959 {
4960         return 0;
4961 }
4962
4963 static bool osd_check_stale(struct dt_object *dt)
4964 {
4965         return false;
4966 }
4967 /*
4968  * Index operations.
4969  */
4970
4971 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4972                                const struct dt_index_features *feat)
4973 {
4974         struct iam_descr *descr;
4975
4976         if (osd_object_is_root(o))
4977                 return feat == &dt_directory_features;
4978
4979         LASSERT(o->oo_dir != NULL);
4980
4981         descr = o->oo_dir->od_container.ic_descr;
4982         if (feat == &dt_directory_features) {
4983                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4984                         return 1;
4985                 else
4986                         return 0;
4987         } else {
4988                 return feat->dif_keysize_min <= descr->id_key_size &&
4989                        descr->id_key_size <= feat->dif_keysize_max &&
4990                        feat->dif_recsize_min <= descr->id_rec_size &&
4991                        descr->id_rec_size <= feat->dif_recsize_max &&
4992                        !(feat->dif_flags & (DT_IND_VARKEY |
4993                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4994                        ergo(feat->dif_flags & DT_IND_UPDATE,
4995                             1 /* XXX check that object (and fs) is writable */);
4996         }
4997 }
4998
4999 static int osd_iam_container_init(const struct lu_env *env,
5000                                   struct osd_object *obj,
5001                                   struct osd_directory *dir)
5002 {
5003         struct iam_container *bag = &dir->od_container;
5004         int result;
5005
5006         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
5007         if (result != 0)
5008                 return result;
5009
5010         result = iam_container_setup(bag);
5011         if (result == 0)
5012                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
5013         else
5014                 iam_container_fini(bag);
5015
5016         return result;
5017 }
5018
5019
5020 /*
5021  * Concurrency: no external locking is necessary.
5022  */
5023 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
5024                          const struct dt_index_features *feat)
5025 {
5026         int result;
5027         int skip_iam = 0;
5028         struct osd_object *obj = osd_dt_obj(dt);
5029
5030         LINVRNT(osd_invariant(obj));
5031
5032         if (osd_object_is_root(obj)) {
5033                 dt->do_index_ops = &osd_index_ea_ops;
5034                 result = 0;
5035         } else if (feat == &dt_directory_features) {
5036                 dt->do_index_ops = &osd_index_ea_ops;
5037                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
5038                         result = 0;
5039                 else
5040                         result = -ENOTDIR;
5041                 skip_iam = 1;
5042         } else if (unlikely(feat == &dt_otable_features)) {
5043                 dt->do_index_ops = &osd_otable_ops;
5044                 return 0;
5045         } else if (unlikely(feat == &dt_acct_features)) {
5046                 dt->do_index_ops = &osd_acct_index_ops;
5047                 result = 0;
5048                 skip_iam = 1;
5049         } else if (!osd_has_index(obj)) {
5050                 struct osd_directory *dir;
5051                 struct osd_device *osd = osd_obj2dev(obj);
5052                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
5053
5054                 OBD_ALLOC_PTR(dir);
5055                 if (dir) {
5056
5057                         spin_lock(&obj->oo_guard);
5058                         if (obj->oo_dir == NULL)
5059                                 obj->oo_dir = dir;
5060                         else
5061                                 /*
5062                                  * Concurrent thread allocated container data.
5063                                  */
5064                                 OBD_FREE_PTR(dir);
5065                         spin_unlock(&obj->oo_guard);
5066                         /*
5067                          * Now, that we have container data, serialize its
5068                          * initialization.
5069                          */
5070                         down_write(&obj->oo_ext_idx_sem);
5071                         /*
5072                          * recheck under lock.
5073                          */
5074
5075                         if (osd_has_index(obj)) {
5076                                 result = 0;
5077                                 goto unlock;
5078                         }
5079
5080                         result = osd_iam_container_init(env, obj, obj->oo_dir);
5081                         if (result || feat == &dt_lfsck_namespace_features ||
5082                             feat == &dt_lfsck_layout_orphan_features ||
5083                             feat == &dt_lfsck_layout_dangling_features)
5084                                 goto unlock;
5085
5086                         result = osd_index_register(osd, fid,
5087                                                     feat->dif_keysize_max,
5088                                                     feat->dif_recsize_max);
5089                         if (result < 0)
5090                                 CWARN("%s: failed to register index "
5091                                       DFID": rc = %d\n",
5092                                       osd_name(osd), PFID(fid), result);
5093                         else if (result > 0)
5094                                 result = 0;
5095                         else
5096                                 CDEBUG(D_LFSCK, "%s: index object "DFID
5097                                        " (%d/%d) registered\n",
5098                                        osd_name(osd), PFID(fid),
5099                                        (int)feat->dif_keysize_max,
5100                                        (int)feat->dif_recsize_max);
5101
5102 unlock:
5103                         up_write(&obj->oo_ext_idx_sem);
5104                 } else {
5105                         result = -ENOMEM;
5106                 }
5107         } else {
5108                 result = 0;
5109         }
5110
5111         if (result == 0 && skip_iam == 0) {
5112                 if (!osd_iam_index_probe(env, obj, feat))
5113                         result = -ENOTDIR;
5114         }
5115         LINVRNT(osd_invariant(obj));
5116
5117         return result;
5118 }
5119
5120 static int osd_otable_it_attr_get(const struct lu_env *env,
5121                                  struct dt_object *dt,
5122                                  struct lu_attr *attr)
5123 {
5124         attr->la_valid = 0;
5125         return 0;
5126 }
5127
5128 static const struct dt_object_operations osd_obj_ops = {
5129         .do_read_lock           = osd_read_lock,
5130         .do_write_lock          = osd_write_lock,
5131         .do_read_unlock         = osd_read_unlock,
5132         .do_write_unlock        = osd_write_unlock,
5133         .do_write_locked        = osd_write_locked,
5134         .do_attr_get            = osd_attr_get,
5135         .do_declare_attr_set    = osd_declare_attr_set,
5136         .do_attr_set            = osd_attr_set,
5137         .do_ah_init             = osd_ah_init,
5138         .do_declare_create      = osd_declare_create,
5139         .do_create              = osd_create,
5140         .do_declare_destroy     = osd_declare_destroy,
5141         .do_destroy             = osd_destroy,
5142         .do_index_try           = osd_index_try,
5143         .do_declare_ref_add     = osd_declare_ref_add,
5144         .do_ref_add             = osd_ref_add,
5145         .do_declare_ref_del     = osd_declare_ref_del,
5146         .do_ref_del             = osd_ref_del,
5147         .do_xattr_get           = osd_xattr_get,
5148         .do_declare_xattr_set   = osd_declare_xattr_set,
5149         .do_xattr_set           = osd_xattr_set,
5150         .do_declare_xattr_del   = osd_declare_xattr_del,
5151         .do_xattr_del           = osd_xattr_del,
5152         .do_xattr_list          = osd_xattr_list,
5153         .do_object_sync         = osd_object_sync,
5154         .do_invalidate          = osd_invalidate,
5155         .do_check_stale         = osd_check_stale,
5156 };
5157
5158 static const struct dt_object_operations osd_obj_otable_it_ops = {
5159         .do_attr_get    = osd_otable_it_attr_get,
5160         .do_index_try   = osd_index_try,
5161 };
5162
5163 static int osd_index_declare_iam_delete(const struct lu_env *env,
5164                                         struct dt_object *dt,
5165                                         const struct dt_key *key,
5166                                         struct thandle *handle)
5167 {
5168         struct osd_thandle *oh;
5169
5170         oh = container_of(handle, struct osd_thandle, ot_super);
5171         LASSERT(oh->ot_handle == NULL);
5172
5173         /* Recycle  may cause additional three blocks to be changed. */
5174         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5175                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5176
5177         return 0;
5178 }
5179
5180 /**
5181  *      delete a (key, value) pair from index \a dt specified by \a key
5182  *
5183  *      \param  dt      osd index object
5184  *      \param  key     key for index
5185  *      \param  rec     record reference
5186  *      \param  handle  transaction handler
5187  *
5188  *      \retval  0  success
5189  *      \retval -ve   failure
5190  */
5191 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5192                                 const struct dt_key *key,
5193                                 struct thandle *handle)
5194 {
5195         struct osd_thread_info *oti = osd_oti_get(env);
5196         struct osd_object *obj = osd_dt_obj(dt);
5197         struct osd_thandle *oh;
5198         struct iam_path_descr *ipd;
5199         struct iam_container *bag = &obj->oo_dir->od_container;
5200         int rc;
5201
5202         ENTRY;
5203
5204         if (!dt_object_exists(dt))
5205                 RETURN(-ENOENT);
5206
5207         LINVRNT(osd_invariant(obj));
5208         LASSERT(!dt_object_remote(dt));
5209         LASSERT(bag->ic_object == obj->oo_inode);
5210         LASSERT(handle != NULL);
5211
5212         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5213
5214         ipd = osd_idx_ipd_get(env, bag);
5215         if (unlikely(ipd == NULL))
5216                 RETURN(-ENOMEM);
5217
5218         oh = container_of(handle, struct osd_thandle, ot_super);
5219         LASSERT(oh->ot_handle != NULL);
5220         LASSERT(oh->ot_handle->h_transaction != NULL);
5221
5222         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5223                 /* swab quota uid/gid provided by caller */
5224                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5225                 key = (const struct dt_key *)&oti->oti_quota_id;
5226         }
5227
5228         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5229         osd_ipd_put(env, bag, ipd);
5230         LINVRNT(osd_invariant(obj));
5231         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5232         RETURN(rc);
5233 }
5234
5235 static int osd_index_declare_ea_delete(const struct lu_env *env,
5236                                        struct dt_object *dt,
5237                                        const struct dt_key *key,
5238                                        struct thandle *handle)
5239 {
5240         struct osd_thandle *oh;
5241         struct inode *inode;
5242         int rc, credits;
5243
5244         ENTRY;
5245
5246         LASSERT(!dt_object_remote(dt));
5247         LASSERT(handle != NULL);
5248
5249         oh = container_of(handle, struct osd_thandle, ot_super);
5250         LASSERT(oh->ot_handle == NULL);
5251
5252         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5253         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5254
5255         inode = osd_dt_obj(dt)->oo_inode;
5256         if (inode == NULL)
5257                 RETURN(-ENOENT);
5258
5259         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5260                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5261                                    NULL, OSD_QID_BLK);
5262         RETURN(rc);
5263 }
5264
5265 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5266                                           struct dt_rec *fid)
5267 {
5268         struct osd_fid_pack *rec;
5269         int rc = -ENODATA;
5270
5271         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5272                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5273                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5274                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5275                         rc = -EINVAL;
5276         }
5277         return rc;
5278 }
5279
5280 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5281                           const struct lu_fid *fid)
5282 {
5283         struct seq_server_site *ss = osd_seq_site(osd);
5284
5285         ENTRY;
5286
5287         /* FID seqs not in FLDB, must be local seq */
5288         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5289                 RETURN(0);
5290
5291         /*
5292          * If FLD is not being initialized yet, it only happens during the
5293          * initialization, likely during mgs initialization, and we assume
5294          * this is local FID.
5295          */
5296         if (ss == NULL || ss->ss_server_fld == NULL)
5297                 RETURN(0);
5298
5299         /* Only check the local FLDB here */
5300         if (osd_seq_exists(env, osd, fid_seq(fid)))
5301                 RETURN(0);
5302
5303         RETURN(1);
5304 }
5305
5306 static void osd_take_care_of_agent(const struct lu_env *env,
5307                                    struct osd_device *osd,
5308                                    struct osd_thandle *oh,
5309                                    struct ldiskfs_dir_entry_2 *de)
5310 {
5311         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5312         struct osd_idmap_cache *idc;
5313         int rc, schedule = 0;
5314
5315         LASSERT(de != NULL);
5316
5317         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5318         if (likely(rc == 0)) {
5319                 idc = osd_idc_find_or_init(env, osd, fid);
5320                 if (IS_ERR(idc) || idc->oic_remote)
5321                         schedule = 1;
5322         } else if (rc == -ENODATA) {
5323                 /*
5324                  * can't get FID, postpone to the end of the
5325                  * transaction when iget() is safe
5326                  */
5327                 schedule = 1;
5328         } else {
5329                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5330         }
5331         if (schedule)
5332                 osd_schedule_agent_inode_removal(env, oh,
5333                                                  le32_to_cpu(de->inode));
5334 }
5335
5336 /**
5337  * Index delete function for interoperability mode (b11826).
5338  * It will remove the directory entry added by osd_index_ea_insert().
5339  * This entry is needed to maintain name->fid mapping.
5340  *
5341  * \param key,  key i.e. file entry to be deleted
5342  *
5343  * \retval   0, on success
5344  * \retval -ve, on error
5345  */
5346 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5347                                const struct dt_key *key, struct thandle *handle)
5348 {
5349         struct osd_object *obj = osd_dt_obj(dt);
5350         struct inode *dir = obj->oo_inode;
5351         struct dentry *dentry;
5352         struct osd_thandle *oh;
5353         struct ldiskfs_dir_entry_2 *de = NULL;
5354         struct buffer_head *bh;
5355         struct htree_lock *hlock = NULL;
5356         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5357         int rc;
5358
5359         ENTRY;
5360
5361         if (!dt_object_exists(dt))
5362                 RETURN(-ENOENT);
5363
5364         LINVRNT(osd_invariant(obj));
5365         LASSERT(!dt_object_remote(dt));
5366         LASSERT(handle != NULL);
5367
5368         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5369
5370         oh = container_of(handle, struct osd_thandle, ot_super);
5371         LASSERT(oh->ot_handle != NULL);
5372         LASSERT(oh->ot_handle->h_transaction != NULL);
5373
5374         dquot_initialize(dir);
5375         dentry = osd_child_dentry_get(env, obj,
5376                                       (char *)key, strlen((char *)key));
5377
5378         if (obj->oo_hl_head != NULL) {
5379                 hlock = osd_oti_get(env)->oti_hlock;
5380                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5381                                    dir, LDISKFS_HLOCK_DEL);
5382         } else {
5383                 down_write(&obj->oo_ext_idx_sem);
5384         }
5385
5386         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5387         if (!IS_ERR(bh)) {
5388                 /*
5389                  * If this is not the ".." entry, it might be a remote DNE
5390                  * entry and  we need to check if the FID is for a remote
5391                  * MDT.  If the FID is  not in the directory entry (e.g.
5392                  * upgraded 1.8 filesystem without dirdata enabled) then
5393                  * we need to get the FID from the LMA. For a remote directory
5394                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5395                  * case.
5396                  *
5397                  * Delete the entry before the agent inode in order to
5398                  * simplify error handling.  At worst an error after deleting
5399                  * the entry first might leak the agent inode afterward. The
5400                  * reverse would need filesystem abort in case of error deleting
5401                  * the entry after the agent had been removed, or leave a
5402                  * dangling entry pointing at a random inode.
5403                  */
5404                 if (strcmp((char *)key, dotdot) != 0)
5405                         osd_take_care_of_agent(env, osd, oh, de);
5406                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5407                 brelse(bh);
5408         } else {
5409                 rc = PTR_ERR(bh);
5410         }
5411
5412         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5413             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5414                 /* NB, dirent count may not be accurate, because it's counted
5415                  * without lock.
5416                  */
5417                 if (obj->oo_dirent_count)
5418                         obj->oo_dirent_count--;
5419                 else
5420                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5421         }
5422         if (hlock != NULL)
5423                 ldiskfs_htree_unlock(hlock);
5424         else
5425                 up_write(&obj->oo_ext_idx_sem);
5426         GOTO(out, rc);
5427 out:
5428         LASSERT(osd_invariant(obj));
5429         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5430         RETURN(rc);
5431 }
5432
5433 /**
5434  *      Lookup index for \a key and copy record to \a rec.
5435  *
5436  *      \param  dt      osd index object
5437  *      \param  key     key for index
5438  *      \param  rec     record reference
5439  *
5440  *      \retval  +ve  success : exact mach
5441  *      \retval  0    return record with key not greater than \a key
5442  *      \retval -ve   failure
5443  */
5444 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5445                                 struct dt_rec *rec, const struct dt_key *key)
5446 {
5447         struct osd_object *obj = osd_dt_obj(dt);
5448         struct iam_path_descr *ipd;
5449         struct iam_container *bag = &obj->oo_dir->od_container;
5450         struct osd_thread_info *oti = osd_oti_get(env);
5451         struct iam_iterator *it = &oti->oti_idx_it;
5452         struct iam_rec *iam_rec;
5453         int rc;
5454
5455         ENTRY;
5456
5457         if (!dt_object_exists(dt))
5458                 RETURN(-ENOENT);
5459
5460         LASSERT(osd_invariant(obj));
5461         LASSERT(!dt_object_remote(dt));
5462         LASSERT(bag->ic_object == obj->oo_inode);
5463
5464         ipd = osd_idx_ipd_get(env, bag);
5465         if (IS_ERR(ipd))
5466                 RETURN(-ENOMEM);
5467
5468         /* got ipd now we can start iterator. */
5469         iam_it_init(it, bag, 0, ipd);
5470
5471         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5472                 /* swab quota uid/gid provided by caller */
5473                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5474                 key = (const struct dt_key *)&oti->oti_quota_id;
5475         }
5476
5477         rc = iam_it_get(it, (struct iam_key *)key);
5478         if (rc >= 0) {
5479                 if (S_ISDIR(obj->oo_inode->i_mode))
5480                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5481                 else
5482                         iam_rec = (struct iam_rec *)rec;
5483
5484                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5485
5486                 if (S_ISDIR(obj->oo_inode->i_mode))
5487                         osd_fid_unpack((struct lu_fid *)rec,
5488                                        (struct osd_fid_pack *)iam_rec);
5489                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5490                         osd_quota_unpack(obj, rec);
5491         }
5492
5493         iam_it_put(it);
5494         iam_it_fini(it);
5495         osd_ipd_put(env, bag, ipd);
5496
5497         LINVRNT(osd_invariant(obj));
5498
5499         RETURN(rc);
5500 }
5501
5502 static int osd_index_declare_iam_insert(const struct lu_env *env,
5503                                         struct dt_object *dt,
5504                                         const struct dt_rec *rec,
5505                                         const struct dt_key *key,
5506                                         struct thandle *handle)
5507 {
5508         struct osd_thandle *oh;
5509
5510         LASSERT(handle != NULL);
5511
5512         oh = container_of(handle, struct osd_thandle, ot_super);
5513         LASSERT(oh->ot_handle == NULL);
5514
5515         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5516                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5517
5518         return 0;
5519 }
5520
5521 /**
5522  *      Inserts (key, value) pair in \a dt index object.
5523  *
5524  *      \param  dt      osd index object
5525  *      \param  key     key for index
5526  *      \param  rec     record reference
5527  *      \param  th      transaction handler
5528  *
5529  *      \retval  0  success
5530  *      \retval -ve failure
5531  */
5532 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5533                                 const struct dt_rec *rec,
5534                                 const struct dt_key *key, struct thandle *th)
5535 {
5536         struct osd_object *obj = osd_dt_obj(dt);
5537         struct iam_path_descr *ipd;
5538         struct osd_thandle *oh;
5539         struct iam_container *bag;
5540         struct osd_thread_info *oti = osd_oti_get(env);
5541         struct iam_rec *iam_rec;
5542         int rc;
5543
5544         ENTRY;
5545
5546         if (!dt_object_exists(dt))
5547                 RETURN(-ENOENT);
5548
5549         LINVRNT(osd_invariant(obj));
5550         LASSERT(!dt_object_remote(dt));
5551
5552         bag = &obj->oo_dir->od_container;
5553         LASSERT(bag->ic_object == obj->oo_inode);
5554         LASSERT(th != NULL);
5555
5556         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5557
5558         ipd = osd_idx_ipd_get(env, bag);
5559         if (unlikely(ipd == NULL))
5560                 RETURN(-ENOMEM);
5561
5562         oh = container_of(th, struct osd_thandle, ot_super);
5563         LASSERT(oh->ot_handle != NULL);
5564         LASSERT(oh->ot_handle->h_transaction != NULL);
5565         if (S_ISDIR(obj->oo_inode->i_mode)) {
5566                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5567                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5568                              &oti->oti_fid);
5569         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5570                 /* pack quota uid/gid */
5571                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5572                 key = (const struct dt_key *)&oti->oti_quota_id;
5573                 /* pack quota record */
5574                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5575                 iam_rec = (struct iam_rec *)rec;
5576         } else {
5577                 iam_rec = (struct iam_rec *)rec;
5578         }
5579
5580         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5581                         iam_rec, ipd);
5582         osd_ipd_put(env, bag, ipd);
5583         LINVRNT(osd_invariant(obj));
5584         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5585         RETURN(rc);
5586 }
5587
5588 /**
5589  * Calls ldiskfs_add_entry() to add directory entry
5590  * into the directory. This is required for
5591  * interoperability mode (b11826)
5592  *
5593  * \retval   0, on success
5594  * \retval -ve, on error
5595  */
5596 static int __osd_ea_add_rec(struct osd_thread_info *info,
5597                             struct osd_object *pobj, struct inode  *cinode,
5598                             const char *name, const struct lu_fid *fid,
5599                             struct htree_lock *hlock, struct thandle *th)
5600 {
5601         struct ldiskfs_dentry_param *ldp;
5602         struct dentry *child;
5603         struct osd_thandle *oth;
5604         int rc;
5605
5606         oth = container_of(th, struct osd_thandle, ot_super);
5607         LASSERT(oth->ot_handle != NULL);
5608         LASSERT(oth->ot_handle->h_transaction != NULL);
5609         LASSERT(pobj->oo_inode);
5610
5611         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5612         if (unlikely(osd_object_is_root(pobj)))
5613                 ldp->edp_magic = 0;
5614         else
5615                 osd_get_ldiskfs_dirent_param(ldp, fid);
5616         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5617         child->d_fsdata = (void *)ldp;
5618         dquot_initialize(pobj->oo_inode);
5619         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5620                                    child, cinode, hlock);
5621         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5622                 struct ldiskfs_dir_entry_2      *de;
5623                 struct buffer_head              *bh;
5624                 int                              rc1;
5625
5626                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5627                                             NULL, hlock);
5628                 if (!IS_ERR(bh)) {
5629                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5630                                                                bh);
5631                         if (rc1 == 0) {
5632                                 if (S_ISDIR(cinode->i_mode))
5633                                         de->file_type = LDISKFS_DIRENT_LUFID |
5634                                                         LDISKFS_FT_REG_FILE;
5635                                 else
5636                                         de->file_type = LDISKFS_DIRENT_LUFID |
5637                                                         LDISKFS_FT_DIR;
5638                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5639                                                               NULL, bh);
5640                         }
5641                         brelse(bh);
5642                 }
5643         }
5644
5645         RETURN(rc);
5646 }
5647
5648 /**
5649  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5650  * into the directory.Also sets flags into osd object to
5651  * indicate dot and dotdot are created. This is required for
5652  * interoperability mode (b11826)
5653  *
5654  * \param dir   directory for dot and dotdot fixup.
5655  * \param obj   child object for linking
5656  *
5657  * \retval   0, on success
5658  * \retval -ve, on error
5659  */
5660 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5661                               struct osd_object *dir,
5662                               struct inode *parent_dir, const char *name,
5663                               const struct lu_fid *dot_fid,
5664                               const struct lu_fid *dot_dot_fid,
5665                               struct thandle *th)
5666 {
5667         struct inode *inode = dir->oo_inode;
5668         struct osd_thandle *oth;
5669         int result = 0;
5670
5671         oth = container_of(th, struct osd_thandle, ot_super);
5672         LASSERT(oth->ot_handle->h_transaction != NULL);
5673         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5674
5675         if (strcmp(name, dot) == 0) {
5676                 if (dir->oo_compat_dot_created) {
5677                         result = -EEXIST;
5678                 } else {
5679                         LASSERT(inode->i_ino == parent_dir->i_ino);
5680                         dir->oo_compat_dot_created = 1;
5681                         result = 0;
5682                 }
5683         } else if (strcmp(name, dotdot) == 0) {
5684                 if (!dir->oo_compat_dot_created)
5685                         return -EINVAL;
5686                 /* in case of rename, dotdot is already created */
5687                 if (dir->oo_compat_dotdot_created) {
5688                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5689                                                 dot_dot_fid, NULL, th);
5690                 }
5691
5692                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5693                         struct lu_fid tfid = *dot_dot_fid;
5694
5695                         tfid.f_oid--;
5696                         result = osd_add_dot_dotdot_internal(info,
5697                                         dir->oo_inode, parent_dir, dot_fid,
5698                                         &tfid, oth);
5699                 } else {
5700                         result = osd_add_dot_dotdot_internal(info,
5701                                         dir->oo_inode, parent_dir, dot_fid,
5702                                         dot_dot_fid, oth);
5703                 }
5704
5705                 if (result == 0)
5706                         dir->oo_compat_dotdot_created = 1;
5707         }
5708
5709         return result;
5710 }
5711
5712
5713 /**
5714  * It will call the appropriate osd_add* function and return the
5715  * value, return by respective functions.
5716  */
5717 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5718                           struct inode *cinode, const char *name,
5719                           const struct lu_fid *fid, struct thandle *th)
5720 {
5721         struct osd_thread_info *info = osd_oti_get(env);
5722         struct htree_lock *hlock;
5723         int rc;
5724
5725         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5726
5727         if (name[0] == '.' && (name[1] == '\0' ||
5728                                (name[1] == '.' && name[2] == '\0'))) {
5729                 if (hlock != NULL) {
5730                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5731                                            pobj->oo_inode, 0);
5732                 } else {
5733                         down_write(&pobj->oo_ext_idx_sem);
5734                 }
5735
5736                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5737                                         lu_object_fid(&pobj->oo_dt.do_lu),
5738                                         fid, th);
5739         } else {
5740                 if (hlock != NULL) {
5741                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5742                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5743                 } else {
5744                         down_write(&pobj->oo_ext_idx_sem);
5745                 }
5746
5747                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5748                         struct lu_fid *tfid = &info->oti_fid;
5749
5750                         *tfid = *fid;
5751                         tfid->f_ver = ~0;
5752                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5753                                               tfid, hlock, th);
5754                 } else {
5755                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5756                                               hlock, th);
5757                 }
5758         }
5759         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5760             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5761                 pobj->oo_dirent_count++;
5762
5763         if (hlock != NULL)
5764                 ldiskfs_htree_unlock(hlock);
5765         else
5766                 up_write(&pobj->oo_ext_idx_sem);
5767
5768         return rc;
5769 }
5770
5771 static int
5772 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5773                       struct osd_idmap_cache *oic)
5774 {
5775         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5776         struct lu_fid *fid = &oic->oic_fid;
5777         struct osd_inode_id *id = &oic->oic_lid;
5778         struct inode *inode = NULL;
5779         int once = 0;
5780         bool insert;
5781         int rc;
5782
5783         ENTRY;
5784
5785         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5786                 RETURN(0);
5787
5788         if (scrub->os_running && scrub->os_pos_current > id->oii_ino)
5789                 RETURN(0);
5790
5791         if (dev->od_auto_scrub_interval == AS_NEVER ||
5792             ktime_get_real_seconds() <
5793             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5794                 RETURN(0);
5795
5796 again:
5797         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5798         if (rc == -ENOENT) {
5799                 __u32 gen = id->oii_gen;
5800
5801                 insert = true;
5802                 if (inode != NULL)
5803                         goto trigger;
5804
5805                 inode = osd_iget(oti, dev, id);
5806                 /* The inode has been removed (by race maybe). */
5807                 if (IS_ERR(inode)) {
5808                         rc = PTR_ERR(inode);
5809
5810                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5811                 }
5812
5813                 /* The OI mapping is lost. */
5814                 if (gen != OSD_OII_NOGEN)
5815                         goto trigger;
5816
5817                 /*
5818                  * The inode may has been reused by others, we do not know,
5819                  * leave it to be handled by subsequent osd_fid_lookup().
5820                  */
5821                 GOTO(out, rc = 0);
5822         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5823                 GOTO(out, rc);
5824         }
5825
5826         insert = false;
5827
5828 trigger:
5829         if (scrub->os_running) {
5830                 if (inode == NULL) {
5831                         inode = osd_iget(oti, dev, id);
5832                         /* The inode has been removed (by race maybe). */
5833                         if (IS_ERR(inode)) {
5834                                 rc = PTR_ERR(inode);
5835
5836                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5837                         }
5838                 }
5839
5840                 rc = osd_oii_insert(dev, oic, insert);
5841                 /*
5842                  * There is race condition between osd_oi_lookup and OI scrub.
5843                  * The OI scrub finished just after osd_oi_lookup() failure.
5844                  * Under such case, it is unnecessary to trigger OI scrub again,
5845                  * but try to call osd_oi_lookup() again.
5846                  */
5847                 if (unlikely(rc == -EAGAIN))
5848                         goto again;
5849
5850                 if (!S_ISDIR(inode->i_mode))
5851                         rc = 0;
5852                 else
5853                         rc = osd_check_lmv(oti, dev, inode, oic);
5854
5855                 GOTO(out, rc);
5856         }
5857
5858         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5859                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5860                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5861                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5862                        "%s: trigger partial OI scrub for RPC inconsistency checking FID "DFID": rc = %d\n",
5863                        osd_dev2name(dev), PFID(fid), rc);
5864                 if (rc == 0 || rc == -EALREADY)
5865                         goto again;
5866         }
5867
5868         GOTO(out, rc);
5869
5870 out:
5871         iput(inode);
5872
5873         RETURN(rc);
5874 }
5875
5876 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5877                                struct osd_device *dev,
5878                                struct osd_idmap_cache *oic,
5879                                struct lu_fid *fid, __u32 ino)
5880 {
5881         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5882         struct inode *inode;
5883         int rc;
5884
5885         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5886         inode = osd_iget(oti, dev, &oic->oic_lid);
5887         if (IS_ERR(inode)) {
5888                 fid_zero(&oic->oic_fid);
5889                 return PTR_ERR(inode);
5890         }
5891
5892         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5893         iput(inode);
5894         if (rc != 0)
5895                 fid_zero(&oic->oic_fid);
5896         else
5897                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5898         return rc;
5899 }
5900
5901 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5902                       struct osd_inode_id *id, const struct lu_fid *fid)
5903 {
5904         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5905                id->oii_ino, id->oii_gen, info);
5906         info->oti_cache.oic_lid = *id;
5907         info->oti_cache.oic_fid = *fid;
5908         info->oti_cache.oic_dev = osd;
5909 }
5910
5911 /**
5912  * Get parent FID from the linkEA.
5913  *
5914  * For a directory which parent resides on remote MDT, to satisfy the
5915  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5916  * the other hand, to make the lookup(..) on the directory can return
5917  * the real parent FID, we append the real parent FID after its ".."
5918  * name entry in the /REMOTE_PARENT_DIR.
5919  *
5920  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5921  * backup. So after the restore, we cannot get the right parent FID from
5922  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5923  * we have stored the real parent FID in the directory object's linkEA,
5924  * we can parse the linkEA for the real parent FID.
5925  *
5926  * \param[in] env       pointer to the thread context
5927  * \param[in] obj       pointer to the object to be handled
5928  * \param[out]fid       pointer to the buffer to hold the parent FID
5929  *
5930  * \retval              0 for getting the real parent FID successfully
5931  * \retval              negative error number on failure
5932  */
5933 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5934                                     struct osd_object *obj,
5935                                     struct lu_fid *fid)
5936 {
5937         struct osd_thread_info *oti = osd_oti_get(env);
5938         struct lu_buf *buf = &oti->oti_big_buf;
5939         struct dentry *dentry = &oti->oti_obj_dentry;
5940         struct inode *inode = obj->oo_inode;
5941         struct linkea_data ldata = { NULL };
5942         int rc;
5943
5944         ENTRY;
5945
5946         fid_zero(fid);
5947         if (!S_ISDIR(inode->i_mode))
5948                 RETURN(-EIO);
5949
5950 again:
5951         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5952                              buf->lb_buf, buf->lb_len);
5953         if (rc == -ERANGE) {
5954                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5955                                      NULL, 0);
5956                 if (rc > 0) {
5957                         lu_buf_realloc(buf, rc);
5958                         if (buf->lb_buf == NULL)
5959                                 RETURN(-ENOMEM);
5960
5961                         goto again;
5962                 }
5963         }
5964
5965         if (unlikely(rc == 0))
5966                 RETURN(-ENODATA);
5967
5968         if (rc < 0)
5969                 RETURN(rc);
5970
5971         if (unlikely(buf->lb_buf == NULL)) {
5972                 lu_buf_realloc(buf, rc);
5973                 if (buf->lb_buf == NULL)
5974                         RETURN(-ENOMEM);
5975
5976                 goto again;
5977         }
5978
5979         ldata.ld_buf = buf;
5980         rc = linkea_init_with_rec(&ldata);
5981         if (!rc) {
5982                 linkea_first_entry(&ldata);
5983                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5984         }
5985
5986         RETURN(rc);
5987 }
5988
5989 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5990                                     struct inode *inode,
5991                                     const struct lu_fid *pfid,
5992                                     const char *name, const int namelen)
5993 {
5994         struct osd_thread_info *oti = osd_oti_get(env);
5995         struct lu_buf *buf = &oti->oti_big_buf;
5996         struct dentry *dentry = &oti->oti_obj_dentry;
5997         struct linkea_data ldata = { NULL };
5998         struct lu_name cname = { .ln_name = name,
5999                                  .ln_namelen = namelen };
6000         int rc;
6001
6002         ENTRY;
6003
6004 again:
6005         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6006                              buf->lb_buf, buf->lb_len);
6007         if (rc == -ERANGE)
6008                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
6009
6010         if (rc < 0)
6011                 RETURN(rc);
6012
6013         if (unlikely(rc == 0))
6014                 RETURN(-ENODATA);
6015
6016         if (buf->lb_len < rc) {
6017                 lu_buf_realloc(buf, rc);
6018                 if (buf->lb_buf == NULL)
6019                         RETURN(-ENOMEM);
6020
6021                 goto again;
6022         }
6023
6024         ldata.ld_buf = buf;
6025         rc = linkea_init_with_rec(&ldata);
6026         if (!rc)
6027                 rc = linkea_links_find(&ldata, &cname, pfid);
6028
6029         RETURN(rc);
6030 }
6031
6032 /**
6033  * Calls ->lookup() to find dentry. From dentry get inode and
6034  * read inode's ea to get fid. This is required for  interoperability
6035  * mode (b11826)
6036  *
6037  * \retval   0, on success
6038  * \retval -ve, on error
6039  */
6040 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
6041                              struct dt_rec *rec, const struct dt_key *key)
6042 {
6043         struct inode *dir = obj->oo_inode;
6044         struct dentry *dentry;
6045         struct ldiskfs_dir_entry_2 *de;
6046         struct buffer_head *bh;
6047         struct lu_fid *fid = (struct lu_fid *)rec;
6048         struct htree_lock *hlock = NULL;
6049         int ino;
6050         int rc;
6051
6052         ENTRY;
6053
6054         LASSERT(dir->i_op != NULL);
6055         LASSERT(dir->i_op->lookup != NULL);
6056
6057         dentry = osd_child_dentry_get(env, obj,
6058                                       (char *)key, strlen((char *)key));
6059
6060         if (obj->oo_hl_head != NULL) {
6061                 hlock = osd_oti_get(env)->oti_hlock;
6062                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6063                                    dir, LDISKFS_HLOCK_LOOKUP);
6064         } else {
6065                 down_read(&obj->oo_ext_idx_sem);
6066         }
6067
6068         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6069         if (!IS_ERR(bh)) {
6070                 struct osd_thread_info *oti = osd_oti_get(env);
6071                 struct osd_inode_id *id = &oti->oti_id;
6072                 struct osd_idmap_cache *oic = &oti->oti_cache;
6073                 struct osd_device *dev = osd_obj2dev(obj);
6074
6075                 ino = le32_to_cpu(de->inode);
6076                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
6077                         brelse(bh);
6078                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
6079                         GOTO(out, rc);
6080                 }
6081
6082                 rc = osd_get_fid_from_dentry(de, rec);
6083
6084                 /* done with de, release bh */
6085                 brelse(bh);
6086                 if (rc != 0) {
6087                         if (unlikely(is_remote_parent_ino(dev, ino))) {
6088                                 const char *name = (const char *)key;
6089
6090                                 /*
6091                                  * If the parent is on remote MDT, and there
6092                                  * is no FID-in-dirent, then we have to get
6093                                  * the parent FID from the linkEA.
6094                                  */
6095                                 if (likely(strlen(name) == 2 &&
6096                                            name[0] == '.' && name[1] == '.'))
6097                                         rc = osd_get_pfid_from_linkea(env, obj,
6098                                                                       fid);
6099                         } else {
6100                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6101                         }
6102                 } else {
6103                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6104                 }
6105
6106                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
6107                         fid_zero(&oic->oic_fid);
6108
6109                         GOTO(out, rc);
6110                 }
6111
6112                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
6113                 rc = osd_consistency_check(oti, dev, oic);
6114                 if (rc == -ENOENT)
6115                         fid_zero(&oic->oic_fid);
6116                 else
6117                         /* Other error should not affect lookup result. */
6118                         rc = 0;
6119         } else {
6120                 rc = PTR_ERR(bh);
6121         }
6122
6123         GOTO(out, rc);
6124
6125 out:
6126         if (hlock != NULL)
6127                 ldiskfs_htree_unlock(hlock);
6128         else
6129                 up_read(&obj->oo_ext_idx_sem);
6130         return rc;
6131 }
6132
6133 static int osd_index_declare_ea_insert(const struct lu_env *env,
6134                                        struct dt_object *dt,
6135                                        const struct dt_rec *rec,
6136                                        const struct dt_key *key,
6137                                        struct thandle *handle)
6138 {
6139         struct osd_thandle *oh;
6140         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6141         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6142         const struct lu_fid *fid = rec1->rec_fid;
6143         int credits, rc = 0;
6144         struct osd_idmap_cache *idc;
6145
6146         ENTRY;
6147
6148         LASSERT(!dt_object_remote(dt));
6149         LASSERT(handle != NULL);
6150         LASSERT(fid != NULL);
6151         LASSERT(rec1->rec_type != 0);
6152
6153         oh = container_of(handle, struct osd_thandle, ot_super);
6154         LASSERT(oh->ot_handle == NULL);
6155
6156         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6157
6158         /*
6159          * we can't call iget() while a transactions is running
6160          * (this can lead to a deadlock), but we need to know
6161          * inum and object type. so we find this information at
6162          * declaration and cache in per-thread info
6163          */
6164         idc = osd_idc_find_or_init(env, osd, fid);
6165         if (IS_ERR(idc))
6166                 RETURN(PTR_ERR(idc));
6167         if (idc->oic_remote) {
6168                 /*
6169                  * a reference to remote inode is represented by an
6170                  * agent inode which we have to create
6171                  */
6172                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6173                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6174         }
6175
6176         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6177
6178         if (osd_dt_obj(dt)->oo_inode != NULL) {
6179                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6180
6181                 /*
6182                  * We ignore block quota on meta pool (MDTs), so needn't
6183                  * calculate how many blocks will be consumed by this index
6184                  * insert
6185                  */
6186                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6187                                            i_gid_read(inode),
6188                                            i_projid_read(inode), 0,
6189                                            oh, osd_dt_obj(dt), NULL,
6190                                            OSD_QID_BLK);
6191                 if (rc)
6192                         RETURN(rc);
6193
6194 #ifdef HAVE_PROJECT_QUOTA
6195                 /*
6196                  * Reserve credits for local agent inode to transfer
6197                  * to 0, quota enforcement is ignored in this case.
6198                  */
6199                 if (idc->oic_remote &&
6200                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6201                     i_projid_read(inode) != 0)
6202                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6203                                                   0, i_projid_read(inode),
6204                                                   0, false, PRJQUOTA, true);
6205 #endif
6206         }
6207
6208         RETURN(rc);
6209 }
6210
6211 /**
6212  * Index add function for interoperability mode (b11826).
6213  * It will add the directory entry.This entry is needed to
6214  * maintain name->fid mapping.
6215  *
6216  * \param key it is key i.e. file entry to be inserted
6217  * \param rec it is value of given key i.e. fid
6218  *
6219  * \retval   0, on success
6220  * \retval -ve, on error
6221  */
6222 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6223                                const struct dt_rec *rec,
6224                                const struct dt_key *key, struct thandle *th)
6225 {
6226         struct osd_object *obj = osd_dt_obj(dt);
6227         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6228         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6229         const struct lu_fid *fid = rec1->rec_fid;
6230         const char *name = (const char *)key;
6231         struct osd_thread_info *oti = osd_oti_get(env);
6232         struct inode *child_inode = NULL;
6233         struct osd_idmap_cache *idc;
6234         int rc;
6235
6236         ENTRY;
6237
6238         if (!dt_object_exists(dt))
6239                 RETURN(-ENOENT);
6240
6241         LASSERT(osd_invariant(obj));
6242         LASSERT(!dt_object_remote(dt));
6243         LASSERT(th != NULL);
6244
6245         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6246
6247         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6248
6249         idc = osd_idc_find(env, osd, fid);
6250         if (unlikely(idc == NULL)) {
6251                 idc = osd_idc_find_or_init(env, osd, fid);
6252                 if (IS_ERR(idc)) {
6253                         /*
6254                          * this dt_insert() wasn't declared properly, so
6255                          * FID is missing in OI cache. we better do not
6256                          * lookup FID in FLDB/OI and don't risk to deadlock,
6257                          * but in some special cases (lfsck testing, etc)
6258                          * it's much simpler than fixing a caller.
6259                          *
6260                          * normally this error should be placed after the first
6261                          * find, but migrate may attach source stripes to
6262                          * target, which doesn't create stripes.
6263                          */
6264                         CERROR("%s: "DFID" wasn't declared for insert\n",
6265                                osd_name(osd), PFID(fid));
6266                         dump_stack();
6267                         RETURN(PTR_ERR(idc));
6268                 }
6269         }
6270
6271         if (idc->oic_remote) {
6272                 /* Insert remote entry */
6273                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6274                         child_inode =
6275                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6276                 } else {
6277                         child_inode = osd_create_local_agent_inode(env, osd,
6278                                         obj, fid, rec1->rec_type & S_IFMT, th);
6279                         if (IS_ERR(child_inode))
6280                                 RETURN(PTR_ERR(child_inode));
6281                 }
6282         } else {
6283                 /* Insert local entry */
6284                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6285                         /* for a reason OI cache wasn't filled properly */
6286                         CERROR("%s: OIC for "DFID" isn't filled\n",
6287                                osd_name(osd), PFID(fid));
6288                         RETURN(-EINVAL);
6289                 }
6290                 child_inode = oti->oti_inode;
6291                 if (unlikely(child_inode == NULL)) {
6292                         struct ldiskfs_inode_info *lii;
6293
6294                         OBD_ALLOC_PTR(lii);
6295                         if (lii == NULL)
6296                                 RETURN(-ENOMEM);
6297                         child_inode = oti->oti_inode = &lii->vfs_inode;
6298                 }
6299                 child_inode->i_sb = osd_sb(osd);
6300                 child_inode->i_ino = idc->oic_lid.oii_ino;
6301                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6302         }
6303
6304         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6305
6306         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6307                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6308
6309         if (child_inode && child_inode != oti->oti_inode)
6310                 iput(child_inode);
6311         LASSERT(osd_invariant(obj));
6312         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6313
6314         RETURN(rc);
6315 }
6316
6317 /**
6318  *  Initialize osd Iterator for given osd index object.
6319  *
6320  *  \param  dt      osd index object
6321  */
6322
6323 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6324                                      struct dt_object *dt,
6325                                      __u32 unused)
6326 {
6327         struct osd_it_iam *it;
6328         struct osd_object *obj = osd_dt_obj(dt);
6329         struct lu_object *lo = &dt->do_lu;
6330         struct iam_path_descr *ipd;
6331         struct iam_container *bag = &obj->oo_dir->od_container;
6332
6333         if (!dt_object_exists(dt))
6334                 return ERR_PTR(-ENOENT);
6335
6336         OBD_ALLOC_PTR(it);
6337         if (it == NULL)
6338                 return ERR_PTR(-ENOMEM);
6339
6340         ipd = osd_it_ipd_get(env, bag);
6341         if (likely(ipd != NULL)) {
6342                 it->oi_obj = obj;
6343                 it->oi_ipd = ipd;
6344                 lu_object_get(lo);
6345                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6346                 return (struct dt_it *)it;
6347         } else {
6348                 OBD_FREE_PTR(it);
6349                 return ERR_PTR(-ENOMEM);
6350         }
6351 }
6352
6353 /**
6354  * free given Iterator.
6355  */
6356 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6357 {
6358         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6359         struct osd_object *obj = it->oi_obj;
6360
6361         iam_it_fini(&it->oi_it);
6362         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6363         osd_object_put(env, obj);
6364         OBD_FREE_PTR(it);
6365 }
6366
6367 /**
6368  *  Move Iterator to record specified by \a key
6369  *
6370  *  \param  di      osd iterator
6371  *  \param  key     key for index
6372  *
6373  *  \retval +ve  di points to record with least key not larger than key
6374  *  \retval  0   di points to exact matched key
6375  *  \retval -ve  failure
6376  */
6377
6378 static int osd_it_iam_get(const struct lu_env *env,
6379                           struct dt_it *di, const struct dt_key *key)
6380 {
6381         struct osd_thread_info *oti = osd_oti_get(env);
6382         struct osd_it_iam *it = (struct osd_it_iam *)di;
6383
6384         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6385                 /* swab quota uid/gid */
6386                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6387                 key = (struct dt_key *)&oti->oti_quota_id;
6388         }
6389
6390         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6391 }
6392
6393 /**
6394  *  Release Iterator
6395  *
6396  *  \param  di      osd iterator
6397  */
6398 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6399 {
6400         struct osd_it_iam *it = (struct osd_it_iam *)di;
6401
6402         iam_it_put(&it->oi_it);
6403 }
6404
6405 /**
6406  *  Move iterator by one record
6407  *
6408  *  \param  di      osd iterator
6409  *
6410  *  \retval +1   end of container reached
6411  *  \retval  0   success
6412  *  \retval -ve  failure
6413  */
6414
6415 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6416 {
6417         struct osd_it_iam *it = (struct osd_it_iam *)di;
6418
6419         return iam_it_next(&it->oi_it);
6420 }
6421
6422 /**
6423  * Return pointer to the key under iterator.
6424  */
6425
6426 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6427                                      const struct dt_it *di)
6428 {
6429         struct osd_thread_info *oti = osd_oti_get(env);
6430         struct osd_it_iam *it = (struct osd_it_iam *)di;
6431         struct osd_object *obj = it->oi_obj;
6432         struct dt_key *key;
6433
6434         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6435
6436         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6437                 /* swab quota uid/gid */
6438                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6439                 key = (struct dt_key *)&oti->oti_quota_id;
6440         }
6441
6442         return key;
6443 }
6444
6445 /**
6446  * Return size of key under iterator (in bytes)
6447  */
6448
6449 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6450 {
6451         struct osd_it_iam *it = (struct osd_it_iam *)di;
6452
6453         return iam_it_key_size(&it->oi_it);
6454 }
6455
6456 static inline void
6457 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6458 {
6459         /* check if file type is required */
6460         if (ent->lde_attrs & LUDA_TYPE) {
6461                 struct luda_type *lt;
6462                 int align = sizeof(*lt) - 1;
6463
6464                 len = (len + align) & ~align;
6465                 lt = (struct luda_type *)(ent->lde_name + len);
6466                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6467         }
6468
6469         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6470 }
6471
6472 /**
6473  * build lu direct from backend fs dirent.
6474  */
6475
6476 static inline void
6477 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6478                    char *name, __u16 namelen, __u16 type, __u32 attr)
6479 {
6480         ent->lde_attrs = attr | LUDA_FID;
6481         fid_cpu_to_le(&ent->lde_fid, fid);
6482
6483         ent->lde_hash = cpu_to_le64(offset);
6484         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6485
6486         strncpy(ent->lde_name, name, namelen);
6487         ent->lde_name[namelen] = '\0';
6488         ent->lde_namelen = cpu_to_le16(namelen);
6489
6490         /* append lustre attributes */
6491         osd_it_append_attrs(ent, namelen, type);
6492 }
6493
6494 /**
6495  * Return pointer to the record under iterator.
6496  */
6497 static int osd_it_iam_rec(const struct lu_env *env,
6498                           const struct dt_it *di,
6499                           struct dt_rec *dtrec, __u32 attr)
6500 {
6501         struct osd_it_iam *it = (struct osd_it_iam *)di;
6502         struct osd_thread_info *info = osd_oti_get(env);
6503
6504         ENTRY;
6505
6506         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6507                 const struct osd_fid_pack *rec;
6508                 struct lu_fid *fid = &info->oti_fid;
6509                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6510                 char *name;
6511                 int namelen;
6512                 __u64 hash;
6513                 int rc;
6514
6515                 name = (char *)iam_it_key_get(&it->oi_it);
6516                 if (IS_ERR(name))
6517                         RETURN(PTR_ERR(name));
6518
6519                 namelen = iam_it_key_size(&it->oi_it);
6520
6521                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6522                 if (IS_ERR(rec))
6523                         RETURN(PTR_ERR(rec));
6524
6525                 rc = osd_fid_unpack(fid, rec);
6526                 if (rc)
6527                         RETURN(rc);
6528
6529                 hash = iam_it_store(&it->oi_it);
6530
6531                 /* IAM does not store object type in IAM index (dir) */
6532                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6533                                    0, LUDA_FID);
6534         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6535                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6536                            (struct iam_rec *)dtrec);
6537                 osd_quota_unpack(it->oi_obj, dtrec);
6538         } else {
6539                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6540                            (struct iam_rec *)dtrec);
6541         }
6542
6543         RETURN(0);
6544 }
6545
6546 /**
6547  * Returns cookie for current Iterator position.
6548  */
6549 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6550 {
6551         struct osd_it_iam *it = (struct osd_it_iam *)di;
6552
6553         return iam_it_store(&it->oi_it);
6554 }
6555
6556 /**
6557  * Restore iterator from cookie.
6558  *
6559  * \param  di      osd iterator
6560  * \param  hash    Iterator location cookie
6561  *
6562  * \retval +ve  di points to record with least key not larger than key.
6563  * \retval  0   di points to exact matched key
6564  * \retval -ve  failure
6565  */
6566
6567 static int osd_it_iam_load(const struct lu_env *env,
6568                            const struct dt_it *di, __u64 hash)
6569 {
6570         struct osd_it_iam *it = (struct osd_it_iam *)di;
6571
6572         return iam_it_load(&it->oi_it, hash);
6573 }
6574
6575 static const struct dt_index_operations osd_index_iam_ops = {
6576         .dio_lookup         = osd_index_iam_lookup,
6577         .dio_declare_insert = osd_index_declare_iam_insert,
6578         .dio_insert         = osd_index_iam_insert,
6579         .dio_declare_delete = osd_index_declare_iam_delete,
6580         .dio_delete         = osd_index_iam_delete,
6581         .dio_it     = {
6582                 .init     = osd_it_iam_init,
6583                 .fini     = osd_it_iam_fini,
6584                 .get      = osd_it_iam_get,
6585                 .put      = osd_it_iam_put,
6586                 .next     = osd_it_iam_next,
6587                 .key      = osd_it_iam_key,
6588                 .key_size = osd_it_iam_key_size,
6589                 .rec      = osd_it_iam_rec,
6590                 .store    = osd_it_iam_store,
6591                 .load     = osd_it_iam_load
6592         }
6593 };
6594
6595
6596 /**
6597  * Creates or initializes iterator context.
6598  *
6599  * \retval struct osd_it_ea, iterator structure on success
6600  *
6601  */
6602 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6603                                     struct dt_object *dt,
6604                                     __u32 attr)
6605 {
6606         struct osd_object *obj = osd_dt_obj(dt);
6607         struct osd_thread_info *info = osd_oti_get(env);
6608         struct osd_it_ea *oie;
6609         struct file *file;
6610         struct lu_object *lo = &dt->do_lu;
6611         struct dentry *obj_dentry;
6612
6613         ENTRY;
6614
6615         if (!dt_object_exists(dt) || obj->oo_destroyed)
6616                 RETURN(ERR_PTR(-ENOENT));
6617
6618         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6619         if (oie == NULL)
6620                 RETURN(ERR_PTR(-ENOMEM));
6621         obj_dentry = &oie->oie_dentry;
6622
6623         obj_dentry->d_inode = obj->oo_inode;
6624         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6625         obj_dentry->d_name.hash = 0;
6626
6627         oie->oie_rd_dirent       = 0;
6628         oie->oie_it_dirent       = 0;
6629         oie->oie_dirent          = NULL;
6630         if (unlikely(!info->oti_it_ea_buf_used)) {
6631                 oie->oie_buf = info->oti_it_ea_buf;
6632                 info->oti_it_ea_buf_used = 1;
6633         } else {
6634                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6635                 if (oie->oie_buf == NULL)
6636                         RETURN(ERR_PTR(-ENOMEM));
6637         }
6638         oie->oie_obj = obj;
6639
6640         file = &oie->oie_file;
6641
6642         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6643         if (attr & LUDA_64BITHASH)
6644                 file->f_mode    = FMODE_64BITHASH;
6645         else
6646                 file->f_mode    = FMODE_32BITHASH;
6647         file->f_path.dentry     = obj_dentry;
6648         file->f_mapping         = obj->oo_inode->i_mapping;
6649         file->f_op              = obj->oo_inode->i_fop;
6650         file->f_inode = obj->oo_inode;
6651
6652         lu_object_get(lo);
6653         RETURN((struct dt_it *)oie);
6654 }
6655
6656 /**
6657  * Destroy or finishes iterator context.
6658  *
6659  * \param di iterator structure to be destroyed
6660  */
6661 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6662 {
6663         struct osd_thread_info *info = osd_oti_get(env);
6664         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6665         struct osd_object *obj = oie->oie_obj;
6666         struct inode *inode = obj->oo_inode;
6667
6668         ENTRY;
6669         oie->oie_file.f_op->release(inode, &oie->oie_file);
6670         osd_object_put(env, obj);
6671         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6672                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6673         else
6674                 info->oti_it_ea_buf_used = 0;
6675         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6676         EXIT;
6677 }
6678
6679 /**
6680  * It position the iterator at given key, so that next lookup continues from
6681  * that key Or it is similar to dio_it->load() but based on a key,
6682  * rather than file position.
6683  *
6684  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6685  * to the beginning.
6686  *
6687  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6688  */
6689 static int osd_it_ea_get(const struct lu_env *env,
6690                          struct dt_it *di, const struct dt_key *key)
6691 {
6692         struct osd_it_ea *it = (struct osd_it_ea *)di;
6693
6694         ENTRY;
6695         LASSERT(((const char *)key)[0] == '\0');
6696         it->oie_file.f_pos = 0;
6697         it->oie_rd_dirent = 0;
6698         it->oie_it_dirent = 0;
6699         it->oie_dirent = NULL;
6700
6701         RETURN(+1);
6702 }
6703
6704 /**
6705  * Does nothing
6706  */
6707 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6708 {
6709 }
6710
6711 struct osd_filldir_cbs {
6712         struct dir_context ctx;
6713         struct osd_it_ea  *it;
6714 };
6715 /**
6716  * It is called internally by ->iterate*(). It fills the
6717  * iterator's in-memory data structure with required
6718  * information i.e. name, namelen, rec_size etc.
6719  *
6720  * \param buf in which information to be filled in.
6721  * \param name name of the file in given dir
6722  *
6723  * \retval 0 on success
6724  * \retval 1 on buffer full
6725  */
6726 #ifdef HAVE_FILLDIR_USE_CTX
6727 static int osd_ldiskfs_filldir(struct dir_context *buf,
6728 #else
6729 static int osd_ldiskfs_filldir(void *buf,
6730 #endif
6731                                const char *name, int namelen,
6732                                loff_t offset, __u64 ino, unsigned int d_type)
6733 {
6734         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6735         struct osd_object *obj = it->oie_obj;
6736         struct osd_it_ea_dirent *ent = it->oie_dirent;
6737         struct lu_fid *fid = &ent->oied_fid;
6738         struct osd_fid_pack *rec;
6739
6740         ENTRY;
6741
6742 /* this should never happen */
6743         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6744                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6745                 RETURN(-EIO);
6746         }
6747
6748         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6749             OSD_IT_EA_BUFSIZE)
6750                 RETURN(1);
6751
6752         /* "." is just the object itself. */
6753         if (namelen == 1 && name[0] == '.') {
6754                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6755         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6756                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6757                 if (osd_fid_unpack(fid, rec) != 0)
6758                         fid_zero(fid);
6759         } else {
6760                 fid_zero(fid);
6761         }
6762         d_type &= ~LDISKFS_DIRENT_LUFID;
6763
6764         /* NOT export local root. */
6765         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6766                 ino = obj->oo_inode->i_ino;
6767                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6768         }
6769
6770         ent->oied_ino     = ino;
6771         ent->oied_off     = offset;
6772         ent->oied_namelen = namelen;
6773         ent->oied_type    = d_type;
6774
6775         memcpy(ent->oied_name, name, namelen);
6776
6777         it->oie_rd_dirent++;
6778         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6779         RETURN(0);
6780 }
6781
6782 /**
6783  * Calls ->iterate*() to load a directory entry at a time
6784  * and stored it in iterator's in-memory data structure.
6785  *
6786  * \param di iterator's in memory structure
6787  *
6788  * \retval   0 on success
6789  * \retval -ve on error
6790  * \retval +1 reach the end of entry
6791  */
6792 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6793                                const struct dt_it *di)
6794 {
6795         struct osd_it_ea *it = (struct osd_it_ea *)di;
6796         struct osd_object *obj = it->oie_obj;
6797         struct inode *inode = obj->oo_inode;
6798         struct htree_lock *hlock = NULL;
6799         struct file *filp = &it->oie_file;
6800         int rc = 0;
6801         struct osd_filldir_cbs buf = {
6802                 .ctx.actor = osd_ldiskfs_filldir,
6803                 .it = it
6804         };
6805
6806         ENTRY;
6807         it->oie_dirent = it->oie_buf;
6808         it->oie_rd_dirent = 0;
6809
6810         if (obj->oo_hl_head != NULL) {
6811                 hlock = osd_oti_get(env)->oti_hlock;
6812                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6813                                    inode, LDISKFS_HLOCK_READDIR);
6814         } else {
6815                 down_read(&obj->oo_ext_idx_sem);
6816         }
6817
6818         filp->f_cred = current_cred();
6819         rc = osd_security_file_alloc(filp);
6820         if (rc)
6821                 GOTO(unlock, rc);
6822
6823         filp->f_flags |= O_NOATIME;
6824         filp->f_mode |= FMODE_NONOTIFY;
6825         rc = iterate_dir(filp, &buf.ctx);
6826         if (rc)
6827                 GOTO(unlock, rc);
6828
6829         if (it->oie_rd_dirent == 0) {
6830                 /*
6831                  * If it does not get any dirent, it means it has been reached
6832                  * to the end of the dir
6833                  */
6834                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6835                 if (rc == 0)
6836                         rc = 1;
6837         } else {
6838                 it->oie_dirent = it->oie_buf;
6839                 it->oie_it_dirent = 1;
6840         }
6841 unlock:
6842         if (hlock != NULL)
6843                 ldiskfs_htree_unlock(hlock);
6844         else
6845                 up_read(&obj->oo_ext_idx_sem);
6846
6847         RETURN(rc);
6848 }
6849
6850 /**
6851  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6852  * to load a directory entry at a time and stored it in
6853  * iterator's in-memory data structure.
6854  *
6855  * \param di iterator's in memory structure
6856  *
6857  * \retval +ve iterator reached to end
6858  * \retval   0 iterator not reached to end
6859  * \retval -ve on error
6860  */
6861 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6862 {
6863         struct osd_it_ea *it = (struct osd_it_ea *)di;
6864         int rc;
6865
6866         ENTRY;
6867
6868         if (it->oie_it_dirent < it->oie_rd_dirent) {
6869                 it->oie_dirent =
6870                         (void *)it->oie_dirent +
6871                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6872                                        it->oie_dirent->oied_namelen);
6873                 it->oie_it_dirent++;
6874                 rc = 0;
6875         } else {
6876                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6877                         rc = 1;
6878                 else
6879                         rc = osd_ldiskfs_it_fill(env, di);
6880         }
6881
6882         RETURN(rc);
6883 }
6884
6885 /**
6886  * Returns the key at current position from iterator's in memory structure.
6887  *
6888  * \param di iterator's in memory structure
6889  *
6890  * \retval key i.e. struct dt_key on success
6891  */
6892 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6893                                     const struct dt_it *di)
6894 {
6895         struct osd_it_ea *it = (struct osd_it_ea *)di;
6896
6897         return (struct dt_key *)it->oie_dirent->oied_name;
6898 }
6899
6900 /**
6901  * Returns key's size at current position from iterator's in memory structure.
6902  *
6903  * \param di iterator's in memory structure
6904  *
6905  * \retval key_size i.e. struct dt_key on success
6906  */
6907 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6908 {
6909         struct osd_it_ea *it = (struct osd_it_ea *)di;
6910
6911         return it->oie_dirent->oied_namelen;
6912 }
6913
6914 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6915 #undef LDISKFS_DIR_REC_LEN
6916 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6917 #endif
6918
6919 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6920 {
6921         if (LDISKFS_DIR_REC_LEN(de) >=
6922             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6923                 return true;
6924
6925         return false;
6926 }
6927
6928 static inline bool
6929 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6930                      unsigned int blocksize, bool dotdot)
6931 {
6932         if (dotdot)
6933                 return osd_dotdot_has_space(de);
6934
6935         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6936             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6937                 return true;
6938
6939         return false;
6940 }
6941
6942 static int
6943 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6944                     handle_t *jh, struct dentry *dentry,
6945                     const struct lu_fid *fid, struct buffer_head *bh,
6946                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6947                     bool dotdot)
6948 {
6949         struct inode *dir = dentry->d_parent->d_inode;
6950         struct inode *inode = dentry->d_inode;
6951         struct osd_fid_pack *rec;
6952         struct ldiskfs_dentry_param *ldp;
6953         int namelen = dentry->d_name.len;
6954         int rc;
6955         struct osd_thread_info *info = osd_oti_get(env);
6956
6957         ENTRY;
6958
6959         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6960                 RETURN(0);
6961
6962         /* There is enough space to hold the FID-in-dirent. */
6963         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6964                 rc = ldiskfs_journal_get_write_access(jh, bh);
6965                 if (rc != 0)
6966                         RETURN(rc);
6967
6968                 de->name[namelen] = 0;
6969                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6970                 rec->fp_len = sizeof(struct lu_fid) + 1;
6971                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6972                 de->file_type |= LDISKFS_DIRENT_LUFID;
6973                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6974
6975                 RETURN(rc);
6976         }
6977
6978         LASSERT(!dotdot);
6979
6980         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6981         if (rc != 0)
6982                 RETURN(rc);
6983
6984         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6985         osd_get_ldiskfs_dirent_param(ldp, fid);
6986         dentry->d_fsdata = (void *)ldp;
6987         dquot_initialize(dir);
6988         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6989         /*
6990          * It is too bad, we cannot reinsert the name entry back.
6991          * That means we lose it!
6992          */
6993         if (rc != 0)
6994                 CDEBUG(D_LFSCK,
6995                        "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6996                        osd_ino2name(inode), dir->i_ino, dir->i_generation,
6997                        namelen, dentry->d_name.name, PFID(fid), rc);
6998
6999         RETURN(rc);
7000 }
7001
7002 static int
7003 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
7004                         struct osd_it_ea *it, struct lu_fid *fid,
7005                         struct osd_inode_id *id, __u32 *attr)
7006 {
7007         struct osd_thread_info *info = osd_oti_get(env);
7008         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
7009         struct osd_device *dev = osd_obj2dev(obj);
7010         struct super_block *sb = osd_sb(dev);
7011         const char *devname = osd_name(dev);
7012         struct osd_it_ea_dirent *ent = it->oie_dirent;
7013         struct inode *dir = obj->oo_inode;
7014         struct htree_lock *hlock = NULL;
7015         struct buffer_head *bh = NULL;
7016         handle_t *jh = NULL;
7017         struct ldiskfs_dir_entry_2 *de;
7018         struct dentry *dentry;
7019         struct inode *inode;
7020         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
7021         int credits;
7022         int rc;
7023         bool dotdot = false;
7024         bool dirty = false;
7025
7026         ENTRY;
7027
7028         if (ent->oied_name[0] == '.') {
7029                 if (ent->oied_namelen == 1)
7030                         RETURN(0);
7031
7032                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
7033                         dotdot = true;
7034         }
7035
7036         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
7037         inode = osd_iget(info, dev, id);
7038         if (IS_ERR(inode)) {
7039                 rc = PTR_ERR(inode);
7040                 if (rc == -ENOENT || rc == -ESTALE) {
7041                         /*
7042                          * Maybe dangling name entry, or
7043                          * corrupted directory entry.
7044                          */
7045                         *attr |= LUDA_UNKNOWN;
7046                         rc = 0;
7047                 } else {
7048                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
7049                                "check_repair, dir = %lu/%u, name = %.*s, "
7050                                "ino = %llu, rc = %d\n",
7051                                devname, dir->i_ino, dir->i_generation,
7052                                ent->oied_namelen, ent->oied_name,
7053                                ent->oied_ino, rc);
7054                 }
7055
7056                 RETURN(rc);
7057         }
7058
7059         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
7060                                            ent->oied_namelen);
7061         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
7062         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
7063                 lma = NULL;
7064         else if (rc != 0)
7065                 GOTO(out, rc);
7066
7067         /*
7068          * We need to ensure that the name entry is still valid.
7069          * Because it may be removed or renamed by other already.
7070          *
7071          * The unlink or rename operation will start journal before PDO lock,
7072          * so to avoid deadlock, here we need to start journal handle before
7073          * related PDO lock also. But because we do not know whether there
7074          * will be something to be repaired before PDO lock, we just start
7075          * journal without conditions.
7076          *
7077          * We may need to remove the name entry firstly, then insert back.
7078          * One credit is for user quota file update.
7079          * One credit is for group quota file update.
7080          * Two credits are for dirty inode.
7081          */
7082         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
7083                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
7084
7085         if (dev->od_dirent_journal != 0) {
7086
7087 again:
7088                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
7089                 if (IS_ERR(jh)) {
7090                         rc = PTR_ERR(jh);
7091                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
7092                                "check_repair, dir = %lu/%u, credits = %d, "
7093                                "name = %.*s, ino = %llu: rc = %d\n",
7094                                devname, dir->i_ino, dir->i_generation, credits,
7095                                ent->oied_namelen, ent->oied_name,
7096                                ent->oied_ino, rc);
7097
7098                         GOTO(out_inode, rc);
7099                 }
7100
7101                 if (obj->oo_hl_head != NULL) {
7102                         hlock = osd_oti_get(env)->oti_hlock;
7103                         /*
7104                          * "0" means exclusive lock for the whole directory.
7105                          * We need to prevent others access such name entry
7106                          * during the delete + insert. Neither HLOCK_ADD nor
7107                          * HLOCK_DEL cannot guarantee the atomicity.
7108                          */
7109                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7110                 } else {
7111                         down_write(&obj->oo_ext_idx_sem);
7112                 }
7113         } else {
7114                 if (obj->oo_hl_head != NULL) {
7115                         hlock = osd_oti_get(env)->oti_hlock;
7116                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7117                                            LDISKFS_HLOCK_LOOKUP);
7118                 } else {
7119                         down_read(&obj->oo_ext_idx_sem);
7120                 }
7121         }
7122
7123         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7124         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7125                 *attr |= LUDA_IGNORE;
7126
7127                 GOTO(out, rc = 0);
7128         }
7129
7130         /*
7131          * For dotdot entry, if there is not enough space to hold the
7132          * FID-in-dirent, just keep them there. It only happens when the
7133          * device upgraded from 1.8 or restored from MDT file-level backup.
7134          * For the whole directory, only dotdot entry have no FID-in-dirent
7135          * and needs to get FID from LMA when readdir, it will not affect the
7136          * performance much.
7137          */
7138         if (dotdot && !osd_dotdot_has_space(de)) {
7139                 *attr |= LUDA_UNKNOWN;
7140
7141                 GOTO(out, rc = 0);
7142         }
7143
7144         if (lma != NULL) {
7145                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7146                         GOTO(out, rc = 0);
7147
7148                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7149                         struct lu_fid *tfid = &lma->lma_self_fid;
7150
7151                         if (likely(dotdot &&
7152                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7153                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7154                                 /*
7155                                  * It must be REMOTE_PARENT_DIR and as the
7156                                  * 'dotdot' entry of remote directory
7157                                  */
7158                                 *attr |= LUDA_IGNORE;
7159                         } else {
7160                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7161                                        "parent directory, but got %.*s under "
7162                                        "dir = %lu/%u with the FID "DFID"\n",
7163                                        devname, ent->oied_namelen,
7164                                        ent->oied_name, dir->i_ino,
7165                                        dir->i_generation, PFID(tfid));
7166
7167                                 *attr |= LUDA_UNKNOWN;
7168                         }
7169
7170                         GOTO(out, rc = 0);
7171                 }
7172         }
7173
7174         if (!fid_is_zero(fid)) {
7175                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7176                                               ent->oied_namelen);
7177                 if (rc == -ENOENT ||
7178                     (rc == -ENODATA &&
7179                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7180                         /*
7181                          * linkEA does not recognize the dirent entry,
7182                          * it may because the dirent entry corruption
7183                          * and points to other's inode.
7184                          */
7185                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7186                                "recognize the dirent, dir = %lu/%u, "
7187                                " name = %.*s, ino = %llu, "
7188                                DFID": rc = %d\n", devname, dir->i_ino,
7189                                dir->i_generation, ent->oied_namelen,
7190                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7191                         *attr |= LUDA_UNKNOWN;
7192
7193                         GOTO(out, rc = 0);
7194                 }
7195
7196                 if (rc && rc != -ENODATA) {
7197                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7198                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7199                                DFID": rc = %d\n", devname, dir->i_ino,
7200                                dir->i_generation, ent->oied_namelen,
7201                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7202                         *attr |= LUDA_UNKNOWN;
7203
7204                         GOTO(out, rc = 0);
7205                 }
7206         }
7207
7208         if (lma != NULL) {
7209                 /*
7210                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7211                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7212                  */
7213                 if (*attr & LUDA_VERIFY_DRYRUN) {
7214                         *fid = lma->lma_self_fid;
7215                         *attr |= LUDA_REPAIR;
7216
7217                         GOTO(out, rc = 0);
7218                 }
7219
7220                 if (jh == NULL) {
7221                         brelse(bh);
7222                         dev->od_dirent_journal = 1;
7223                         if (hlock != NULL) {
7224                                 ldiskfs_htree_unlock(hlock);
7225                                 hlock = NULL;
7226                         } else {
7227                                 up_read(&obj->oo_ext_idx_sem);
7228                         }
7229
7230                         goto again;
7231                 }
7232
7233                 *fid = lma->lma_self_fid;
7234                 dirty = true;
7235                 /* Update or append the FID-in-dirent. */
7236                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7237                                          bh, de, hlock, dotdot);
7238                 if (rc == 0)
7239                         *attr |= LUDA_REPAIR;
7240                 else
7241                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7242                                "the dirent, dir = %lu/%u, name = %.*s, "
7243                                "ino = %llu, "DFID": rc = %d\n",
7244                                devname, dir->i_ino, dir->i_generation,
7245                                ent->oied_namelen, ent->oied_name,
7246                                ent->oied_ino, PFID(fid), rc);
7247         } else {
7248                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7249                 if (*attr & LUDA_VERIFY_DRYRUN) {
7250                         if (fid_is_sane(fid)) {
7251                                 *attr |= LUDA_REPAIR;
7252                         } else if (dev->od_index == 0) {
7253                                 lu_igif_build(fid, inode->i_ino,
7254                                               inode->i_generation);
7255                                 *attr |= LUDA_UPGRADE;
7256                         }
7257
7258                         GOTO(out, rc = 0);
7259                 }
7260
7261                 if (jh == NULL) {
7262                         brelse(bh);
7263                         dev->od_dirent_journal = 1;
7264                         if (hlock != NULL) {
7265                                 ldiskfs_htree_unlock(hlock);
7266                                 hlock = NULL;
7267                         } else {
7268                                 up_read(&obj->oo_ext_idx_sem);
7269                         }
7270
7271                         goto again;
7272                 }
7273
7274                 dirty = true;
7275                 if (unlikely(fid_is_sane(fid))) {
7276                         /*
7277                          * FID-in-dirent exists, but FID-in-LMA is lost.
7278                          * Trust the FID-in-dirent, and add FID-in-LMA.
7279                          */
7280                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7281                         if (rc == 0)
7282                                 *attr |= LUDA_REPAIR;
7283                         else
7284                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7285                                        "update dirent, dir = %lu/%u, "
7286                                        "name = %.*s, ino = %llu, "
7287                                        DFID": rc = %d\n",
7288                                        devname, dir->i_ino, dir->i_generation,
7289                                        ent->oied_namelen, ent->oied_name,
7290                                        ent->oied_ino, PFID(fid), rc);
7291                 } else if (dev->od_index == 0) {
7292                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7293                         /*
7294                          * It is probably IGIF object. Only aappend the
7295                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7296                          */
7297                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7298                                                  bh, de, hlock, dotdot);
7299                         if (rc == 0)
7300                                 *attr |= LUDA_UPGRADE;
7301                         else
7302                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7303                                        "after the dirent, dir = %lu/%u, "
7304                                        "name = %.*s, ino = %llu, "
7305                                        DFID": rc = %d\n",
7306                                        devname, dir->i_ino, dir->i_generation,
7307                                        ent->oied_namelen, ent->oied_name,
7308                                        ent->oied_ino, PFID(fid), rc);
7309                 }
7310         }
7311
7312         GOTO(out, rc);
7313
7314 out:
7315         if (!IS_ERR(bh))
7316                 brelse(bh);
7317         if (hlock != NULL) {
7318                 ldiskfs_htree_unlock(hlock);
7319         } else {
7320                 if (dev->od_dirent_journal != 0)
7321                         up_write(&obj->oo_ext_idx_sem);
7322                 else
7323                         up_read(&obj->oo_ext_idx_sem);
7324         }
7325
7326         if (jh != NULL)
7327                 ldiskfs_journal_stop(jh);
7328
7329 out_inode:
7330         iput(inode);
7331         if (rc >= 0 && !dirty)
7332                 dev->od_dirent_journal = 0;
7333
7334         return rc;
7335 }
7336
7337 /**
7338  * Returns the value at current position from iterator's in memory structure.
7339  *
7340  * \param di struct osd_it_ea, iterator's in memory structure
7341  * \param attr attr requested for dirent.
7342  * \param lde lustre dirent
7343  *
7344  * \retval   0 no error and \param lde has correct lustre dirent.
7345  * \retval -ve on error
7346  */
7347 static inline int osd_it_ea_rec(const struct lu_env *env,
7348                                 const struct dt_it *di,
7349                                 struct dt_rec *dtrec, __u32 attr)
7350 {
7351         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7352         struct osd_object      *obj   = it->oie_obj;
7353         struct osd_device      *dev   = osd_obj2dev(obj);
7354         struct osd_thread_info *oti   = osd_oti_get(env);
7355         struct osd_inode_id    *id    = &oti->oti_id;
7356         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7357         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7358         __u32 ino = it->oie_dirent->oied_ino;
7359         int rc = 0;
7360
7361         ENTRY;
7362
7363         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7364
7365         if (attr & LUDA_VERIFY) {
7366                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7367                         attr |= LUDA_IGNORE;
7368                         /*
7369                          * If the parent is on remote MDT, and there
7370                          * is no FID-in-dirent, then we have to get
7371                          * the parent FID from the linkEA.
7372                          */
7373                         if (!fid_is_sane(fid) &&
7374                             it->oie_dirent->oied_namelen == 2 &&
7375                             it->oie_dirent->oied_name[0] == '.' &&
7376                             it->oie_dirent->oied_name[1] == '.')
7377                                 osd_get_pfid_from_linkea(env, obj, fid);
7378                 } else {
7379                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7380                                                      &attr);
7381                 }
7382
7383                 if (!fid_is_sane(fid))
7384                         attr |= LUDA_UNKNOWN;
7385         } else {
7386                 attr &= ~LU_DIRENT_ATTRS_MASK;
7387                 if (!fid_is_sane(fid)) {
7388                         bool is_dotdot = false;
7389
7390                         if (it->oie_dirent->oied_namelen == 2 &&
7391                             it->oie_dirent->oied_name[0] == '.' &&
7392                             it->oie_dirent->oied_name[1] == '.')
7393                                 is_dotdot = true;
7394                         /*
7395                          * If the parent is on remote MDT, and there
7396                          * is no FID-in-dirent, then we have to get
7397                          * the parent FID from the linkEA.
7398                          */
7399                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7400                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7401                         } else {
7402                                 if (is_dotdot == false &&
7403                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7404                                         RETURN(-ENOENT);
7405
7406                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7407                         }
7408                 }
7409         }
7410
7411         /* Pack the entry anyway, at least the offset is right. */
7412         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7413                            it->oie_dirent->oied_name,
7414                            it->oie_dirent->oied_namelen,
7415                            it->oie_dirent->oied_type, attr);
7416
7417         RETURN(rc > 0 ? 0 : rc);
7418 }
7419
7420 /**
7421  * Returns the record size size at current position.
7422  *
7423  * This function will return record(lu_dirent) size in bytes.
7424  *
7425  * \param[in] env       execution environment
7426  * \param[in] di        iterator's in memory structure
7427  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7428  *                      calculate the lu_dirent size.
7429  *
7430  * \retval      record size(in bytes & in memory) of the current lu_dirent
7431  *              entry.
7432  */
7433 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7434                               __u32 attr)
7435 {
7436         struct osd_it_ea *it = (struct osd_it_ea *)di;
7437
7438         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7439 }
7440
7441 /**
7442  * Returns a cookie for current position of the iterator head, so that
7443  * user can use this cookie to load/start the iterator next time.
7444  *
7445  * \param di iterator's in memory structure
7446  *
7447  * \retval cookie for current position, on success
7448  */
7449 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7450 {
7451         struct osd_it_ea *it = (struct osd_it_ea *)di;
7452
7453         return it->oie_dirent->oied_off;
7454 }
7455
7456 /**
7457  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7458  * to load a directory entry at a time and stored it i inn,
7459  * in iterator's in-memory data structure.
7460  *
7461  * \param di struct osd_it_ea, iterator's in memory structure
7462  *
7463  * \retval +ve on success
7464  * \retval -ve on error
7465  */
7466 static int osd_it_ea_load(const struct lu_env *env,
7467                           const struct dt_it *di, __u64 hash)
7468 {
7469         struct osd_it_ea *it = (struct osd_it_ea *)di;
7470         int rc;
7471
7472         ENTRY;
7473         it->oie_file.f_pos = hash;
7474
7475         rc =  osd_ldiskfs_it_fill(env, di);
7476         if (rc > 0)
7477                 rc = -ENODATA;
7478
7479         if (rc == 0)
7480                 rc = 1;
7481
7482         RETURN(rc);
7483 }
7484
7485 /**
7486  * Index lookup function for interoperability mode (b11826).
7487  *
7488  * \param key,  key i.e. file name to be searched
7489  *
7490  * \retval +ve, on success
7491  * \retval -ve, on error
7492  */
7493 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7494                                struct dt_rec *rec, const struct dt_key *key)
7495 {
7496         struct osd_object *obj = osd_dt_obj(dt);
7497         int rc = 0;
7498
7499         ENTRY;
7500
7501         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7502         LINVRNT(osd_invariant(obj));
7503
7504         rc = osd_ea_lookup_rec(env, obj, rec, key);
7505         if (rc == 0)
7506                 rc = 1;
7507         RETURN(rc);
7508 }
7509
7510 /**
7511  * Index and Iterator operations for interoperability
7512  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7513  */
7514 static const struct dt_index_operations osd_index_ea_ops = {
7515         .dio_lookup         = osd_index_ea_lookup,
7516         .dio_declare_insert = osd_index_declare_ea_insert,
7517         .dio_insert         = osd_index_ea_insert,
7518         .dio_declare_delete = osd_index_declare_ea_delete,
7519         .dio_delete         = osd_index_ea_delete,
7520         .dio_it     = {
7521                 .init     = osd_it_ea_init,
7522                 .fini     = osd_it_ea_fini,
7523                 .get      = osd_it_ea_get,
7524                 .put      = osd_it_ea_put,
7525                 .next     = osd_it_ea_next,
7526                 .key      = osd_it_ea_key,
7527                 .key_size = osd_it_ea_key_size,
7528                 .rec      = osd_it_ea_rec,
7529                 .rec_size = osd_it_ea_rec_size,
7530                 .store    = osd_it_ea_store,
7531                 .load     = osd_it_ea_load
7532         }
7533 };
7534
7535 static void *osd_key_init(const struct lu_context *ctx,
7536                           struct lu_context_key *key)
7537 {
7538         struct osd_thread_info *info;
7539
7540         OBD_ALLOC_PTR(info);
7541         if (info == NULL)
7542                 return ERR_PTR(-ENOMEM);
7543
7544         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7545         if (info->oti_it_ea_buf == NULL)
7546                 goto out_free_info;
7547
7548         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7549
7550         info->oti_hlock = ldiskfs_htree_lock_alloc();
7551         if (info->oti_hlock == NULL)
7552                 goto out_free_ea;
7553
7554         return info;
7555
7556 out_free_ea:
7557         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7558 out_free_info:
7559         OBD_FREE_PTR(info);
7560         return ERR_PTR(-ENOMEM);
7561 }
7562
7563 static void osd_key_fini(const struct lu_context *ctx,
7564                          struct lu_context_key *key, void *data)
7565 {
7566         struct osd_thread_info *info = data;
7567         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7568         struct osd_idmap_cache *idc = info->oti_ins_cache;
7569
7570         if (info->oti_dio_pages) {
7571                 int i;
7572                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7573                         struct page *page = info->oti_dio_pages[i];
7574                         if (page) {
7575                                 LASSERT(PagePrivate2(page));
7576                                 LASSERT(PageLocked(page));
7577                                 ClearPagePrivate2(page);
7578                                 unlock_page(page);
7579                                 __free_page(page);
7580                         }
7581                 }
7582                 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7583                                          PTLRPC_MAX_BRW_PAGES);
7584         }
7585
7586         if (info->oti_inode != NULL)
7587                 OBD_FREE_PTR(lli);
7588         if (info->oti_hlock != NULL)
7589                 ldiskfs_htree_lock_free(info->oti_hlock);
7590         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7591         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7592         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7593         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7594         lu_buf_free(&info->oti_big_buf);
7595         if (idc != NULL) {
7596                 LASSERT(info->oti_ins_cache_size > 0);
7597                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7598                 info->oti_ins_cache = NULL;
7599                 info->oti_ins_cache_size = 0;
7600         }
7601         OBD_FREE_PTR(info);
7602 }
7603
7604 static void osd_key_exit(const struct lu_context *ctx,
7605                          struct lu_context_key *key, void *data)
7606 {
7607         struct osd_thread_info *info = data;
7608
7609         LASSERT(info->oti_r_locks == 0);
7610         LASSERT(info->oti_w_locks == 0);
7611         LASSERT(info->oti_txns    == 0);
7612 }
7613
7614 /* type constructor/destructor: osd_type_init, osd_type_fini */
7615 LU_TYPE_INIT_FINI(osd, &osd_key);
7616
7617 struct lu_context_key osd_key = {
7618         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7619         .lct_init = osd_key_init,
7620         .lct_fini = osd_key_fini,
7621         .lct_exit = osd_key_exit
7622 };
7623
7624
7625 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7626                            const char *name, struct lu_device *next)
7627 {
7628         struct osd_device *osd = osd_dev(d);
7629
7630         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7631             sizeof(osd->od_svname))
7632                 return -E2BIG;
7633         return osd_procfs_init(osd, name);
7634 }
7635
7636 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7637 {
7638         struct seq_server_site *ss = osd_seq_site(osd);
7639         int rc = 0;
7640
7641         ENTRY;
7642
7643         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7644                 RETURN(0);
7645
7646         if (unlikely(ss == NULL))
7647                 RETURN(-ENODEV);
7648
7649         OBD_ALLOC_PTR(osd->od_cl_seq);
7650         if (osd->od_cl_seq == NULL)
7651                 RETURN(-ENOMEM);
7652
7653         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7654                         osd->od_svname, ss->ss_server_seq);
7655
7656         if (ss->ss_node_id == 0) {
7657                 /*
7658                  * If the OSD on the sequence controller(MDT0), then allocate
7659                  * sequence here, otherwise allocate sequence after connected
7660                  * to MDT0 (see mdt_register_lwp_callback()).
7661                  */
7662                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7663                                    &osd->od_cl_seq->lcs_space, env);
7664         }
7665
7666         RETURN(rc);
7667 }
7668
7669 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7670 {
7671         if (osd->od_cl_seq == NULL)
7672                 return;
7673
7674         seq_client_fini(osd->od_cl_seq);
7675         OBD_FREE_PTR(osd->od_cl_seq);
7676         osd->od_cl_seq = NULL;
7677 }
7678
7679 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7680 {
7681         ENTRY;
7682
7683         /* shutdown quota slave instance associated with the device */
7684         if (o->od_quota_slave_md != NULL) {
7685                 struct qsd_instance *qsd = o->od_quota_slave_md;
7686
7687                 o->od_quota_slave_md = NULL;
7688                 qsd_fini(env, qsd);
7689         }
7690
7691         if (o->od_quota_slave_dt != NULL) {
7692                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7693
7694                 o->od_quota_slave_dt = NULL;
7695                 qsd_fini(env, qsd);
7696         }
7697
7698         osd_fid_fini(env, o);
7699         osd_scrub_cleanup(env, o);
7700
7701         RETURN(0);
7702 }
7703
7704 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7705 {
7706         ENTRY;
7707
7708         if (o->od_mnt != NULL) {
7709                 shrink_dcache_sb(osd_sb(o));
7710                 osd_sync(env, &o->od_dt_dev);
7711                 wait_event(o->od_commit_cb_done,
7712                           !atomic_read(&o->od_commit_cb_in_flight));
7713
7714                 mntput(o->od_mnt);
7715                 o->od_mnt = NULL;
7716         }
7717
7718         EXIT;
7719 }
7720
7721 static int osd_mount(const struct lu_env *env,
7722                      struct osd_device *o, struct lustre_cfg *cfg)
7723 {
7724         const char *name = lustre_cfg_string(cfg, 0);
7725         const char *dev = lustre_cfg_string(cfg, 1);
7726         const char *opts;
7727         unsigned long page, s_flags = 0, lmd_flags = 0;
7728         struct page *__page;
7729         struct file_system_type *type;
7730         char *options = NULL;
7731         const char *str;
7732         struct osd_thread_info *info = osd_oti_get(env);
7733         struct lu_fid *fid = &info->oti_fid;
7734         struct inode *inode;
7735         int rc = 0, force_over_1024tb = 0;
7736
7737         ENTRY;
7738
7739         if (o->od_mnt != NULL)
7740                 RETURN(0);
7741
7742         if (strlen(dev) >= sizeof(o->od_mntdev))
7743                 RETURN(-E2BIG);
7744         strcpy(o->od_mntdev, dev);
7745
7746         str = lustre_cfg_buf(cfg, 2);
7747         sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
7748
7749         opts = lustre_cfg_string(cfg, 3);
7750 #ifdef __BIG_ENDIAN
7751         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7752                 CERROR("%s: device %s extents feature is not guaranteed to "
7753                        "work on big-endian systems. Use \"bigendian_extents\" "
7754                        "mount option to override.\n", name, dev);
7755                 RETURN(-EINVAL);
7756         }
7757 #endif
7758 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7759         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7760                 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");
7761         }
7762 #endif
7763 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7764         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7765                 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");
7766         }
7767 #endif
7768 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7769         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7770                 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");
7771         }
7772 #endif
7773
7774         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7775                 force_over_1024tb = 1;
7776
7777         __page = alloc_page(GFP_KERNEL);
7778         if (__page == NULL)
7779                 GOTO(out, rc = -ENOMEM);
7780         page = (unsigned long)page_address(__page);
7781         options = (char *)page;
7782         *options = '\0';
7783         if (opts != NULL) {
7784                 /* strip out the options for back compatiblity */
7785                 static const char * const sout[] = {
7786                         "mballoc",
7787                         "iopen",
7788                         "noiopen",
7789                         "iopen_nopriv",
7790                         "extents",
7791                         "noextents",
7792                         /* strip out option we processed in osd */
7793                         "bigendian_extents",
7794                         "force_over_128tb",
7795                         "force_over_256tb",
7796                         "force_over_512tb",
7797                         "force_over_1024tb",
7798                         NULL
7799                 };
7800                 strncat(options, opts, PAGE_SIZE);
7801                 for (rc = 0, str = options; sout[rc]; ) {
7802                         char *op = strstr(str, sout[rc]);
7803
7804                         if (op == NULL) {
7805                                 rc++;
7806                                 str = options;
7807                                 continue;
7808                         }
7809                         if (op == options || *(op - 1) == ',') {
7810                                 str = op + strlen(sout[rc]);
7811                                 if (*str == ',' || *str == '\0') {
7812                                         *str == ',' ? str++ : str;
7813                                         memmove(op, str, strlen(str) + 1);
7814                                 }
7815                         }
7816                         for (str = op; *str != ',' && *str != '\0'; str++)
7817                                 ;
7818                 }
7819         } else {
7820                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7821         }
7822
7823         /* Glom up mount options */
7824         if (*options != '\0')
7825                 strncat(options, ",", PAGE_SIZE);
7826         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7827
7828         type = get_fs_type("ldiskfs");
7829         if (!type) {
7830                 CERROR("%s: cannot find ldiskfs module\n", name);
7831                 GOTO(out, rc = -ENODEV);
7832         }
7833
7834         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7835         module_put(type->owner);
7836
7837         if (IS_ERR(o->od_mnt)) {
7838                 rc = PTR_ERR(o->od_mnt);
7839                 o->od_mnt = NULL;
7840                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7841                 GOTO(out, rc);
7842         }
7843
7844         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7845                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7846                                  force_over_1024tb == 0) {
7847                 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",
7848                        name, dev);
7849                 GOTO(out_mnt, rc = -EINVAL);
7850         }
7851
7852         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7853                 LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
7854                               name);
7855
7856                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7857         }
7858
7859         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7860                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7861                 GOTO(out_mnt, rc = -EINVAL);
7862         }
7863
7864 #ifdef LDISKFS_MOUNT_DIRDATA
7865         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7866                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7867         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7868                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7869                       "enabling the dirdata feature. If you do not want to "
7870                       "downgrade to Lustre-1.x again, you can enable it via "
7871                       "'tune2fs -O dirdata device'\n", name, dev);
7872 #endif
7873         inode = osd_sb(o)->s_root->d_inode;
7874         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7875         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7876         if (rc != 0) {
7877                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7878                 GOTO(out_mnt, rc);
7879         }
7880
7881         if (lmd_flags & LMD_FLG_NOSCRUB)
7882                 o->od_auto_scrub_interval = AS_NEVER;
7883
7884         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7885                 /* do not use pagecache with flash-backed storage */
7886                 o->od_writethrough_cache = 0;
7887                 o->od_read_cache = 0;
7888         }
7889
7890         GOTO(out, rc = 0);
7891
7892 out_mnt:
7893         mntput(o->od_mnt);
7894         o->od_mnt = NULL;
7895
7896 out:
7897         if (__page)
7898                 __free_page(__page);
7899
7900         return rc;
7901 }
7902
7903 static struct lu_device *osd_device_fini(const struct lu_env *env,
7904                                          struct lu_device *d)
7905 {
7906         struct osd_device *o = osd_dev(d);
7907
7908         ENTRY;
7909
7910         osd_index_backup(env, o, false);
7911         osd_shutdown(env, o);
7912         osd_procfs_fini(o);
7913         if (o->od_oi_table != NULL)
7914                 osd_oi_fini(osd_oti_get(env), o);
7915         if (o->od_extent_bytes_percpu)
7916                 free_percpu(o->od_extent_bytes_percpu);
7917         osd_obj_map_fini(o);
7918         osd_umount(env, o);
7919
7920         RETURN(NULL);
7921 }
7922
7923 static int osd_device_init0(const struct lu_env *env,
7924                             struct osd_device *o,
7925                             struct lustre_cfg *cfg)
7926 {
7927         struct lu_device *l = osd2lu_dev(o);
7928         struct osd_thread_info *info;
7929         int rc;
7930         int cplen = 0;
7931
7932         /* if the module was re-loaded, env can loose its keys */
7933         rc = lu_env_refill((struct lu_env *)env);
7934         if (rc)
7935                 GOTO(out, rc);
7936         info = osd_oti_get(env);
7937         LASSERT(info);
7938
7939         l->ld_ops = &osd_lu_ops;
7940         o->od_dt_dev.dd_ops = &osd_dt_ops;
7941
7942         spin_lock_init(&o->od_osfs_lock);
7943         mutex_init(&o->od_otable_mutex);
7944         INIT_LIST_HEAD(&o->od_orphan_list);
7945         INIT_LIST_HEAD(&o->od_index_backup_list);
7946         INIT_LIST_HEAD(&o->od_index_restore_list);
7947         spin_lock_init(&o->od_lock);
7948         o->od_index_backup_policy = LIBP_NONE;
7949         o->od_t10_type = 0;
7950         init_waitqueue_head(&o->od_commit_cb_done);
7951
7952         o->od_read_cache = 1;
7953         o->od_writethrough_cache = 1;
7954         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7955         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7956         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7957         o->od_auto_scrub_interval = AS_DEFAULT;
7958         /* default fallocate to unwritten extents: LU-14326/LU-14333 */
7959         o->od_fallocate_zero_blocks = 0;
7960
7961         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7962                         sizeof(o->od_svname));
7963         if (cplen >= sizeof(o->od_svname)) {
7964                 rc = -E2BIG;
7965                 GOTO(out, rc);
7966         }
7967
7968         o->od_index_backup_stop = 0;
7969         o->od_index = -1; /* -1 means index is invalid */
7970         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7971         if (rc == LDD_F_SV_TYPE_OST)
7972                 o->od_is_ost = 1;
7973
7974         o->od_full_scrub_ratio = OFSR_DEFAULT;
7975         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7976         rc = osd_mount(env, o, cfg);
7977         if (rc != 0)
7978                 GOTO(out, rc);
7979
7980         /* Can only check block device after mount */
7981         o->od_nonrotational =
7982                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7983
7984         rc = osd_obj_map_init(env, o);
7985         if (rc != 0)
7986                 GOTO(out_mnt, rc);
7987
7988         rc = lu_site_init(&o->od_site, l);
7989         if (rc != 0)
7990                 GOTO(out_compat, rc);
7991         o->od_site.ls_bottom_dev = l;
7992
7993         rc = lu_site_init_finish(&o->od_site);
7994         if (rc != 0)
7995                 GOTO(out_site, rc);
7996
7997         INIT_LIST_HEAD(&o->od_ios_list);
7998         /* setup scrub, including OI files initialization */
7999         o->od_in_init = 1;
8000         rc = osd_scrub_setup(env, o);
8001         o->od_in_init = 0;
8002         if (rc < 0)
8003                 GOTO(out_site, rc);
8004
8005         rc = osd_procfs_init(o, o->od_svname);
8006         if (rc != 0) {
8007                 CERROR("%s: can't initialize procfs: rc = %d\n",
8008                        o->od_svname, rc);
8009                 GOTO(out_scrub, rc);
8010         }
8011
8012         LASSERT(l->ld_site->ls_linkage.next != NULL);
8013         LASSERT(l->ld_site->ls_linkage.prev != NULL);
8014
8015         /* initialize quota slave instance */
8016         /* currently it's no need to prepare qsd_instance_md for OST */
8017         if (!o->od_is_ost) {
8018                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
8019                                                 &o->od_dt_dev,
8020                                                 o->od_proc_entry, true);
8021                 if (IS_ERR(o->od_quota_slave_md)) {
8022                         rc = PTR_ERR(o->od_quota_slave_md);
8023                         o->od_quota_slave_md = NULL;
8024                         GOTO(out_procfs, rc);
8025                 }
8026         }
8027
8028         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
8029                                         o->od_proc_entry, false);
8030
8031         if (IS_ERR(o->od_quota_slave_dt)) {
8032                 if (o->od_quota_slave_md != NULL) {
8033                         qsd_fini(env, o->od_quota_slave_md);
8034                         o->od_quota_slave_md = NULL;
8035                 }
8036
8037                 rc = PTR_ERR(o->od_quota_slave_dt);
8038                 o->od_quota_slave_dt = NULL;
8039                 GOTO(out_procfs, rc);
8040         }
8041
8042         o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
8043         if (!o->od_extent_bytes_percpu) {
8044                 rc = -ENOMEM;
8045                 GOTO(out_procfs, rc);
8046         }
8047
8048         RETURN(0);
8049
8050 out_procfs:
8051         osd_procfs_fini(o);
8052 out_scrub:
8053         osd_scrub_cleanup(env, o);
8054 out_site:
8055         lu_site_fini(&o->od_site);
8056 out_compat:
8057         osd_obj_map_fini(o);
8058 out_mnt:
8059         osd_umount(env, o);
8060 out:
8061         return rc;
8062 }
8063
8064 static struct lu_device *osd_device_alloc(const struct lu_env *env,
8065                                           struct lu_device_type *t,
8066                                           struct lustre_cfg *cfg)
8067 {
8068         struct osd_device *o;
8069         int rc;
8070
8071         OBD_ALLOC_PTR(o);
8072         if (o == NULL)
8073                 return ERR_PTR(-ENOMEM);
8074
8075         rc = dt_device_init(&o->od_dt_dev, t);
8076         if (rc == 0) {
8077                 /*
8078                  * Because the ctx might be revived in dt_device_init,
8079                  * refill the env here
8080                  */
8081                 lu_env_refill((struct lu_env *)env);
8082                 rc = osd_device_init0(env, o, cfg);
8083                 if (rc)
8084                         dt_device_fini(&o->od_dt_dev);
8085         }
8086
8087         if (unlikely(rc != 0))
8088                 OBD_FREE_PTR(o);
8089
8090         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
8091 }
8092
8093 static struct lu_device *osd_device_free(const struct lu_env *env,
8094                                          struct lu_device *d)
8095 {
8096         struct osd_device *o = osd_dev(d);
8097
8098         ENTRY;
8099
8100         /* XXX: make osd top device in order to release reference */
8101         d->ld_site->ls_top_dev = d;
8102         lu_site_purge(env, d->ld_site, -1);
8103         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8104                       D_ERROR, lu_cdebug_printer);
8105         lu_site_fini(&o->od_site);
8106         dt_device_fini(&o->od_dt_dev);
8107         OBD_FREE_PTR(o);
8108         RETURN(NULL);
8109 }
8110
8111 static int osd_process_config(const struct lu_env *env,
8112                               struct lu_device *d, struct lustre_cfg *cfg)
8113 {
8114         struct osd_device *o = osd_dev(d);
8115         ssize_t count;
8116         int rc;
8117
8118         ENTRY;
8119
8120         switch (cfg->lcfg_command) {
8121         case LCFG_SETUP:
8122                 rc = osd_mount(env, o, cfg);
8123                 break;
8124         case LCFG_CLEANUP:
8125                 /*
8126                  * For the case LCFG_PRE_CLEANUP is not called in advance,
8127                  * that may happend if hit failure during mount process.
8128                  */
8129                 osd_index_backup(env, o, false);
8130                 lu_dev_del_linkage(d->ld_site, d);
8131                 rc = osd_shutdown(env, o);
8132                 break;
8133         case LCFG_PARAM:
8134                 LASSERT(&o->od_dt_dev);
8135                 count  = class_modify_config(cfg, PARAM_OSD,
8136                                              &o->od_dt_dev.dd_kobj);
8137                 if (count < 0)
8138                         count = class_modify_config(cfg, PARAM_OST,
8139                                                     &o->od_dt_dev.dd_kobj);
8140                 rc = count > 0 ? 0 : count;
8141                 break;
8142         case LCFG_PRE_CLEANUP:
8143                 osd_scrub_stop(o);
8144                 osd_index_backup(env, o,
8145                                  o->od_index_backup_policy != LIBP_NONE);
8146                 rc = 0;
8147                 break;
8148         default:
8149                 rc = -ENOSYS;
8150         }
8151
8152         RETURN(rc);
8153 }
8154
8155 static int osd_recovery_complete(const struct lu_env *env,
8156                                  struct lu_device *d)
8157 {
8158         struct osd_device *osd = osd_dev(d);
8159         int rc = 0;
8160
8161         ENTRY;
8162
8163         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8164                 RETURN(0);
8165
8166         /*
8167          * start qsd instance on recovery completion, this notifies the quota
8168          * slave code that we are about to process new requests now
8169          */
8170         rc = qsd_start(env, osd->od_quota_slave_dt);
8171         if (rc == 0 && osd->od_quota_slave_md != NULL)
8172                 rc = qsd_start(env, osd->od_quota_slave_md);
8173
8174         RETURN(rc);
8175 }
8176
8177 /*
8178  * we use exports to track all osd users
8179  */
8180 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8181                            struct obd_device *obd, struct obd_uuid *cluuid,
8182                            struct obd_connect_data *data, void *localdata)
8183 {
8184         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8185         struct lustre_handle conn;
8186         int rc;
8187
8188         ENTRY;
8189
8190         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8191
8192         rc = class_connect(&conn, obd, cluuid);
8193         if (rc)
8194                 RETURN(rc);
8195
8196         *exp = class_conn2export(&conn);
8197
8198         spin_lock(&osd->od_osfs_lock);
8199         osd->od_connects++;
8200         spin_unlock(&osd->od_osfs_lock);
8201
8202         RETURN(0);
8203 }
8204
8205 /*
8206  * once last export (we don't count self-export) disappeared
8207  * osd can be released
8208  */
8209 static int osd_obd_disconnect(struct obd_export *exp)
8210 {
8211         struct obd_device *obd = exp->exp_obd;
8212         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8213         int rc, release = 0;
8214
8215         ENTRY;
8216
8217         /* Only disconnect the underlying layers on the final disconnect. */
8218         spin_lock(&osd->od_osfs_lock);
8219         osd->od_connects--;
8220         if (osd->od_connects == 0)
8221                 release = 1;
8222         spin_unlock(&osd->od_osfs_lock);
8223
8224         rc = class_disconnect(exp); /* bz 9811 */
8225
8226         if (rc == 0 && release)
8227                 class_manual_cleanup(obd);
8228         RETURN(rc);
8229 }
8230
8231 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8232                        struct lu_device *dev)
8233 {
8234         struct osd_device *osd = osd_dev(dev);
8235         struct lr_server_data *lsd =
8236                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8237         int result = 0;
8238
8239         ENTRY;
8240
8241         if (osd->od_quota_slave_md != NULL) {
8242                 /* set up quota slave objects for inode */
8243                 result = qsd_prepare(env, osd->od_quota_slave_md);
8244                 if (result != 0)
8245                         RETURN(result);
8246         }
8247
8248         if (osd->od_quota_slave_dt != NULL) {
8249                 /* set up quota slave objects for block */
8250                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8251                 if (result != 0)
8252                         RETURN(result);
8253         }
8254
8255
8256         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8257 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8258                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8259                         osd->od_index_in_idif = 1;
8260                 } else {
8261                         osd->od_index_in_idif = 0;
8262                         result = osd_register_proc_index_in_idif(osd);
8263                         if (result != 0)
8264                                 RETURN(result);
8265                 }
8266 #else
8267                 osd->od_index_in_idif = 1;
8268 #endif
8269         }
8270
8271         result = osd_fid_init(env, osd);
8272
8273         RETURN(result);
8274 }
8275
8276 /**
8277  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8278  *
8279  * Allocate FID.
8280  *
8281  * see include/lu_object.h for the details.
8282  */
8283 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8284                          struct lu_fid *fid, struct lu_object *parent,
8285                          const struct lu_name *name)
8286 {
8287         struct osd_device *osd = osd_dev(d);
8288
8289         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8290 }
8291
8292 static const struct lu_object_operations osd_lu_obj_ops = {
8293         .loo_object_init      = osd_object_init,
8294         .loo_object_delete    = osd_object_delete,
8295         .loo_object_release   = osd_object_release,
8296         .loo_object_free      = osd_object_free,
8297         .loo_object_print     = osd_object_print,
8298         .loo_object_invariant = osd_object_invariant
8299 };
8300
8301 const struct lu_device_operations osd_lu_ops = {
8302         .ldo_object_alloc      = osd_object_alloc,
8303         .ldo_process_config    = osd_process_config,
8304         .ldo_recovery_complete = osd_recovery_complete,
8305         .ldo_prepare           = osd_prepare,
8306         .ldo_fid_alloc         = osd_fid_alloc,
8307 };
8308
8309 static const struct lu_device_type_operations osd_device_type_ops = {
8310         .ldto_init = osd_type_init,
8311         .ldto_fini = osd_type_fini,
8312
8313         .ldto_start = osd_type_start,
8314         .ldto_stop  = osd_type_stop,
8315
8316         .ldto_device_alloc = osd_device_alloc,
8317         .ldto_device_free  = osd_device_free,
8318
8319         .ldto_device_init = osd_device_init,
8320         .ldto_device_fini = osd_device_fini
8321 };
8322
8323 static struct lu_device_type osd_device_type = {
8324         .ldt_tags     = LU_DEVICE_DT,
8325         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8326         .ldt_ops      = &osd_device_type_ops,
8327         .ldt_ctx_tags = LCT_LOCAL,
8328 };
8329
8330 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8331 {
8332         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8333         struct super_block *sb = osd_sb(osd);
8334
8335         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8336 }
8337
8338 /*
8339  * lprocfs legacy support.
8340  */
8341 static const struct obd_ops osd_obd_device_ops = {
8342         .o_owner = THIS_MODULE,
8343         .o_connect      = osd_obd_connect,
8344         .o_disconnect   = osd_obd_disconnect,
8345         .o_health_check = osd_health_check,
8346 };
8347
8348 static ssize_t track_declares_assert_show(struct kobject *kobj,
8349                                    struct attribute *attr,
8350                                    char *buf)
8351 {
8352         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8353 }
8354
8355 static ssize_t track_declares_assert_store(struct kobject *kobj,
8356                                            struct attribute *attr,
8357                                            const char *buffer, size_t count)
8358 {
8359         bool track_declares_assert;
8360         int rc;
8361
8362         rc = kstrtobool(buffer, &track_declares_assert);
8363         if (rc)
8364                 return rc;
8365
8366         ldiskfs_track_declares_assert = track_declares_assert;
8367
8368         return count;
8369 }
8370 LUSTRE_RW_ATTR(track_declares_assert);
8371
8372 static int __init osd_init(void)
8373 {
8374         struct kobject *kobj;
8375         int rc;
8376
8377         BUILD_BUG_ON(BH_DXLock >=
8378                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8379 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8380         /* please, try to keep osd_thread_info smaller than a page */
8381         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8382 #endif
8383
8384         osd_oi_mod_init();
8385
8386         rc = lu_kmem_init(ldiskfs_caches);
8387         if (rc)
8388                 return rc;
8389
8390 #ifdef CONFIG_KALLSYMS
8391         priv_security_file_alloc =
8392                 (void *)kallsyms_lookup_name("security_file_alloc");
8393 #endif
8394
8395         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8396                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8397         if (rc) {
8398                 lu_kmem_fini(ldiskfs_caches);
8399                 return rc;
8400         }
8401
8402         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8403         if (kobj) {
8404                 rc = sysfs_create_file(kobj,
8405                                        &lustre_attr_track_declares_assert.attr);
8406                 kobject_put(kobj);
8407                 if (rc) {
8408                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8409                         rc = 0;
8410                 }
8411         }
8412         return rc;
8413 }
8414
8415 static void __exit osd_exit(void)
8416 {
8417         struct kobject *kobj;
8418
8419         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8420         if (kobj) {
8421                 sysfs_remove_file(kobj,
8422                                   &lustre_attr_track_declares_assert.attr);
8423                 kobject_put(kobj);
8424         }
8425         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8426         lu_kmem_fini(ldiskfs_caches);
8427 }
8428
8429 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8430 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8431 MODULE_VERSION(LUSTRE_VERSION_STRING);
8432 MODULE_LICENSE("GPL");
8433
8434 module_init(osd_init);
8435 module_exit(osd_exit);