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