lustre/osd-ldiskfs/osd_handler.c

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