lustre/osd-ldiskfs/osd_handler.c

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