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 = &oti->oti_file;
 961         const struct file_operations *fops;
 962         struct lmv_mds_md_v1 *lmv1;
 963         struct osd_check_lmv_buf oclb = {
 964                 .ctx.actor = osd_stripe_dir_filldir,
 965                 .oclb_info = oti,
 966                 .oclb_dev = dev,
 967                 .oclb_oic = oic,
 968                 .oclb_found = false,
 969         };
 970         int rc = 0;
 971
 972         ENTRY;
 973
 974 again:
 975         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 976                              buf->lb_len);
 977         if (rc == -ERANGE) {
 978                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 979                 if (rc > 0) {
 980                         lu_buf_realloc(buf, rc);
 981                         if (buf->lb_buf == NULL)
 982                                 GOTO(out, rc = -ENOMEM);
 983
 984                         goto again;
 985                 }
 986         }
 987
 988         if (unlikely(rc == 0 || rc == -ENODATA))
 989                 GOTO(out, rc = 0);
 990
 991         if (rc < 0)
 992                 GOTO(out, rc);
 993
 994         if (unlikely(buf->lb_buf == NULL)) {
 995                 lu_buf_realloc(buf, rc);
 996                 if (buf->lb_buf == NULL)
 997                         GOTO(out, rc = -ENOMEM);
 998
 999                 goto again;
1000         }
1001
1002         lmv1 = buf->lb_buf;
1003         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1004                 GOTO(out, rc = 0);
1005
1006         fops = inode->i_fop;
1007         dentry->d_inode = inode;
1008         dentry->d_sb = inode->i_sb;
1009         filp->f_pos = 0;
1010         filp->f_path.dentry = dentry;
1011         filp->f_flags |= O_NOATIME;
1012         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1013         filp->f_mapping = inode->i_mapping;
1014         filp->f_op = fops;
1015         filp->private_data = NULL;
1016         filp->f_cred = current_cred();
1017         filp->f_inode = inode;
1018         rc = osd_security_file_alloc(filp);
1019         if (rc)
1020                 goto out;
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024                 rc = iterate_dir(filp, &oclb.ctx);
1025         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1026                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1027         fops->release(inode, filp);
1028
1029 out:
1030         if (rc < 0)
1031                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1032                        DFID": rc = %d\n", osd_ino2name(inode),
1033                        inode->i_ino, inode->i_generation,
1034                        PFID(&oic->oic_fid), rc);
1035         else
1036                 rc = 0;
1037
1038         RETURN(rc);
1039 }
1040
1041 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1042                           const struct lu_fid *fid,
1043                           const struct lu_object_conf *conf)
1044 {
1045         struct osd_thread_info *info;
1046         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1047         struct osd_device *dev;
1048         struct osd_idmap_cache *oic;
1049         struct osd_inode_id *id;
1050         struct inode *inode = NULL;
1051         struct lustre_scrub *scrub;
1052         struct scrub_file *sf;
1053         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1054         __u32 saved_ino;
1055         __u32 saved_gen;
1056         int result = 0;
1057         int rc1 = 0;
1058         bool remote = false;
1059         bool trusted = true;
1060         bool updated = false;
1061         bool checked = false;
1062
1063         ENTRY;
1064
1065         LINVRNT(osd_invariant(obj));
1066         LASSERT(obj->oo_inode == NULL);
1067         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1068
1069         dev = osd_dev(ldev);
1070         scrub = &dev->od_scrub.os_scrub;
1071         sf = &scrub->os_file;
1072         info = osd_oti_get(env);
1073         LASSERT(info);
1074         oic = &info->oti_cache;
1075
1076         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1077                 RETURN(-ENOENT);
1078
1079         /*
1080          * For the object is created as locking anchor, or for the object to
1081          * be created on disk. No need to osd_oi_lookup() at here because FID
1082          * shouldn't never be re-used, if it's really a duplicate FID from
1083          * unexpected reason, we should be able to detect it later by calling
1084          * do_create->osd_oi_insert().
1085          */
1086         if (conf && conf->loc_flags & LOC_F_NEW)
1087                 GOTO(out, result = 0);
1088
1089         /* Search order: 1. per-thread cache. */
1090         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1091                 id = &oic->oic_lid;
1092                 goto iget;
1093         }
1094
1095         id = &info->oti_id;
1096         if (!list_empty(&scrub->os_inconsistent_items)) {
1097                 /* Search order: 2. OI scrub pending list. */
1098                 result = osd_oii_lookup(dev, fid, id);
1099                 if (!result)
1100                         goto iget;
1101         }
1102
1103         /*
1104          * The OI mapping in the OI file can be updated by the OI scrub
1105          * when we locate the inode via FID. So it may be not trustable.
1106          */
1107         trusted = false;
1108
1109         /* Search order: 3. OI files. */
1110         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1111         if (result == -ENOENT) {
1112                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1113                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1114                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1115                                       sf->sf_oi_bitmap))
1116                         GOTO(out, result = 0);
1117
1118                 goto trigger;
1119         }
1120
1121         /* -ESTALE is returned if inode of OST object doesn't exist */
1122         if (result == -ESTALE &&
1123             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1124                 GOTO(out, result = 0);
1125         }
1126
1127         if (result)
1128                 GOTO(out, result);
1129
1130 iget:
1131         obj->oo_inode = NULL;
1132         /* for later passes through checks, not true on first pass */
1133         if (!IS_ERR_OR_NULL(inode))
1134                 iput(inode);
1135
1136         inode = osd_iget_check(info, dev, fid, id, trusted);
1137         if (!IS_ERR(inode)) {
1138                 obj->oo_inode = inode;
1139                 result = 0;
1140                 if (remote)
1141                         goto trigger;
1142
1143                 goto check_lma;
1144         }
1145
1146         result = PTR_ERR(inode);
1147         if (result == -ENOENT || result == -ESTALE)
1148                 GOTO(out, result = 0);
1149
1150         if (result != -EREMCHG)
1151                 GOTO(out, result);
1152
1153 trigger:
1154         /*
1155          * We still have chance to get the valid inode: for the
1156          * object which is referenced by remote name entry, the
1157          * object on the local MDT will be linked under the dir
1158          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1159          *
1160          * We do not know whether the object for the given FID
1161          * is referenced by some remote name entry or not, and
1162          * especially for DNE II, a multiple-linked object may
1163          * have many name entries reside on many MDTs.
1164          *
1165          * To simplify the operation, OSD will not distinguish
1166          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1167          * only happened for the RPC from other MDT during the
1168          * OI scrub, or for the client side RPC with FID only,
1169          * such as FID to path, or from old connected client.
1170          */
1171         if (!remote) {
1172                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1173                 if (!rc1) {
1174                         remote = true;
1175                         trusted = true;
1176                         flags |= SS_AUTO_PARTIAL;
1177                         flags &= ~SS_AUTO_FULL;
1178                         goto iget;
1179                 }
1180         }
1181
1182         if (thread_is_running(&scrub->os_thread)) {
1183                 if (scrub->os_partial_scan && !scrub->os_in_join)
1184                         goto join;
1185
1186                 if (IS_ERR_OR_NULL(inode) || result)
1187                         GOTO(out, result = -EINPROGRESS);
1188
1189                 LASSERT(remote);
1190                 LASSERT(obj->oo_inode == inode);
1191
1192                 osd_add_oi_cache(info, dev, id, fid);
1193                 osd_oii_insert(dev, oic, true);
1194                 goto found;
1195         }
1196
1197         if (dev->od_auto_scrub_interval == AS_NEVER) {
1198                 if (!remote)
1199                         GOTO(out, result = -EREMCHG);
1200
1201                 LASSERT(!result);
1202                 LASSERT(obj->oo_inode == inode);
1203
1204                 osd_add_oi_cache(info, dev, id, fid);
1205                 goto found;
1206         }
1207
1208 join:
1209         rc1 = osd_scrub_start(env, dev, flags);
1210         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1211                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1212         if (rc1 && rc1 != -EALREADY)
1213                 GOTO(out, result = -EREMCHG);
1214
1215         if (IS_ERR_OR_NULL(inode) || result)
1216                 GOTO(out, result = -EINPROGRESS);
1217
1218         LASSERT(remote);
1219         LASSERT(obj->oo_inode == inode);
1220
1221         osd_add_oi_cache(info, dev, id, fid);
1222         osd_oii_insert(dev, oic, true);
1223         goto found;
1224
1225 check_lma:
1226         checked = true;
1227         if (unlikely(obj->oo_header))
1228                 goto found;
1229
1230         result = osd_check_lma(env, obj);
1231         if (!result)
1232                 goto found;
1233
1234         LASSERTF(id->oii_ino == inode->i_ino &&
1235                  id->oii_gen == inode->i_generation,
1236                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1237                  PFID(fid), id->oii_ino, id->oii_gen,
1238                  inode->i_ino, inode->i_generation);
1239
1240         saved_ino = inode->i_ino;
1241         saved_gen = inode->i_generation;
1242
1243         if (unlikely(result == -ENODATA)) {
1244                 /*
1245                  * If the OI scrub updated the OI mapping by race, it
1246                  * must be valid. Trust the inode that has no LMA EA.
1247                  */
1248                 if (updated)
1249                         goto found;
1250
1251                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1252                 if (!result) {
1253                         /*
1254                          * The OI mapping is still there, the inode is still
1255                          * valid. It is just becaues the inode has no LMA EA.
1256                          */
1257                         if (saved_ino == id->oii_ino &&
1258                             saved_gen == id->oii_gen)
1259                                 goto found;
1260
1261                         /*
1262                          * It is the OI scrub updated the OI mapping by race.
1263                          * The new OI mapping must be valid.
1264                          */
1265                         trusted = true;
1266                         updated = true;
1267                         goto iget;
1268                 }
1269
1270                 /*
1271                  * "result == -ENOENT" means that the OI mappinghas been
1272                  * removed by race, so the inode belongs to other object.
1273                  *
1274                  * Others error can be returned  directly.
1275                  */
1276                 if (result == -ENOENT) {
1277                         LASSERT(trusted);
1278
1279                         obj->oo_inode = NULL;
1280                         result = 0;
1281                 }
1282         }
1283
1284         if (result != -EREMCHG)
1285                 GOTO(out, result);
1286
1287         LASSERT(!updated);
1288
1289         /*
1290          * if two OST objects map to the same inode, and inode mode is
1291          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1292          * reserved by precreate, and not written yet, in this case, don't
1293          * set inode for the object whose FID mismatch, so that it can create
1294          * inode and not block precreate.
1295          */
1296         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1297             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1298                 obj->oo_inode = NULL;
1299                 GOTO(out, result = 0);
1300         }
1301
1302         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1303         /*
1304          * "result == -ENOENT" means the cached OI mapping has been removed
1305          * from the OI file by race, above inode belongs to other object.
1306          */
1307         if (result == -ENOENT) {
1308                 LASSERT(trusted);
1309
1310                 obj->oo_inode = NULL;
1311                 GOTO(out, result = 0);
1312         }
1313
1314         if (result)
1315                 GOTO(out, result);
1316
1317         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1318                 result = -EREMCHG;
1319                 goto trigger;
1320         }
1321
1322         /*
1323          * It is the OI scrub updated the OI mapping by race.
1324          * The new OI mapping must be valid.
1325          */
1326         trusted = true;
1327         updated = true;
1328         goto iget;
1329
1330 found:
1331         if (!checked) {
1332                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1333                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1334
1335                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1336                 if (!result) {
1337                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1338                             dev->od_is_ost)
1339                                 obj->oo_pfid_in_lma = 1;
1340                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1341                             !dev->od_is_ost)
1342                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1343                 } else if (result != -ENODATA) {
1344                         GOTO(out, result);
1345                 }
1346         }
1347
1348         obj->oo_compat_dot_created = 1;
1349         obj->oo_compat_dotdot_created = 1;
1350
1351         if (S_ISDIR(inode->i_mode) &&
1352             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1353                 osd_check_lmv(info, dev, inode, oic);
1354
1355         result = osd_attach_jinode(inode);
1356         if (result)
1357                 GOTO(out, result);
1358
1359         if (!ldiskfs_pdo)
1360                 GOTO(out, result = 0);
1361
1362         LASSERT(!obj->oo_hl_head);
1363         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1364
1365         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1366
1367 out:
1368         if (result || !obj->oo_inode) {
1369                 if (!IS_ERR_OR_NULL(inode))
1370                         iput(inode);
1371
1372                 obj->oo_inode = NULL;
1373                 if (trusted)
1374                         fid_zero(&oic->oic_fid);
1375         }
1376
1377         LINVRNT(osd_invariant(obj));
1378         return result;
1379 }
1380
1381 /*
1382  * Concurrency: shouldn't matter.
1383  */
1384 static void osd_object_init0(struct osd_object *obj)
1385 {
1386         LASSERT(obj->oo_inode != NULL);
1387         obj->oo_dt.do_body_ops = &osd_body_ops;
1388         obj->oo_dt.do_lu.lo_header->loh_attr |=
1389                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1390 }
1391
1392 /*
1393  * Concurrency: no concurrent access is possible that early in object
1394  * life-cycle.
1395  */
1396 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1397                            const struct lu_object_conf *conf)
1398 {
1399         struct osd_object *obj = osd_obj(l);
1400         int result;
1401
1402         LINVRNT(osd_invariant(obj));
1403
1404         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1405             cfs_fail_val == 2) {
1406                 struct osd_thread_info *info = osd_oti_get(env);
1407                 struct osd_idmap_cache *oic = &info->oti_cache;
1408                 /* invalidate thread cache */
1409                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1410         }
1411         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1412                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1413                 l->lo_header->loh_attr |= LOHA_EXISTS;
1414                 return 0;
1415         }
1416
1417         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1418         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1419         if (result == 0 && obj->oo_inode != NULL) {
1420                 struct osd_thread_info *oti = osd_oti_get(env);
1421                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1422
1423                 osd_object_init0(obj);
1424                 if (unlikely(obj->oo_header))
1425                         return 0;
1426
1427                 result = osd_get_lma(oti, obj->oo_inode,
1428                                      &oti->oti_obj_dentry, loa);
1429                 if (!result) {
1430                         /*
1431                          * Convert LMAI flags to lustre LMA flags
1432                          * and cache it to oo_lma_flags
1433                          */
1434                         obj->oo_lma_flags =
1435                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1436                 } else if (result == -ENODATA) {
1437                         result = 0;
1438                 }
1439         }
1440         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1441
1442         LINVRNT(osd_invariant(obj));
1443         return result;
1444 }
1445
1446 /*
1447  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1448  * The left part is for value, binary mode.
1449  */
1450 struct osd_xattr_entry {
1451         struct list_head        oxe_list;
1452         size_t                  oxe_len;
1453         size_t                  oxe_namelen;
1454         bool                    oxe_exist;
1455         struct rcu_head         oxe_rcu;
1456         char                    oxe_buf[0];
1457 };
1458
1459 static int osd_oxc_get(struct osd_object *obj, const char *name,
1460                        struct lu_buf *buf)
1461 {
1462         struct osd_xattr_entry *tmp;
1463         struct osd_xattr_entry *oxe = NULL;
1464         size_t namelen = strlen(name);
1465         int rc;
1466
1467         rcu_read_lock();
1468         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1469                 if (namelen == tmp->oxe_namelen &&
1470                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1471                         oxe = tmp;
1472                         break;
1473                 }
1474         }
1475
1476         if (oxe == NULL)
1477                 GOTO(out, rc = -ENOENT);
1478
1479         if (!oxe->oxe_exist)
1480                 GOTO(out, rc = -ENODATA);
1481
1482         /* vallen */
1483         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1484         LASSERT(rc > 0);
1485
1486         if (buf->lb_buf == NULL)
1487                 GOTO(out, rc);
1488
1489         if (buf->lb_len < rc)
1490                 GOTO(out, rc = -ERANGE);
1491
1492         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1493 out:
1494         rcu_read_unlock();
1495
1496         return rc;
1497 }
1498
1499 static void osd_oxc_free(struct rcu_head *head)
1500 {
1501         struct osd_xattr_entry *oxe;
1502
1503         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1504         OBD_FREE(oxe, oxe->oxe_len);
1505 }
1506
1507 static void osd_oxc_add(struct osd_object *obj, const char *name,
1508                         const char *buf, int buflen)
1509 {
1510         struct osd_xattr_entry *oxe;
1511         struct osd_xattr_entry *old = NULL;
1512         struct osd_xattr_entry *tmp;
1513         size_t namelen = strlen(name);
1514         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1515
1516         OBD_ALLOC(oxe, len);
1517         if (oxe == NULL)
1518                 return;
1519
1520         INIT_LIST_HEAD(&oxe->oxe_list);
1521         oxe->oxe_len = len;
1522         oxe->oxe_namelen = namelen;
1523         memcpy(oxe->oxe_buf, name, namelen);
1524         if (buflen > 0) {
1525                 LASSERT(buf != NULL);
1526                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1527                 oxe->oxe_exist = true;
1528         } else {
1529                 oxe->oxe_exist = false;
1530         }
1531
1532         /* this should be rarely called, just remove old and add new */
1533         spin_lock(&obj->oo_guard);
1534         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1535                 if (namelen == tmp->oxe_namelen &&
1536                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1537                         old = tmp;
1538                         break;
1539                 }
1540         }
1541         if (old != NULL) {
1542                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1543                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1544         } else {
1545                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1546         }
1547         spin_unlock(&obj->oo_guard);
1548 }
1549
1550 static void osd_oxc_del(struct osd_object *obj, const char *name)
1551 {
1552         struct osd_xattr_entry *oxe;
1553         size_t namelen = strlen(name);
1554
1555         spin_lock(&obj->oo_guard);
1556         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1557                 if (namelen == oxe->oxe_namelen &&
1558                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1559                         list_del_rcu(&oxe->oxe_list);
1560                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1561                         break;
1562                 }
1563         }
1564         spin_unlock(&obj->oo_guard);
1565 }
1566
1567 static void osd_oxc_fini(struct osd_object *obj)
1568 {
1569         struct osd_xattr_entry *oxe, *next;
1570
1571         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1572                 list_del(&oxe->oxe_list);
1573                 OBD_FREE(oxe, oxe->oxe_len);
1574         }
1575 }
1576
1577 /*
1578  * Concurrency: no concurrent access is possible that late in object
1579  * life-cycle.
1580  */
1581 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1582 {
1583         struct osd_object *obj = osd_obj(l);
1584         struct lu_object_header *h = obj->oo_header;
1585
1586         LINVRNT(osd_invariant(obj));
1587
1588         osd_oxc_fini(obj);
1589         dt_object_fini(&obj->oo_dt);
1590         if (obj->oo_hl_head != NULL)
1591                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1592         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1593         OBD_FREE_PTR(obj);
1594         if (unlikely(h)) {
1595                 lu_object_header_fini(h);
1596                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1597                 kfree_rcu(h, loh_rcu);
1598         }
1599 }
1600
1601 /*
1602  * Concurrency: no concurrent access is possible that late in object
1603  * life-cycle.
1604  */
1605 static void osd_index_fini(struct osd_object *o)
1606 {
1607         struct iam_container *bag;
1608
1609         if (o->oo_dir != NULL) {
1610                 bag = &o->oo_dir->od_container;
1611                 if (o->oo_inode != NULL) {
1612                         if (bag->ic_object == o->oo_inode)
1613                                 iam_container_fini(bag);
1614                 }
1615                 OBD_FREE_PTR(o->oo_dir);
1616                 o->oo_dir = NULL;
1617         }
1618 }
1619
1620 /*
1621  * Concurrency: no concurrent access is possible that late in object
1622  * life-cycle (for all existing callers, that is. New callers have to provide
1623  * their own locking.)
1624  */
1625 static int osd_inode_unlinked(const struct inode *inode)
1626 {
1627         return inode->i_nlink == 0;
1628 }
1629
1630 enum {
1631         OSD_TXN_OI_DELETE_CREDITS    = 20,
1632         OSD_TXN_INODE_DELETE_CREDITS = 20
1633 };
1634
1635 /*
1636  * Journal
1637  */
1638
1639 #if OSD_THANDLE_STATS
1640 /**
1641  * Set time when the handle is allocated
1642  */
1643 static void osd_th_alloced(struct osd_thandle *oth)
1644 {
1645         oth->oth_alloced = ktime_get();
1646 }
1647
1648 /**
1649  * Set time when the handle started
1650  */
1651 static void osd_th_started(struct osd_thandle *oth)
1652 {
1653         oth->oth_started = ktime_get();
1654 }
1655
1656 /**
1657  * Check whether the we deal with this handle for too long.
1658  */
1659 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1660                                 ktime_t alloced, ktime_t started,
1661                                 ktime_t closed)
1662 {
1663         ktime_t now = ktime_get();
1664
1665         LASSERT(dev != NULL);
1666
1667         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1668                             ktime_us_delta(started, alloced));
1669         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1670                             ktime_us_delta(closed, started));
1671         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1672                             ktime_us_delta(now, closed));
1673
1674         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1675                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1676                                 oth, now, alloced, started, closed);
1677                 libcfs_debug_dumpstack(NULL);
1678         }
1679 }
1680
1681 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1682 {                                                                       \
1683         ktime_t __closed = ktime_get();                                 \
1684         ktime_t __alloced = oth->oth_alloced;                           \
1685         ktime_t __started = oth->oth_started;                           \
1686                                                                         \
1687         expr;                                                           \
1688         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1689 }
1690
1691 #else /* OSD_THANDLE_STATS */
1692
1693 #define osd_th_alloced(h)                  do {} while(0)
1694 #define osd_th_started(h)                  do {} while(0)
1695 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1696
1697 #endif /* OSD_THANDLE_STATS */
1698
1699 /*
1700  * Concurrency: doesn't access mutable data.
1701  */
1702 static int osd_param_is_not_sane(const struct osd_device *dev,
1703                                  const struct thandle *th)
1704 {
1705         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1706
1707         return oh->ot_credits > osd_transaction_size(dev);
1708 }
1709
1710 /*
1711  * Concurrency: shouldn't matter.
1712  */
1713 static void osd_trans_commit_cb(struct super_block *sb,
1714                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1715 {
1716         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1717         struct thandle *th = &oh->ot_super;
1718         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1719         struct dt_txn_commit_cb *dcb, *tmp;
1720
1721         LASSERT(oh->ot_handle == NULL);
1722
1723         if (error)
1724                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1725
1726         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1727         /* call per-transaction callbacks if any */
1728         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1729                                  dcb_linkage) {
1730                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1731                          "commit callback entry: magic=%x name='%s'\n",
1732                          dcb->dcb_magic, dcb->dcb_name);
1733                 list_del_init(&dcb->dcb_linkage);
1734                 dcb->dcb_func(NULL, th, dcb, error);
1735         }
1736
1737         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1738         lu_device_put(lud);
1739         th->th_dev = NULL;
1740
1741         OBD_FREE_PTR(oh);
1742 }
1743
1744 static struct thandle *osd_trans_create(const struct lu_env *env,
1745                                         struct dt_device *d)
1746 {
1747         struct osd_thread_info *oti = osd_oti_get(env);
1748         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1749         struct osd_thandle *oh;
1750         struct thandle *th;
1751
1752         ENTRY;
1753
1754         if (d->dd_rdonly) {
1755                 CERROR("%s: someone try to start transaction under "
1756                        "readonly mode, should be disabled.\n",
1757                        osd_name(osd_dt_dev(d)));
1758                 dump_stack();
1759                 RETURN(ERR_PTR(-EROFS));
1760         }
1761
1762         /* on pending IO in this thread should left from prev. request */
1763         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1764
1765         sb_start_write(osd_sb(osd_dt_dev(d)));
1766
1767         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1768         if (!oh) {
1769                 sb_end_write(osd_sb(osd_dt_dev(d)));
1770                 RETURN(ERR_PTR(-ENOMEM));
1771         }
1772
1773         oh->ot_quota_trans = &oti->oti_quota_trans;
1774         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1775         th = &oh->ot_super;
1776         th->th_dev = d;
1777         th->th_result = 0;
1778         oh->ot_credits = 0;
1779         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1780         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1781         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1782         osd_th_alloced(oh);
1783
1784         memset(oti->oti_declare_ops, 0,
1785                sizeof(oti->oti_declare_ops));
1786         memset(oti->oti_declare_ops_cred, 0,
1787                sizeof(oti->oti_declare_ops_cred));
1788         memset(oti->oti_declare_ops_used, 0,
1789                sizeof(oti->oti_declare_ops_used));
1790
1791         oti->oti_ins_cache_depth++;
1792
1793         RETURN(th);
1794 }
1795
1796 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1797 {
1798         struct osd_thread_info *oti = osd_oti_get(env);
1799         struct osd_thandle *oh;
1800
1801         oh = container_of(th, struct osd_thandle, ot_super);
1802         LASSERT(oh != NULL);
1803
1804         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1805               oti->oti_declare_ops[OSD_OT_CREATE],
1806               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1807               oti->oti_declare_ops_used[OSD_OT_CREATE],
1808               oti->oti_declare_ops[OSD_OT_DESTROY],
1809               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1810               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1811         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1812               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1813               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1814               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1815               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1816               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1817               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1818         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1819               oti->oti_declare_ops[OSD_OT_WRITE],
1820               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1821               oti->oti_declare_ops_used[OSD_OT_WRITE],
1822               oti->oti_declare_ops[OSD_OT_PUNCH],
1823               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1824               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1825               oti->oti_declare_ops[OSD_OT_QUOTA],
1826               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1827               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1828         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1829               oti->oti_declare_ops[OSD_OT_INSERT],
1830               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1831               oti->oti_declare_ops_used[OSD_OT_INSERT],
1832               oti->oti_declare_ops[OSD_OT_DELETE],
1833               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1834               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1835         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1836               oti->oti_declare_ops[OSD_OT_REF_ADD],
1837               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1838               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1839               oti->oti_declare_ops[OSD_OT_REF_DEL],
1840               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1841               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1842 }
1843
1844 /*
1845  * Concurrency: shouldn't matter.
1846  */
1847 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1848                            struct thandle *th)
1849 {
1850         struct osd_thread_info *oti = osd_oti_get(env);
1851         struct osd_device *dev = osd_dt_dev(d);
1852         handle_t *jh;
1853         struct osd_thandle *oh;
1854         int rc;
1855
1856         ENTRY;
1857
1858         LASSERT(current->journal_info == NULL);
1859
1860         oh = container_of(th, struct osd_thandle, ot_super);
1861         LASSERT(oh != NULL);
1862         LASSERT(oh->ot_handle == NULL);
1863
1864         rc = dt_txn_hook_start(env, d, th);
1865         if (rc != 0)
1866                 GOTO(out, rc);
1867
1868         if (unlikely(osd_param_is_not_sane(dev, th))) {
1869                 static unsigned long last_printed;
1870                 static int last_credits;
1871
1872                 /*
1873                  * don't make noise on a tiny testing systems
1874                  * actual credits misuse will be caught anyway
1875                  */
1876                 if (last_credits != oh->ot_credits &&
1877                     time_after(jiffies, last_printed +
1878                                cfs_time_seconds(60)) &&
1879                     osd_transaction_size(dev) > 512) {
1880                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1881                               oh->ot_credits, osd_transaction_size(dev));
1882                         osd_trans_dump_creds(env, th);
1883                         libcfs_debug_dumpstack(NULL);
1884                         last_credits = oh->ot_credits;
1885                         last_printed = jiffies;
1886                 }
1887                 /*
1888                  * XXX Limit the credits to 'max_transaction_buffers', and
1889                  *     let the underlying filesystem to catch the error if
1890                  *     we really need so many credits.
1891                  *
1892                  *     This should be removed when we can calculate the
1893                  *     credits precisely.
1894                  */
1895                 oh->ot_credits = osd_transaction_size(dev);
1896         } else if (ldiskfs_track_declares_assert != 0) {
1897                 /*
1898                  * reserve few credits to prevent an assertion in JBD
1899                  * our debugging mechanism will be able to detected
1900                  * overuse. this can help to debug single-update
1901                  * transactions
1902                  */
1903                 oh->ot_credits += 10;
1904                 if (unlikely(osd_param_is_not_sane(dev, th)))
1905                         oh->ot_credits = osd_transaction_size(dev);
1906         }
1907
1908         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1909                 GOTO(out, rc = -EIO);
1910
1911         /*
1912          * XXX temporary stuff. Some abstraction layer should
1913          * be used.
1914          */
1915         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1916         osd_th_started(oh);
1917         if (!IS_ERR(jh)) {
1918                 oh->ot_handle = jh;
1919                 LASSERT(oti->oti_txns == 0);
1920
1921                 lu_device_get(&d->dd_lu_dev);
1922                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1923                               "osd-tx", th);
1924                 oti->oti_txns++;
1925                 rc = 0;
1926         } else {
1927                 rc = PTR_ERR(jh);
1928         }
1929 out:
1930         RETURN(rc);
1931 }
1932
1933 static int osd_seq_exists(const struct lu_env *env,
1934                           struct osd_device *osd, u64 seq)
1935 {
1936         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1937         struct seq_server_site *ss = osd_seq_site(osd);
1938         int rc;
1939
1940         ENTRY;
1941
1942         LASSERT(ss != NULL);
1943         LASSERT(ss->ss_server_fld != NULL);
1944
1945         rc = osd_fld_lookup(env, osd, seq, range);
1946         if (rc != 0) {
1947                 if (rc != -ENOENT)
1948                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1949                                osd_name(osd), seq, rc);
1950                 RETURN(0);
1951         }
1952
1953         RETURN(ss->ss_node_id == range->lsr_index);
1954 }
1955
1956 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1957 {
1958         struct dt_txn_commit_cb *dcb;
1959         struct dt_txn_commit_cb *tmp;
1960
1961         /* call per-transaction stop callbacks if any */
1962         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1963                                  dcb_linkage) {
1964                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1965                          "commit callback entry: magic=%x name='%s'\n",
1966                          dcb->dcb_magic, dcb->dcb_name);
1967                 list_del_init(&dcb->dcb_linkage);
1968                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1969         }
1970 }
1971
1972 /*
1973  * Concurrency: shouldn't matter.
1974  */
1975 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1976                           struct thandle *th)
1977 {
1978         struct osd_thread_info *oti = osd_oti_get(env);
1979         struct osd_thandle *oh;
1980         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1981         struct osd_device *osd = osd_dt_dev(th->th_dev);
1982         struct qsd_instance *qsd = osd_def_qsd(osd);
1983         struct lquota_trans *qtrans;
1984         LIST_HEAD(truncates);
1985         int rc = 0, remove_agents = 0;
1986
1987         ENTRY;
1988
1989         oh = container_of(th, struct osd_thandle, ot_super);
1990
1991         remove_agents = oh->ot_remove_agents;
1992
1993         qtrans = oh->ot_quota_trans;
1994         oh->ot_quota_trans = NULL;
1995
1996         /* move locks to local list, stop tx, execute truncates */
1997         list_splice(&oh->ot_trunc_locks, &truncates);
1998
1999         if (oh->ot_handle != NULL) {
2000                 int rc2;
2001
2002                 handle_t *hdl = oh->ot_handle;
2003
2004                 /*
2005                  * add commit callback
2006                  * notice we don't do this in osd_trans_start()
2007                  * as underlying transaction can change during truncate
2008                  */
2009                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2010                                              &oh->ot_jcb);
2011
2012                 LASSERT(oti->oti_txns == 1);
2013                 oti->oti_txns--;
2014
2015                 rc = dt_txn_hook_stop(env, th);
2016                 if (rc != 0)
2017                         CERROR("%s: failed in transaction hook: rc = %d\n",
2018                                osd_name(osd), rc);
2019
2020                 osd_trans_stop_cb(oh, rc);
2021                 /* hook functions might modify th_sync */
2022                 hdl->h_sync = th->th_sync;
2023
2024                 oh->ot_handle = NULL;
2025                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2026                 if (rc2 != 0)
2027                         CERROR("%s: failed to stop transaction: rc = %d\n",
2028                                osd_name(osd), rc2);
2029                 if (!rc)
2030                         rc = rc2;
2031
2032                 osd_process_truncates(&truncates);
2033         } else {
2034                 osd_trans_stop_cb(oh, th->th_result);
2035                 OBD_FREE_PTR(oh);
2036         }
2037
2038         osd_trunc_unlock_all(env, &truncates);
2039
2040         /* inform the quota slave device that the transaction is stopping */
2041         qsd_op_end(env, qsd, qtrans);
2042
2043         /*
2044          * as we want IO to journal and data IO be concurrent, we don't block
2045          * awaiting data IO completion in osd_do_bio(), instead we wait here
2046          * once transaction is submitted to the journal. all reqular requests
2047          * don't do direct IO (except read/write), thus this wait_event becomes
2048          * no-op for them.
2049          *
2050          * IMPORTANT: we have to wait till any IO submited by the thread is
2051          * completed otherwise iobuf may be corrupted by different request
2052          */
2053         wait_event(iobuf->dr_wait,
2054                        atomic_read(&iobuf->dr_numreqs) == 0);
2055         osd_fini_iobuf(osd, iobuf);
2056         if (!rc)
2057                 rc = iobuf->dr_error;
2058
2059         if (unlikely(remove_agents != 0))
2060                 osd_process_scheduled_agent_removals(env, osd);
2061
2062         oti->oti_ins_cache_depth--;
2063         /* reset OI cache for safety */
2064         if (oti->oti_ins_cache_depth == 0)
2065                 oti->oti_ins_cache_used = 0;
2066
2067         sb_end_write(osd_sb(osd));
2068
2069         RETURN(rc);
2070 }
2071
2072 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2073 {
2074         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2075                                               ot_super);
2076
2077         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2078         LASSERT(&dcb->dcb_func != NULL);
2079         if (dcb->dcb_flags & DCB_TRANS_STOP)
2080                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2081         else
2082                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2083
2084         return 0;
2085 }
2086
2087 /*
2088  * Called just before object is freed. Releases all resources except for
2089  * object itself (that is released by osd_object_free()).
2090  *
2091  * Concurrency: no concurrent access is possible that late in object
2092  * life-cycle.
2093  */
2094 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2095 {
2096         struct osd_object *obj = osd_obj(l);
2097         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2098         struct inode *inode = obj->oo_inode;
2099         __u64 projid;
2100         qid_t uid;
2101         qid_t gid;
2102
2103         LINVRNT(osd_invariant(obj));
2104
2105         /*
2106          * If object is unlinked remove fid->ino mapping from object index.
2107          */
2108
2109         osd_index_fini(obj);
2110
2111         if (!inode)
2112                 return;
2113
2114         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2115                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2116
2117         uid = i_uid_read(inode);
2118         gid = i_gid_read(inode);
2119         projid = i_projid_read(inode);
2120
2121         obj->oo_inode = NULL;
2122         iput(inode);
2123
2124         /* do not rebalance quota if the caller needs to release memory
2125          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2126          * transaction and risk to deadlock - LU-12178 */
2127         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2128                 return;
2129
2130         if (!obj->oo_header && qsd) {
2131                 struct osd_thread_info *info = osd_oti_get(env);
2132                 struct lquota_id_info *qi = &info->oti_qi;
2133
2134                 /* Release granted quota to master if necessary */
2135                 qi->lqi_id.qid_uid = uid;
2136                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2137
2138                 qi->lqi_id.qid_uid = gid;
2139                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2140
2141                 qi->lqi_id.qid_uid = projid;
2142                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2143         }
2144 }
2145
2146 /*
2147  * Concurrency: ->loo_object_release() is called under site spin-lock.
2148  */
2149 static void osd_object_release(const struct lu_env *env,
2150                                struct lu_object *l)
2151 {
2152         struct osd_object *o = osd_obj(l);
2153
2154         /*
2155          * nobody should be releasing a non-destroyed object with nlink=0
2156          * the API allows this, but ldiskfs doesn't like and then report
2157          * this inode as deleted
2158          */
2159         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2160                   o->oo_inode->i_nlink == 0));
2161 }
2162
2163 /*
2164  * Concurrency: shouldn't matter.
2165  */
2166 static int osd_object_print(const struct lu_env *env, void *cookie,
2167                             lu_printer_t p, const struct lu_object *l)
2168 {
2169         struct osd_object *o = osd_obj(l);
2170         struct iam_descr *d;
2171
2172         if (o->oo_dir != NULL)
2173                 d = o->oo_dir->od_container.ic_descr;
2174         else
2175                 d = NULL;
2176         return (*p)(env, cookie,
2177                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2178                     o, o->oo_inode,
2179                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2180                     o->oo_inode ? o->oo_inode->i_generation : 0,
2181                     d ? d->id_ops->id_name : "plain");
2182 }
2183
2184 /*
2185  * Concurrency: shouldn't matter.
2186  */
2187 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2188                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2189 {
2190         struct osd_device *osd = osd_dt_dev(d);
2191         struct super_block *sb = osd_sb(osd);
2192         struct kstatfs *ksfs;
2193         __u64 reserved;
2194         int result = 0;
2195
2196         if (unlikely(osd->od_mnt == NULL))
2197                 return -EINPROGRESS;
2198
2199         /* osd_lproc.c call this without env, allocate ksfs for that case */
2200         if (unlikely(env == NULL)) {
2201                 OBD_ALLOC_PTR(ksfs);
2202                 if (ksfs == NULL)
2203                         return -ENOMEM;
2204         } else {
2205                 ksfs = &osd_oti_get(env)->oti_ksfs;
2206         }
2207
2208         result = sb->s_op->statfs(sb->s_root, ksfs);
2209         if (result)
2210                 goto out;
2211
2212         statfs_pack(sfs, ksfs);
2213         if (unlikely(sb->s_flags & SB_RDONLY))
2214                 sfs->os_state |= OS_STATFS_READONLY;
2215
2216         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2217
2218         if (ldiskfs_has_feature_extents(sb))
2219                 sfs->os_maxbytes = sb->s_maxbytes;
2220         else
2221                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2222
2223         /*
2224          * Reserve some space so to avoid fragmenting the filesystem too much.
2225          * Fragmentation not only impacts performance, but can also increase
2226          * metadata overhead significantly, causing grant calculation to be
2227          * wrong.
2228          *
2229          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2230          */
2231         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2232         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2233         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2234                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2235
2236         sfs->os_blocks -= reserved;
2237         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2238         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2239
2240 out:
2241         if (unlikely(env == NULL))
2242                 OBD_FREE_PTR(ksfs);
2243         return result;
2244 }
2245
2246 /**
2247  * Estimate space needed for file creations. We assume the largest filename
2248  * which is 2^64 - 1, hence a filename of 20 chars.
2249  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2250  */
2251 #ifdef __LDISKFS_DIR_REC_LEN
2252 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2253 #else
2254 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2255 #endif
2256
2257 /*
2258  * Concurrency: doesn't access mutable data.
2259  */
2260 static void osd_conf_get(const struct lu_env *env,
2261                          const struct dt_device *dev,
2262                          struct dt_device_param *param)
2263 {
2264         struct osd_device *d = osd_dt_dev(dev);
2265         struct super_block *sb = osd_sb(d);
2266         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2267         const char *name;
2268         int ea_overhead;
2269
2270         /*
2271          * XXX should be taken from not-yet-existing fs abstraction layer.
2272          */
2273         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2274         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2275         param->ddp_symlink_max  = sb->s_blocksize;
2276         param->ddp_mount_type   = LDD_MT_LDISKFS;
2277         if (ldiskfs_has_feature_extents(sb))
2278                 param->ddp_maxbytes = sb->s_maxbytes;
2279         else
2280                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2281         /*
2282          * inode are statically allocated, so per-inode space consumption
2283          * is the space consumed by the directory entry
2284          */
2285         param->ddp_inodespace     = PER_OBJ_USAGE;
2286         /*
2287          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2288          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2289          * maximum length to not under-count the actual number of extents
2290          * needed for writing a file if there are sub-optimal block allocations.
2291          */
2292         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2293         /* worst-case extent insertion metadata overhead */
2294         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2295         param->ddp_mntopts = 0;
2296         if (test_opt(sb, XATTR_USER))
2297                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2298         if (test_opt(sb, POSIX_ACL))
2299                 param->ddp_mntopts |= MNTOPT_ACL;
2300
2301         /*
2302          * LOD might calculate the max stripe count based on max_ea_size,
2303          * so we need take account in the overhead as well,
2304          * xattr_header + magic + xattr_entry_head
2305          */
2306         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2307                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2308
2309 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2310         if (ldiskfs_has_feature_ea_inode(sb))
2311                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2312                                                                 ea_overhead;
2313         else
2314 #endif
2315                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2316
2317         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2318                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2319
2320         /*
2321          * Preferred RPC size for efficient disk IO.  4MB shows good
2322          * all-around performance for ldiskfs, but use bigalloc chunk size
2323          * by default if larger.
2324          */
2325 #if defined(LDISKFS_CLUSTER_SIZE)
2326         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2327                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2328         else
2329 #endif
2330                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2331
2332         param->ddp_t10_cksum_type = 0;
2333         if (bi) {
2334                 unsigned short interval = blk_integrity_interval(bi);
2335                 name = blk_integrity_name(bi);
2336                 /*
2337                  * Expected values:
2338                  * T10-DIF-TYPE1-CRC
2339                  * T10-DIF-TYPE3-CRC
2340                  * T10-DIF-TYPE1-IP
2341                  * T10-DIF-TYPE3-IP
2342                  */
2343                 if (strncmp(name, "T10-DIF-TYPE",
2344                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2345                         /* also skip "1/3-" at end */
2346                         const int type_off = sizeof("T10-DIF-TYPE.");
2347                         char type_number = name[type_off - 2];
2348
2349                         if (interval != 512 && interval != 4096) {
2350                                 CERROR("%s: unsupported T10PI sector size %u\n",
2351                                        d->od_svname, interval);
2352                         } else if (type_number != '1' && type_number != '3') {
2353                                 CERROR("%s: unsupported T10PI type %s\n",
2354                                        d->od_svname, name);
2355                         } else if (strcmp(name + type_off, "CRC") == 0) {
2356                                 d->od_t10_type = type_number == '1' ?
2357                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2358                                 param->ddp_t10_cksum_type = interval == 512 ?
2359                                         OBD_CKSUM_T10CRC512 :
2360                                         OBD_CKSUM_T10CRC4K;
2361                         } else if (strcmp(name + type_off, "IP") == 0) {
2362                                 d->od_t10_type = type_number == '1' ?
2363                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2364                                 param->ddp_t10_cksum_type = interval == 512 ?
2365                                         OBD_CKSUM_T10IP512 :
2366                                         OBD_CKSUM_T10IP4K;
2367                         } else {
2368                                 CERROR("%s: unsupported checksum type of "
2369                                        "T10PI type '%s'",
2370                                        d->od_svname, name);
2371                         }
2372
2373                 } else {
2374                         CERROR("%s: unsupported T10PI type '%s'",
2375                                d->od_svname, name);
2376                 }
2377         }
2378 }
2379
2380 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2381 {
2382         return osd_sb(osd_dt_dev(d));
2383 }
2384
2385 /*
2386  * Concurrency: shouldn't matter.
2387  */
2388 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2389 {
2390         int rc;
2391         struct super_block *s = osd_sb(osd_dt_dev(d));
2392         ENTRY;
2393
2394         down_read(&s->s_umount);
2395         rc = s->s_op->sync_fs(s, 1);
2396         up_read(&s->s_umount);
2397
2398         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2399                rc);
2400
2401         return rc;
2402 }
2403
2404 /**
2405  * Start commit for OSD device.
2406  *
2407  * An implementation of dt_commit_async method for OSD device.
2408  * Asychronously starts underlayng fs sync and thereby a transaction
2409  * commit.
2410  *
2411  * \param env environment
2412  * \param d dt device
2413  *
2414  * \see dt_device_operations
2415  */
2416 static int osd_commit_async(const struct lu_env *env,
2417                             struct dt_device *d)
2418 {
2419         struct super_block *s = osd_sb(osd_dt_dev(d));
2420         int rc;
2421
2422         ENTRY;
2423
2424         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2425         down_read(&s->s_umount);
2426         rc = s->s_op->sync_fs(s, 0);
2427         up_read(&s->s_umount);
2428
2429         RETURN(rc);
2430 }
2431
2432 static int (*priv_security_file_alloc)(struct file *file);
2433
2434 int osd_security_file_alloc(struct file *file)
2435 {
2436         if (priv_security_file_alloc)
2437                 return priv_security_file_alloc(file);
2438         return 0;
2439 }
2440
2441 /*
2442  * Concurrency: shouldn't matter.
2443  */
2444 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2445 {
2446         struct super_block *sb = osd_sb(osd_dt_dev(d));
2447         struct block_device *dev = sb->s_bdev;
2448         int rc = -EOPNOTSUPP;
2449
2450         ENTRY;
2451
2452         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2453                osd_dt_dev(d)->od_svname, (long)dev, rc);
2454
2455         RETURN(rc);
2456 }
2457
2458 /**
2459  * Note: we do not count into QUOTA here.
2460  * If we mount with --data_journal we may need more.
2461  */
2462 const int osd_dto_credits_noquota[DTO_NR] = {
2463         /**
2464          * Insert.
2465          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2466          * SINGLEDATA_TRANS_BLOCKS(8)
2467          * XXX Note: maybe iam need more, since iam have more level than
2468          *           EXT3 htree.
2469          */
2470         [DTO_INDEX_INSERT]  = 16,
2471         /**
2472          * Delete
2473          * just modify a single entry, probably merge few within a block
2474          */
2475         [DTO_INDEX_DELETE]  = 1,
2476         /**
2477          * Used for OI scrub
2478          */
2479         [DTO_INDEX_UPDATE]  = 16,
2480         /**
2481          * 4(inode, inode bits, groups, GDT)
2482          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2483          */
2484         [DTO_OBJECT_CREATE] = 4,
2485         /**
2486          * 4(inode, inode bits, groups, GDT)
2487          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2488          */
2489         [DTO_OBJECT_DELETE] = 4,
2490         /**
2491          * Attr set credits (inode)
2492          */
2493         [DTO_ATTR_SET_BASE] = 1,
2494         /**
2495          * Xattr set. The same as xattr of EXT3.
2496          * DATA_TRANS_BLOCKS(14)
2497          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2498          * are also counted in. Do not know why?
2499          */
2500         [DTO_XATTR_SET]     = 14,
2501         /**
2502          * credits for inode change during write.
2503          */
2504         [DTO_WRITE_BASE]    = 3,
2505         /**
2506          * credits for single block write.
2507          */
2508         [DTO_WRITE_BLOCK]   = 14,
2509         /**
2510          * Attr set credits for chown.
2511          * This is extra credits for setattr, and it is null without quota
2512          */
2513         [DTO_ATTR_SET_CHOWN] = 0
2514 };
2515
2516 static const struct dt_device_operations osd_dt_ops = {
2517         .dt_root_get       = osd_root_get,
2518         .dt_statfs         = osd_statfs,
2519         .dt_trans_create   = osd_trans_create,
2520         .dt_trans_start    = osd_trans_start,
2521         .dt_trans_stop     = osd_trans_stop,
2522         .dt_trans_cb_add   = osd_trans_cb_add,
2523         .dt_conf_get       = osd_conf_get,
2524         .dt_mnt_sb_get     = osd_mnt_sb_get,
2525         .dt_sync           = osd_sync,
2526         .dt_ro             = osd_ro,
2527         .dt_commit_async   = osd_commit_async,
2528 };
2529
2530 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2531                           unsigned int role)
2532 {
2533         struct osd_object *obj = osd_dt_obj(dt);
2534         struct osd_thread_info *oti = osd_oti_get(env);
2535
2536         LINVRNT(osd_invariant(obj));
2537
2538         LASSERT(obj->oo_owner != env);
2539         down_read_nested(&obj->oo_sem, role);
2540
2541         LASSERT(obj->oo_owner == NULL);
2542         oti->oti_r_locks++;
2543 }
2544
2545 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2546                            unsigned int role)
2547 {
2548         struct osd_object *obj = osd_dt_obj(dt);
2549         struct osd_thread_info *oti = osd_oti_get(env);
2550
2551         LINVRNT(osd_invariant(obj));
2552
2553         LASSERT(obj->oo_owner != env);
2554         down_write_nested(&obj->oo_sem, role);
2555
2556         LASSERT(obj->oo_owner == NULL);
2557         obj->oo_owner = env;
2558         oti->oti_w_locks++;
2559 }
2560
2561 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2562 {
2563         struct osd_object *obj = osd_dt_obj(dt);
2564         struct osd_thread_info *oti = osd_oti_get(env);
2565
2566         LINVRNT(osd_invariant(obj));
2567
2568         LASSERT(oti->oti_r_locks > 0);
2569         oti->oti_r_locks--;
2570         up_read(&obj->oo_sem);
2571 }
2572
2573 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2574 {
2575         struct osd_object *obj = osd_dt_obj(dt);
2576         struct osd_thread_info *oti = osd_oti_get(env);
2577
2578         LINVRNT(osd_invariant(obj));
2579
2580         LASSERT(obj->oo_owner == env);
2581         LASSERT(oti->oti_w_locks > 0);
2582         oti->oti_w_locks--;
2583         obj->oo_owner = NULL;
2584         up_write(&obj->oo_sem);
2585 }
2586
2587 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2588 {
2589         struct osd_object *obj = osd_dt_obj(dt);
2590
2591         LINVRNT(osd_invariant(obj));
2592
2593         return obj->oo_owner == env;
2594 }
2595
2596 static void osd_inode_getattr(const struct lu_env *env,
2597                               struct inode *inode, struct lu_attr *attr)
2598 {
2599         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2600                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2601                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2602                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2603
2604         attr->la_atime = inode->i_atime.tv_sec;
2605         attr->la_mtime = inode->i_mtime.tv_sec;
2606         attr->la_ctime = inode->i_ctime.tv_sec;
2607         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2608         attr->la_mode    = inode->i_mode;
2609         attr->la_size    = i_size_read(inode);
2610         attr->la_blocks  = inode->i_blocks;
2611         attr->la_uid     = i_uid_read(inode);
2612         attr->la_gid     = i_gid_read(inode);
2613         attr->la_projid  = i_projid_read(inode);
2614         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2615         attr->la_nlink   = inode->i_nlink;
2616         attr->la_rdev    = inode->i_rdev;
2617         attr->la_blksize = 1 << inode->i_blkbits;
2618         attr->la_blkbits = inode->i_blkbits;
2619         /*
2620          * Ext4 did not transfer inherit flags from raw inode
2621          * to inode flags, and ext4 internally test raw inode
2622          * @i_flags directly. Instead of patching ext4, we do it here.
2623          */
2624         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2625                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2626 }
2627
2628 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2629                             u64 *count)
2630 {
2631         struct osd_object *obj = osd_dt_obj(dt);
2632         const struct dt_it_ops *iops;
2633         struct dt_it *it;
2634         int rc;
2635
2636         ENTRY;
2637
2638         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2639         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2640
2641         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2642                 *count = obj->oo_dirent_count;
2643                 RETURN(0);
2644         }
2645
2646         /* directory not initialized yet */
2647         if (!dt->do_index_ops) {
2648                 *count = 0;
2649                 RETURN(0);
2650         }
2651
2652         iops = &dt->do_index_ops->dio_it;
2653         it = iops->init(env, dt, LUDA_64BITHASH);
2654         if (IS_ERR(it))
2655                 RETURN(PTR_ERR(it));
2656
2657         rc = iops->load(env, it, 0);
2658         if (rc < 0) {
2659                 if (rc == -ENODATA) {
2660                         rc = 0;
2661                         *count = 0;
2662                 }
2663                 GOTO(out, rc);
2664         }
2665         if (rc > 0)
2666                 rc = iops->next(env, it);
2667
2668         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2669                 if (rc == -ESTALE)
2670                         continue;
2671
2672                 if (iops->key_size(env, it) == 0)
2673                         continue;
2674
2675                 (*count)++;
2676         }
2677         if (rc == 1) {
2678                 obj->oo_dirent_count = *count;
2679                 rc = 0;
2680         }
2681 out:
2682         iops->put(env, it);
2683         iops->fini(env, it);
2684
2685         RETURN(rc);
2686 }
2687
2688 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2689                         struct lu_attr *attr)
2690 {
2691         struct osd_object *obj = osd_dt_obj(dt);
2692         int rc = 0;
2693
2694         if (unlikely(!dt_object_exists(dt)))
2695                 return -ENOENT;
2696         if (unlikely(obj->oo_destroyed))
2697                 return -ENOENT;
2698
2699         LASSERT(!dt_object_remote(dt));
2700         LINVRNT(osd_invariant(obj));
2701
2702         spin_lock(&obj->oo_guard);
2703         osd_inode_getattr(env, obj->oo_inode, attr);
2704         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2705                 attr->la_valid |= LA_FLAGS;
2706                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2707         }
2708         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2709                 attr->la_valid |= LA_FLAGS;
2710                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2711         }
2712         spin_unlock(&obj->oo_guard);
2713
2714         if (S_ISDIR(obj->oo_inode->i_mode) &&
2715             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2716                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2717
2718         return rc;
2719 }
2720
2721 static int osd_declare_attr_qid(const struct lu_env *env,
2722                                 struct osd_object *obj,
2723                                 struct osd_thandle *oh, long long bspace,
2724                                 qid_t old_id, qid_t new_id, bool enforce,
2725                                 unsigned int type, bool ignore_edquot)
2726 {
2727         int rc;
2728         struct osd_thread_info *info = osd_oti_get(env);
2729         struct lquota_id_info  *qi = &info->oti_qi;
2730
2731         qi->lqi_type = type;
2732         /* inode accounting */
2733         qi->lqi_is_blk = false;
2734
2735         /* one more inode for the new id ... */
2736         qi->lqi_id.qid_uid = new_id;
2737         qi->lqi_space      = 1;
2738         /* Reserve credits for the new id */
2739         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2740         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2741                 rc = 0;
2742         if (rc)
2743                 RETURN(rc);
2744
2745         /* and one less inode for the current id */
2746         qi->lqi_id.qid_uid = old_id;
2747         qi->lqi_space = -1;
2748         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2749         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2750                 rc = 0;
2751         if (rc)
2752                 RETURN(rc);
2753
2754         /* block accounting */
2755         qi->lqi_is_blk = true;
2756
2757         /* more blocks for the new id ... */
2758         qi->lqi_id.qid_uid = new_id;
2759         qi->lqi_space      = bspace;
2760         /*
2761          * Credits for the new uid has been reserved, re-use "obj"
2762          * to save credit reservation.
2763          */
2764         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2765         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2766                 rc = 0;
2767         if (rc)
2768                 RETURN(rc);
2769
2770         /* and finally less blocks for the current uid */
2771         qi->lqi_id.qid_uid = old_id;
2772         qi->lqi_space      = -bspace;
2773         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2774         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2775                 rc = 0;
2776
2777         RETURN(rc);
2778 }
2779
2780 static int osd_declare_attr_set(const struct lu_env *env,
2781                                 struct dt_object *dt,
2782                                 const struct lu_attr *attr,
2783                                 struct thandle *handle)
2784 {
2785         struct osd_thandle *oh;
2786         struct osd_object *obj;
2787         qid_t uid;
2788         qid_t gid;
2789         long long bspace;
2790         int rc = 0;
2791         bool enforce;
2792
2793         ENTRY;
2794
2795         LASSERT(dt != NULL);
2796         LASSERT(handle != NULL);
2797
2798         obj = osd_dt_obj(dt);
2799         LASSERT(osd_invariant(obj));
2800
2801         oh = container_of(handle, struct osd_thandle, ot_super);
2802         LASSERT(oh->ot_handle == NULL);
2803
2804         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2805                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2806
2807         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2808                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2809
2810         if (attr == NULL || obj->oo_inode == NULL)
2811                 RETURN(rc);
2812
2813         bspace   = obj->oo_inode->i_blocks << 9;
2814         bspace   = toqb(bspace);
2815
2816         /*
2817          * Changing ownership is always preformed by super user, it should not
2818          * fail with EDQUOT unless required explicitly.
2819          *
2820          * We still need to call the osd_declare_qid() to calculate the journal
2821          * credits for updating quota accounting files and to trigger quota
2822          * space adjustment once the operation is completed.
2823          */
2824         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2825                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2826
2827                 if (!ignore_edquot)
2828                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2829                                "(the quota space is %lld)\n",
2830                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2831                                attr->la_gid, bspace);
2832
2833                 /* USERQUOTA */
2834                 uid = i_uid_read(obj->oo_inode);
2835                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2836                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2837                                           attr->la_uid, enforce, USRQUOTA,
2838                                           true);
2839                 if (rc)
2840                         RETURN(rc);
2841
2842                 gid = i_gid_read(obj->oo_inode);
2843                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2844                        attr->la_uid, gid, attr->la_gid);
2845                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2846                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2847                                           attr->la_gid, enforce, GRPQUOTA,
2848                                           ignore_edquot);
2849                 if (rc)
2850                         RETURN(rc);
2851
2852         }
2853 #ifdef HAVE_PROJECT_QUOTA
2854         if (attr->la_valid & LA_PROJID) {
2855                 __u32 projid = i_projid_read(obj->oo_inode);
2856
2857                 enforce = (attr->la_valid & LA_PROJID) &&
2858                                         (attr->la_projid != projid);
2859                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2860                                           (qid_t)projid, (qid_t)attr->la_projid,
2861                                           enforce, PRJQUOTA, true);
2862                 if (rc)
2863                         RETURN(rc);
2864         }
2865 #endif
2866         RETURN(rc);
2867 }
2868
2869 static int osd_inode_setattr(const struct lu_env *env,
2870                              struct inode *inode, const struct lu_attr *attr)
2871 {
2872         __u64 bits = attr->la_valid;
2873
2874         /* Only allow set size for regular file */
2875         if (!S_ISREG(inode->i_mode))
2876                 bits &= ~(LA_SIZE | LA_BLOCKS);
2877
2878         if (bits == 0)
2879                 return 0;
2880
2881         if (bits & LA_ATIME)
2882                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2883         if (bits & LA_CTIME)
2884                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2885         if (bits & LA_MTIME)
2886                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2887         if (bits & LA_SIZE) {
2888                 spin_lock(&inode->i_lock);
2889                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2890                 i_size_write(inode, attr->la_size);
2891                 spin_unlock(&inode->i_lock);
2892         }
2893
2894         /*
2895          * OSD should not change "i_blocks" which is used by quota.
2896          * "i_blocks" should be changed by ldiskfs only.
2897          */
2898         if (bits & LA_MODE)
2899                 inode->i_mode = (inode->i_mode & S_IFMT) |
2900                                 (attr->la_mode & ~S_IFMT);
2901         if (bits & LA_UID)
2902                 i_uid_write(inode, attr->la_uid);
2903         if (bits & LA_GID)
2904                 i_gid_write(inode, attr->la_gid);
2905         if (bits & LA_PROJID)
2906                 i_projid_write(inode, attr->la_projid);
2907         if (bits & LA_NLINK)
2908                 set_nlink(inode, attr->la_nlink);
2909         if (bits & LA_RDEV)
2910                 inode->i_rdev = attr->la_rdev;
2911
2912         if (bits & LA_FLAGS) {
2913                 /* always keep S_NOCMTIME */
2914                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2915                                  S_NOCMTIME;
2916                 /*
2917                  * Ext4 did not transfer inherit flags from
2918                  * @inode->i_flags to raw inode i_flags when writing
2919                  * flags, we do it explictly here.
2920                  */
2921                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2922                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2923                 else
2924                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2925         }
2926         return 0;
2927 }
2928
2929 #ifdef HAVE_PROJECT_QUOTA
2930 static int osd_transfer_project(struct inode *inode, __u32 projid,
2931                                 struct thandle *handle)
2932 {
2933         struct super_block *sb = inode->i_sb;
2934         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2935         int err;
2936         kprojid_t kprojid;
2937         struct ldiskfs_iloc iloc;
2938         struct ldiskfs_inode *raw_inode;
2939         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2940
2941         if (!ldiskfs_has_feature_project(sb)) {
2942                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2943                         == LDISKFS_DEF_PROJID);
2944                 if (projid != LDISKFS_DEF_PROJID)
2945                         return -EOPNOTSUPP;
2946                 else
2947                         return 0;
2948         }
2949
2950         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2951                 return -EOPNOTSUPP;
2952
2953         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2954         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2955                 return 0;
2956
2957         err = ldiskfs_get_inode_loc(inode, &iloc);
2958         if (err)
2959                 return err;
2960
2961         raw_inode = ldiskfs_raw_inode(&iloc);
2962         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2963                 struct osd_thandle *oh =
2964                                 container_of0(handle, struct osd_thandle,
2965                                               ot_super);
2966                 /**
2967                  * try to expand inode size automatically.
2968                  */
2969                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
2970                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2971                         err = -EOVERFLOW;
2972                         brelse(iloc.bh);
2973                         return err;
2974                 }
2975         }
2976         brelse(iloc.bh);
2977
2978         dquot_initialize(inode);
2979         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2980         if (transfer_to[PRJQUOTA]) {
2981                 err = __dquot_transfer(inode, transfer_to);
2982                 dqput(transfer_to[PRJQUOTA]);
2983                 if (err)
2984                         return err;
2985         }
2986
2987         return err;
2988 }
2989 #endif
2990
2991 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
2992                               struct thandle *handle)
2993 {
2994         int rc;
2995
2996         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2997             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2998                 struct iattr iattr;
2999
3000                 CDEBUG(D_QUOTA,
3001                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3002                        inode->i_ino, i_uid_read(inode), attr->la_uid,
3003                        i_gid_read(inode), attr->la_gid);
3004
3005                 dquot_initialize(inode);
3006                 iattr.ia_valid = 0;
3007                 if (attr->la_valid & LA_UID)
3008                         iattr.ia_valid |= ATTR_UID;
3009                 if (attr->la_valid & LA_GID)
3010                         iattr.ia_valid |= ATTR_GID;
3011                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3012                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3013
3014                 rc = dquot_transfer(inode, &iattr);
3015                 if (rc) {
3016                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
3017                                "enforcement enabled on the ldiskfs "
3018                                "filesystem?\n", inode->i_sb->s_id, rc);
3019                         return rc;
3020                 }
3021         }
3022
3023         /* Handle project id transfer here properly */
3024         if (attr->la_valid & LA_PROJID &&
3025             attr->la_projid != i_projid_read(inode)) {
3026 #ifdef HAVE_PROJECT_QUOTA
3027                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3028 #else
3029                 rc = -ENOTSUPP;
3030 #endif
3031                 if (rc) {
3032                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3033                                "enforcement enabled on the ldiskfs "
3034                                "filesystem?\n", inode->i_sb->s_id, rc);
3035                         return rc;
3036                 }
3037         }
3038         return 0;
3039 }
3040
3041 static int osd_attr_set(const struct lu_env *env,
3042                         struct dt_object *dt,
3043                         const struct lu_attr *attr,
3044                         struct thandle *handle)
3045 {
3046         struct osd_object *obj = osd_dt_obj(dt);
3047         struct inode *inode;
3048         int rc;
3049
3050         if (!dt_object_exists(dt))
3051                 return -ENOENT;
3052
3053         LASSERT(handle != NULL);
3054         LASSERT(!dt_object_remote(dt));
3055         LASSERT(osd_invariant(obj));
3056
3057         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3058
3059         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3060             !osd_obj2dev(obj)->od_is_ost) {
3061                 struct osd_thread_info *oti = osd_oti_get(env);
3062                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3063                 struct lu_fid *fid1 = &oti->oti_fid;
3064                 struct osd_inode_id *id = &oti->oti_id;
3065                 struct iam_path_descr *ipd;
3066                 struct iam_container *bag;
3067                 struct osd_thandle *oh;
3068                 int rc;
3069
3070                 fid_cpu_to_be(fid1, fid0);
3071                 memset(id, 1, sizeof(*id));
3072                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3073                                   fid0)->oi_dir.od_container;
3074                 ipd = osd_idx_ipd_get(env, bag);
3075                 if (unlikely(ipd == NULL))
3076                         RETURN(-ENOMEM);
3077
3078                 oh = container_of(handle, struct osd_thandle, ot_super);
3079                 rc = iam_update(oh->ot_handle, bag,
3080                                 (const struct iam_key *)fid1,
3081                                 (const struct iam_rec *)id, ipd);
3082                 osd_ipd_put(env, bag, ipd);
3083                 return(rc > 0 ? 0 : rc);
3084         }
3085
3086         inode = obj->oo_inode;
3087
3088         rc = osd_quota_transfer(inode, attr, handle);
3089         if (rc)
3090                 return rc;
3091
3092         spin_lock(&obj->oo_guard);
3093         rc = osd_inode_setattr(env, inode, attr);
3094         spin_unlock(&obj->oo_guard);
3095         if (rc != 0)
3096                 GOTO(out, rc);
3097
3098         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3099
3100         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3101
3102         if (!(attr->la_valid & LA_FLAGS))
3103                 GOTO(out, rc);
3104
3105         /* Let's check if there are extra flags need to be set into LMA */
3106         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3107                 struct osd_thread_info *info = osd_oti_get(env);
3108                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3109
3110                 LASSERT(!obj->oo_pfid_in_lma);
3111
3112                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3113                                  &info->oti_ost_attrs);
3114                 if (rc)
3115                         GOTO(out, rc);
3116
3117                 lma->lma_incompat |=
3118                         lustre_to_lma_flags(attr->la_flags);
3119                 lustre_lma_swab(lma);
3120
3121                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3122
3123                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3124                                      lma, sizeof(*lma), XATTR_REPLACE);
3125                 if (rc != 0) {
3126                         struct osd_device *osd = osd_obj2dev(obj);
3127
3128                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3129                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3130                               lma->lma_incompat, rc);
3131                 } else {
3132                         obj->oo_lma_flags =
3133                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3134                 }
3135                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3136         }
3137 out:
3138
3139         return rc;
3140 }
3141
3142 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3143                                            struct osd_object *obj,
3144                                            const char *name, const int namelen)
3145 {
3146         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3147 }
3148
3149 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3150                       umode_t mode, struct dt_allocation_hint *hint,
3151                       struct thandle *th, struct lu_attr *attr)
3152 {
3153         int result;
3154         struct osd_device *osd = osd_obj2dev(obj);
3155         struct osd_thandle *oth;
3156         struct dt_object *parent = NULL;
3157         struct inode *inode;
3158         uid_t owner[2] = {0, 0};
3159
3160         if (attr->la_valid & LA_UID)
3161                 owner[0] = attr->la_uid;
3162         if (attr->la_valid & LA_GID)
3163                 owner[1] = attr->la_gid;
3164
3165         LINVRNT(osd_invariant(obj));
3166         LASSERT(obj->oo_inode == NULL);
3167         LASSERT(obj->oo_hl_head == NULL);
3168
3169         if (S_ISDIR(mode) && ldiskfs_pdo) {
3170                 obj->oo_hl_head =
3171                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3172                 if (obj->oo_hl_head == NULL)
3173                         return -ENOMEM;
3174         }
3175
3176         oth = container_of(th, struct osd_thandle, ot_super);
3177         LASSERT(oth->ot_handle->h_transaction != NULL);
3178
3179         if (hint != NULL && hint->dah_parent != NULL &&
3180             !dt_object_remote(hint->dah_parent))
3181                 parent = hint->dah_parent;
3182
3183         inode = ldiskfs_create_inode(oth->ot_handle,
3184                                      parent ? osd_dt_obj(parent)->oo_inode :
3185                                               osd_sb(osd)->s_root->d_inode,
3186                                      mode, owner);
3187         if (!IS_ERR(inode)) {
3188                 /* Do not update file c/mtime in ldiskfs. */
3189                 inode->i_flags |= S_NOCMTIME;
3190
3191                 /*
3192                  * For new created object, it must be consistent,
3193                  * and it is unnecessary to scrub against it.
3194                  */
3195                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3196
3197                 obj->oo_inode = inode;
3198                 result = 0;
3199         } else {
3200                 if (obj->oo_hl_head != NULL) {
3201                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3202                         obj->oo_hl_head = NULL;
3203                 }
3204                 result = PTR_ERR(inode);
3205         }
3206         LINVRNT(osd_invariant(obj));
3207         return result;
3208 }
3209
3210 enum {
3211         OSD_NAME_LEN = 255
3212 };
3213
3214 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3215                      struct lu_attr *attr,
3216                      struct dt_allocation_hint *hint,
3217                      struct dt_object_format *dof,
3218                      struct thandle *th)
3219 {
3220         int result;
3221         struct osd_thandle *oth;
3222         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3223
3224         LASSERT(S_ISDIR(attr->la_mode));
3225
3226         oth = container_of(th, struct osd_thandle, ot_super);
3227         LASSERT(oth->ot_handle->h_transaction != NULL);
3228         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3229                 obj->oo_dirent_count = 0;
3230         result = osd_mkfile(info, obj, mode, hint, th, attr);
3231
3232         return result;
3233 }
3234
3235 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3236                         struct lu_attr *attr,
3237                         struct dt_allocation_hint *hint,
3238                         struct dt_object_format *dof,
3239                         struct thandle *th)
3240 {
3241         int result;
3242         struct osd_thandle *oth;
3243         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3244
3245         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3246
3247         LASSERT(S_ISREG(attr->la_mode));
3248
3249         oth = container_of(th, struct osd_thandle, ot_super);
3250         LASSERT(oth->ot_handle->h_transaction != NULL);
3251
3252         result = osd_mkfile(info, obj, mode, hint, th, attr);
3253         if (result == 0) {
3254                 LASSERT(obj->oo_inode != NULL);
3255                 if (feat->dif_flags & DT_IND_VARKEY)
3256                         result = iam_lvar_create(obj->oo_inode,
3257                                                  feat->dif_keysize_max,
3258                                                  feat->dif_ptrsize,
3259                                                  feat->dif_recsize_max,
3260                                                  oth->ot_handle);
3261                 else
3262                         result = iam_lfix_create(obj->oo_inode,
3263                                                  feat->dif_keysize_max,
3264                                                  feat->dif_ptrsize,
3265                                                  feat->dif_recsize_max,
3266                                                  oth->ot_handle);
3267         }
3268         return result;
3269 }
3270
3271 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3272                      struct lu_attr *attr,
3273                      struct dt_allocation_hint *hint,
3274                      struct dt_object_format *dof,
3275                      struct thandle *th)
3276 {
3277         LASSERT(S_ISREG(attr->la_mode));
3278         return osd_mkfile(info, obj, (attr->la_mode &
3279                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3280                           attr);
3281 }
3282
3283 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3284                      struct lu_attr *attr,
3285                      struct dt_allocation_hint *hint,
3286                      struct dt_object_format *dof,
3287                      struct thandle *th)
3288 {
3289         LASSERT(S_ISLNK(attr->la_mode));
3290         return osd_mkfile(info, obj, (attr->la_mode &
3291                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3292                           attr);
3293 }
3294
3295 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3296                      struct lu_attr *attr,
3297                      struct dt_allocation_hint *hint,
3298                      struct dt_object_format *dof,
3299                      struct thandle *th)
3300 {
3301         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3302         int result;
3303
3304         LINVRNT(osd_invariant(obj));
3305         LASSERT(obj->oo_inode == NULL);
3306         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3307                 S_ISFIFO(mode) || S_ISSOCK(mode));
3308
3309         result = osd_mkfile(info, obj, mode, hint, th, attr);
3310         if (result == 0) {
3311                 LASSERT(obj->oo_inode != NULL);
3312                 /*
3313                  * This inode should be marked dirty for i_rdev.  Currently
3314                  * that is done in the osd_attr_init().
3315                  */
3316                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3317                                    attr->la_rdev);
3318         }
3319         LINVRNT(osd_invariant(obj));
3320         return result;
3321 }
3322
3323 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3324                               struct lu_attr *,
3325                               struct dt_allocation_hint *hint,
3326                               struct dt_object_format *dof,
3327                               struct thandle *);
3328
3329 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3330 {
3331         osd_obj_type_f result;
3332
3333         switch (type) {
3334         case DFT_DIR:
3335                 result = osd_mkdir;
3336                 break;
3337         case DFT_REGULAR:
3338                 result = osd_mkreg;
3339                 break;
3340         case DFT_SYM:
3341                 result = osd_mksym;
3342                 break;
3343         case DFT_NODE:
3344                 result = osd_mknod;
3345                 break;
3346         case DFT_INDEX:
3347                 result = osd_mk_index;
3348                 break;
3349
3350         default:
3351                 LBUG();
3352                 break;
3353         }
3354         return result;
3355 }
3356
3357
3358 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3359                         struct dt_object *parent, struct dt_object *child,
3360                         umode_t child_mode)
3361 {
3362         LASSERT(ah);
3363
3364         ah->dah_parent = parent;
3365         ah->dah_mode = child_mode;
3366
3367         if (parent != NULL && !dt_object_remote(parent)) {
3368                 /* will help to find FID->ino at dt_insert("..") */
3369                 struct osd_object *pobj = osd_dt_obj(parent);
3370
3371                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3372         }
3373 }
3374
3375 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3376                           struct lu_attr *attr, struct dt_object_format *dof,
3377                           struct thandle *handle)
3378 {
3379         struct inode *inode = obj->oo_inode;
3380         __u64 valid = attr->la_valid;
3381         int result;
3382
3383         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3384
3385         if (dof->dof_type != DFT_NODE)
3386                 attr->la_valid &= ~LA_RDEV;
3387         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3388                 attr->la_valid &= ~LA_ATIME;
3389         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3390                 attr->la_valid &= ~LA_CTIME;
3391         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3392                 attr->la_valid &= ~LA_MTIME;
3393
3394         result = osd_quota_transfer(inode, attr, handle);
3395         if (result)
3396                 return;
3397
3398         if (attr->la_valid != 0) {
3399                 result = osd_inode_setattr(info->oti_env, inode, attr);
3400                 /*
3401                  * The osd_inode_setattr() should always succeed here.  The
3402                  * only error that could be returned is EDQUOT when we are
3403                  * trying to change the UID or GID of the inode. However, this
3404                  * should not happen since quota enforcement is no longer
3405                  * enabled on ldiskfs (lquota takes care of it).
3406                  */
3407                 LASSERTF(result == 0, "%d\n", result);
3408                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3409         }
3410
3411         attr->la_valid = valid;
3412 }
3413
3414 /**
3415  * Helper function for osd_create()
3416  *
3417  * \retval 0, on success
3418  */
3419 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3420                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3421                         struct dt_object_format *dof, struct thandle *th)
3422 {
3423         int result;
3424         __u32 umask;
3425
3426         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3427
3428         /* we drop umask so that permissions we pass are not affected */
3429         umask = current->fs->umask;
3430         current->fs->umask = 0;
3431
3432         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3433                                                   th);
3434         if (likely(obj->oo_inode != NULL)) {
3435                 LASSERT(obj->oo_inode->i_state & I_NEW);
3436
3437                 /*
3438                  * Unlock the inode before attr initialization to avoid
3439                  * unnecessary dqget operations. LU-6378
3440                  */
3441                 unlock_new_inode(obj->oo_inode);
3442         }
3443
3444         if (likely(result == 0)) {
3445                 osd_attr_init(info, obj, attr, dof, th);
3446                 osd_object_init0(obj);
3447         }
3448
3449         /* restore previous umask value */
3450         current->fs->umask = umask;
3451
3452         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3453
3454         return result;
3455 }
3456
3457 /**
3458  * Helper function for osd_create()
3459  *
3460  * \retval 0, on success
3461  */
3462 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3463                            const struct lu_fid *fid, struct thandle *th)
3464 {
3465         struct osd_thread_info *info = osd_oti_get(env);
3466         struct osd_inode_id    *id   = &info->oti_id;
3467         struct osd_device      *osd  = osd_obj2dev(obj);
3468         struct osd_thandle     *oh;
3469         int rc;
3470
3471         LASSERT(obj->oo_inode != NULL);
3472
3473         oh = container_of(th, struct osd_thandle, ot_super);
3474         LASSERT(oh->ot_handle);
3475         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3476
3477         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3478         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3479                            OI_CHECK_FLD, NULL);
3480         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3481                 struct lu_fid next_fid = *fid;
3482
3483                 /* insert next object in advance, and map to the same inode */
3484                 next_fid.f_oid++;
3485                 if (next_fid.f_oid != 0) {
3486                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3487                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3488                                       OI_CHECK_FLD, NULL);
3489                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3490                 }
3491         }
3492
3493         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3494
3495         return rc;
3496 }
3497
3498 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3499                    u64 seq, struct lu_seq_range *range)
3500 {
3501         struct seq_server_site *ss = osd_seq_site(osd);
3502
3503         if (fid_seq_is_idif(seq)) {
3504                 fld_range_set_ost(range);
3505                 range->lsr_index = idif_ost_idx(seq);
3506                 return 0;
3507         }
3508
3509         if (!fid_seq_in_fldb(seq)) {
3510                 fld_range_set_mdt(range);
3511                 if (ss != NULL)
3512                         /*
3513                          * FIXME: If ss is NULL, it suppose not get lsr_index
3514                          * at all
3515                          */
3516                         range->lsr_index = ss->ss_node_id;
3517                 return 0;
3518         }
3519
3520         LASSERT(ss != NULL);
3521         fld_range_set_any(range);
3522         /* OSD will only do local fld lookup */
3523         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3524 }
3525
3526 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3527                               struct lu_attr *attr,
3528                               struct dt_allocation_hint *hint,
3529                               struct dt_object_format *dof,
3530                               struct thandle *handle)
3531 {
3532         struct osd_thandle *oh;
3533         int rc;
3534
3535         ENTRY;
3536
3537         LASSERT(handle != NULL);
3538
3539         oh = container_of(handle, struct osd_thandle, ot_super);
3540         LASSERT(oh->ot_handle == NULL);
3541
3542         /*
3543          * EA object consumes more credits than regular object: osd_mk_index
3544          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3545          * leaf_node, could involves the block, block bitmap, groups, GDT
3546          * change for each block, so add 4 * 2 credits in that case.
3547          */
3548         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3549                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3550                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3551         /*
3552          * Reuse idle OI block may cause additional one OI block
3553          * to be changed.
3554          */
3555         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3556                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3557         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3558                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3559                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3560
3561         /* will help to find FID->ino mapping at dt_insert() */
3562         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3563                                    osd_dt_obj(dt));
3564         if (rc != 0)
3565                 RETURN(rc);
3566
3567         if (!attr)
3568                 RETURN(0);
3569
3570         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3571                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3572                                    NULL, OSD_QID_INODE);
3573         if (rc != 0)
3574                 RETURN(rc);
3575
3576         RETURN(rc);
3577 }
3578
3579 /**
3580  * Called to destroy on-disk representation of the object
3581  *
3582  * Concurrency: must be locked
3583  */
3584 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3585                                struct thandle *th)
3586 {
3587         struct osd_object *obj = osd_dt_obj(dt);
3588         struct inode *inode = obj->oo_inode;
3589         struct osd_thandle *oh;
3590         int rc;
3591
3592         ENTRY;
3593
3594         if (inode == NULL)
3595                 RETURN(-ENOENT);
3596
3597         oh = container_of(th, struct osd_thandle, ot_super);
3598         LASSERT(oh->ot_handle == NULL);
3599
3600         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3601                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3602
3603         /* For removing agent entry */
3604         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3605                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3606
3607         /*
3608          * Recycle idle OI leaf may cause additional three OI blocks
3609          * to be changed.
3610          */
3611         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3612                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3613                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3614         /* one less inode */
3615         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3616                                    i_projid_read(inode), -1, oh, obj, NULL,
3617                                    OSD_QID_INODE);
3618         if (rc)
3619                 RETURN(rc);
3620         /* data to be truncated */
3621         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3622                                    i_projid_read(inode), 0, oh, obj, NULL,
3623                                    OSD_QID_BLK);
3624         if (rc)
3625                 RETURN(rc);
3626
3627         /*
3628          * will help to find FID->ino when this object is being
3629          * added to PENDING
3630          */
3631         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3632
3633         RETURN(rc);
3634 }
3635
3636 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3637                        struct thandle *th)
3638 {
3639         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3640         struct osd_object *obj = osd_dt_obj(dt);
3641         struct inode *inode = obj->oo_inode;
3642         struct osd_device *osd = osd_obj2dev(obj);
3643         struct osd_thandle *oh;
3644         int result;
3645
3646         ENTRY;
3647
3648         oh = container_of(th, struct osd_thandle, ot_super);
3649         LASSERT(oh->ot_handle);
3650         LASSERT(inode);
3651         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3652
3653         if (unlikely(fid_is_acct(fid)))
3654                 RETURN(-EPERM);
3655
3656         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3657                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3658                 if (result != 0)
3659                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3660                                osd_name(osd), PFID(fid), result);
3661         }
3662
3663         if (S_ISDIR(inode->i_mode)) {
3664                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3665                         inode->i_nlink == 2);
3666
3667                 spin_lock(&obj->oo_guard);
3668                 clear_nlink(inode);
3669                 spin_unlock(&obj->oo_guard);
3670                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3671         }
3672
3673         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3674
3675         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3676
3677         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3678                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3679                                        oh->ot_handle, OI_CHECK_FLD);
3680
3681         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3682         /* XXX: add to ext3 orphan list */
3683         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3684
3685         /* not needed in the cache anymore */
3686         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3687         obj->oo_destroyed = 1;
3688
3689         RETURN(0);
3690 }
3691
3692 /**
3693  * Put the fid into lustre_mdt_attrs, and then place the structure
3694  * inode's ea. This fid should not be altered during the life time
3695  * of the inode.
3696  *
3697  * \retval +ve, on success
3698  * \retval -ve, on error
3699  *
3700  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3701  */
3702 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3703                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3704 {
3705         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3706         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3707         int rc;
3708
3709         ENTRY;
3710
3711         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3712                 RETURN(0);
3713
3714         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3715                 rc = -ENOMEM;
3716
3717         lustre_loa_init(loa, fid, compat, incompat);
3718         lustre_loa_swab(loa, false);
3719
3720         /*
3721          * For the OST device with 256 bytes inode size by default,
3722          * the PFID EA will be stored together with LMA EA to avoid
3723          * performance trouble. Otherwise the PFID EA can be stored
3724          * independently. LU-8998
3725          */
3726         if ((compat & LMAC_FID_ON_OST) &&
3727             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3728                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3729                                      sizeof(*loa), XATTR_CREATE);
3730         else
3731                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3732                                      sizeof(*lma), XATTR_CREATE);
3733         /*
3734          * LMA may already exist, but we need to check that all the
3735          * desired compat/incompat flags have been added.
3736          */
3737         if (unlikely(rc == -EEXIST)) {
3738                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3739                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3740                 if (rc < 0)
3741                         RETURN(rc);
3742
3743                 if (rc < sizeof(*lma))
3744                         RETURN(-EINVAL);
3745
3746                 lustre_loa_swab(loa, true);
3747                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3748                     ((compat == 0 && incompat == 0) ||
3749                      (!(~lma->lma_compat & compat) &&
3750                       !(~lma->lma_incompat & incompat))))
3751                         RETURN(0);
3752
3753                 lma->lma_self_fid = *fid;
3754                 lma->lma_compat |= compat;
3755                 lma->lma_incompat |= incompat;
3756                 if (rc == sizeof(*lma)) {
3757                         lustre_lma_swab(lma);
3758                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3759                                              sizeof(*lma), XATTR_REPLACE);
3760                 } else {
3761                         lustre_loa_swab(loa, false);
3762                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3763                                              sizeof(*loa), XATTR_REPLACE);
3764                 }
3765         }
3766
3767         RETURN(rc);
3768 }
3769
3770 /**
3771  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3772  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3773  * To have compatilibility with 1.8 ldiskfs driver we need to have
3774  * magic number at start of fid data.
3775  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3776  * its inmemory API.
3777  */
3778 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3779                                          const struct lu_fid *fid)
3780 {
3781         if (!fid_is_namespace_visible(fid) ||
3782             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3783                 param->edp_magic = 0;
3784                 return;
3785         }
3786
3787         param->edp_magic = LDISKFS_LUFID_MAGIC;
3788         param->edp_len =  sizeof(struct lu_fid) + 1;
3789         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3790 }
3791
3792 /**
3793  * Try to read the fid from inode ea into dt_rec.
3794  *
3795  * \param fid object fid.
3796  *
3797  * \retval 0 on success
3798  */
3799 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3800                           __u32 ino, struct lu_fid *fid,
3801                           struct osd_inode_id *id)
3802 {
3803         struct osd_thread_info *info  = osd_oti_get(env);
3804         struct inode *inode;
3805
3806         ENTRY;
3807
3808         osd_id_gen(id, ino, OSD_OII_NOGEN);
3809         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3810         if (IS_ERR(inode))
3811                 RETURN(PTR_ERR(inode));
3812
3813         iput(inode);
3814         RETURN(0);
3815 }
3816
3817 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3818                                         struct inode *dir,
3819                                         struct inode *parent_dir,
3820                                         const struct lu_fid *dot_fid,
3821                                         const struct lu_fid *dot_dot_fid,
3822                                         struct osd_thandle *oth)
3823 {
3824         struct ldiskfs_dentry_param *dot_ldp;
3825         struct ldiskfs_dentry_param *dot_dot_ldp;
3826         __u32 saved_nlink = dir->i_nlink;
3827         int rc;
3828
3829         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3830         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3831
3832         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3833         dot_ldp->edp_magic = 0;
3834
3835         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3836                                     dir, dot_ldp, dot_dot_ldp);
3837         /*
3838          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3839          * the subseqent ref_add() will increase the dir->i_nlink
3840          * as 3. That is incorrect for new created directory.
3841          *
3842          * It looks like hack, because we want to make the OSD API
3843          * to be order-independent for new created directory object
3844          * between dt_insert(..) and ref_add() operations.
3845          *
3846          * Here, we only restore the in-RAM dir-inode's nlink attr,
3847          * becuase if the nlink attr is not 2, then there will be
3848          * ref_add() called following the dt_insert(..), such call
3849          * will make both the in-RAM and on-disk dir-inode's nlink
3850          * attr to be set as 2. LU-7447
3851          */
3852         set_nlink(dir, saved_nlink);
3853         return rc;
3854 }
3855
3856 /**
3857  * Create an local agent inode for remote entry
3858  */
3859 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3860                                                   struct osd_device *osd,
3861                                                   struct osd_object *pobj,
3862                                                   const struct lu_fid *fid,
3863                                                   __u32 type,
3864                                                   struct thandle *th)
3865 {
3866         struct osd_thread_info *info = osd_oti_get(env);
3867         struct inode *local;
3868         struct osd_thandle *oh;
3869         int rc;
3870
3871         ENTRY;
3872
3873         LASSERT(th);
3874         oh = container_of(th, struct osd_thandle, ot_super);
3875         LASSERT(oh->ot_handle->h_transaction != NULL);
3876
3877         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3878                                      NULL);
3879         if (IS_ERR(local)) {
3880                 CERROR("%s: create local error %d\n", osd_name(osd),
3881                        (int)PTR_ERR(local));
3882                 RETURN(local);
3883         }
3884
3885         /*
3886          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3887          * correct gid on remote file, not agent here
3888          */
3889         local->i_gid = current_fsgid();
3890         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3891
3892         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3893          * That gives e2fsck something to look at and be happy, and allows
3894          * debugging if we need to determine where this symlink came from.
3895          */
3896         if (S_ISLNK(type)) {
3897                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3898                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3899                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3900
3901                 i_size_write(local, rc);
3902                 LDISKFS_I(local)->i_disksize = rc;
3903         }
3904         unlock_new_inode(local);
3905
3906         /* Agent inode should not have project ID */
3907 #ifdef  HAVE_PROJECT_QUOTA
3908         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3909             i_projid_read(pobj->oo_inode) != 0) {
3910                 rc = osd_transfer_project(local, 0, th);
3911                 if (rc) {
3912                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3913                                "quota enforcement enabled on the ldiskfs "
3914                                "filesystem?\n", local->i_sb->s_id, rc);
3915                         RETURN(ERR_PTR(rc));
3916                 }
3917         }
3918 #endif
3919         /* Set special LMA flag for local agent inode */
3920         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3921         if (rc != 0) {
3922                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3923                        osd_name(osd), PFID(fid), rc);
3924                 RETURN(ERR_PTR(rc));
3925         }
3926
3927         if (!S_ISDIR(type))
3928                 RETURN(local);
3929
3930         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3931                                          lu_object_fid(&pobj->oo_dt.do_lu),
3932                                          fid, oh);
3933         if (rc != 0) {
3934                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3935                         osd_name(osd), PFID(fid), rc);
3936                 RETURN(ERR_PTR(rc));
3937         }
3938
3939         RETURN(local);
3940 }
3941
3942 /**
3943  * when direntry is deleted, we have to take care of possible agent inode
3944  * referenced by that. unfortunately we can't do this at that point:
3945  * iget() within a running transaction leads to deadlock and we better do
3946  * not call that every delete declaration to save performance. so we put
3947  * a potention agent inode on a list and process that once the transaction
3948  * is over. Notice it's not any worse in terms of real orphans as regular
3949  * object destroy doesn't put inodes on the on-disk orphan list. this should
3950  * be addressed separately
3951  */
3952 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3953                                             struct osd_thandle *oh,
3954                                             __u32 ino)
3955 {
3956         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3957         struct osd_obj_orphan *oor;
3958
3959         OBD_ALLOC_PTR(oor);
3960         if (oor == NULL)
3961                 return -ENOMEM;
3962
3963         oor->oor_ino = ino;
3964         oor->oor_env = (struct lu_env *)env;
3965         spin_lock(&osd->od_osfs_lock);
3966         list_add(&oor->oor_list, &osd->od_orphan_list);
3967         spin_unlock(&osd->od_osfs_lock);
3968
3969         oh->ot_remove_agents = 1;
3970
3971         return 0;
3972
3973 }
3974
3975 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3976                                                 struct osd_device *osd)
3977 {
3978         struct osd_thread_info *info = osd_oti_get(env);
3979         struct osd_obj_orphan *oor, *tmp;
3980         struct osd_inode_id id;
3981         LIST_HEAD(list);
3982         struct inode *inode;
3983         struct lu_fid fid;
3984         handle_t *jh;
3985         __u32 ino;
3986
3987         spin_lock(&osd->od_osfs_lock);
3988         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3989                 if (oor->oor_env == env)
3990                         list_move(&oor->oor_list, &list);
3991         }
3992         spin_unlock(&osd->od_osfs_lock);
3993
3994         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3995
3996                 ino = oor->oor_ino;
3997
3998                 list_del(&oor->oor_list);
3999                 OBD_FREE_PTR(oor);
4000
4001                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4002                 inode = osd_iget_fid(info, osd, &id, &fid);
4003                 if (IS_ERR(inode))
4004                         continue;
4005
4006                 if (!osd_remote_fid(env, osd, &fid)) {
4007                         iput(inode);
4008                         continue;
4009                 }
4010
4011                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4012                 clear_nlink(inode);
4013                 mark_inode_dirty(inode);
4014                 ldiskfs_journal_stop(jh);
4015                 iput(inode);
4016         }
4017
4018         return 0;
4019 }
4020
4021 /**
4022  * OSD layer object create function for OST objects (b=11826).
4023  *
4024  * The FID is inserted into inode xattr here.
4025  *
4026  * \retval   0, on success
4027  * \retval -ve, on error
4028  */
4029 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4030                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4031                       struct dt_object_format *dof, struct thandle *th)
4032 {
4033         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4034         struct osd_object *obj = osd_dt_obj(dt);
4035         struct osd_thread_info *info = osd_oti_get(env);
4036         int result, on_ost = 0;
4037
4038         ENTRY;
4039
4040         if (dt_object_exists(dt))
4041                 RETURN(-EEXIST);
4042
4043         LINVRNT(osd_invariant(obj));
4044         LASSERT(!dt_object_remote(dt));
4045         LASSERT(osd_is_write_locked(env, obj));
4046         LASSERT(th != NULL);
4047
4048         if (unlikely(fid_is_acct(fid)))
4049                 /*
4050                  * Quota files can't be created from the kernel any more,
4051                  * 'tune2fs -O quota' will take care of creating them
4052                  */
4053                 RETURN(-EPERM);
4054
4055         result = __osd_create(info, obj, attr, hint, dof, th);
4056         if (result == 0) {
4057                 if (fid_is_idif(fid) &&
4058                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4059                         struct lu_fid *tfid = &info->oti_fid;
4060                         struct ost_id *oi   = &info->oti_ostid;
4061
4062                         fid_to_ostid(fid, oi);
4063                         ostid_to_fid(tfid, oi, 0);
4064                         on_ost = 1;
4065                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4066                                                 LMAC_FID_ON_OST, 0);
4067                 } else {
4068                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4069                                                fid, OI_CHECK_FLD);
4070                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4071                                                 on_ost ? LMAC_FID_ON_OST : 0,
4072                                                 0);
4073                 }
4074                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4075                         obj->oo_dt.do_body_ops = &osd_body_ops;
4076         }
4077
4078         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4079                 result = __osd_oi_insert(env, obj, fid, th);
4080
4081         /*
4082          * a small optimization - dt_insert() isn't usually applied
4083          * to OST objects, so we don't need to cache OI mapping for
4084          * OST objects
4085          */
4086         if (result == 0 && on_ost == 0) {
4087                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4088
4089                 result = osd_idc_find_and_init(env, osd, obj);
4090                 LASSERT(result == 0);
4091         }
4092
4093         LASSERT(ergo(result == 0,
4094                      dt_object_exists(dt) && !dt_object_remote(dt)));
4095         LINVRNT(osd_invariant(obj));
4096         RETURN(result);
4097 }
4098
4099 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4100                                struct thandle *handle)
4101 {
4102         struct osd_thandle *oh;
4103
4104         /* it's possible that object doesn't exist yet */
4105         LASSERT(handle != NULL);
4106
4107         oh = container_of(handle, struct osd_thandle, ot_super);
4108         LASSERT(oh->ot_handle == NULL);
4109
4110         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4111                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4112
4113         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4114
4115         return 0;
4116 }
4117
4118 /*
4119  * Concurrency: @dt is write locked.
4120  */
4121 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4122                        struct thandle *th)
4123 {
4124         struct osd_object *obj = osd_dt_obj(dt);
4125         struct inode *inode = obj->oo_inode;
4126         struct osd_thandle *oh;
4127         int rc = 0;
4128
4129         if (!dt_object_exists(dt) || obj->oo_destroyed)
4130                 return -ENOENT;
4131
4132         LINVRNT(osd_invariant(obj));
4133         LASSERT(!dt_object_remote(dt));
4134         LASSERT(osd_is_write_locked(env, obj));
4135         LASSERT(th != NULL);
4136
4137         oh = container_of(th, struct osd_thandle, ot_super);
4138         LASSERT(oh->ot_handle != NULL);
4139
4140         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4141
4142         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4143                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4144         /*
4145          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4146          * (65000) subdirectories by storing "1" in i_nlink if the link count
4147          * would otherwise overflow. Directory tranversal tools understand
4148          * that (st_nlink == 1) indicates that the filesystem dose not track
4149          * hard links count on the directory, and will not abort subdirectory
4150          * scanning early once (st_nlink - 2) subdirs have been found.
4151          *
4152          * This also has to properly handle the case of inodes with nlink == 0
4153          * in case they are being linked into the PENDING directory
4154          */
4155         spin_lock(&obj->oo_guard);
4156         if (unlikely(inode->i_nlink == 0))
4157                 /* inc_nlink from 0 may cause WARN_ON */
4158                 set_nlink(inode, 1);
4159         else {
4160                 ldiskfs_inc_count(oh->ot_handle, inode);
4161                 if (!S_ISDIR(inode->i_mode))
4162                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4163         }
4164         spin_unlock(&obj->oo_guard);
4165
4166         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4167         LINVRNT(osd_invariant(obj));
4168
4169         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4170
4171         return rc;
4172 }
4173
4174 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4175                                struct thandle *handle)
4176 {
4177         struct osd_thandle *oh;
4178
4179         if (!dt_object_exists(dt))
4180                 return -ENOENT;
4181
4182         LASSERT(!dt_object_remote(dt));
4183         LASSERT(handle != NULL);
4184
4185         oh = container_of(handle, struct osd_thandle, ot_super);
4186         LASSERT(oh->ot_handle == NULL);
4187
4188         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4189                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4190
4191         return 0;
4192 }
4193
4194 /*
4195  * Concurrency: @dt is write locked.
4196  */
4197 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4198                        struct thandle *th)
4199 {
4200         struct osd_object *obj = osd_dt_obj(dt);
4201         struct inode *inode = obj->oo_inode;
4202         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4203         struct osd_thandle *oh;
4204
4205         if (!dt_object_exists(dt))
4206                 return -ENOENT;
4207
4208         LINVRNT(osd_invariant(obj));
4209         LASSERT(!dt_object_remote(dt));
4210         LASSERT(osd_is_write_locked(env, obj));
4211         LASSERT(th != NULL);
4212
4213         oh = container_of(th, struct osd_thandle, ot_super);
4214         LASSERT(oh->ot_handle != NULL);
4215
4216         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4217
4218         spin_lock(&obj->oo_guard);
4219         /*
4220          * That can be result of upgrade from old Lustre version and
4221          * applied only to local files.  Just skip this ref_del call.
4222          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4223          */
4224         if (inode->i_nlink == 0) {
4225                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4226                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4227                              ", maybe an upgraded file? (LU-3915)\n",
4228                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4229                 spin_unlock(&obj->oo_guard);
4230                 return 0;
4231         }
4232
4233         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4234                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4235
4236         ldiskfs_dec_count(oh->ot_handle, inode);
4237         spin_unlock(&obj->oo_guard);
4238
4239         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4240         LINVRNT(osd_invariant(obj));
4241
4242         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4243
4244         return 0;
4245 }
4246
4247 /*
4248  * Concurrency: @dt is read locked.
4249  */
4250 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4251                          struct lu_buf *buf, const char *name)
4252 {
4253         struct osd_object      *obj    = osd_dt_obj(dt);
4254         struct inode           *inode  = obj->oo_inode;
4255         struct osd_thread_info *info   = osd_oti_get(env);
4256         struct dentry          *dentry = &info->oti_obj_dentry;
4257         bool cache_xattr = false;
4258         int rc;
4259
4260         LASSERT(buf);
4261
4262         /* version get is not real XATTR but uses xattr API */
4263         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4264                 dt_obj_version_t *ver = buf->lb_buf;
4265
4266                 /*
4267                  * for version we are just using xattr API but change inode
4268                  * field instead
4269                  */
4270                 if (buf->lb_len == 0)
4271                         return sizeof(dt_obj_version_t);
4272
4273                 if (buf->lb_len < sizeof(dt_obj_version_t))
4274                         return -ERANGE;
4275
4276                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4277                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4278
4279                 *ver = LDISKFS_I(inode)->i_fs_version;
4280
4281                 return sizeof(dt_obj_version_t);
4282         }
4283
4284         if (!dt_object_exists(dt))
4285                 return -ENOENT;
4286
4287         LASSERT(!dt_object_remote(dt));
4288         LASSERT(inode->i_op != NULL);
4289 #ifdef HAVE_IOP_XATTR
4290         LASSERT(inode->i_op->getxattr != NULL);
4291 #endif
4292
4293         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4294             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4295                 cache_xattr = true;
4296
4297         if (cache_xattr) {
4298                 rc = osd_oxc_get(obj, name, buf);
4299                 if (rc != -ENOENT)
4300                         return rc;
4301         }
4302
4303         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4304                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4305                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4306                 struct filter_fid *ff;
4307                 struct ost_layout *ol;
4308
4309                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4310                 if (rc)
4311                         return rc;
4312
4313                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4314
4315                 rc = sizeof(*ff);
4316                 if (buf->lb_len == 0 || !buf->lb_buf)
4317                         return rc;
4318
4319                 if (buf->lb_len < rc)
4320                         return -ERANGE;
4321
4322                 ff = buf->lb_buf;
4323                 ol = &ff->ff_layout;
4324                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4325                                                   PFID_STRIPE_IDX_BITS);
4326                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4327                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4328                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4329                 if (lma->lma_compat & LMAC_COMP_INFO) {
4330                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4331                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4332                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4333                 } else {
4334                         ol->ol_comp_start = 0;
4335                         ol->ol_comp_end = 0;
4336                         ol->ol_comp_id = 0;
4337                 }
4338         } else {
4339                 rc = __osd_xattr_get(inode, dentry, name,
4340                                      buf->lb_buf, buf->lb_len);
4341         }
4342
4343         if (cache_xattr) {
4344                 if (rc == -ENOENT || rc == -ENODATA)
4345                         osd_oxc_add(obj, name, NULL, 0);
4346                 else if (rc > 0 && buf->lb_buf != NULL)
4347                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4348         }
4349
4350         return rc;
4351 }
4352
4353 static int osd_declare_xattr_set(const struct lu_env *env,
4354                                  struct dt_object *dt,
4355                                  const struct lu_buf *buf, const char *name,
4356                                  int fl, struct thandle *handle)
4357 {
4358         struct osd_thandle *oh;
4359         int credits = 0;
4360         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4361
4362         LASSERT(handle != NULL);
4363
4364         oh = container_of(handle, struct osd_thandle, ot_super);
4365         LASSERT(oh->ot_handle == NULL);
4366
4367         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4368                 /*
4369                  * For non-upgrading case, the LMA is set first and
4370                  * usually fit inode. But for upgrade case, the LMA
4371                  * may be in another separated EA block.
4372                  */
4373                 if (dt_object_exists(dt)) {
4374                         if (fl == LU_XATTR_REPLACE)
4375                                 credits = 1;
4376                         else
4377                                 goto upgrade;
4378                 }
4379         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4380                 credits = 1;
4381         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4382                 /* We may need to delete the old PFID EA. */
4383                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4384                 if (fl == LU_XATTR_REPLACE)
4385                         credits += 1;
4386                 else
4387                         goto upgrade;
4388         } else {
4389                 /*
4390                  * If some name entry resides on remote MDT, then will create
4391                  * agent entry under remote parent. On the other hand, if the
4392                  * remote entry will be removed, then related agent entry may
4393                  * need to be removed from the remote parent. So there may be
4394                  * kinds of cases, let's declare enough credits. The credits
4395                  * for create agent entry is enough for remove case.
4396                  */
4397                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4398                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4399                         if (dt_object_exists(dt))
4400                                 credits += 1; /* For updating LMA */
4401                 }
4402
4403 upgrade:
4404                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4405
4406                 if (buf != NULL) {
4407                         ssize_t buflen;
4408
4409                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4410                                 /*
4411                                  * learn xattr size from osd_xattr_get if
4412                                  * attribute has not been read yet
4413                                  */
4414                                 buflen = __osd_xattr_get(
4415                                     osd_dt_obj(dt)->oo_inode,
4416                                     &osd_oti_get(env)->oti_obj_dentry,
4417                                     name, NULL, 0);
4418                                 if (buflen < 0)
4419                                         buflen = 0;
4420                         } else {
4421                                 buflen = buf->lb_len;
4422                         }
4423
4424                         if (buflen > sb->s_blocksize) {
4425                                 credits += osd_calc_bkmap_credits(
4426                                     sb, NULL, 0, -1,
4427                                     (buflen + sb->s_blocksize - 1) >>
4428                                     sb->s_blocksize_bits);
4429                         }
4430                 }
4431                 /*
4432                  * xattr set may involve inode quota change, reserve credits for
4433                  * dquot_initialize()
4434                  */
4435                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4436         }
4437
4438         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4439
4440         return 0;
4441 }
4442
4443 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4444                               const struct lu_buf *buf, int fl,
4445                               struct thandle *handle)
4446 {
4447         struct osd_thread_info *info = osd_oti_get(env);
4448         struct dentry *dentry = &info->oti_obj_dentry;
4449         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4450         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4451         struct inode *inode = obj->oo_inode;
4452         struct filter_fid *ff = buf->lb_buf;
4453         struct ost_layout *ol = &ff->ff_layout;
4454         int flags = XATTR_REPLACE;
4455         int rc;
4456
4457         ENTRY;
4458
4459         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4460                 RETURN(-EINVAL);
4461
4462         rc = osd_get_lma(info, inode, dentry, loa);
4463         if (rc == -ENODATA) {
4464                 /* Usually for upgarding from old device */
4465                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4466                                 LMAC_FID_ON_OST, 0);
4467                 flags = XATTR_CREATE;
4468         } else if (rc) {
4469                 RETURN(rc);
4470         }
4471
4472         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4473                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4474                         RETURN(-EEXIST);
4475
4476                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4477                         /* Separate PFID EA from LMA */
4478                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4479                         lustre_lma_swab(lma);
4480                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4481                                              sizeof(*lma), XATTR_REPLACE);
4482                         if (!rc) {
4483                                 obj->oo_pfid_in_lma = 0;
4484                                 rc = LU_XATTR_CREATE;
4485                         }
4486
4487                         RETURN(rc);
4488                 }
4489         } else {
4490                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4491                         RETURN(fl);
4492
4493                 /*
4494                  * Old client does not send stripe information,
4495                  * then store the PFID EA on disk separatedly.
4496                  */
4497                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4498                              ol->ol_stripe_size == 0))
4499                         RETURN(fl);
4500
4501                 /* Remove old PFID EA entry firstly. */
4502                 dquot_initialize(inode);
4503                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4504                 if (rc == -ENODATA) {
4505                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4506                                 RETURN(rc);
4507                 } else if (rc) {
4508                         RETURN(rc);
4509                 }
4510         }
4511
4512         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4513         if (likely(ol->ol_stripe_size != 0)) {
4514                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4515                                              PFID_STRIPE_IDX_BITS;
4516                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4517                 lma->lma_compat |= LMAC_STRIPE_INFO;
4518                 if (ol->ol_comp_id != 0) {
4519                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4520                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4521                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4522                         lma->lma_compat |= LMAC_COMP_INFO;
4523                 }
4524         }
4525
4526         lustre_loa_swab(loa, false);
4527
4528         /* Store the PFID EA inside LMA. */
4529         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4530                              flags);
4531         if (!rc)
4532                 obj->oo_pfid_in_lma = 1;
4533
4534         RETURN(rc);
4535 }
4536
4537 /*
4538  * In DNE environment, the object (in spite of regular file or directory)
4539  * and its name entry may reside on different MDTs. Under such case, we will
4540  * create an agent entry on the MDT where the object resides. The agent entry
4541  * references the object locally, that makes the object to be visible to the
4542  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4543  * such as 'tar' can handle the object properly.
4544  *
4545  * We handle the agent entry during set linkEA that is the common interface
4546  * for both regular file and directroy, can handle kinds of cases, such as
4547  * create/link/unlink/rename, and so on.
4548  *
4549  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4550  *
4551  * XXX: There are two known issues:
4552  * 1. For one object, we will create at most one agent entry even if there
4553  *    may be more than one cross-MDTs hard links on the object. So the local
4554  *    e2fsck may claim that the object's nlink is larger than the name entries
4555  *    that reference such inode. And in further, the e2fsck will fix the nlink
4556  *    attribute to match the local references. Then it will cause the object's
4557  *    nlink attribute to be inconsistent with the global references. it is bad
4558  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4559  *    other hand, the global namespace LFSCK can repair the object's attribute
4560  *    according to the linkEA.
4561  * 2. There may be too many hard links on the object as to its linkEA overflow,
4562  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4563  *    case happened, then at this point, we do not know whether there are some
4564  *    cross-MDTs reference. But there are local references, it guarantees that
4565  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4566  */
4567 static int osd_xattr_handle_linkea(const struct lu_env *env,
4568                                    struct osd_device *osd,
4569                                    struct osd_object *obj,
4570                                    const struct lu_buf *buf,
4571                                    struct thandle *handle)
4572 {
4573         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4574         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4575         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4576         struct lu_name tmpname;
4577         struct osd_thandle *oh;
4578         int rc;
4579         bool remote = false;
4580
4581         ENTRY;
4582
4583         oh = container_of(handle, struct osd_thandle, ot_super);
4584         LASSERT(oh->ot_handle != NULL);
4585
4586         rc = linkea_init_with_rec(&ldata);
4587         if (!rc) {
4588                 linkea_first_entry(&ldata);
4589                 while (ldata.ld_lee != NULL && !remote) {
4590                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4591                                             &tmpname, tfid);
4592                         if (osd_remote_fid(env, osd, tfid) > 0)
4593                                 remote = true;
4594                         else
4595                                 linkea_next_entry(&ldata);
4596                 }
4597         } else if (rc == -ENODATA) {
4598                 rc = 0;
4599         } else {
4600                 RETURN(rc);
4601         }
4602
4603         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4604                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4605                 if (rc)
4606                         CERROR("%s: failed to remove agent entry for "DFID
4607                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4608         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4609                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4610                 if (rc)
4611                         CERROR("%s: failed to create agent entry for "DFID
4612                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4613         }
4614
4615         RETURN(rc);
4616 }
4617
4618 /*
4619  * Concurrency: @dt is write locked.
4620  */
4621 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4622                          const struct lu_buf *buf, const char *name, int fl,
4623                          struct thandle *handle)
4624 {
4625         struct osd_object *obj = osd_dt_obj(dt);
4626         struct osd_device *osd = osd_obj2dev(obj);
4627         struct inode *inode = obj->oo_inode;
4628         struct osd_thread_info *info = osd_oti_get(env);
4629         int fs_flags = 0;
4630         int len;
4631         int rc;
4632
4633         ENTRY;
4634
4635         LASSERT(handle);
4636         LASSERT(buf);
4637
4638         /* version set is not real XATTR */
4639         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4640                 dt_obj_version_t *version = buf->lb_buf;
4641
4642                 /*
4643                  * for version we are just using xattr API but change inode
4644                  * field instead
4645                  */
4646                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4647
4648                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4649                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4650
4651                 LDISKFS_I(inode)->i_fs_version = *version;
4652                 /*
4653                  * Version is set after all inode operations are finished,
4654                  * so we should mark it dirty here
4655                  */
4656                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4657
4658                 RETURN(0);
4659         }
4660
4661         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4662                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4663
4664         len = buf->lb_len;
4665         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4666
4667         /*
4668          * For the OST device with 256 bytes inode size by default,
4669          * the PFID EA will be stored together with LMA EA to avoid
4670          * performance trouble. Otherwise the PFID EA can be stored
4671          * independently. LU-8998
4672          */
4673         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4674             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4675                 LASSERT(buf->lb_buf);
4676
4677                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4678                 if (fl <= 0)
4679                         RETURN(fl);
4680         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4681                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4682                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4683
4684                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4685                 if (rc)
4686                         RETURN(rc);
4687
4688                 lma->lma_incompat |= LMAI_STRIPED;
4689                 lustre_lma_swab(lma);
4690                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4691                                      sizeof(*lma), XATTR_REPLACE);
4692                 if (rc != 0)
4693                         RETURN(rc);
4694         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4695                 LASSERT(!osd->od_is_ost);
4696
4697                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4698                 if (rc)
4699                         RETURN(rc);
4700         }
4701
4702         if (fl & LU_XATTR_REPLACE)
4703                 fs_flags |= XATTR_REPLACE;
4704
4705         if (fl & LU_XATTR_CREATE)
4706                 fs_flags |= XATTR_CREATE;
4707
4708         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4709         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4710
4711         if (rc == 0 &&
4712             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4713              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4714                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4715
4716         return rc;
4717 }
4718
4719 /*
4720  * Concurrency: @dt is read locked.
4721  */
4722 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4723                           const struct lu_buf *buf)
4724 {
4725         struct osd_object *obj = osd_dt_obj(dt);
4726         struct inode *inode = obj->oo_inode;
4727         struct osd_thread_info *info = osd_oti_get(env);
4728         struct dentry *dentry = &info->oti_obj_dentry;
4729
4730         if (!dt_object_exists(dt))
4731                 return -ENOENT;
4732
4733         LASSERT(!dt_object_remote(dt));
4734         LASSERT(inode->i_op != NULL);
4735         LASSERT(inode->i_op->listxattr != NULL);
4736
4737         dentry->d_inode = inode;
4738         dentry->d_sb = inode->i_sb;
4739         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4740 }
4741
4742 static int osd_declare_xattr_del(const struct lu_env *env,
4743                                  struct dt_object *dt, const char *name,
4744                                  struct thandle *handle)
4745 {
4746         struct osd_thandle *oh;
4747         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4748
4749         LASSERT(!dt_object_remote(dt));
4750         LASSERT(handle != NULL);
4751
4752         oh = container_of(handle, struct osd_thandle, ot_super);
4753         LASSERT(oh->ot_handle == NULL);
4754
4755         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4756                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4757         /*
4758          * xattr del may involve inode quota change, reserve credits for
4759          * dquot_initialize()
4760          */
4761         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4762
4763         return 0;
4764 }
4765
4766 /*
4767  * Concurrency: @dt is write locked.
4768  */
4769 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4770                          const char *name, struct thandle *handle)
4771 {
4772         struct osd_object *obj = osd_dt_obj(dt);
4773         struct inode *inode = obj->oo_inode;
4774         struct osd_thread_info *info = osd_oti_get(env);
4775         struct dentry *dentry = &info->oti_obj_dentry;
4776         int rc;
4777
4778         if (!dt_object_exists(dt))
4779                 return -ENOENT;
4780
4781         LASSERT(!dt_object_remote(dt));
4782         LASSERT(inode->i_op != NULL);
4783         LASSERT(handle != NULL);
4784 #ifdef HAVE_IOP_XATTR
4785         LASSERT(inode->i_op->removexattr != NULL);
4786 #endif
4787
4788         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4789
4790         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4791                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4792
4793                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4794                                  &info->oti_ost_attrs);
4795                 if (!rc) {
4796                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4797
4798                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4799                         lustre_lma_swab(lma);
4800                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4801                                              sizeof(*lma), XATTR_REPLACE);
4802                         if (!rc)
4803                                 obj->oo_pfid_in_lma = 0;
4804                 }
4805         } else {
4806                 dquot_initialize(inode);
4807                 dentry->d_inode = inode;
4808                 dentry->d_sb = inode->i_sb;
4809                 rc = ll_vfs_removexattr(dentry, inode, name);
4810         }
4811
4812         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4813
4814         if (rc == 0 &&
4815             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4816              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4817                 osd_oxc_del(obj, name);
4818
4819         return rc;
4820 }
4821
4822 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4823                            __u64 start, __u64 end)
4824 {
4825         struct osd_object *obj = osd_dt_obj(dt);
4826         struct inode *inode = obj->oo_inode;
4827         struct osd_thread_info *info = osd_oti_get(env);
4828         struct dentry *dentry = &info->oti_obj_dentry;
4829         struct file *file = &info->oti_file;
4830         int rc;
4831
4832         ENTRY;
4833
4834         dentry->d_inode = inode;
4835         dentry->d_sb = inode->i_sb;
4836         file->f_path.dentry = dentry;
4837         file->f_mapping = inode->i_mapping;
4838         file->f_op = inode->i_fop;
4839         file->f_inode = inode;
4840
4841         rc = vfs_fsync_range(file, start, end, 0);
4842
4843         RETURN(rc);
4844 }
4845
4846 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4847 {
4848         return 0;
4849 }
4850
4851 /*
4852  * Index operations.
4853  */
4854
4855 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4856                                const struct dt_index_features *feat)
4857 {
4858         struct iam_descr *descr;
4859
4860         if (osd_object_is_root(o))
4861                 return feat == &dt_directory_features;
4862
4863         LASSERT(o->oo_dir != NULL);
4864
4865         descr = o->oo_dir->od_container.ic_descr;
4866         if (feat == &dt_directory_features) {
4867                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4868                         return 1;
4869                 else
4870                         return 0;
4871         } else {
4872                 return feat->dif_keysize_min <= descr->id_key_size &&
4873                        descr->id_key_size <= feat->dif_keysize_max &&
4874                        feat->dif_recsize_min <= descr->id_rec_size &&
4875                        descr->id_rec_size <= feat->dif_recsize_max &&
4876                        !(feat->dif_flags & (DT_IND_VARKEY |
4877                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4878                        ergo(feat->dif_flags & DT_IND_UPDATE,
4879                             1 /* XXX check that object (and fs) is writable */);
4880         }
4881 }
4882
4883 static int osd_iam_container_init(const struct lu_env *env,
4884                                   struct osd_object *obj,
4885                                   struct osd_directory *dir)
4886 {
4887         struct iam_container *bag = &dir->od_container;
4888         int result;
4889
4890         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4891         if (result != 0)
4892                 return result;
4893
4894         result = iam_container_setup(bag);
4895         if (result == 0)
4896                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4897         else
4898                 iam_container_fini(bag);
4899
4900         return result;
4901 }
4902
4903
4904 /*
4905  * Concurrency: no external locking is necessary.
4906  */
4907 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4908                          const struct dt_index_features *feat)
4909 {
4910         int result;
4911         int skip_iam = 0;
4912         struct osd_object *obj = osd_dt_obj(dt);
4913
4914         LINVRNT(osd_invariant(obj));
4915
4916         if (osd_object_is_root(obj)) {
4917                 dt->do_index_ops = &osd_index_ea_ops;
4918                 result = 0;
4919         } else if (feat == &dt_directory_features) {
4920                 dt->do_index_ops = &osd_index_ea_ops;
4921                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4922                         result = 0;
4923                 else
4924                         result = -ENOTDIR;
4925                 skip_iam = 1;
4926         } else if (unlikely(feat == &dt_otable_features)) {
4927                 dt->do_index_ops = &osd_otable_ops;
4928                 return 0;
4929         } else if (unlikely(feat == &dt_acct_features)) {
4930                 dt->do_index_ops = &osd_acct_index_ops;
4931                 result = 0;
4932                 skip_iam = 1;
4933         } else if (!osd_has_index(obj)) {
4934                 struct osd_directory *dir;
4935                 struct osd_device *osd = osd_obj2dev(obj);
4936                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4937
4938                 OBD_ALLOC_PTR(dir);
4939                 if (dir) {
4940
4941                         spin_lock(&obj->oo_guard);
4942                         if (obj->oo_dir == NULL)
4943                                 obj->oo_dir = dir;
4944                         else
4945                                 /*
4946                                  * Concurrent thread allocated container data.
4947                                  */
4948                                 OBD_FREE_PTR(dir);
4949                         spin_unlock(&obj->oo_guard);
4950                         /*
4951                          * Now, that we have container data, serialize its
4952                          * initialization.
4953                          */
4954                         down_write(&obj->oo_ext_idx_sem);
4955                         /*
4956                          * recheck under lock.
4957                          */
4958
4959                         if (osd_has_index(obj)) {
4960                                 result = 0;
4961                                 goto unlock;
4962                         }
4963
4964                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4965                         if (result || feat == &dt_lfsck_namespace_features ||
4966                             feat == &dt_lfsck_layout_orphan_features ||
4967                             feat == &dt_lfsck_layout_dangling_features)
4968                                 goto unlock;
4969
4970                         result = osd_index_register(osd, fid,
4971                                                     feat->dif_keysize_max,
4972                                                     feat->dif_recsize_max);
4973                         if (result < 0)
4974                                 CWARN("%s: failed to register index "
4975                                       DFID": rc = %d\n",
4976                                       osd_name(osd), PFID(fid), result);
4977                         else if (result > 0)
4978                                 result = 0;
4979                         else
4980                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4981                                        " (%d/%d) registered\n",
4982                                        osd_name(osd), PFID(fid),
4983                                        (int)feat->dif_keysize_max,
4984                                        (int)feat->dif_recsize_max);
4985
4986 unlock:
4987                         up_write(&obj->oo_ext_idx_sem);
4988                 } else {
4989                         result = -ENOMEM;
4990                 }
4991         } else {
4992                 result = 0;
4993         }
4994
4995         if (result == 0 && skip_iam == 0) {
4996                 if (!osd_iam_index_probe(env, obj, feat))
4997                         result = -ENOTDIR;
4998         }
4999         LINVRNT(osd_invariant(obj));
5000
5001         return result;
5002 }
5003
5004 static int osd_otable_it_attr_get(const struct lu_env *env,
5005                                  struct dt_object *dt,
5006                                  struct lu_attr *attr)
5007 {
5008         attr->la_valid = 0;
5009         return 0;
5010 }
5011
5012 static const struct dt_object_operations osd_obj_ops = {
5013         .do_read_lock           = osd_read_lock,
5014         .do_write_lock          = osd_write_lock,
5015         .do_read_unlock         = osd_read_unlock,
5016         .do_write_unlock        = osd_write_unlock,
5017         .do_write_locked        = osd_write_locked,
5018         .do_attr_get            = osd_attr_get,
5019         .do_declare_attr_set    = osd_declare_attr_set,
5020         .do_attr_set            = osd_attr_set,
5021         .do_ah_init             = osd_ah_init,
5022         .do_declare_create      = osd_declare_create,
5023         .do_create              = osd_create,
5024         .do_declare_destroy     = osd_declare_destroy,
5025         .do_destroy             = osd_destroy,
5026         .do_index_try           = osd_index_try,
5027         .do_declare_ref_add     = osd_declare_ref_add,
5028         .do_ref_add             = osd_ref_add,
5029         .do_declare_ref_del     = osd_declare_ref_del,
5030         .do_ref_del             = osd_ref_del,
5031         .do_xattr_get           = osd_xattr_get,
5032         .do_declare_xattr_set   = osd_declare_xattr_set,
5033         .do_xattr_set           = osd_xattr_set,
5034         .do_declare_xattr_del   = osd_declare_xattr_del,
5035         .do_xattr_del           = osd_xattr_del,
5036         .do_xattr_list          = osd_xattr_list,
5037         .do_object_sync         = osd_object_sync,
5038         .do_invalidate          = osd_invalidate,
5039 };
5040
5041 static const struct dt_object_operations osd_obj_otable_it_ops = {
5042         .do_attr_get    = osd_otable_it_attr_get,
5043         .do_index_try   = osd_index_try,
5044 };
5045
5046 static int osd_index_declare_iam_delete(const struct lu_env *env,
5047                                         struct dt_object *dt,
5048                                         const struct dt_key *key,
5049                                         struct thandle *handle)
5050 {
5051         struct osd_thandle *oh;
5052
5053         oh = container_of(handle, struct osd_thandle, ot_super);
5054         LASSERT(oh->ot_handle == NULL);
5055
5056         /* Recycle  may cause additional three blocks to be changed. */
5057         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5058                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5059
5060         return 0;
5061 }
5062
5063 /**
5064  *      delete a (key, value) pair from index \a dt specified by \a key
5065  *
5066  *      \param  dt      osd index object
5067  *      \param  key     key for index
5068  *      \param  rec     record reference
5069  *      \param  handle  transaction handler
5070  *
5071  *      \retval  0  success
5072  *      \retval -ve   failure
5073  */
5074 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5075                                 const struct dt_key *key,
5076                                 struct thandle *handle)
5077 {
5078         struct osd_thread_info *oti = osd_oti_get(env);
5079         struct osd_object *obj = osd_dt_obj(dt);
5080         struct osd_thandle *oh;
5081         struct iam_path_descr *ipd;
5082         struct iam_container *bag = &obj->oo_dir->od_container;
5083         int rc;
5084
5085         ENTRY;
5086
5087         if (!dt_object_exists(dt))
5088                 RETURN(-ENOENT);
5089
5090         LINVRNT(osd_invariant(obj));
5091         LASSERT(!dt_object_remote(dt));
5092         LASSERT(bag->ic_object == obj->oo_inode);
5093         LASSERT(handle != NULL);
5094
5095         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5096
5097         ipd = osd_idx_ipd_get(env, bag);
5098         if (unlikely(ipd == NULL))
5099                 RETURN(-ENOMEM);
5100
5101         oh = container_of(handle, struct osd_thandle, ot_super);
5102         LASSERT(oh->ot_handle != NULL);
5103         LASSERT(oh->ot_handle->h_transaction != NULL);
5104
5105         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5106                 /* swab quota uid/gid provided by caller */
5107                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5108                 key = (const struct dt_key *)&oti->oti_quota_id;
5109         }
5110
5111         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5112         osd_ipd_put(env, bag, ipd);
5113         LINVRNT(osd_invariant(obj));
5114         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5115         RETURN(rc);
5116 }
5117
5118 static int osd_index_declare_ea_delete(const struct lu_env *env,
5119                                        struct dt_object *dt,
5120                                        const struct dt_key *key,
5121                                        struct thandle *handle)
5122 {
5123         struct osd_thandle *oh;
5124         struct inode *inode;
5125         int rc, credits;
5126
5127         ENTRY;
5128
5129         LASSERT(!dt_object_remote(dt));
5130         LASSERT(handle != NULL);
5131
5132         oh = container_of(handle, struct osd_thandle, ot_super);
5133         LASSERT(oh->ot_handle == NULL);
5134
5135         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5136         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5137
5138         inode = osd_dt_obj(dt)->oo_inode;
5139         if (inode == NULL)
5140                 RETURN(-ENOENT);
5141
5142         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5143                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5144                                    NULL, OSD_QID_BLK);
5145         RETURN(rc);
5146 }
5147
5148 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5149                                           struct dt_rec *fid)
5150 {
5151         struct osd_fid_pack *rec;
5152         int rc = -ENODATA;
5153
5154         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5155                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5156                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5157                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5158                         rc = -EINVAL;
5159         }
5160         return rc;
5161 }
5162
5163 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5164                           const struct lu_fid *fid)
5165 {
5166         struct seq_server_site *ss = osd_seq_site(osd);
5167
5168         ENTRY;
5169
5170         /* FID seqs not in FLDB, must be local seq */
5171         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5172                 RETURN(0);
5173
5174         /*
5175          * If FLD is not being initialized yet, it only happens during the
5176          * initialization, likely during mgs initialization, and we assume
5177          * this is local FID.
5178          */
5179         if (ss == NULL || ss->ss_server_fld == NULL)
5180                 RETURN(0);
5181
5182         /* Only check the local FLDB here */
5183         if (osd_seq_exists(env, osd, fid_seq(fid)))
5184                 RETURN(0);
5185
5186         RETURN(1);
5187 }
5188
5189 static void osd_take_care_of_agent(const struct lu_env *env,
5190                                    struct osd_device *osd,
5191                                    struct osd_thandle *oh,
5192                                    struct ldiskfs_dir_entry_2 *de)
5193 {
5194         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5195         struct osd_idmap_cache *idc;
5196         int rc, schedule = 0;
5197
5198         LASSERT(de != NULL);
5199
5200         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5201         if (likely(rc == 0)) {
5202                 idc = osd_idc_find_or_init(env, osd, fid);
5203                 if (IS_ERR(idc) || idc->oic_remote)
5204                         schedule = 1;
5205         } else if (rc == -ENODATA) {
5206                 /*
5207                  * can't get FID, postpone to the end of the
5208                  * transaction when iget() is safe
5209                  */
5210                 schedule = 1;
5211         } else {
5212                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5213         }
5214         if (schedule)
5215                 osd_schedule_agent_inode_removal(env, oh,
5216                                                  le32_to_cpu(de->inode));
5217 }
5218
5219 /**
5220  * Index delete function for interoperability mode (b11826).
5221  * It will remove the directory entry added by osd_index_ea_insert().
5222  * This entry is needed to maintain name->fid mapping.
5223  *
5224  * \param key,  key i.e. file entry to be deleted
5225  *
5226  * \retval   0, on success
5227  * \retval -ve, on error
5228  */
5229 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5230                                const struct dt_key *key, struct thandle *handle)
5231 {
5232         struct osd_object *obj = osd_dt_obj(dt);
5233         struct inode *dir = obj->oo_inode;
5234         struct dentry *dentry;
5235         struct osd_thandle *oh;
5236         struct ldiskfs_dir_entry_2 *de = NULL;
5237         struct buffer_head *bh;
5238         struct htree_lock *hlock = NULL;
5239         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5240         int rc;
5241
5242         ENTRY;
5243
5244         if (!dt_object_exists(dt))
5245                 RETURN(-ENOENT);
5246
5247         LINVRNT(osd_invariant(obj));
5248         LASSERT(!dt_object_remote(dt));
5249         LASSERT(handle != NULL);
5250
5251         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5252
5253         oh = container_of(handle, struct osd_thandle, ot_super);
5254         LASSERT(oh->ot_handle != NULL);
5255         LASSERT(oh->ot_handle->h_transaction != NULL);
5256
5257         dquot_initialize(dir);
5258         dentry = osd_child_dentry_get(env, obj,
5259                                       (char *)key, strlen((char *)key));
5260
5261         if (obj->oo_hl_head != NULL) {
5262                 hlock = osd_oti_get(env)->oti_hlock;
5263                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5264                                    dir, LDISKFS_HLOCK_DEL);
5265         } else {
5266                 down_write(&obj->oo_ext_idx_sem);
5267         }
5268
5269         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5270         if (!IS_ERR(bh)) {
5271                 /*
5272                  * If this is not the ".." entry, it might be a remote DNE
5273                  * entry and  we need to check if the FID is for a remote
5274                  * MDT.  If the FID is  not in the directory entry (e.g.
5275                  * upgraded 1.8 filesystem without dirdata enabled) then
5276                  * we need to get the FID from the LMA. For a remote directory
5277                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5278                  * case.
5279                  *
5280                  * Delete the entry before the agent inode in order to
5281                  * simplify error handling.  At worst an error after deleting
5282                  * the entry first might leak the agent inode afterward. The
5283                  * reverse would need filesystem abort in case of error deleting
5284                  * the entry after the agent had been removed, or leave a
5285                  * dangling entry pointing at a random inode.
5286                  */
5287                 if (strcmp((char *)key, dotdot) != 0)
5288                         osd_take_care_of_agent(env, osd, oh, de);
5289                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5290                 brelse(bh);
5291         } else {
5292                 rc = PTR_ERR(bh);
5293         }
5294
5295         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5296             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5297                 /* NB, dirent count may not be accurate, because it's counted
5298                  * without lock.
5299                  */
5300                 if (obj->oo_dirent_count)
5301                         obj->oo_dirent_count--;
5302                 else
5303                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5304         }
5305         if (hlock != NULL)
5306                 ldiskfs_htree_unlock(hlock);
5307         else
5308                 up_write(&obj->oo_ext_idx_sem);
5309         GOTO(out, rc);
5310 out:
5311         LASSERT(osd_invariant(obj));
5312         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5313         RETURN(rc);
5314 }
5315
5316 /**
5317  *      Lookup index for \a key and copy record to \a rec.
5318  *
5319  *      \param  dt      osd index object
5320  *      \param  key     key for index
5321  *      \param  rec     record reference
5322  *
5323  *      \retval  +ve  success : exact mach
5324  *      \retval  0    return record with key not greater than \a key
5325  *      \retval -ve   failure
5326  */
5327 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5328                                 struct dt_rec *rec, const struct dt_key *key)
5329 {
5330         struct osd_object *obj = osd_dt_obj(dt);
5331         struct iam_path_descr *ipd;
5332         struct iam_container *bag = &obj->oo_dir->od_container;
5333         struct osd_thread_info *oti = osd_oti_get(env);
5334         struct iam_iterator *it = &oti->oti_idx_it;
5335         struct iam_rec *iam_rec;
5336         int rc;
5337
5338         ENTRY;
5339
5340         if (!dt_object_exists(dt))
5341                 RETURN(-ENOENT);
5342
5343         LASSERT(osd_invariant(obj));
5344         LASSERT(!dt_object_remote(dt));
5345         LASSERT(bag->ic_object == obj->oo_inode);
5346
5347         ipd = osd_idx_ipd_get(env, bag);
5348         if (IS_ERR(ipd))
5349                 RETURN(-ENOMEM);
5350
5351         /* got ipd now we can start iterator. */
5352         iam_it_init(it, bag, 0, ipd);
5353
5354         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5355                 /* swab quota uid/gid provided by caller */
5356                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5357                 key = (const struct dt_key *)&oti->oti_quota_id;
5358         }
5359
5360         rc = iam_it_get(it, (struct iam_key *)key);
5361         if (rc >= 0) {
5362                 if (S_ISDIR(obj->oo_inode->i_mode))
5363                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5364                 else
5365                         iam_rec = (struct iam_rec *)rec;
5366
5367                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5368
5369                 if (S_ISDIR(obj->oo_inode->i_mode))
5370                         osd_fid_unpack((struct lu_fid *)rec,
5371                                        (struct osd_fid_pack *)iam_rec);
5372                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5373                         osd_quota_unpack(obj, rec);
5374         }
5375
5376         iam_it_put(it);
5377         iam_it_fini(it);
5378         osd_ipd_put(env, bag, ipd);
5379
5380         LINVRNT(osd_invariant(obj));
5381
5382         RETURN(rc);
5383 }
5384
5385 static int osd_index_declare_iam_insert(const struct lu_env *env,
5386                                         struct dt_object *dt,
5387                                         const struct dt_rec *rec,
5388                                         const struct dt_key *key,
5389                                         struct thandle *handle)
5390 {
5391         struct osd_thandle *oh;
5392
5393         LASSERT(handle != NULL);
5394
5395         oh = container_of(handle, struct osd_thandle, ot_super);
5396         LASSERT(oh->ot_handle == NULL);
5397
5398         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5399                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5400
5401         return 0;
5402 }
5403
5404 /**
5405  *      Inserts (key, value) pair in \a dt index object.
5406  *
5407  *      \param  dt      osd index object
5408  *      \param  key     key for index
5409  *      \param  rec     record reference
5410  *      \param  th      transaction handler
5411  *
5412  *      \retval  0  success
5413  *      \retval -ve failure
5414  */
5415 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5416                                 const struct dt_rec *rec,
5417                                 const struct dt_key *key, struct thandle *th)
5418 {
5419         struct osd_object *obj = osd_dt_obj(dt);
5420         struct iam_path_descr *ipd;
5421         struct osd_thandle *oh;
5422         struct iam_container *bag;
5423         struct osd_thread_info *oti = osd_oti_get(env);
5424         struct iam_rec *iam_rec;
5425         int rc;
5426
5427         ENTRY;
5428
5429         if (!dt_object_exists(dt))
5430                 RETURN(-ENOENT);
5431
5432         LINVRNT(osd_invariant(obj));
5433         LASSERT(!dt_object_remote(dt));
5434
5435         bag = &obj->oo_dir->od_container;
5436         LASSERT(bag->ic_object == obj->oo_inode);
5437         LASSERT(th != NULL);
5438
5439         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5440
5441         ipd = osd_idx_ipd_get(env, bag);
5442         if (unlikely(ipd == NULL))
5443                 RETURN(-ENOMEM);
5444
5445         oh = container_of(th, struct osd_thandle, ot_super);
5446         LASSERT(oh->ot_handle != NULL);
5447         LASSERT(oh->ot_handle->h_transaction != NULL);
5448         if (S_ISDIR(obj->oo_inode->i_mode)) {
5449                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5450                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5451                              &oti->oti_fid);
5452         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5453                 /* pack quota uid/gid */
5454                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5455                 key = (const struct dt_key *)&oti->oti_quota_id;
5456                 /* pack quota record */
5457                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5458                 iam_rec = (struct iam_rec *)rec;
5459         } else {
5460                 iam_rec = (struct iam_rec *)rec;
5461         }
5462
5463         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5464                         iam_rec, ipd);
5465         osd_ipd_put(env, bag, ipd);
5466         LINVRNT(osd_invariant(obj));
5467         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5468         RETURN(rc);
5469 }
5470
5471 /**
5472  * Calls ldiskfs_add_entry() to add directory entry
5473  * into the directory. This is required for
5474  * interoperability mode (b11826)
5475  *
5476  * \retval   0, on success
5477  * \retval -ve, on error
5478  */
5479 static int __osd_ea_add_rec(struct osd_thread_info *info,
5480                             struct osd_object *pobj, struct inode  *cinode,
5481                             const char *name, const struct lu_fid *fid,
5482                             struct htree_lock *hlock, struct thandle *th)
5483 {
5484         struct ldiskfs_dentry_param *ldp;
5485         struct dentry *child;
5486         struct osd_thandle *oth;
5487         int rc;
5488
5489         oth = container_of(th, struct osd_thandle, ot_super);
5490         LASSERT(oth->ot_handle != NULL);
5491         LASSERT(oth->ot_handle->h_transaction != NULL);
5492         LASSERT(pobj->oo_inode);
5493
5494         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5495         if (unlikely(pobj->oo_inode ==
5496                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5497                 ldp->edp_magic = 0;
5498         else
5499                 osd_get_ldiskfs_dirent_param(ldp, fid);
5500         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5501         child->d_fsdata = (void *)ldp;
5502         dquot_initialize(pobj->oo_inode);
5503         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5504                                    child, cinode, hlock);
5505         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5506                 struct ldiskfs_dir_entry_2      *de;
5507                 struct buffer_head              *bh;
5508                 int                              rc1;
5509
5510                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5511                                             NULL, hlock);
5512                 if (!IS_ERR(bh)) {
5513                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5514                                                                bh);
5515                         if (rc1 == 0) {
5516                                 if (S_ISDIR(cinode->i_mode))
5517                                         de->file_type = LDISKFS_DIRENT_LUFID |
5518                                                         LDISKFS_FT_REG_FILE;
5519                                 else
5520                                         de->file_type = LDISKFS_DIRENT_LUFID |
5521                                                         LDISKFS_FT_DIR;
5522                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5523                                                               NULL, bh);
5524                         }
5525                         brelse(bh);
5526                 }
5527         }
5528
5529         RETURN(rc);
5530 }
5531
5532 /**
5533  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5534  * into the directory.Also sets flags into osd object to
5535  * indicate dot and dotdot are created. This is required for
5536  * interoperability mode (b11826)
5537  *
5538  * \param dir   directory for dot and dotdot fixup.
5539  * \param obj   child object for linking
5540  *
5541  * \retval   0, on success
5542  * \retval -ve, on error
5543  */
5544 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5545                               struct osd_object *dir,
5546                               struct inode *parent_dir, const char *name,
5547                               const struct lu_fid *dot_fid,
5548                               const struct lu_fid *dot_dot_fid,
5549                               struct thandle *th)
5550 {
5551         struct inode *inode = dir->oo_inode;
5552         struct osd_thandle *oth;
5553         int result = 0;
5554
5555         oth = container_of(th, struct osd_thandle, ot_super);
5556         LASSERT(oth->ot_handle->h_transaction != NULL);
5557         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5558
5559         if (strcmp(name, dot) == 0) {
5560                 if (dir->oo_compat_dot_created) {
5561                         result = -EEXIST;
5562                 } else {
5563                         LASSERT(inode->i_ino == parent_dir->i_ino);
5564                         dir->oo_compat_dot_created = 1;
5565                         result = 0;
5566                 }
5567         } else if (strcmp(name, dotdot) == 0) {
5568                 if (!dir->oo_compat_dot_created)
5569                         return -EINVAL;
5570                 /* in case of rename, dotdot is already created */
5571                 if (dir->oo_compat_dotdot_created) {
5572                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5573                                                 dot_dot_fid, NULL, th);
5574                 }
5575
5576                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5577                         struct lu_fid tfid = *dot_dot_fid;
5578
5579                         tfid.f_oid--;
5580                         result = osd_add_dot_dotdot_internal(info,
5581                                         dir->oo_inode, parent_dir, dot_fid,
5582                                         &tfid, oth);
5583                 } else {
5584                         result = osd_add_dot_dotdot_internal(info,
5585                                         dir->oo_inode, parent_dir, dot_fid,
5586                                         dot_dot_fid, oth);
5587                 }
5588
5589                 if (result == 0)
5590                         dir->oo_compat_dotdot_created = 1;
5591         }
5592
5593         return result;
5594 }
5595
5596
5597 /**
5598  * It will call the appropriate osd_add* function and return the
5599  * value, return by respective functions.
5600  */
5601 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5602                           struct inode *cinode, const char *name,
5603                           const struct lu_fid *fid, struct thandle *th)
5604 {
5605         struct osd_thread_info *info = osd_oti_get(env);
5606         struct htree_lock *hlock;
5607         int rc;
5608
5609         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5610
5611         if (name[0] == '.' && (name[1] == '\0' ||
5612                                (name[1] == '.' && name[2] == '\0'))) {
5613                 if (hlock != NULL) {
5614                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5615                                            pobj->oo_inode, 0);
5616                 } else {
5617                         down_write(&pobj->oo_ext_idx_sem);
5618                 }
5619
5620                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5621                                         lu_object_fid(&pobj->oo_dt.do_lu),
5622                                         fid, th);
5623         } else {
5624                 if (hlock != NULL) {
5625                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5626                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5627                 } else {
5628                         down_write(&pobj->oo_ext_idx_sem);
5629                 }
5630
5631                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5632                         struct lu_fid *tfid = &info->oti_fid;
5633
5634                         *tfid = *fid;
5635                         tfid->f_ver = ~0;
5636                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5637                                               tfid, hlock, th);
5638                 } else {
5639                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5640                                               hlock, th);
5641                 }
5642         }
5643         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5644             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5645                 pobj->oo_dirent_count++;
5646
5647         if (hlock != NULL)
5648                 ldiskfs_htree_unlock(hlock);
5649         else
5650                 up_write(&pobj->oo_ext_idx_sem);
5651
5652         return rc;
5653 }
5654
5655 static int
5656 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5657                       struct osd_idmap_cache *oic)
5658 {
5659         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5660         struct lu_fid *fid = &oic->oic_fid;
5661         struct osd_inode_id *id = &oic->oic_lid;
5662         struct inode *inode = NULL;
5663         int once = 0;
5664         bool insert;
5665         int rc;
5666
5667         ENTRY;
5668
5669         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5670                 RETURN(0);
5671
5672         if (thread_is_running(&scrub->os_thread) &&
5673             scrub->os_pos_current > id->oii_ino)
5674                 RETURN(0);
5675
5676         if (dev->od_auto_scrub_interval == AS_NEVER ||
5677             ktime_get_real_seconds() <
5678             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5679                 RETURN(0);
5680
5681 again:
5682         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5683         if (rc == -ENOENT) {
5684                 __u32 gen = id->oii_gen;
5685
5686                 insert = true;
5687                 if (inode != NULL)
5688                         goto trigger;
5689
5690                 inode = osd_iget(oti, dev, id);
5691                 /* The inode has been removed (by race maybe). */
5692                 if (IS_ERR(inode)) {
5693                         rc = PTR_ERR(inode);
5694
5695                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5696                 }
5697
5698                 /* The OI mapping is lost. */
5699                 if (gen != OSD_OII_NOGEN)
5700                         goto trigger;
5701
5702                 /*
5703                  * The inode may has been reused by others, we do not know,
5704                  * leave it to be handled by subsequent osd_fid_lookup().
5705                  */
5706                 GOTO(out, rc = 0);
5707         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5708                 GOTO(out, rc);
5709         }
5710
5711         insert = false;
5712
5713 trigger:
5714         if (thread_is_running(&scrub->os_thread)) {
5715                 if (inode == NULL) {
5716                         inode = osd_iget(oti, dev, id);
5717                         /* The inode has been removed (by race maybe). */
5718                         if (IS_ERR(inode)) {
5719                                 rc = PTR_ERR(inode);
5720
5721                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5722                         }
5723                 }
5724
5725                 rc = osd_oii_insert(dev, oic, insert);
5726                 /*
5727                  * There is race condition between osd_oi_lookup and OI scrub.
5728                  * The OI scrub finished just after osd_oi_lookup() failure.
5729                  * Under such case, it is unnecessary to trigger OI scrub again,
5730                  * but try to call osd_oi_lookup() again.
5731                  */
5732                 if (unlikely(rc == -EAGAIN))
5733                         goto again;
5734
5735                 if (!S_ISDIR(inode->i_mode))
5736                         rc = 0;
5737                 else
5738                         rc = osd_check_lmv(oti, dev, inode, oic);
5739
5740                 GOTO(out, rc);
5741         }
5742
5743         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5744                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5745                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5746                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5747                        "%s: trigger partial OI scrub for RPC inconsistency "
5748                        "checking FID "DFID": rc = %d\n",
5749                        osd_dev2name(dev), PFID(fid), rc);
5750                 if (rc == 0 || rc == -EALREADY)
5751                         goto again;
5752         }
5753
5754         GOTO(out, rc);
5755
5756 out:
5757         if (inode)
5758                 iput(inode);
5759
5760         RETURN(rc);
5761 }
5762
5763 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5764                                struct osd_device *dev,
5765                                struct osd_idmap_cache *oic,
5766                                struct lu_fid *fid, __u32 ino)
5767 {
5768         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5769         struct inode *inode;
5770         int rc;
5771
5772         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5773         inode = osd_iget(oti, dev, &oic->oic_lid);
5774         if (IS_ERR(inode)) {
5775                 fid_zero(&oic->oic_fid);
5776                 return PTR_ERR(inode);
5777         }
5778
5779         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5780         iput(inode);
5781         if (rc != 0)
5782                 fid_zero(&oic->oic_fid);
5783         else
5784                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5785         return rc;
5786 }
5787
5788 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5789                       struct osd_inode_id *id, const struct lu_fid *fid)
5790 {
5791         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5792                id->oii_ino, id->oii_gen, info);
5793         info->oti_cache.oic_lid = *id;
5794         info->oti_cache.oic_fid = *fid;
5795         info->oti_cache.oic_dev = osd;
5796 }
5797
5798 /**
5799  * Get parent FID from the linkEA.
5800  *
5801  * For a directory which parent resides on remote MDT, to satisfy the
5802  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5803  * the other hand, to make the lookup(..) on the directory can return
5804  * the real parent FID, we append the real parent FID after its ".."
5805  * name entry in the /REMOTE_PARENT_DIR.
5806  *
5807  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5808  * backup. So after the restore, we cannot get the right parent FID from
5809  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5810  * we have stored the real parent FID in the directory object's linkEA,
5811  * we can parse the linkEA for the real parent FID.
5812  *
5813  * \param[in] env       pointer to the thread context
5814  * \param[in] obj       pointer to the object to be handled
5815  * \param[out]fid       pointer to the buffer to hold the parent FID
5816  *
5817  * \retval              0 for getting the real parent FID successfully
5818  * \retval              negative error number on failure
5819  */
5820 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5821                                     struct osd_object *obj,
5822                                     struct lu_fid *fid)
5823 {
5824         struct osd_thread_info *oti = osd_oti_get(env);
5825         struct lu_buf *buf = &oti->oti_big_buf;
5826         struct dentry *dentry = &oti->oti_obj_dentry;
5827         struct inode *inode = obj->oo_inode;
5828         struct linkea_data ldata = { NULL };
5829         int rc;
5830
5831         ENTRY;
5832
5833         fid_zero(fid);
5834         if (!S_ISDIR(inode->i_mode))
5835                 RETURN(-EIO);
5836
5837 again:
5838         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5839                              buf->lb_buf, buf->lb_len);
5840         if (rc == -ERANGE) {
5841                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5842                                      NULL, 0);
5843                 if (rc > 0) {
5844                         lu_buf_realloc(buf, rc);
5845                         if (buf->lb_buf == NULL)
5846                                 RETURN(-ENOMEM);
5847
5848                         goto again;
5849                 }
5850         }
5851
5852         if (unlikely(rc == 0))
5853                 RETURN(-ENODATA);
5854
5855         if (rc < 0)
5856                 RETURN(rc);
5857
5858         if (unlikely(buf->lb_buf == NULL)) {
5859                 lu_buf_realloc(buf, rc);
5860                 if (buf->lb_buf == NULL)
5861                         RETURN(-ENOMEM);
5862
5863                 goto again;
5864         }
5865
5866         ldata.ld_buf = buf;
5867         rc = linkea_init_with_rec(&ldata);
5868         if (!rc) {
5869                 linkea_first_entry(&ldata);
5870                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5871         }
5872
5873         RETURN(rc);
5874 }
5875
5876 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5877                                     struct inode *inode,
5878                                     const struct lu_fid *pfid,
5879                                     const char *name, const int namelen)
5880 {
5881         struct osd_thread_info *oti = osd_oti_get(env);
5882         struct lu_buf *buf = &oti->oti_big_buf;
5883         struct dentry *dentry = &oti->oti_obj_dentry;
5884         struct linkea_data ldata = { NULL };
5885         struct lu_name cname = { .ln_name = name,
5886                                  .ln_namelen = namelen };
5887         int rc;
5888
5889         ENTRY;
5890
5891 again:
5892         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5893                              buf->lb_buf, buf->lb_len);
5894         if (rc == -ERANGE)
5895                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5896
5897         if (rc < 0)
5898                 RETURN(rc);
5899
5900         if (unlikely(rc == 0))
5901                 RETURN(-ENODATA);
5902
5903         if (buf->lb_len < rc) {
5904                 lu_buf_realloc(buf, rc);
5905                 if (buf->lb_buf == NULL)
5906                         RETURN(-ENOMEM);
5907
5908                 goto again;
5909         }
5910
5911         ldata.ld_buf = buf;
5912         rc = linkea_init_with_rec(&ldata);
5913         if (!rc)
5914                 rc = linkea_links_find(&ldata, &cname, pfid);
5915
5916         RETURN(rc);
5917 }
5918
5919 /**
5920  * Calls ->lookup() to find dentry. From dentry get inode and
5921  * read inode's ea to get fid. This is required for  interoperability
5922  * mode (b11826)
5923  *
5924  * \retval   0, on success
5925  * \retval -ve, on error
5926  */
5927 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5928                              struct dt_rec *rec, const struct dt_key *key)
5929 {
5930         struct inode *dir = obj->oo_inode;
5931         struct dentry *dentry;
5932         struct ldiskfs_dir_entry_2 *de;
5933         struct buffer_head *bh;
5934         struct lu_fid *fid = (struct lu_fid *)rec;
5935         struct htree_lock *hlock = NULL;
5936         int ino;
5937         int rc;
5938
5939         ENTRY;
5940
5941         LASSERT(dir->i_op != NULL);
5942         LASSERT(dir->i_op->lookup != NULL);
5943
5944         dentry = osd_child_dentry_get(env, obj,
5945                                       (char *)key, strlen((char *)key));
5946
5947         if (obj->oo_hl_head != NULL) {
5948                 hlock = osd_oti_get(env)->oti_hlock;
5949                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5950                                    dir, LDISKFS_HLOCK_LOOKUP);
5951         } else {
5952                 down_read(&obj->oo_ext_idx_sem);
5953         }
5954
5955         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5956         if (!IS_ERR(bh)) {
5957                 struct osd_thread_info *oti = osd_oti_get(env);
5958                 struct osd_inode_id *id = &oti->oti_id;
5959                 struct osd_idmap_cache *oic = &oti->oti_cache;
5960                 struct osd_device *dev = osd_obj2dev(obj);
5961
5962                 ino = le32_to_cpu(de->inode);
5963                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5964                         brelse(bh);
5965                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5966                         GOTO(out, rc);
5967                 }
5968
5969                 rc = osd_get_fid_from_dentry(de, rec);
5970
5971                 /* done with de, release bh */
5972                 brelse(bh);
5973                 if (rc != 0) {
5974                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5975                                 const char *name = (const char *)key;
5976
5977                                 /*
5978                                  * If the parent is on remote MDT, and there
5979                                  * is no FID-in-dirent, then we have to get
5980                                  * the parent FID from the linkEA.
5981                                  */
5982                                 if (likely(strlen(name) == 2 &&
5983                                            name[0] == '.' && name[1] == '.'))
5984                                         rc = osd_get_pfid_from_linkea(env, obj,
5985                                                                       fid);
5986                         } else {
5987                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5988                         }
5989                 } else {
5990                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5991                 }
5992
5993                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5994                         fid_zero(&oic->oic_fid);
5995
5996                         GOTO(out, rc);
5997                 }
5998
5999                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
6000                 rc = osd_consistency_check(oti, dev, oic);
6001                 if (rc == -ENOENT)
6002                         fid_zero(&oic->oic_fid);
6003                 else
6004                         /* Other error should not affect lookup result. */
6005                         rc = 0;
6006         } else {
6007                 rc = PTR_ERR(bh);
6008         }
6009
6010         GOTO(out, rc);
6011
6012 out:
6013         if (hlock != NULL)
6014                 ldiskfs_htree_unlock(hlock);
6015         else
6016                 up_read(&obj->oo_ext_idx_sem);
6017         return rc;
6018 }
6019
6020 static int osd_index_declare_ea_insert(const struct lu_env *env,
6021                                        struct dt_object *dt,
6022                                        const struct dt_rec *rec,
6023                                        const struct dt_key *key,
6024                                        struct thandle *handle)
6025 {
6026         struct osd_thandle *oh;
6027         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6028         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6029         const struct lu_fid *fid = rec1->rec_fid;
6030         int credits, rc = 0;
6031         struct osd_idmap_cache *idc;
6032
6033         ENTRY;
6034
6035         LASSERT(!dt_object_remote(dt));
6036         LASSERT(handle != NULL);
6037         LASSERT(fid != NULL);
6038         LASSERT(rec1->rec_type != 0);
6039
6040         oh = container_of(handle, struct osd_thandle, ot_super);
6041         LASSERT(oh->ot_handle == NULL);
6042
6043         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6044
6045         /*
6046          * we can't call iget() while a transactions is running
6047          * (this can lead to a deadlock), but we need to know
6048          * inum and object type. so we find this information at
6049          * declaration and cache in per-thread info
6050          */
6051         idc = osd_idc_find_or_init(env, osd, fid);
6052         if (IS_ERR(idc))
6053                 RETURN(PTR_ERR(idc));
6054         if (idc->oic_remote) {
6055                 /*
6056                  * a reference to remote inode is represented by an
6057                  * agent inode which we have to create
6058                  */
6059                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6060                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6061         }
6062
6063         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6064
6065         if (osd_dt_obj(dt)->oo_inode != NULL) {
6066                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6067
6068                 /*
6069                  * We ignore block quota on meta pool (MDTs), so needn't
6070                  * calculate how many blocks will be consumed by this index
6071                  * insert
6072                  */
6073                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6074                                            i_gid_read(inode),
6075                                            i_projid_read(inode), 0,
6076                                            oh, osd_dt_obj(dt), NULL,
6077                                            OSD_QID_BLK);
6078                 if (rc)
6079                         RETURN(rc);
6080
6081 #ifdef HAVE_PROJECT_QUOTA
6082                 /*
6083                  * Reserve credits for local agent inode to transfer
6084                  * to 0, quota enforcement is ignored in this case.
6085                  */
6086                 if (idc->oic_remote &&
6087                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6088                     i_projid_read(inode) != 0)
6089                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6090                                                   0, i_projid_read(inode),
6091                                                   0, false, PRJQUOTA, true);
6092 #endif
6093         }
6094
6095         RETURN(rc);
6096 }
6097
6098 /**
6099  * Index add function for interoperability mode (b11826).
6100  * It will add the directory entry.This entry is needed to
6101  * maintain name->fid mapping.
6102  *
6103  * \param key it is key i.e. file entry to be inserted
6104  * \param rec it is value of given key i.e. fid
6105  *
6106  * \retval   0, on success
6107  * \retval -ve, on error
6108  */
6109 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6110                                const struct dt_rec *rec,
6111                                const struct dt_key *key, struct thandle *th)
6112 {
6113         struct osd_object *obj = osd_dt_obj(dt);
6114         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6115         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6116         const struct lu_fid *fid = rec1->rec_fid;
6117         const char *name = (const char *)key;
6118         struct osd_thread_info *oti = osd_oti_get(env);
6119         struct inode *child_inode = NULL;
6120         struct osd_idmap_cache *idc;
6121         int rc;
6122
6123         ENTRY;
6124
6125         if (!dt_object_exists(dt))
6126                 RETURN(-ENOENT);
6127
6128         LASSERT(osd_invariant(obj));
6129         LASSERT(!dt_object_remote(dt));
6130         LASSERT(th != NULL);
6131
6132         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6133
6134         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6135
6136         idc = osd_idc_find(env, osd, fid);
6137         if (unlikely(idc == NULL)) {
6138                 idc = osd_idc_find_or_init(env, osd, fid);
6139                 if (IS_ERR(idc)) {
6140                         /*
6141                          * this dt_insert() wasn't declared properly, so
6142                          * FID is missing in OI cache. we better do not
6143                          * lookup FID in FLDB/OI and don't risk to deadlock,
6144                          * but in some special cases (lfsck testing, etc)
6145                          * it's much simpler than fixing a caller.
6146                          *
6147                          * normally this error should be placed after the first
6148                          * find, but migrate may attach source stripes to
6149                          * target, which doesn't create stripes.
6150                          */
6151                         CERROR("%s: "DFID" wasn't declared for insert\n",
6152                                osd_name(osd), PFID(fid));
6153                         dump_stack();
6154                         RETURN(PTR_ERR(idc));
6155                 }
6156         }
6157
6158         if (idc->oic_remote) {
6159                 /* Insert remote entry */
6160                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6161                         child_inode =
6162                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6163                 } else {
6164                         child_inode = osd_create_local_agent_inode(env, osd,
6165                                         obj, fid, rec1->rec_type & S_IFMT, th);
6166                         if (IS_ERR(child_inode))
6167                                 RETURN(PTR_ERR(child_inode));
6168                 }
6169         } else {
6170                 /* Insert local entry */
6171                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6172                         /* for a reason OI cache wasn't filled properly */
6173                         CERROR("%s: OIC for "DFID" isn't filled\n",
6174                                osd_name(osd), PFID(fid));
6175                         RETURN(-EINVAL);
6176                 }
6177                 child_inode = oti->oti_inode;
6178                 if (unlikely(child_inode == NULL)) {
6179                         struct ldiskfs_inode_info *lii;
6180
6181                         OBD_ALLOC_PTR(lii);
6182                         if (lii == NULL)
6183                                 RETURN(-ENOMEM);
6184                         child_inode = oti->oti_inode = &lii->vfs_inode;
6185                 }
6186                 child_inode->i_sb = osd_sb(osd);
6187                 child_inode->i_ino = idc->oic_lid.oii_ino;
6188                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6189         }
6190
6191         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6192
6193         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6194                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6195
6196         if (child_inode && child_inode != oti->oti_inode)
6197                 iput(child_inode);
6198         LASSERT(osd_invariant(obj));
6199         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6200
6201         RETURN(rc);
6202 }
6203
6204 /**
6205  *  Initialize osd Iterator for given osd index object.
6206  *
6207  *  \param  dt      osd index object
6208  */
6209
6210 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6211                                      struct dt_object *dt,
6212                                      __u32 unused)
6213 {
6214         struct osd_it_iam *it;
6215         struct osd_object *obj = osd_dt_obj(dt);
6216         struct lu_object *lo = &dt->do_lu;
6217         struct iam_path_descr *ipd;
6218         struct iam_container *bag = &obj->oo_dir->od_container;
6219
6220         if (!dt_object_exists(dt))
6221                 return ERR_PTR(-ENOENT);
6222
6223         OBD_ALLOC_PTR(it);
6224         if (it == NULL)
6225                 return ERR_PTR(-ENOMEM);
6226
6227         ipd = osd_it_ipd_get(env, bag);
6228         if (likely(ipd != NULL)) {
6229                 it->oi_obj = obj;
6230                 it->oi_ipd = ipd;
6231                 lu_object_get(lo);
6232                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6233                 return (struct dt_it *)it;
6234         } else {
6235                 OBD_FREE_PTR(it);
6236                 return ERR_PTR(-ENOMEM);
6237         }
6238 }
6239
6240 /**
6241  * free given Iterator.
6242  */
6243 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6244 {
6245         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6246         struct osd_object *obj = it->oi_obj;
6247
6248         iam_it_fini(&it->oi_it);
6249         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6250         osd_object_put(env, obj);
6251         OBD_FREE_PTR(it);
6252 }
6253
6254 /**
6255  *  Move Iterator to record specified by \a key
6256  *
6257  *  \param  di      osd iterator
6258  *  \param  key     key for index
6259  *
6260  *  \retval +ve  di points to record with least key not larger than key
6261  *  \retval  0   di points to exact matched key
6262  *  \retval -ve  failure
6263  */
6264
6265 static int osd_it_iam_get(const struct lu_env *env,
6266                           struct dt_it *di, const struct dt_key *key)
6267 {
6268         struct osd_thread_info *oti = osd_oti_get(env);
6269         struct osd_it_iam *it = (struct osd_it_iam *)di;
6270
6271         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6272                 /* swab quota uid/gid */
6273                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6274                 key = (struct dt_key *)&oti->oti_quota_id;
6275         }
6276
6277         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6278 }
6279
6280 /**
6281  *  Release Iterator
6282  *
6283  *  \param  di      osd iterator
6284  */
6285 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6286 {
6287         struct osd_it_iam *it = (struct osd_it_iam *)di;
6288
6289         iam_it_put(&it->oi_it);
6290 }
6291
6292 /**
6293  *  Move iterator by one record
6294  *
6295  *  \param  di      osd iterator
6296  *
6297  *  \retval +1   end of container reached
6298  *  \retval  0   success
6299  *  \retval -ve  failure
6300  */
6301
6302 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6303 {
6304         struct osd_it_iam *it = (struct osd_it_iam *)di;
6305
6306         return iam_it_next(&it->oi_it);
6307 }
6308
6309 /**
6310  * Return pointer to the key under iterator.
6311  */
6312
6313 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6314                                      const struct dt_it *di)
6315 {
6316         struct osd_thread_info *oti = osd_oti_get(env);
6317         struct osd_it_iam *it = (struct osd_it_iam *)di;
6318         struct osd_object *obj = it->oi_obj;
6319         struct dt_key *key;
6320
6321         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6322
6323         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6324                 /* swab quota uid/gid */
6325                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6326                 key = (struct dt_key *)&oti->oti_quota_id;
6327         }
6328
6329         return key;
6330 }
6331
6332 /**
6333  * Return size of key under iterator (in bytes)
6334  */
6335
6336 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6337 {
6338         struct osd_it_iam *it = (struct osd_it_iam *)di;
6339
6340         return iam_it_key_size(&it->oi_it);
6341 }
6342
6343 static inline void
6344 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6345 {
6346         /* check if file type is required */
6347         if (ent->lde_attrs & LUDA_TYPE) {
6348                 struct luda_type *lt;
6349                 int align = sizeof(*lt) - 1;
6350
6351                 len = (len + align) & ~align;
6352                 lt = (struct luda_type *)(ent->lde_name + len);
6353                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6354         }
6355
6356         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6357 }
6358
6359 /**
6360  * build lu direct from backend fs dirent.
6361  */
6362
6363 static inline void
6364 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6365                    char *name, __u16 namelen, __u16 type, __u32 attr)
6366 {
6367         ent->lde_attrs = attr | LUDA_FID;
6368         fid_cpu_to_le(&ent->lde_fid, fid);
6369
6370         ent->lde_hash = cpu_to_le64(offset);
6371         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6372
6373         strncpy(ent->lde_name, name, namelen);
6374         ent->lde_name[namelen] = '\0';
6375         ent->lde_namelen = cpu_to_le16(namelen);
6376
6377         /* append lustre attributes */
6378         osd_it_append_attrs(ent, namelen, type);
6379 }
6380
6381 /**
6382  * Return pointer to the record under iterator.
6383  */
6384 static int osd_it_iam_rec(const struct lu_env *env,
6385                           const struct dt_it *di,
6386                           struct dt_rec *dtrec, __u32 attr)
6387 {
6388         struct osd_it_iam *it = (struct osd_it_iam *)di;
6389         struct osd_thread_info *info = osd_oti_get(env);
6390
6391         ENTRY;
6392
6393         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6394                 const struct osd_fid_pack *rec;
6395                 struct lu_fid *fid = &info->oti_fid;
6396                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6397                 char *name;
6398                 int namelen;
6399                 __u64 hash;
6400                 int rc;
6401
6402                 name = (char *)iam_it_key_get(&it->oi_it);
6403                 if (IS_ERR(name))
6404                         RETURN(PTR_ERR(name));
6405
6406                 namelen = iam_it_key_size(&it->oi_it);
6407
6408                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6409                 if (IS_ERR(rec))
6410                         RETURN(PTR_ERR(rec));
6411
6412                 rc = osd_fid_unpack(fid, rec);
6413                 if (rc)
6414                         RETURN(rc);
6415
6416                 hash = iam_it_store(&it->oi_it);
6417
6418                 /* IAM does not store object type in IAM index (dir) */
6419                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6420                                    0, LUDA_FID);
6421         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6422                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6423                            (struct iam_rec *)dtrec);
6424                 osd_quota_unpack(it->oi_obj, dtrec);
6425         } else {
6426                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6427                            (struct iam_rec *)dtrec);
6428         }
6429
6430         RETURN(0);
6431 }
6432
6433 /**
6434  * Returns cookie for current Iterator position.
6435  */
6436 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6437 {
6438         struct osd_it_iam *it = (struct osd_it_iam *)di;
6439
6440         return iam_it_store(&it->oi_it);
6441 }
6442
6443 /**
6444  * Restore iterator from cookie.
6445  *
6446  * \param  di      osd iterator
6447  * \param  hash    Iterator location cookie
6448  *
6449  * \retval +ve  di points to record with least key not larger than key.
6450  * \retval  0   di points to exact matched key
6451  * \retval -ve  failure
6452  */
6453
6454 static int osd_it_iam_load(const struct lu_env *env,
6455                            const struct dt_it *di, __u64 hash)
6456 {
6457         struct osd_it_iam *it = (struct osd_it_iam *)di;
6458
6459         return iam_it_load(&it->oi_it, hash);
6460 }
6461
6462 static const struct dt_index_operations osd_index_iam_ops = {
6463         .dio_lookup         = osd_index_iam_lookup,
6464         .dio_declare_insert = osd_index_declare_iam_insert,
6465         .dio_insert         = osd_index_iam_insert,
6466         .dio_declare_delete = osd_index_declare_iam_delete,
6467         .dio_delete         = osd_index_iam_delete,
6468         .dio_it     = {
6469                 .init     = osd_it_iam_init,
6470                 .fini     = osd_it_iam_fini,
6471                 .get      = osd_it_iam_get,
6472                 .put      = osd_it_iam_put,
6473                 .next     = osd_it_iam_next,
6474                 .key      = osd_it_iam_key,
6475                 .key_size = osd_it_iam_key_size,
6476                 .rec      = osd_it_iam_rec,
6477                 .store    = osd_it_iam_store,
6478                 .load     = osd_it_iam_load
6479         }
6480 };
6481
6482
6483 /**
6484  * Creates or initializes iterator context.
6485  *
6486  * \retval struct osd_it_ea, iterator structure on success
6487  *
6488  */
6489 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6490                                     struct dt_object *dt,
6491                                     __u32 attr)
6492 {
6493         struct osd_object *obj = osd_dt_obj(dt);
6494         struct osd_thread_info *info = osd_oti_get(env);
6495         struct osd_it_ea *oie;
6496         struct file *file;
6497         struct lu_object *lo = &dt->do_lu;
6498         struct dentry *obj_dentry;
6499
6500         ENTRY;
6501
6502         if (!dt_object_exists(dt) || obj->oo_destroyed)
6503                 RETURN(ERR_PTR(-ENOENT));
6504
6505         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6506         if (oie == NULL)
6507                 RETURN(ERR_PTR(-ENOMEM));
6508         obj_dentry = &oie->oie_dentry;
6509
6510         obj_dentry->d_inode = obj->oo_inode;
6511         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6512         obj_dentry->d_name.hash = 0;
6513
6514         oie->oie_rd_dirent       = 0;
6515         oie->oie_it_dirent       = 0;
6516         oie->oie_dirent          = NULL;
6517         if (unlikely(!info->oti_it_ea_buf_used)) {
6518                 oie->oie_buf = info->oti_it_ea_buf;
6519                 info->oti_it_ea_buf_used = 1;
6520         } else {
6521                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6522                 if (oie->oie_buf == NULL)
6523                         RETURN(ERR_PTR(-ENOMEM));
6524         }
6525         oie->oie_obj = obj;
6526
6527         file = &oie->oie_file;
6528
6529         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6530         if (attr & LUDA_64BITHASH)
6531                 file->f_mode    = FMODE_64BITHASH;
6532         else
6533                 file->f_mode    = FMODE_32BITHASH;
6534         file->f_path.dentry     = obj_dentry;
6535         file->f_mapping         = obj->oo_inode->i_mapping;
6536         file->f_op              = obj->oo_inode->i_fop;
6537         file->f_inode = obj->oo_inode;
6538
6539         lu_object_get(lo);
6540         RETURN((struct dt_it *)oie);
6541 }
6542
6543 /**
6544  * Destroy or finishes iterator context.
6545  *
6546  * \param di iterator structure to be destroyed
6547  */
6548 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6549 {
6550         struct osd_thread_info *info = osd_oti_get(env);
6551         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6552         struct osd_object *obj = oie->oie_obj;
6553         struct inode *inode = obj->oo_inode;
6554
6555         ENTRY;
6556         oie->oie_file.f_op->release(inode, &oie->oie_file);
6557         osd_object_put(env, obj);
6558         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6559                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6560         else
6561                 info->oti_it_ea_buf_used = 0;
6562         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6563         EXIT;
6564 }
6565
6566 /**
6567  * It position the iterator at given key, so that next lookup continues from
6568  * that key Or it is similar to dio_it->load() but based on a key,
6569  * rather than file position.
6570  *
6571  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6572  * to the beginning.
6573  *
6574  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6575  */
6576 static int osd_it_ea_get(const struct lu_env *env,
6577                          struct dt_it *di, const struct dt_key *key)
6578 {
6579         struct osd_it_ea *it = (struct osd_it_ea *)di;
6580
6581         ENTRY;
6582         LASSERT(((const char *)key)[0] == '\0');
6583         it->oie_file.f_pos = 0;
6584         it->oie_rd_dirent = 0;
6585         it->oie_it_dirent = 0;
6586         it->oie_dirent = NULL;
6587
6588         RETURN(+1);
6589 }
6590
6591 /**
6592  * Does nothing
6593  */
6594 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6595 {
6596 }
6597
6598 struct osd_filldir_cbs {
6599         struct dir_context ctx;
6600         struct osd_it_ea  *it;
6601 };
6602 /**
6603  * It is called internally by ->iterate*(). It fills the
6604  * iterator's in-memory data structure with required
6605  * information i.e. name, namelen, rec_size etc.
6606  *
6607  * \param buf in which information to be filled in.
6608  * \param name name of the file in given dir
6609  *
6610  * \retval 0 on success
6611  * \retval 1 on buffer full
6612  */
6613 #ifdef HAVE_FILLDIR_USE_CTX
6614 static int osd_ldiskfs_filldir(struct dir_context *buf,
6615 #else
6616 static int osd_ldiskfs_filldir(void *buf,
6617 #endif
6618                                const char *name, int namelen,
6619                                loff_t offset, __u64 ino, unsigned int d_type)
6620 {
6621         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6622         struct osd_object *obj = it->oie_obj;
6623         struct osd_it_ea_dirent *ent = it->oie_dirent;
6624         struct lu_fid *fid = &ent->oied_fid;
6625         struct osd_fid_pack *rec;
6626
6627         ENTRY;
6628
6629 /* this should never happen */
6630         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6631                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6632                 RETURN(-EIO);
6633         }
6634
6635         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6636             OSD_IT_EA_BUFSIZE)
6637                 RETURN(1);
6638
6639         /* "." is just the object itself. */
6640         if (namelen == 1 && name[0] == '.') {
6641                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6642         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6643                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6644                 if (osd_fid_unpack(fid, rec) != 0)
6645                         fid_zero(fid);
6646         } else {
6647                 fid_zero(fid);
6648         }
6649         d_type &= ~LDISKFS_DIRENT_LUFID;
6650
6651         /* NOT export local root. */
6652         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6653                 ino = obj->oo_inode->i_ino;
6654                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6655         }
6656
6657         ent->oied_ino     = ino;
6658         ent->oied_off     = offset;
6659         ent->oied_namelen = namelen;
6660         ent->oied_type    = d_type;
6661
6662         memcpy(ent->oied_name, name, namelen);
6663
6664         it->oie_rd_dirent++;
6665         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6666         RETURN(0);
6667 }
6668
6669 /**
6670  * Calls ->iterate*() to load a directory entry at a time
6671  * and stored it in iterator's in-memory data structure.
6672  *
6673  * \param di iterator's in memory structure
6674  *
6675  * \retval   0 on success
6676  * \retval -ve on error
6677  * \retval +1 reach the end of entry
6678  */
6679 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6680                                const struct dt_it *di)
6681 {
6682         struct osd_it_ea *it = (struct osd_it_ea *)di;
6683         struct osd_object *obj = it->oie_obj;
6684         struct inode *inode = obj->oo_inode;
6685         struct htree_lock *hlock = NULL;
6686         struct file *filp = &it->oie_file;
6687         int rc = 0;
6688         struct osd_filldir_cbs buf = {
6689                 .ctx.actor = osd_ldiskfs_filldir,
6690                 .it = it
6691         };
6692
6693         ENTRY;
6694         it->oie_dirent = it->oie_buf;
6695         it->oie_rd_dirent = 0;
6696
6697         if (obj->oo_hl_head != NULL) {
6698                 hlock = osd_oti_get(env)->oti_hlock;
6699                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6700                                    inode, LDISKFS_HLOCK_READDIR);
6701         } else {
6702                 down_read(&obj->oo_ext_idx_sem);
6703         }
6704
6705         filp->f_cred = current_cred();
6706         rc = osd_security_file_alloc(filp);
6707         if (rc)
6708                 RETURN(rc);
6709
6710         filp->f_flags |= O_NOATIME;
6711         filp->f_mode |= FMODE_NONOTIFY;
6712         rc = iterate_dir(filp, &buf.ctx);
6713         if (rc)
6714                 RETURN(rc);
6715
6716         if (hlock != NULL)
6717                 ldiskfs_htree_unlock(hlock);
6718         else
6719                 up_read(&obj->oo_ext_idx_sem);
6720
6721         if (it->oie_rd_dirent == 0) {
6722                 /*
6723                  * If it does not get any dirent, it means it has been reached
6724                  * to the end of the dir
6725                  */
6726                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6727                 if (rc == 0)
6728                         rc = 1;
6729         } else {
6730                 it->oie_dirent = it->oie_buf;
6731                 it->oie_it_dirent = 1;
6732         }
6733
6734         RETURN(rc);
6735 }
6736
6737 /**
6738  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6739  * to load a directory entry at a time and stored it in
6740  * iterator's in-memory data structure.
6741  *
6742  * \param di iterator's in memory structure
6743  *
6744  * \retval +ve iterator reached to end
6745  * \retval   0 iterator not reached to end
6746  * \retval -ve on error
6747  */
6748 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6749 {
6750         struct osd_it_ea *it = (struct osd_it_ea *)di;
6751         int rc;
6752
6753         ENTRY;
6754
6755         if (it->oie_it_dirent < it->oie_rd_dirent) {
6756                 it->oie_dirent =
6757                         (void *)it->oie_dirent +
6758                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6759                                        it->oie_dirent->oied_namelen);
6760                 it->oie_it_dirent++;
6761                 rc = 0;
6762         } else {
6763                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6764                         rc = 1;
6765                 else
6766                         rc = osd_ldiskfs_it_fill(env, di);
6767         }
6768
6769         RETURN(rc);
6770 }
6771
6772 /**
6773  * Returns the key at current position from iterator's in memory structure.
6774  *
6775  * \param di iterator's in memory structure
6776  *
6777  * \retval key i.e. struct dt_key on success
6778  */
6779 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6780                                     const struct dt_it *di)
6781 {
6782         struct osd_it_ea *it = (struct osd_it_ea *)di;
6783
6784         return (struct dt_key *)it->oie_dirent->oied_name;
6785 }
6786
6787 /**
6788  * Returns key's size at current position from iterator's in memory structure.
6789  *
6790  * \param di iterator's in memory structure
6791  *
6792  * \retval key_size i.e. struct dt_key on success
6793  */
6794 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6795 {
6796         struct osd_it_ea *it = (struct osd_it_ea *)di;
6797
6798         return it->oie_dirent->oied_namelen;
6799 }
6800
6801 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6802 #undef LDISKFS_DIR_REC_LEN
6803 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6804 #endif
6805
6806 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6807 {
6808         if (LDISKFS_DIR_REC_LEN(de) >=
6809             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6810                 return true;
6811
6812         return false;
6813 }
6814
6815 static inline bool
6816 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6817                      unsigned int blocksize, bool dotdot)
6818 {
6819         if (dotdot)
6820                 return osd_dotdot_has_space(de);
6821
6822         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6823             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6824                 return true;
6825
6826         return false;
6827 }
6828
6829 static int
6830 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6831                     handle_t *jh, struct dentry *dentry,
6832                     const struct lu_fid *fid, struct buffer_head *bh,
6833                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6834                     bool dotdot)
6835 {
6836         struct inode *dir = dentry->d_parent->d_inode;
6837         struct inode *inode = dentry->d_inode;
6838         struct osd_fid_pack *rec;
6839         struct ldiskfs_dentry_param *ldp;
6840         int namelen = dentry->d_name.len;
6841         int rc;
6842         struct osd_thread_info *info = osd_oti_get(env);
6843
6844         ENTRY;
6845
6846         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6847                 RETURN(0);
6848
6849         /* There is enough space to hold the FID-in-dirent. */
6850         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6851                 rc = ldiskfs_journal_get_write_access(jh, bh);
6852                 if (rc != 0)
6853                         RETURN(rc);
6854
6855                 de->name[namelen] = 0;
6856                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6857                 rec->fp_len = sizeof(struct lu_fid) + 1;
6858                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6859                 de->file_type |= LDISKFS_DIRENT_LUFID;
6860                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6861
6862                 RETURN(rc);
6863         }
6864
6865         LASSERT(!dotdot);
6866
6867         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6868         if (rc != 0)
6869                 RETURN(rc);
6870
6871         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6872         osd_get_ldiskfs_dirent_param(ldp, fid);
6873         dentry->d_fsdata = (void *)ldp;
6874         dquot_initialize(dir);
6875         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6876         /*
6877          * It is too bad, we cannot reinsert the name entry back.
6878          * That means we lose it!
6879          */
6880         if (rc != 0)
6881                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6882                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6883                        osd_ino2name(inode),
6884                        dir->i_ino, dir->i_generation, namelen,
6885                        dentry->d_name.name, PFID(fid), rc);
6886
6887         RETURN(rc);
6888 }
6889
6890 static int
6891 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6892                         struct osd_it_ea *it, struct lu_fid *fid,
6893                         struct osd_inode_id *id, __u32 *attr)
6894 {
6895         struct osd_thread_info *info = osd_oti_get(env);
6896         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6897         struct osd_device *dev = osd_obj2dev(obj);
6898         struct super_block *sb = osd_sb(dev);
6899         const char *devname = osd_name(dev);
6900         struct osd_it_ea_dirent *ent = it->oie_dirent;
6901         struct inode *dir = obj->oo_inode;
6902         struct htree_lock *hlock = NULL;
6903         struct buffer_head *bh = NULL;
6904         handle_t *jh = NULL;
6905         struct ldiskfs_dir_entry_2 *de;
6906         struct dentry *dentry;
6907         struct inode *inode;
6908         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6909         int credits;
6910         int rc;
6911         bool dotdot = false;
6912         bool dirty = false;
6913
6914         ENTRY;
6915
6916         if (ent->oied_name[0] == '.') {
6917                 if (ent->oied_namelen == 1)
6918                         RETURN(0);
6919
6920                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6921                         dotdot = true;
6922         }
6923
6924         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6925         inode = osd_iget(info, dev, id);
6926         if (IS_ERR(inode)) {
6927                 rc = PTR_ERR(inode);
6928                 if (rc == -ENOENT || rc == -ESTALE) {
6929                         /*
6930                          * Maybe dangling name entry, or
6931                          * corrupted directory entry.
6932                          */
6933                         *attr |= LUDA_UNKNOWN;
6934                         rc = 0;
6935                 } else {
6936                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6937                                "check_repair, dir = %lu/%u, name = %.*s, "
6938                                "ino = %llu, rc = %d\n",
6939                                devname, dir->i_ino, dir->i_generation,
6940                                ent->oied_namelen, ent->oied_name,
6941                                ent->oied_ino, rc);
6942                 }
6943
6944                 RETURN(rc);
6945         }
6946
6947         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6948                                            ent->oied_namelen);
6949         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6950         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6951                 lma = NULL;
6952         else if (rc != 0)
6953                 GOTO(out, rc);
6954
6955         /*
6956          * We need to ensure that the name entry is still valid.
6957          * Because it may be removed or renamed by other already.
6958          *
6959          * The unlink or rename operation will start journal before PDO lock,
6960          * so to avoid deadlock, here we need to start journal handle before
6961          * related PDO lock also. But because we do not know whether there
6962          * will be something to be repaired before PDO lock, we just start
6963          * journal without conditions.
6964          *
6965          * We may need to remove the name entry firstly, then insert back.
6966          * One credit is for user quota file update.
6967          * One credit is for group quota file update.
6968          * Two credits are for dirty inode.
6969          */
6970         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6971                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6972
6973         if (dev->od_dirent_journal != 0) {
6974
6975 again:
6976                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6977                 if (IS_ERR(jh)) {
6978                         rc = PTR_ERR(jh);
6979                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6980                                "check_repair, dir = %lu/%u, credits = %d, "
6981                                "name = %.*s, ino = %llu: rc = %d\n",
6982                                devname, dir->i_ino, dir->i_generation, credits,
6983                                ent->oied_namelen, ent->oied_name,
6984                                ent->oied_ino, rc);
6985
6986                         GOTO(out_inode, rc);
6987                 }
6988
6989                 if (obj->oo_hl_head != NULL) {
6990                         hlock = osd_oti_get(env)->oti_hlock;
6991                         /*
6992                          * "0" means exclusive lock for the whole directory.
6993                          * We need to prevent others access such name entry
6994                          * during the delete + insert. Neither HLOCK_ADD nor
6995                          * HLOCK_DEL cannot guarantee the atomicity.
6996                          */
6997                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6998                 } else {
6999                         down_write(&obj->oo_ext_idx_sem);
7000                 }
7001         } else {
7002                 if (obj->oo_hl_head != NULL) {
7003                         hlock = osd_oti_get(env)->oti_hlock;
7004                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7005                                            LDISKFS_HLOCK_LOOKUP);
7006                 } else {
7007                         down_read(&obj->oo_ext_idx_sem);
7008                 }
7009         }
7010
7011         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7012         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7013                 *attr |= LUDA_IGNORE;
7014
7015                 GOTO(out, rc = 0);
7016         }
7017
7018         /*
7019          * For dotdot entry, if there is not enough space to hold the
7020          * FID-in-dirent, just keep them there. It only happens when the
7021          * device upgraded from 1.8 or restored from MDT file-level backup.
7022          * For the whole directory, only dotdot entry have no FID-in-dirent
7023          * and needs to get FID from LMA when readdir, it will not affect the
7024          * performance much.
7025          */
7026         if (dotdot && !osd_dotdot_has_space(de)) {
7027                 *attr |= LUDA_UNKNOWN;
7028
7029                 GOTO(out, rc = 0);
7030         }
7031
7032         if (lma != NULL) {
7033                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7034                         GOTO(out, rc = 0);
7035
7036                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7037                         struct lu_fid *tfid = &lma->lma_self_fid;
7038
7039                         if (likely(dotdot &&
7040                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7041                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7042                                 /*
7043                                  * It must be REMOTE_PARENT_DIR and as the
7044                                  * 'dotdot' entry of remote directory
7045                                  */
7046                                 *attr |= LUDA_IGNORE;
7047                         } else {
7048                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7049                                        "parent directory, but got %.*s under "
7050                                        "dir = %lu/%u with the FID "DFID"\n",
7051                                        devname, ent->oied_namelen,
7052                                        ent->oied_name, dir->i_ino,
7053                                        dir->i_generation, PFID(tfid));
7054
7055                                 *attr |= LUDA_UNKNOWN;
7056                         }
7057
7058                         GOTO(out, rc = 0);
7059                 }
7060         }
7061
7062         if (!fid_is_zero(fid)) {
7063                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7064                                               ent->oied_namelen);
7065                 if (rc == -ENOENT ||
7066                     (rc == -ENODATA &&
7067                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7068                         /*
7069                          * linkEA does not recognize the dirent entry,
7070                          * it may because the dirent entry corruption
7071                          * and points to other's inode.
7072                          */
7073                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7074                                "recognize the dirent, dir = %lu/%u, "
7075                                " name = %.*s, ino = %llu, "
7076                                DFID": rc = %d\n", devname, dir->i_ino,
7077                                dir->i_generation, ent->oied_namelen,
7078                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7079                         *attr |= LUDA_UNKNOWN;
7080
7081                         GOTO(out, rc = 0);
7082                 }
7083
7084                 if (rc && rc != -ENODATA) {
7085                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7086                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7087                                DFID": rc = %d\n", devname, dir->i_ino,
7088                                dir->i_generation, ent->oied_namelen,
7089                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7090                         *attr |= LUDA_UNKNOWN;
7091
7092                         GOTO(out, rc = 0);
7093                 }
7094         }
7095
7096         if (lma != NULL) {
7097                 /*
7098                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7099                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7100                  */
7101                 if (*attr & LUDA_VERIFY_DRYRUN) {
7102                         *fid = lma->lma_self_fid;
7103                         *attr |= LUDA_REPAIR;
7104
7105                         GOTO(out, rc = 0);
7106                 }
7107
7108                 if (jh == NULL) {
7109                         brelse(bh);
7110                         dev->od_dirent_journal = 1;
7111                         if (hlock != NULL) {
7112                                 ldiskfs_htree_unlock(hlock);
7113                                 hlock = NULL;
7114                         } else {
7115                                 up_read(&obj->oo_ext_idx_sem);
7116                         }
7117
7118                         goto again;
7119                 }
7120
7121                 *fid = lma->lma_self_fid;
7122                 dirty = true;
7123                 /* Update or append the FID-in-dirent. */
7124                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7125                                          bh, de, hlock, dotdot);
7126                 if (rc == 0)
7127                         *attr |= LUDA_REPAIR;
7128                 else
7129                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7130                                "the dirent, dir = %lu/%u, name = %.*s, "
7131                                "ino = %llu, "DFID": rc = %d\n",
7132                                devname, dir->i_ino, dir->i_generation,
7133                                ent->oied_namelen, ent->oied_name,
7134                                ent->oied_ino, PFID(fid), rc);
7135         } else {
7136                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7137                 if (*attr & LUDA_VERIFY_DRYRUN) {
7138                         if (fid_is_sane(fid)) {
7139                                 *attr |= LUDA_REPAIR;
7140                         } else if (dev->od_index == 0) {
7141                                 lu_igif_build(fid, inode->i_ino,
7142                                               inode->i_generation);
7143                                 *attr |= LUDA_UPGRADE;
7144                         }
7145
7146                         GOTO(out, rc = 0);
7147                 }
7148
7149                 if (jh == NULL) {
7150                         brelse(bh);
7151                         dev->od_dirent_journal = 1;
7152                         if (hlock != NULL) {
7153                                 ldiskfs_htree_unlock(hlock);
7154                                 hlock = NULL;
7155                         } else {
7156                                 up_read(&obj->oo_ext_idx_sem);
7157                         }
7158
7159                         goto again;
7160                 }
7161
7162                 dirty = true;
7163                 if (unlikely(fid_is_sane(fid))) {
7164                         /*
7165                          * FID-in-dirent exists, but FID-in-LMA is lost.
7166                          * Trust the FID-in-dirent, and add FID-in-LMA.
7167                          */
7168                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7169                         if (rc == 0)
7170                                 *attr |= LUDA_REPAIR;
7171                         else
7172                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7173                                        "update dirent, dir = %lu/%u, "
7174                                        "name = %.*s, ino = %llu, "
7175                                        DFID": rc = %d\n",
7176                                        devname, dir->i_ino, dir->i_generation,
7177                                        ent->oied_namelen, ent->oied_name,
7178                                        ent->oied_ino, PFID(fid), rc);
7179                 } else if (dev->od_index == 0) {
7180                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7181                         /*
7182                          * It is probably IGIF object. Only aappend the
7183                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7184                          */
7185                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7186                                                  bh, de, hlock, dotdot);
7187                         if (rc == 0)
7188                                 *attr |= LUDA_UPGRADE;
7189                         else
7190                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7191                                        "after the dirent, dir = %lu/%u, "
7192                                        "name = %.*s, ino = %llu, "
7193                                        DFID": rc = %d\n",
7194                                        devname, dir->i_ino, dir->i_generation,
7195                                        ent->oied_namelen, ent->oied_name,
7196                                        ent->oied_ino, PFID(fid), rc);
7197                 }
7198         }
7199
7200         GOTO(out, rc);
7201
7202 out:
7203         if (!IS_ERR(bh))
7204                 brelse(bh);
7205         if (hlock != NULL) {
7206                 ldiskfs_htree_unlock(hlock);
7207         } else {
7208                 if (dev->od_dirent_journal != 0)
7209                         up_write(&obj->oo_ext_idx_sem);
7210                 else
7211                         up_read(&obj->oo_ext_idx_sem);
7212         }
7213
7214         if (jh != NULL)
7215                 ldiskfs_journal_stop(jh);
7216
7217 out_inode:
7218         iput(inode);
7219         if (rc >= 0 && !dirty)
7220                 dev->od_dirent_journal = 0;
7221
7222         return rc;
7223 }
7224
7225 /**
7226  * Returns the value at current position from iterator's in memory structure.
7227  *
7228  * \param di struct osd_it_ea, iterator's in memory structure
7229  * \param attr attr requested for dirent.
7230  * \param lde lustre dirent
7231  *
7232  * \retval   0 no error and \param lde has correct lustre dirent.
7233  * \retval -ve on error
7234  */
7235 static inline int osd_it_ea_rec(const struct lu_env *env,
7236                                 const struct dt_it *di,
7237                                 struct dt_rec *dtrec, __u32 attr)
7238 {
7239         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7240         struct osd_object      *obj   = it->oie_obj;
7241         struct osd_device      *dev   = osd_obj2dev(obj);
7242         struct osd_thread_info *oti   = osd_oti_get(env);
7243         struct osd_inode_id    *id    = &oti->oti_id;
7244         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7245         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7246         __u32 ino = it->oie_dirent->oied_ino;
7247         int rc = 0;
7248
7249         ENTRY;
7250
7251         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7252
7253         if (attr & LUDA_VERIFY) {
7254                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7255                         attr |= LUDA_IGNORE;
7256                         /*
7257                          * If the parent is on remote MDT, and there
7258                          * is no FID-in-dirent, then we have to get
7259                          * the parent FID from the linkEA.
7260                          */
7261                         if (!fid_is_sane(fid) &&
7262                             it->oie_dirent->oied_namelen == 2 &&
7263                             it->oie_dirent->oied_name[0] == '.' &&
7264                             it->oie_dirent->oied_name[1] == '.')
7265                                 osd_get_pfid_from_linkea(env, obj, fid);
7266                 } else {
7267                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7268                                                      &attr);
7269                 }
7270
7271                 if (!fid_is_sane(fid))
7272                         attr |= LUDA_UNKNOWN;
7273         } else {
7274                 attr &= ~LU_DIRENT_ATTRS_MASK;
7275                 if (!fid_is_sane(fid)) {
7276                         bool is_dotdot = false;
7277
7278                         if (it->oie_dirent->oied_namelen == 2 &&
7279                             it->oie_dirent->oied_name[0] == '.' &&
7280                             it->oie_dirent->oied_name[1] == '.')
7281                                 is_dotdot = true;
7282                         /*
7283                          * If the parent is on remote MDT, and there
7284                          * is no FID-in-dirent, then we have to get
7285                          * the parent FID from the linkEA.
7286                          */
7287                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7288                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7289                         } else {
7290                                 if (is_dotdot == false &&
7291                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7292                                         RETURN(-ENOENT);
7293
7294                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7295                         }
7296                 } else {
7297                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7298                 }
7299         }
7300
7301         /* Pack the entry anyway, at least the offset is right. */
7302         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7303                            it->oie_dirent->oied_name,
7304                            it->oie_dirent->oied_namelen,
7305                            it->oie_dirent->oied_type, attr);
7306
7307         if (rc < 0)
7308                 RETURN(rc);
7309
7310         if (osd_remote_fid(env, dev, fid))
7311                 RETURN(0);
7312
7313         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7314                 osd_add_oi_cache(oti, dev, id, fid);
7315
7316         RETURN(rc > 0 ? 0 : rc);
7317 }
7318
7319 /**
7320  * Returns the record size size at current position.
7321  *
7322  * This function will return record(lu_dirent) size in bytes.
7323  *
7324  * \param[in] env       execution environment
7325  * \param[in] di        iterator's in memory structure
7326  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7327  *                      calculate the lu_dirent size.
7328  *
7329  * \retval      record size(in bytes & in memory) of the current lu_dirent
7330  *              entry.
7331  */
7332 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7333                               __u32 attr)
7334 {
7335         struct osd_it_ea *it = (struct osd_it_ea *)di;
7336
7337         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7338 }
7339
7340 /**
7341  * Returns a cookie for current position of the iterator head, so that
7342  * user can use this cookie to load/start the iterator next time.
7343  *
7344  * \param di iterator's in memory structure
7345  *
7346  * \retval cookie for current position, on success
7347  */
7348 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7349 {
7350         struct osd_it_ea *it = (struct osd_it_ea *)di;
7351
7352         return it->oie_dirent->oied_off;
7353 }
7354
7355 /**
7356  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7357  * to load a directory entry at a time and stored it i inn,
7358  * in iterator's in-memory data structure.
7359  *
7360  * \param di struct osd_it_ea, iterator's in memory structure
7361  *
7362  * \retval +ve on success
7363  * \retval -ve on error
7364  */
7365 static int osd_it_ea_load(const struct lu_env *env,
7366                           const struct dt_it *di, __u64 hash)
7367 {
7368         struct osd_it_ea *it = (struct osd_it_ea *)di;
7369         int rc;
7370
7371         ENTRY;
7372         it->oie_file.f_pos = hash;
7373
7374         rc =  osd_ldiskfs_it_fill(env, di);
7375         if (rc > 0)
7376                 rc = -ENODATA;
7377
7378         if (rc == 0)
7379                 rc = 1;
7380
7381         RETURN(rc);
7382 }
7383
7384 /**
7385  * Index lookup function for interoperability mode (b11826).
7386  *
7387  * \param key,  key i.e. file name to be searched
7388  *
7389  * \retval +ve, on success
7390  * \retval -ve, on error
7391  */
7392 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7393                                struct dt_rec *rec, const struct dt_key *key)
7394 {
7395         struct osd_object *obj = osd_dt_obj(dt);
7396         int rc = 0;
7397
7398         ENTRY;
7399
7400         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7401         LINVRNT(osd_invariant(obj));
7402
7403         rc = osd_ea_lookup_rec(env, obj, rec, key);
7404         if (rc == 0)
7405                 rc = 1;
7406         RETURN(rc);
7407 }
7408
7409 /**
7410  * Index and Iterator operations for interoperability
7411  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7412  */
7413 static const struct dt_index_operations osd_index_ea_ops = {
7414         .dio_lookup         = osd_index_ea_lookup,
7415         .dio_declare_insert = osd_index_declare_ea_insert,
7416         .dio_insert         = osd_index_ea_insert,
7417         .dio_declare_delete = osd_index_declare_ea_delete,
7418         .dio_delete         = osd_index_ea_delete,
7419         .dio_it     = {
7420                 .init     = osd_it_ea_init,
7421                 .fini     = osd_it_ea_fini,
7422                 .get      = osd_it_ea_get,
7423                 .put      = osd_it_ea_put,
7424                 .next     = osd_it_ea_next,
7425                 .key      = osd_it_ea_key,
7426                 .key_size = osd_it_ea_key_size,
7427                 .rec      = osd_it_ea_rec,
7428                 .rec_size = osd_it_ea_rec_size,
7429                 .store    = osd_it_ea_store,
7430                 .load     = osd_it_ea_load
7431         }
7432 };
7433
7434 static void *osd_key_init(const struct lu_context *ctx,
7435                           struct lu_context_key *key)
7436 {
7437         struct osd_thread_info *info;
7438
7439         OBD_ALLOC_PTR(info);
7440         if (info == NULL)
7441                 return ERR_PTR(-ENOMEM);
7442
7443         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7444         if (info->oti_it_ea_buf == NULL)
7445                 goto out_free_info;
7446
7447         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7448
7449         info->oti_hlock = ldiskfs_htree_lock_alloc();
7450         if (info->oti_hlock == NULL)
7451                 goto out_free_ea;
7452
7453         return info;
7454
7455 out_free_ea:
7456         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7457 out_free_info:
7458         OBD_FREE_PTR(info);
7459         return ERR_PTR(-ENOMEM);
7460 }
7461
7462 static void osd_key_fini(const struct lu_context *ctx,
7463                          struct lu_context_key *key, void *data)
7464 {
7465         struct osd_thread_info *info = data;
7466         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7467         struct osd_idmap_cache *idc = info->oti_ins_cache;
7468
7469         if (info->oti_dio_pages) {
7470                 int i;
7471                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7472                         struct page *page = info->oti_dio_pages[i];
7473                         if (page) {
7474                                 LASSERT(PagePrivate2(page));
7475                                 LASSERT(PageLocked(page));
7476                                 ClearPagePrivate2(page);
7477                                 unlock_page(page);
7478                                 __free_page(page);
7479                         }
7480                 }
7481                 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7482                                          PTLRPC_MAX_BRW_PAGES);
7483         }
7484
7485         if (info->oti_inode != NULL)
7486                 OBD_FREE_PTR(lli);
7487         if (info->oti_hlock != NULL)
7488                 ldiskfs_htree_lock_free(info->oti_hlock);
7489         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7490         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7491         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7492         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7493         lu_buf_free(&info->oti_big_buf);
7494         if (idc != NULL) {
7495                 LASSERT(info->oti_ins_cache_size > 0);
7496                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7497                 info->oti_ins_cache = NULL;
7498                 info->oti_ins_cache_size = 0;
7499         }
7500         OBD_FREE_PTR(info);
7501 }
7502
7503 static void osd_key_exit(const struct lu_context *ctx,
7504                          struct lu_context_key *key, void *data)
7505 {
7506         struct osd_thread_info *info = data;
7507
7508         LASSERT(info->oti_r_locks == 0);
7509         LASSERT(info->oti_w_locks == 0);
7510         LASSERT(info->oti_txns    == 0);
7511 }
7512
7513 /* type constructor/destructor: osd_type_init, osd_type_fini */
7514 LU_TYPE_INIT_FINI(osd, &osd_key);
7515
7516 struct lu_context_key osd_key = {
7517         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7518         .lct_init = osd_key_init,
7519         .lct_fini = osd_key_fini,
7520         .lct_exit = osd_key_exit
7521 };
7522
7523
7524 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7525                            const char *name, struct lu_device *next)
7526 {
7527         struct osd_device *osd = osd_dev(d);
7528
7529         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7530             sizeof(osd->od_svname))
7531                 return -E2BIG;
7532         return osd_procfs_init(osd, name);
7533 }
7534
7535 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7536 {
7537         struct seq_server_site *ss = osd_seq_site(osd);
7538         int rc = 0;
7539
7540         ENTRY;
7541
7542         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7543                 RETURN(0);
7544
7545         if (unlikely(ss == NULL))
7546                 RETURN(-ENODEV);
7547
7548         OBD_ALLOC_PTR(osd->od_cl_seq);
7549         if (osd->od_cl_seq == NULL)
7550                 RETURN(-ENOMEM);
7551
7552         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7553                         osd->od_svname, ss->ss_server_seq);
7554
7555         if (ss->ss_node_id == 0) {
7556                 /*
7557                  * If the OSD on the sequence controller(MDT0), then allocate
7558                  * sequence here, otherwise allocate sequence after connected
7559                  * to MDT0 (see mdt_register_lwp_callback()).
7560                  */
7561                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7562                                    &osd->od_cl_seq->lcs_space, env);
7563         }
7564
7565         RETURN(rc);
7566 }
7567
7568 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7569 {
7570         if (osd->od_cl_seq == NULL)
7571                 return;
7572
7573         seq_client_fini(osd->od_cl_seq);
7574         OBD_FREE_PTR(osd->od_cl_seq);
7575         osd->od_cl_seq = NULL;
7576 }
7577
7578 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7579 {
7580         ENTRY;
7581
7582         /* shutdown quota slave instance associated with the device */
7583         if (o->od_quota_slave_md != NULL) {
7584                 struct qsd_instance *qsd = o->od_quota_slave_md;
7585
7586                 o->od_quota_slave_md = NULL;
7587                 qsd_fini(env, qsd);
7588         }
7589
7590         if (o->od_quota_slave_dt != NULL) {
7591                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7592
7593                 o->od_quota_slave_dt = NULL;
7594                 qsd_fini(env, qsd);
7595         }
7596
7597         osd_fid_fini(env, o);
7598         osd_scrub_cleanup(env, o);
7599
7600         RETURN(0);
7601 }
7602
7603 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7604 {
7605         ENTRY;
7606
7607         if (o->od_mnt != NULL) {
7608                 shrink_dcache_sb(osd_sb(o));
7609                 osd_sync(env, &o->od_dt_dev);
7610
7611                 mntput(o->od_mnt);
7612                 o->od_mnt = NULL;
7613         }
7614
7615         EXIT;
7616 }
7617
7618 static int osd_mount(const struct lu_env *env,
7619                      struct osd_device *o, struct lustre_cfg *cfg)
7620 {
7621         const char *name = lustre_cfg_string(cfg, 0);
7622         const char *dev = lustre_cfg_string(cfg, 1);
7623         const char *opts;
7624         unsigned long page, s_flags, lmd_flags = 0;
7625         struct page *__page;
7626         struct file_system_type *type;
7627         char *options = NULL;
7628         char *str;
7629         struct osd_thread_info *info = osd_oti_get(env);
7630         struct lu_fid *fid = &info->oti_fid;
7631         struct inode *inode;
7632         int rc = 0, force_over_1024tb = 0;
7633
7634         ENTRY;
7635
7636         if (o->od_mnt != NULL)
7637                 RETURN(0);
7638
7639         if (strlen(dev) >= sizeof(o->od_mntdev))
7640                 RETURN(-E2BIG);
7641         strcpy(o->od_mntdev, dev);
7642
7643         str = lustre_cfg_string(cfg, 2);
7644         s_flags = simple_strtoul(str, NULL, 0);
7645         str = strstr(str, ":");
7646         if (str)
7647                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7648         opts = lustre_cfg_string(cfg, 3);
7649 #ifdef __BIG_ENDIAN
7650         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7651                 CERROR("%s: device %s extents feature is not guaranteed to "
7652                        "work on big-endian systems. Use \"bigendian_extents\" "
7653                        "mount option to override.\n", name, dev);
7654                 RETURN(-EINVAL);
7655         }
7656 #endif
7657 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7658         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7659                 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");
7660         }
7661 #endif
7662 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7663         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7664                 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");
7665         }
7666 #endif
7667 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7668         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7669                 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");
7670         }
7671 #endif
7672
7673         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7674                 force_over_1024tb = 1;
7675
7676         __page = alloc_page(GFP_KERNEL);
7677         if (__page == NULL)
7678                 GOTO(out, rc = -ENOMEM);
7679         page = (unsigned long)page_address(__page);
7680         options = (char *)page;
7681         *options = '\0';
7682         if (opts != NULL) {
7683                 /* strip out the options for back compatiblity */
7684                 static const char * const sout[] = {
7685                         "mballoc",
7686                         "iopen",
7687                         "noiopen",
7688                         "iopen_nopriv",
7689                         "extents",
7690                         "noextents",
7691                         /* strip out option we processed in osd */
7692                         "bigendian_extents",
7693                         "force_over_128tb",
7694                         "force_over_256tb",
7695                         "force_over_512tb",
7696                         "force_over_1024tb",
7697                         NULL
7698                 };
7699                 strncat(options, opts, PAGE_SIZE);
7700                 for (rc = 0, str = options; sout[rc]; ) {
7701                         char *op = strstr(str, sout[rc]);
7702
7703                         if (op == NULL) {
7704                                 rc++;
7705                                 str = options;
7706                                 continue;
7707                         }
7708                         if (op == options || *(op - 1) == ',') {
7709                                 str = op + strlen(sout[rc]);
7710                                 if (*str == ',' || *str == '\0') {
7711                                         *str == ',' ? str++ : str;
7712                                         memmove(op, str, strlen(str) + 1);
7713                                 }
7714                         }
7715                         for (str = op; *str != ',' && *str != '\0'; str++)
7716                                 ;
7717                 }
7718         } else {
7719                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7720         }
7721
7722         /* Glom up mount options */
7723         if (*options != '\0')
7724                 strncat(options, ",", PAGE_SIZE);
7725         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7726
7727         type = get_fs_type("ldiskfs");
7728         if (!type) {
7729                 CERROR("%s: cannot find ldiskfs module\n", name);
7730                 GOTO(out, rc = -ENODEV);
7731         }
7732
7733         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7734         module_put(type->owner);
7735
7736         if (IS_ERR(o->od_mnt)) {
7737                 rc = PTR_ERR(o->od_mnt);
7738                 o->od_mnt = NULL;
7739                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7740                 GOTO(out, rc);
7741         }
7742
7743         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7744                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7745                                  force_over_1024tb == 0) {
7746                 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",
7747                        name, dev);
7748                 GOTO(out_mnt, rc = -EINVAL);
7749         }
7750
7751         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7752                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7753                               name);
7754
7755                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7756         }
7757
7758         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7759                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7760                 GOTO(out_mnt, rc = -EINVAL);
7761         }
7762
7763 #ifdef LDISKFS_MOUNT_DIRDATA
7764         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7765                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7766         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7767                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7768                       "enabling the dirdata feature. If you do not want to "
7769                       "downgrade to Lustre-1.x again, you can enable it via "
7770                       "'tune2fs -O dirdata device'\n", name, dev);
7771 #endif
7772         inode = osd_sb(o)->s_root->d_inode;
7773         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7774         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7775         if (rc != 0) {
7776                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7777                 GOTO(out_mnt, rc);
7778         }
7779
7780         if (lmd_flags & LMD_FLG_NOSCRUB)
7781                 o->od_auto_scrub_interval = AS_NEVER;
7782
7783         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7784                 /* do not use pagecache with flash-backed storage */
7785                 o->od_writethrough_cache = 0;
7786                 o->od_read_cache = 0;
7787         }
7788
7789         GOTO(out, rc = 0);
7790
7791 out_mnt:
7792         mntput(o->od_mnt);
7793         o->od_mnt = NULL;
7794
7795 out:
7796         if (__page)
7797                 __free_page(__page);
7798
7799         return rc;
7800 }
7801
7802 static struct lu_device *osd_device_fini(const struct lu_env *env,
7803                                          struct lu_device *d)
7804 {
7805         struct osd_device *o = osd_dev(d);
7806
7807         ENTRY;
7808
7809         osd_index_backup(env, o, false);
7810         osd_shutdown(env, o);
7811         osd_procfs_fini(o);
7812         if (o->od_oi_table != NULL)
7813                 osd_oi_fini(osd_oti_get(env), o);
7814         osd_obj_map_fini(o);
7815         osd_umount(env, o);
7816
7817         RETURN(NULL);
7818 }
7819
7820 static int osd_device_init0(const struct lu_env *env,
7821                             struct osd_device *o,
7822                             struct lustre_cfg *cfg)
7823 {
7824         struct lu_device *l = osd2lu_dev(o);
7825         struct osd_thread_info *info;
7826         int rc;
7827         int cplen = 0;
7828
7829         /* if the module was re-loaded, env can loose its keys */
7830         rc = lu_env_refill((struct lu_env *)env);
7831         if (rc)
7832                 GOTO(out, rc);
7833         info = osd_oti_get(env);
7834         LASSERT(info);
7835
7836         l->ld_ops = &osd_lu_ops;
7837         o->od_dt_dev.dd_ops = &osd_dt_ops;
7838
7839         spin_lock_init(&o->od_osfs_lock);
7840         mutex_init(&o->od_otable_mutex);
7841         INIT_LIST_HEAD(&o->od_orphan_list);
7842         INIT_LIST_HEAD(&o->od_index_backup_list);
7843         INIT_LIST_HEAD(&o->od_index_restore_list);
7844         spin_lock_init(&o->od_lock);
7845         o->od_index_backup_policy = LIBP_NONE;
7846         o->od_t10_type = 0;
7847
7848         o->od_read_cache = 1;
7849         o->od_writethrough_cache = 1;
7850         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7851         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7852         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7853         o->od_auto_scrub_interval = AS_DEFAULT;
7854
7855         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7856                         sizeof(o->od_svname));
7857         if (cplen >= sizeof(o->od_svname)) {
7858                 rc = -E2BIG;
7859                 GOTO(out, rc);
7860         }
7861
7862         o->od_index_backup_stop = 0;
7863         o->od_index = -1; /* -1 means index is invalid */
7864         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7865         if (rc == LDD_F_SV_TYPE_OST)
7866                 o->od_is_ost = 1;
7867
7868         o->od_full_scrub_ratio = OFSR_DEFAULT;
7869         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7870         rc = osd_mount(env, o, cfg);
7871         if (rc != 0)
7872                 GOTO(out, rc);
7873
7874         /* Can only check block device after mount */
7875         o->od_nonrotational =
7876                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7877
7878         rc = osd_obj_map_init(env, o);
7879         if (rc != 0)
7880                 GOTO(out_mnt, rc);
7881
7882         rc = lu_site_init(&o->od_site, l);
7883         if (rc != 0)
7884                 GOTO(out_compat, rc);
7885         o->od_site.ls_bottom_dev = l;
7886
7887         rc = lu_site_init_finish(&o->od_site);
7888         if (rc != 0)
7889                 GOTO(out_site, rc);
7890
7891         INIT_LIST_HEAD(&o->od_ios_list);
7892         /* setup scrub, including OI files initialization */
7893         o->od_in_init = 1;
7894         rc = osd_scrub_setup(env, o);
7895         o->od_in_init = 0;
7896         if (rc < 0)
7897                 GOTO(out_site, rc);
7898
7899         rc = osd_procfs_init(o, o->od_svname);
7900         if (rc != 0) {
7901                 CERROR("%s: can't initialize procfs: rc = %d\n",
7902                        o->od_svname, rc);
7903                 GOTO(out_scrub, rc);
7904         }
7905
7906         LASSERT(l->ld_site->ls_linkage.next != NULL);
7907         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7908
7909         /* initialize quota slave instance */
7910         /* currently it's no need to prepare qsd_instance_md for OST */
7911         if (!o->od_is_ost) {
7912                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7913                                                 &o->od_dt_dev,
7914                                                 o->od_proc_entry, true);
7915                 if (IS_ERR(o->od_quota_slave_md)) {
7916                         rc = PTR_ERR(o->od_quota_slave_md);
7917                         o->od_quota_slave_md = NULL;
7918                         GOTO(out_procfs, rc);
7919                 }
7920         }
7921
7922         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7923                                         o->od_proc_entry, false);
7924
7925         if (IS_ERR(o->od_quota_slave_dt)) {
7926                 if (o->od_quota_slave_md != NULL) {
7927                         qsd_fini(env, o->od_quota_slave_md);
7928                         o->od_quota_slave_md = NULL;
7929                 }
7930
7931                 rc = PTR_ERR(o->od_quota_slave_dt);
7932                 o->od_quota_slave_dt = NULL;
7933                 GOTO(out_procfs, rc);
7934         }
7935
7936         RETURN(0);
7937
7938 out_procfs:
7939         osd_procfs_fini(o);
7940 out_scrub:
7941         osd_scrub_cleanup(env, o);
7942 out_site:
7943         lu_site_fini(&o->od_site);
7944 out_compat:
7945         osd_obj_map_fini(o);
7946 out_mnt:
7947         osd_umount(env, o);
7948 out:
7949         return rc;
7950 }
7951
7952 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7953                                           struct lu_device_type *t,
7954                                           struct lustre_cfg *cfg)
7955 {
7956         struct osd_device *o;
7957         int rc;
7958
7959         OBD_ALLOC_PTR(o);
7960         if (o == NULL)
7961                 return ERR_PTR(-ENOMEM);
7962
7963         rc = dt_device_init(&o->od_dt_dev, t);
7964         if (rc == 0) {
7965                 /*
7966                  * Because the ctx might be revived in dt_device_init,
7967                  * refill the env here
7968                  */
7969                 lu_env_refill((struct lu_env *)env);
7970                 rc = osd_device_init0(env, o, cfg);
7971                 if (rc)
7972                         dt_device_fini(&o->od_dt_dev);
7973         }
7974
7975         if (unlikely(rc != 0))
7976                 OBD_FREE_PTR(o);
7977
7978         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7979 }
7980
7981 static struct lu_device *osd_device_free(const struct lu_env *env,
7982                                          struct lu_device *d)
7983 {
7984         struct osd_device *o = osd_dev(d);
7985
7986         ENTRY;
7987
7988         /* XXX: make osd top device in order to release reference */
7989         d->ld_site->ls_top_dev = d;
7990         lu_site_purge(env, d->ld_site, -1);
7991         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
7992                       D_ERROR, lu_cdebug_printer);
7993         lu_site_fini(&o->od_site);
7994         dt_device_fini(&o->od_dt_dev);
7995         OBD_FREE_PTR(o);
7996         RETURN(NULL);
7997 }
7998
7999 static int osd_process_config(const struct lu_env *env,
8000                               struct lu_device *d, struct lustre_cfg *cfg)
8001 {
8002         struct osd_device *o = osd_dev(d);
8003         ssize_t count;
8004         int rc;
8005
8006         ENTRY;
8007
8008         switch (cfg->lcfg_command) {
8009         case LCFG_SETUP:
8010                 rc = osd_mount(env, o, cfg);
8011                 break;
8012         case LCFG_CLEANUP:
8013                 /*
8014                  * For the case LCFG_PRE_CLEANUP is not called in advance,
8015                  * that may happend if hit failure during mount process.
8016                  */
8017                 osd_index_backup(env, o, false);
8018                 lu_dev_del_linkage(d->ld_site, d);
8019                 rc = osd_shutdown(env, o);
8020                 break;
8021         case LCFG_PARAM:
8022                 LASSERT(&o->od_dt_dev);
8023                 count  = class_modify_config(cfg, PARAM_OSD,
8024                                              &o->od_dt_dev.dd_kobj);
8025                 if (count < 0)
8026                         count = class_modify_config(cfg, PARAM_OST,
8027                                                     &o->od_dt_dev.dd_kobj);
8028                 rc = count > 0 ? 0 : count;
8029                 break;
8030         case LCFG_PRE_CLEANUP:
8031                 osd_scrub_stop(o);
8032                 osd_index_backup(env, o,
8033                                  o->od_index_backup_policy != LIBP_NONE);
8034                 rc = 0;
8035                 break;
8036         default:
8037                 rc = -ENOSYS;
8038         }
8039
8040         RETURN(rc);
8041 }
8042
8043 static int osd_recovery_complete(const struct lu_env *env,
8044                                  struct lu_device *d)
8045 {
8046         struct osd_device *osd = osd_dev(d);
8047         int rc = 0;
8048
8049         ENTRY;
8050
8051         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8052                 RETURN(0);
8053
8054         /*
8055          * start qsd instance on recovery completion, this notifies the quota
8056          * slave code that we are about to process new requests now
8057          */
8058         rc = qsd_start(env, osd->od_quota_slave_dt);
8059         if (rc == 0 && osd->od_quota_slave_md != NULL)
8060                 rc = qsd_start(env, osd->od_quota_slave_md);
8061
8062         RETURN(rc);
8063 }
8064
8065 /*
8066  * we use exports to track all osd users
8067  */
8068 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8069                            struct obd_device *obd, struct obd_uuid *cluuid,
8070                            struct obd_connect_data *data, void *localdata)
8071 {
8072         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8073         struct lustre_handle conn;
8074         int rc;
8075
8076         ENTRY;
8077
8078         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8079
8080         rc = class_connect(&conn, obd, cluuid);
8081         if (rc)
8082                 RETURN(rc);
8083
8084         *exp = class_conn2export(&conn);
8085
8086         spin_lock(&osd->od_osfs_lock);
8087         osd->od_connects++;
8088         spin_unlock(&osd->od_osfs_lock);
8089
8090         RETURN(0);
8091 }
8092
8093 /*
8094  * once last export (we don't count self-export) disappeared
8095  * osd can be released
8096  */
8097 static int osd_obd_disconnect(struct obd_export *exp)
8098 {
8099         struct obd_device *obd = exp->exp_obd;
8100         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8101         int rc, release = 0;
8102
8103         ENTRY;
8104
8105         /* Only disconnect the underlying layers on the final disconnect. */
8106         spin_lock(&osd->od_osfs_lock);
8107         osd->od_connects--;
8108         if (osd->od_connects == 0)
8109                 release = 1;
8110         spin_unlock(&osd->od_osfs_lock);
8111
8112         rc = class_disconnect(exp); /* bz 9811 */
8113
8114         if (rc == 0 && release)
8115                 class_manual_cleanup(obd);
8116         RETURN(rc);
8117 }
8118
8119 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8120                        struct lu_device *dev)
8121 {
8122         struct osd_device *osd = osd_dev(dev);
8123         struct lr_server_data *lsd =
8124                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8125         int result = 0;
8126
8127         ENTRY;
8128
8129         if (osd->od_quota_slave_md != NULL) {
8130                 /* set up quota slave objects for inode */
8131                 result = qsd_prepare(env, osd->od_quota_slave_md);
8132                 if (result != 0)
8133                         RETURN(result);
8134         }
8135
8136         if (osd->od_quota_slave_dt != NULL) {
8137                 /* set up quota slave objects for block */
8138                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8139                 if (result != 0)
8140                         RETURN(result);
8141         }
8142
8143
8144         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8145 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8146                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8147                         osd->od_index_in_idif = 1;
8148                 } else {
8149                         osd->od_index_in_idif = 0;
8150                         result = osd_register_proc_index_in_idif(osd);
8151                         if (result != 0)
8152                                 RETURN(result);
8153                 }
8154 #else
8155                 osd->od_index_in_idif = 1;
8156 #endif
8157         }
8158
8159         result = osd_fid_init(env, osd);
8160
8161         RETURN(result);
8162 }
8163
8164 /**
8165  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8166  *
8167  * Allocate FID.
8168  *
8169  * see include/lu_object.h for the details.
8170  */
8171 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8172                          struct lu_fid *fid, struct lu_object *parent,
8173                          const struct lu_name *name)
8174 {
8175         struct osd_device *osd = osd_dev(d);
8176
8177         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8178 }
8179
8180 static const struct lu_object_operations osd_lu_obj_ops = {
8181         .loo_object_init      = osd_object_init,
8182         .loo_object_delete    = osd_object_delete,
8183         .loo_object_release   = osd_object_release,
8184         .loo_object_free      = osd_object_free,
8185         .loo_object_print     = osd_object_print,
8186         .loo_object_invariant = osd_object_invariant
8187 };
8188
8189 const struct lu_device_operations osd_lu_ops = {
8190         .ldo_object_alloc      = osd_object_alloc,
8191         .ldo_process_config    = osd_process_config,
8192         .ldo_recovery_complete = osd_recovery_complete,
8193         .ldo_prepare           = osd_prepare,
8194         .ldo_fid_alloc         = osd_fid_alloc,
8195 };
8196
8197 static const struct lu_device_type_operations osd_device_type_ops = {
8198         .ldto_init = osd_type_init,
8199         .ldto_fini = osd_type_fini,
8200
8201         .ldto_start = osd_type_start,
8202         .ldto_stop  = osd_type_stop,
8203
8204         .ldto_device_alloc = osd_device_alloc,
8205         .ldto_device_free  = osd_device_free,
8206
8207         .ldto_device_init = osd_device_init,
8208         .ldto_device_fini = osd_device_fini
8209 };
8210
8211 static struct lu_device_type osd_device_type = {
8212         .ldt_tags     = LU_DEVICE_DT,
8213         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8214         .ldt_ops      = &osd_device_type_ops,
8215         .ldt_ctx_tags = LCT_LOCAL,
8216 };
8217
8218 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8219 {
8220         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8221         struct super_block *sb = osd_sb(osd);
8222
8223         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8224 }
8225
8226 /*
8227  * lprocfs legacy support.
8228  */
8229 static const struct obd_ops osd_obd_device_ops = {
8230         .o_owner = THIS_MODULE,
8231         .o_connect      = osd_obd_connect,
8232         .o_disconnect   = osd_obd_disconnect,
8233         .o_health_check = osd_health_check,
8234 };
8235
8236 static ssize_t track_declares_assert_show(struct kobject *kobj,
8237                                    struct attribute *attr,
8238                                    char *buf)
8239 {
8240         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8241 }
8242
8243 static ssize_t track_declares_assert_store(struct kobject *kobj,
8244                                            struct attribute *attr,
8245                                            const char *buffer, size_t count)
8246 {
8247         bool track_declares_assert;
8248         int rc;
8249
8250         rc = kstrtobool(buffer, &track_declares_assert);
8251         if (rc)
8252                 return rc;
8253
8254         ldiskfs_track_declares_assert = track_declares_assert;
8255
8256         return count;
8257 }
8258 LUSTRE_RW_ATTR(track_declares_assert);
8259
8260 static int __init osd_init(void)
8261 {
8262         struct kobject *kobj;
8263         int rc;
8264
8265         BUILD_BUG_ON(BH_DXLock >=
8266                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8267 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8268         /* please, try to keep osd_thread_info smaller than a page */
8269         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8270 #endif
8271
8272         osd_oi_mod_init();
8273
8274         rc = lu_kmem_init(ldiskfs_caches);
8275         if (rc)
8276                 return rc;
8277
8278 #ifdef CONFIG_KALLSYMS
8279         priv_security_file_alloc =
8280                 (void *)kallsyms_lookup_name("security_file_alloc");
8281 #endif
8282
8283         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8284                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8285         if (rc) {
8286                 lu_kmem_fini(ldiskfs_caches);
8287                 return rc;
8288         }
8289
8290         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8291         if (kobj) {
8292                 rc = sysfs_create_file(kobj,
8293                                        &lustre_attr_track_declares_assert.attr);
8294                 kobject_put(kobj);
8295                 if (rc) {
8296                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8297                         rc = 0;
8298                 }
8299         }
8300         return rc;
8301 }
8302
8303 static void __exit osd_exit(void)
8304 {
8305         struct kobject *kobj;
8306
8307         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8308         if (kobj) {
8309                 sysfs_remove_file(kobj,
8310                                   &lustre_attr_track_declares_assert.attr);
8311                 kobject_put(kobj);
8312         }
8313         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8314         lu_kmem_fini(ldiskfs_caches);
8315 }
8316
8317 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8318 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8319 MODULE_VERSION(LUSTRE_VERSION_STRING);
8320 MODULE_LICENSE("GPL");
8321
8322 module_init(osd_init);
8323 module_exit(osd_exit);