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/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 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                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 499                               "maximum entry limit\n",
 500                               osd_name(osd), parent->i_ino, PFID(fid),
 501                               rc == -ENOSPC ? "has reached" : "is approaching");
 502                 else
 503                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 504                               "maximum entry limit\n",
 505                               osd_name(osd), parent->i_ino,
 506                               rc == -ENOSPC ? "has reached" : "is approaching");
 507
 508                 /* ignore such error now */
 509                 if (rc == -ENOBUFS)
 510                         rc = 0;
 511         }
 512
 513         return rc;
 514 }
 515
 516
 517 struct inode *
 518 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 519              struct osd_inode_id *id, struct lu_fid *fid)
 520 {
 521         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 522         struct inode *inode;
 523         int rc;
 524
 525         inode = osd_iget(info, dev, id);
 526         if (IS_ERR(inode))
 527                 return inode;
 528
 529         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 530         if (!rc) {
 531                 *fid = loa->loa_lma.lma_self_fid;
 532         } else if (rc == -ENODATA) {
 533                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 534                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 535                 else
 536                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 537         } else {
 538                 iput(inode);
 539                 inode = ERR_PTR(rc);
 540         }
 541         return inode;
 542 }
 543
 544 static struct inode *osd_iget_check(struct osd_thread_info *info,
 545                                     struct osd_device *dev,
 546                                     const struct lu_fid *fid,
 547                                     struct osd_inode_id *id,
 548                                     bool trusted)
 549 {
 550         struct inode *inode;
 551         int rc = 0;
 552
 553         ENTRY;
 554
 555         /*
 556          * The cached OI mapping is trustable. If we cannot locate the inode
 557          * via the cached OI mapping, then return the failure to the caller
 558          * directly without further OI checking.
 559          */
 560
 561 again:
 562         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_18_23, contains self-fid
 711  * \retval -v: other failure cases
 712  */
 713 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 714                  struct dentry *dentry, struct lu_fid *fid)
 715 {
 716         struct filter_fid *ff = &info->oti_ff;
 717         struct ost_id *ostid = &info->oti_ostid;
 718         int rc;
 719
 720         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 721         if (rc == sizeof(struct filter_fid_18_23)) {
 722                 struct filter_fid_18_23 *ff_old = (void *)ff;
 723
 724                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 725                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 726                 /*
 727                  * XXX: use 0 as the index for compatibility, the caller will
 728                  * handle index related issues when necessary.
 729                  */
 730                 if (!rc)
 731                         ostid_to_fid(fid, ostid, 0);
 732         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 733                 rc = 1;
 734         } else if (rc >= 0) {
 735                 rc = -EINVAL;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int osd_lma_self_repair(struct osd_thread_info *info,
 742                                struct osd_device *osd, struct inode *inode,
 743                                const struct lu_fid *fid, __u32 compat)
 744 {
 745         handle_t *jh;
 746         int rc;
 747
 748         LASSERT(current->journal_info == NULL);
 749
 750         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 751                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 752         if (IS_ERR(jh)) {
 753                 rc = PTR_ERR(jh);
 754                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 755                       osd_name(osd), rc);
 756                 return rc;
 757         }
 758
 759         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 760         if (rc != 0)
 761                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 762                       osd_name(osd), rc);
 763         ldiskfs_journal_stop(jh);
 764         return rc;
 765 }
 766
 767 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 768 {
 769         struct osd_thread_info *info = osd_oti_get(env);
 770         struct osd_device *osd = osd_obj2dev(obj);
 771         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 772         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 773         struct inode *inode = obj->oo_inode;
 774         struct dentry *dentry = &info->oti_obj_dentry;
 775         struct lu_fid *fid = NULL;
 776         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 777         int rc;
 778
 779         ENTRY;
 780
 781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 782                              (void *)loa, sizeof(*loa));
 783         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 784                 fid = &lma->lma_self_fid;
 785                 rc = osd_get_idif(info, inode, dentry, fid);
 786                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 787                         /*
 788                          * For the given OST-object, if it has neither LMA nor
 789                          * FID in XATTR_NAME_FID, then the given FID (which is
 790                          * contained in the @obj, from client RPC for locating
 791                          * the OST-object) is trusted. We use it to generate
 792                          * the LMA.
 793                          */
 794                         osd_lma_self_repair(info, osd, inode, rfid,
 795                                             LMAC_FID_ON_OST);
 796                         RETURN(0);
 797                 }
 798         }
 799
 800         if (rc < 0)
 801                 RETURN(rc);
 802
 803         if (rc > 0) {
 804                 rc = 0;
 805                 lustre_lma_swab(lma);
 806                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 807                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 808                               S_ISREG(inode->i_mode)))) {
 809                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 810                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 811                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 812                               PFID(rfid), inode->i_ino);
 813                         rc = -EOPNOTSUPP;
 814                 } else {
 815                         fid = &lma->lma_self_fid;
 816                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 817                             osd->od_is_ost)
 818                                 obj->oo_pfid_in_lma = 1;
 819                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 820                             !osd->od_is_ost)
 821                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 822                 }
 823         }
 824
 825         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 826                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 827                         struct ost_id   *oi   = &info->oti_ostid;
 828                         struct lu_fid   *fid1 = &info->oti_fid3;
 829                         __u32            idx  = fid_idif_ost_idx(rfid);
 830
 831                         /*
 832                          * For old IDIF, the OST index is not part of the IDIF,
 833                          * Means that different OSTs may have the same IDIFs.
 834                          * Under such case, we need to make some compatible
 835                          * check to make sure to trigger OI scrub properly.
 836                          */
 837                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 838                                 /* Given @rfid is new, LMA is old. */
 839                                 fid_to_ostid(fid, oi);
 840                                 ostid_to_fid(fid1, oi, idx);
 841                                 if (lu_fid_eq(fid1, rfid)) {
 842                                         if (osd->od_index_in_idif)
 843                                                 osd_lma_self_repair(info, osd,
 844                                                         inode, rfid,
 845                                                         LMAC_FID_ON_OST);
 846                                         RETURN(0);
 847                                 }
 848                         }
 849                 }
 850
 851                 rc = -EREMCHG;
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 struct osd_check_lmv_buf {
 858 #ifdef HAVE_DIR_CONTEXT
 859         /* please keep it as first member */
 860         struct dir_context ctx;
 861 #endif
 862         struct osd_thread_info *oclb_info;
 863         struct osd_device *oclb_dev;
 864         struct osd_idmap_cache *oclb_oic;
 865         int oclb_items;
 866         bool oclb_found;
 867 };
 868
 869 /**
 870  * It is called internally by ->readdir() to filter out the
 871  * local slave object's FID of the striped directory.
 872  *
 873  * \retval      1 found the local slave's FID
 874  * \retval      0 continue to check next item
 875  * \retval      -ve for failure
 876  */
 877 #ifdef HAVE_FILLDIR_USE_CTX
 878 static int osd_stripe_dir_filldir(struct dir_context *buf,
 879 #else
 880 static int osd_stripe_dir_filldir(void *buf,
 881 #endif
 882                                   const char *name, int namelen,
 883                                   loff_t offset, __u64 ino, unsigned int d_type)
 884 {
 885         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 886         struct osd_thread_info *oti = oclb->oclb_info;
 887         struct lu_fid *fid = &oti->oti_fid3;
 888         struct osd_inode_id *id = &oti->oti_id3;
 889         struct osd_device *dev = oclb->oclb_dev;
 890         struct osd_idmap_cache *oic = oclb->oclb_oic;
 891         struct inode *inode;
 892
 893         oclb->oclb_items++;
 894
 895         if (name[0] == '.')
 896                 return 0;
 897
 898         fid_zero(fid);
 899         sscanf(name + 1, SFID, RFID(fid));
 900         if (!fid_is_sane(fid))
 901                 return 0;
 902
 903         if (osd_remote_fid(oti->oti_env, dev, fid))
 904                 return 0;
 905
 906         osd_id_gen(id, ino, OSD_OII_NOGEN);
 907         inode = osd_iget(oti, dev, id);
 908         if (IS_ERR(inode))
 909                 return PTR_ERR(inode);
 910
 911         iput(inode);
 912         osd_add_oi_cache(oti, dev, id, fid);
 913         oic->oic_fid = *fid;
 914         oic->oic_lid = *id;
 915         oic->oic_dev = dev;
 916         osd_oii_insert(dev, oic, true);
 917         oclb->oclb_found = true;
 918
 919         return 1;
 920 }
 921
 922 /*
 923  * When lookup item under striped directory, we need to locate the master
 924  * MDT-object of the striped directory firstly, then the client will send
 925  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 926  * and the item's name. If the system is restored from MDT file level backup,
 927  * then before the OI scrub completely built the OI files, the OI mappings of
 928  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 929  * not a problem for the master MDT-object. Because when locate the master
 930  * MDT-object, we will do name based lookup (for the striped directory itself)
 931  * firstly, during such process we can setup the correct OI mapping for the
 932  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 933  * the client will not trigger name based lookup on the MDT to locate the slave
 934  * MDT-object before locating item under the striped directory, then when
 935  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 936  * is invalid and does not know what the right OI mapping is, then the MDT has
 937  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 938  * the OI file, related OI mapping is unknown yet, please try again later. And
 939  * then client will re-try the RPC again and again until related OI mapping has
 940  * been updated. That is quite inefficient.
 941  *
 942  * To resolve above trouble, we will handle it as the following two cases:
 943  *
 944  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 945  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 946  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 947  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 948  *    directly. Please check osd_fid_lookup().
 949  *
 950  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 951  *    Under such case, during lookup the master MDT-object, we will lookup the
 952  *    slave MDT-object via readdir against the master MDT-object, because the
 953  *    slave MDT-objects information are stored as sub-directories with the name
 954  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 955  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 956  *    the correct OI mapping for the slave MDT-object.
 957  */
 958 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 959                          struct inode *inode, struct osd_idmap_cache *oic)
 960 {
 961         struct lu_buf *buf = &oti->oti_big_buf;
 962         struct dentry *dentry = &oti->oti_obj_dentry;
 963         struct file *filp = &oti->oti_file;
 964         const struct file_operations *fops;
 965         struct lmv_mds_md_v1 *lmv1;
 966         struct osd_check_lmv_buf oclb = {
 967 #ifdef HAVE_DIR_CONTEXT
 968                 .ctx.actor = osd_stripe_dir_filldir,
 969 #endif
 970                 .oclb_info = oti,
 971                 .oclb_dev = dev,
 972                 .oclb_oic = oic,
 973                 .oclb_found = false,
 974         };
 975         int rc = 0;
 976
 977         ENTRY;
 978
 979 again:
 980         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 981                              buf->lb_len);
 982         if (rc == -ERANGE) {
 983                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 984                 if (rc > 0) {
 985                         lu_buf_realloc(buf, rc);
 986                         if (buf->lb_buf == NULL)
 987                                 GOTO(out, rc = -ENOMEM);
 988
 989                         goto again;
 990                 }
 991         }
 992
 993         if (unlikely(rc == 0 || rc == -ENODATA))
 994                 GOTO(out, rc = 0);
 995
 996         if (rc < 0)
 997                 GOTO(out, rc);
 998
 999         if (unlikely(buf->lb_buf == NULL)) {
1000                 lu_buf_realloc(buf, rc);
1001                 if (buf->lb_buf == NULL)
1002                         GOTO(out, rc = -ENOMEM);
1003
1004                 goto again;
1005         }
1006
1007         lmv1 = buf->lb_buf;
1008         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1009                 GOTO(out, rc = 0);
1010
1011         fops = inode->i_fop;
1012         dentry->d_inode = inode;
1013         dentry->d_sb = inode->i_sb;
1014         filp->f_pos = 0;
1015         filp->f_path.dentry = dentry;
1016         filp->f_mode = FMODE_64BITHASH;
1017         filp->f_mapping = inode->i_mapping;
1018         filp->f_op = fops;
1019         filp->private_data = NULL;
1020         set_file_inode(filp, inode);
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024 #ifdef HAVE_DIR_CONTEXT
1025                 oclb.ctx.pos = filp->f_pos;
1026 #ifdef HAVE_ITERATE_SHARED
1027                 rc = fops->iterate_shared(filp, &oclb.ctx);
1028 #else
1029                 rc = fops->iterate(filp, &oclb.ctx);
1030 #endif
1031                 filp->f_pos = oclb.ctx.pos;
1032 #else
1033                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1034 #endif
1035         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1036                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1037         fops->release(inode, filp);
1038
1039 out:
1040         if (rc < 0)
1041                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1042                        DFID": rc = %d\n", osd_ino2name(inode),
1043                        inode->i_ino, inode->i_generation,
1044                        PFID(&oic->oic_fid), rc);
1045         else
1046                 rc = 0;
1047
1048         RETURN(rc);
1049 }
1050
1051 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1052                           const struct lu_fid *fid,
1053                           const struct lu_object_conf *conf)
1054 {
1055         struct osd_thread_info *info;
1056         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1057         struct osd_device *dev;
1058         struct osd_idmap_cache *oic;
1059         struct osd_inode_id *id;
1060         struct inode *inode = NULL;
1061         struct lustre_scrub *scrub;
1062         struct scrub_file *sf;
1063         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1064         __u32 saved_ino;
1065         __u32 saved_gen;
1066         int result = 0;
1067         int rc1 = 0;
1068         bool remote = false;
1069         bool trusted = true;
1070         bool updated = false;
1071         bool checked = false;
1072
1073         ENTRY;
1074
1075         LINVRNT(osd_invariant(obj));
1076         LASSERT(obj->oo_inode == NULL);
1077         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1078
1079         dev = osd_dev(ldev);
1080         scrub = &dev->od_scrub.os_scrub;
1081         sf = &scrub->os_file;
1082         info = osd_oti_get(env);
1083         LASSERT(info);
1084         oic = &info->oti_cache;
1085
1086         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1087                 RETURN(-ENOENT);
1088
1089         /*
1090          * For the object is created as locking anchor, or for the object to
1091          * be created on disk. No need to osd_oi_lookup() at here because FID
1092          * shouldn't never be re-used, if it's really a duplicate FID from
1093          * unexpected reason, we should be able to detect it later by calling
1094          * do_create->osd_oi_insert().
1095          */
1096         if (conf && conf->loc_flags & LOC_F_NEW)
1097                 GOTO(out, result = 0);
1098
1099         /* Search order: 1. per-thread cache. */
1100         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1101                 id = &oic->oic_lid;
1102                 goto iget;
1103         }
1104
1105         id = &info->oti_id;
1106         if (!list_empty(&scrub->os_inconsistent_items)) {
1107                 /* Search order: 2. OI scrub pending list. */
1108                 result = osd_oii_lookup(dev, fid, id);
1109                 if (!result)
1110                         goto iget;
1111         }
1112
1113         /*
1114          * The OI mapping in the OI file can be updated by the OI scrub
1115          * when we locate the inode via FID. So it may be not trustable.
1116          */
1117         trusted = false;
1118
1119         /* Search order: 3. OI files. */
1120         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1121         if (result == -ENOENT) {
1122                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1123                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1124                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1125                                       sf->sf_oi_bitmap))
1126                         GOTO(out, result = 0);
1127
1128                 goto trigger;
1129         }
1130
1131         if (result)
1132                 GOTO(out, result);
1133
1134 iget:
1135         obj->oo_inode = NULL;
1136         /* for later passes through checks, not true on first pass */
1137         if (!IS_ERR_OR_NULL(inode))
1138                 iput(inode);
1139
1140         inode = osd_iget_check(info, dev, fid, id, trusted);
1141         if (!IS_ERR(inode)) {
1142                 obj->oo_inode = inode;
1143                 result = 0;
1144                 if (remote)
1145                         goto trigger;
1146
1147                 goto check_lma;
1148         }
1149
1150         result = PTR_ERR(inode);
1151         if (result == -ENOENT || result == -ESTALE)
1152                 GOTO(out, result = 0);
1153
1154         if (result != -EREMCHG)
1155                 GOTO(out, result);
1156
1157 trigger:
1158         /*
1159          * We still have chance to get the valid inode: for the
1160          * object which is referenced by remote name entry, the
1161          * object on the local MDT will be linked under the dir
1162          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1163          *
1164          * We do not know whether the object for the given FID
1165          * is referenced by some remote name entry or not, and
1166          * especially for DNE II, a multiple-linked object may
1167          * have many name entries reside on many MDTs.
1168          *
1169          * To simplify the operation, OSD will not distinguish
1170          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1171          * only happened for the RPC from other MDT during the
1172          * OI scrub, or for the client side RPC with FID only,
1173          * such as FID to path, or from old connected client.
1174          */
1175         if (!remote) {
1176                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1177                 if (!rc1) {
1178                         remote = true;
1179                         trusted = true;
1180                         flags |= SS_AUTO_PARTIAL;
1181                         flags &= ~SS_AUTO_FULL;
1182                         goto iget;
1183                 }
1184         }
1185
1186         if (thread_is_running(&scrub->os_thread)) {
1187                 if (scrub->os_partial_scan && !scrub->os_in_join)
1188                         goto join;
1189
1190                 if (IS_ERR_OR_NULL(inode) || result)
1191                         GOTO(out, result = -EINPROGRESS);
1192
1193                 LASSERT(remote);
1194                 LASSERT(obj->oo_inode == inode);
1195
1196                 osd_add_oi_cache(info, dev, id, fid);
1197                 osd_oii_insert(dev, oic, true);
1198                 goto found;
1199         }
1200
1201         if (dev->od_auto_scrub_interval == AS_NEVER) {
1202                 if (!remote)
1203                         GOTO(out, result = -EREMCHG);
1204
1205                 LASSERT(!result);
1206                 LASSERT(obj->oo_inode == inode);
1207
1208                 osd_add_oi_cache(info, dev, id, fid);
1209                 goto found;
1210         }
1211
1212 join:
1213         rc1 = osd_scrub_start(env, dev, flags);
1214         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1215                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1216         if (rc1 && rc1 != -EALREADY)
1217                 GOTO(out, result = -EREMCHG);
1218
1219         if (IS_ERR_OR_NULL(inode) || result)
1220                 GOTO(out, result = -EINPROGRESS);
1221
1222         LASSERT(remote);
1223         LASSERT(obj->oo_inode == inode);
1224
1225         osd_add_oi_cache(info, dev, id, fid);
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         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1294         /*
1295          * "result == -ENOENT" means the cached OI mapping has been removed
1296          * from the OI file by race, above inode belongs to other object.
1297          */
1298         if (result == -ENOENT) {
1299                 LASSERT(trusted);
1300
1301                 obj->oo_inode = NULL;
1302                 GOTO(out, result = 0);
1303         }
1304
1305         if (result)
1306                 GOTO(out, result);
1307
1308         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1309                 result = -EREMCHG;
1310                 goto trigger;
1311         }
1312
1313         /*
1314          * It is the OI scrub updated the OI mapping by race.
1315          * The new OI mapping must be valid.
1316          */
1317         trusted = true;
1318         updated = true;
1319         goto iget;
1320
1321 found:
1322         if (!checked) {
1323                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1324                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1325
1326                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1327                 if (!result) {
1328                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1329                             dev->od_is_ost)
1330                                 obj->oo_pfid_in_lma = 1;
1331                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1332                             !dev->od_is_ost)
1333                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1334                 } else if (result != -ENODATA) {
1335                         GOTO(out, result);
1336                 }
1337         }
1338
1339         obj->oo_compat_dot_created = 1;
1340         obj->oo_compat_dotdot_created = 1;
1341
1342         if (S_ISDIR(inode->i_mode) &&
1343             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1344                 osd_check_lmv(info, dev, inode, oic);
1345
1346         result = osd_attach_jinode(inode);
1347         if (result)
1348                 GOTO(out, result);
1349
1350         if (!ldiskfs_pdo)
1351                 GOTO(out, result = 0);
1352
1353         LASSERT(!obj->oo_hl_head);
1354         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1355
1356         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1357
1358 out:
1359         if (result || !obj->oo_inode) {
1360                 if (!IS_ERR_OR_NULL(inode))
1361                         iput(inode);
1362
1363                 obj->oo_inode = NULL;
1364                 if (trusted)
1365                         fid_zero(&oic->oic_fid);
1366         }
1367
1368         LINVRNT(osd_invariant(obj));
1369         return result;
1370 }
1371
1372 /*
1373  * Concurrency: shouldn't matter.
1374  */
1375 static void osd_object_init0(struct osd_object *obj)
1376 {
1377         LASSERT(obj->oo_inode != NULL);
1378         obj->oo_dt.do_body_ops = &osd_body_ops;
1379         obj->oo_dt.do_lu.lo_header->loh_attr |=
1380                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1381 }
1382
1383 /*
1384  * Concurrency: no concurrent access is possible that early in object
1385  * life-cycle.
1386  */
1387 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1388                            const struct lu_object_conf *conf)
1389 {
1390         struct osd_object *obj = osd_obj(l);
1391         int result;
1392
1393         LINVRNT(osd_invariant(obj));
1394
1395         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1396                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1397                 l->lo_header->loh_attr |= LOHA_EXISTS;
1398                 return 0;
1399         }
1400
1401         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1402         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1403         if (result == 0 && obj->oo_inode != NULL) {
1404                 struct osd_thread_info *oti = osd_oti_get(env);
1405                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1406
1407                 osd_object_init0(obj);
1408                 if (unlikely(obj->oo_header))
1409                         return 0;
1410
1411                 result = osd_get_lma(oti, obj->oo_inode,
1412                                      &oti->oti_obj_dentry, loa);
1413                 if (!result) {
1414                         /*
1415                          * Convert LMAI flags to lustre LMA flags
1416                          * and cache it to oo_lma_flags
1417                          */
1418                         obj->oo_lma_flags =
1419                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1420                 } else if (result == -ENODATA) {
1421                         result = 0;
1422                 }
1423         }
1424
1425         LINVRNT(osd_invariant(obj));
1426         return result;
1427 }
1428
1429 /*
1430  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1431  * The left part is for value, binary mode.
1432  */
1433 struct osd_xattr_entry {
1434         struct list_head        oxe_list;
1435         size_t                  oxe_len;
1436         size_t                  oxe_namelen;
1437         bool                    oxe_exist;
1438         struct rcu_head         oxe_rcu;
1439         char                    oxe_buf[0];
1440 };
1441
1442 static int osd_oxc_get(struct osd_object *obj, const char *name,
1443                        struct lu_buf *buf)
1444 {
1445         struct osd_xattr_entry *tmp;
1446         struct osd_xattr_entry *oxe = NULL;
1447         size_t namelen = strlen(name);
1448         int rc;
1449
1450         ENTRY;
1451
1452         rcu_read_lock();
1453         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1454                 if (namelen == tmp->oxe_namelen &&
1455                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1456                         oxe = tmp;
1457                         break;
1458                 }
1459         }
1460
1461         if (oxe == NULL)
1462                 GOTO(out, rc = -ENOENT);
1463
1464         if (!oxe->oxe_exist)
1465                 GOTO(out, rc = -ENODATA);
1466
1467         /* vallen */
1468         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1469         LASSERT(rc > 0);
1470
1471         if (buf->lb_buf == NULL)
1472                 GOTO(out, rc);
1473
1474         if (buf->lb_len < rc)
1475                 GOTO(out, rc = -ERANGE);
1476
1477         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1478         EXIT;
1479 out:
1480         rcu_read_unlock();
1481
1482         return rc;
1483 }
1484
1485 static void osd_oxc_free(struct rcu_head *head)
1486 {
1487         struct osd_xattr_entry *oxe;
1488
1489         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1490         OBD_FREE(oxe, oxe->oxe_len);
1491 }
1492
1493 static void osd_oxc_add(struct osd_object *obj, const char *name,
1494                         const char *buf, int buflen)
1495 {
1496         struct osd_xattr_entry *oxe;
1497         struct osd_xattr_entry *old = NULL;
1498         struct osd_xattr_entry *tmp;
1499         size_t namelen = strlen(name);
1500         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1501
1502         OBD_ALLOC(oxe, len);
1503         if (oxe == NULL)
1504                 return;
1505
1506         INIT_LIST_HEAD(&oxe->oxe_list);
1507         oxe->oxe_len = len;
1508         oxe->oxe_namelen = namelen;
1509         memcpy(oxe->oxe_buf, name, namelen);
1510         if (buflen > 0) {
1511                 LASSERT(buf != NULL);
1512                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1513                 oxe->oxe_exist = true;
1514         } else {
1515                 oxe->oxe_exist = false;
1516         }
1517
1518         /* this should be rarely called, just remove old and add new */
1519         spin_lock(&obj->oo_guard);
1520         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1521                 if (namelen == tmp->oxe_namelen &&
1522                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1523                         old = tmp;
1524                         break;
1525                 }
1526         }
1527         if (old != NULL) {
1528                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1529                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1530         } else {
1531                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1532         }
1533         spin_unlock(&obj->oo_guard);
1534 }
1535
1536 static void osd_oxc_del(struct osd_object *obj, const char *name)
1537 {
1538         struct osd_xattr_entry *oxe;
1539         size_t namelen = strlen(name);
1540
1541         spin_lock(&obj->oo_guard);
1542         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1543                 if (namelen == oxe->oxe_namelen &&
1544                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1545                         list_del_rcu(&oxe->oxe_list);
1546                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1547                         break;
1548                 }
1549         }
1550         spin_unlock(&obj->oo_guard);
1551 }
1552
1553 static void osd_oxc_fini(struct osd_object *obj)
1554 {
1555         struct osd_xattr_entry *oxe, *next;
1556
1557         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1558                 list_del(&oxe->oxe_list);
1559                 OBD_FREE(oxe, oxe->oxe_len);
1560         }
1561 }
1562
1563 /*
1564  * Concurrency: no concurrent access is possible that late in object
1565  * life-cycle.
1566  */
1567 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1568 {
1569         struct osd_object *obj = osd_obj(l);
1570         struct lu_object_header *h = obj->oo_header;
1571
1572         LINVRNT(osd_invariant(obj));
1573
1574         osd_oxc_fini(obj);
1575         dt_object_fini(&obj->oo_dt);
1576         if (obj->oo_hl_head != NULL)
1577                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1578         OBD_FREE_PTR(obj);
1579         if (unlikely(h)) {
1580                 lu_object_header_fini(h);
1581                 OBD_FREE_PTR(h);
1582         }
1583 }
1584
1585 /*
1586  * Concurrency: no concurrent access is possible that late in object
1587  * life-cycle.
1588  */
1589 static void osd_index_fini(struct osd_object *o)
1590 {
1591         struct iam_container *bag;
1592
1593         if (o->oo_dir != NULL) {
1594                 bag = &o->oo_dir->od_container;
1595                 if (o->oo_inode != NULL) {
1596                         if (bag->ic_object == o->oo_inode)
1597                                 iam_container_fini(bag);
1598                 }
1599                 OBD_FREE_PTR(o->oo_dir);
1600                 o->oo_dir = NULL;
1601         }
1602 }
1603
1604 /*
1605  * Concurrency: no concurrent access is possible that late in object
1606  * life-cycle (for all existing callers, that is. New callers have to provide
1607  * their own locking.)
1608  */
1609 static int osd_inode_unlinked(const struct inode *inode)
1610 {
1611         return inode->i_nlink == 0;
1612 }
1613
1614 enum {
1615         OSD_TXN_OI_DELETE_CREDITS    = 20,
1616         OSD_TXN_INODE_DELETE_CREDITS = 20
1617 };
1618
1619 /*
1620  * Journal
1621  */
1622
1623 #if OSD_THANDLE_STATS
1624 /**
1625  * Set time when the handle is allocated
1626  */
1627 static void osd_th_alloced(struct osd_thandle *oth)
1628 {
1629         oth->oth_alloced = ktime_get();
1630 }
1631
1632 /**
1633  * Set time when the handle started
1634  */
1635 static void osd_th_started(struct osd_thandle *oth)
1636 {
1637         oth->oth_started = ktime_get();
1638 }
1639
1640 /**
1641  * Check whether the we deal with this handle for too long.
1642  */
1643 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1644                                 ktime_t alloced, ktime_t started,
1645                                 ktime_t closed)
1646 {
1647         ktime_t now = ktime_get();
1648
1649         LASSERT(dev != NULL);
1650
1651         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1652                             ktime_us_delta(started, alloced));
1653         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1654                             ktime_us_delta(closed, started));
1655         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1656                             ktime_us_delta(now, closed));
1657
1658         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1659                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1660                                 oth, now, alloced, started, closed);
1661                 libcfs_debug_dumpstack(NULL);
1662         }
1663 }
1664
1665 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1666 {                                                                       \
1667         ktime_t __closed = ktime_get();                                 \
1668         ktime_t __alloced = oth->oth_alloced;                           \
1669         ktime_t __started = oth->oth_started;                           \
1670                                                                         \
1671         expr;                                                           \
1672         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1673 }
1674
1675 #else /* OSD_THANDLE_STATS */
1676
1677 #define osd_th_alloced(h)                  do {} while(0)
1678 #define osd_th_started(h)                  do {} while(0)
1679 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1680
1681 #endif /* OSD_THANDLE_STATS */
1682
1683 /*
1684  * Concurrency: doesn't access mutable data.
1685  */
1686 static int osd_param_is_not_sane(const struct osd_device *dev,
1687                                  const struct thandle *th)
1688 {
1689         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1690
1691         return oh->ot_credits > osd_transaction_size(dev);
1692 }
1693
1694 /*
1695  * Concurrency: shouldn't matter.
1696  */
1697 static void osd_trans_commit_cb(struct super_block *sb,
1698                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1699 {
1700         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1701         struct thandle *th = &oh->ot_super;
1702         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1703         struct dt_txn_commit_cb *dcb, *tmp;
1704
1705         LASSERT(oh->ot_handle == NULL);
1706
1707         if (error)
1708                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1709
1710         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1711         /* call per-transaction callbacks if any */
1712         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1713                                  dcb_linkage) {
1714                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1715                          "commit callback entry: magic=%x name='%s'\n",
1716                          dcb->dcb_magic, dcb->dcb_name);
1717                 list_del_init(&dcb->dcb_linkage);
1718                 dcb->dcb_func(NULL, th, dcb, error);
1719         }
1720
1721         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1722         lu_device_put(lud);
1723         th->th_dev = NULL;
1724
1725         OBD_FREE_PTR(oh);
1726 }
1727
1728 #ifndef HAVE_SB_START_WRITE
1729 # define sb_start_write(sb) do {} while (0)
1730 # define sb_end_write(sb) do {} while (0)
1731 #endif
1732
1733 static struct thandle *osd_trans_create(const struct lu_env *env,
1734                                         struct dt_device *d)
1735 {
1736         struct osd_thread_info *oti = osd_oti_get(env);
1737         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1738         struct osd_thandle *oh;
1739         struct thandle *th;
1740
1741         ENTRY;
1742
1743         if (d->dd_rdonly) {
1744                 CERROR("%s: someone try to start transaction under "
1745                        "readonly mode, should be disabled.\n",
1746                        osd_name(osd_dt_dev(d)));
1747                 dump_stack();
1748                 RETURN(ERR_PTR(-EROFS));
1749         }
1750
1751         /* on pending IO in this thread should left from prev. request */
1752         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1753
1754         sb_start_write(osd_sb(osd_dt_dev(d)));
1755
1756         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1757         if (!oh) {
1758                 sb_end_write(osd_sb(osd_dt_dev(d)));
1759                 RETURN(ERR_PTR(-ENOMEM));
1760         }
1761
1762         oh->ot_quota_trans = &oti->oti_quota_trans;
1763         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1764         th = &oh->ot_super;
1765         th->th_dev = d;
1766         th->th_result = 0;
1767         oh->ot_credits = 0;
1768         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1769         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1770         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1771         osd_th_alloced(oh);
1772
1773         memset(oti->oti_declare_ops, 0,
1774                sizeof(oti->oti_declare_ops));
1775         memset(oti->oti_declare_ops_cred, 0,
1776                sizeof(oti->oti_declare_ops_cred));
1777         memset(oti->oti_declare_ops_used, 0,
1778                sizeof(oti->oti_declare_ops_used));
1779
1780         oti->oti_ins_cache_depth++;
1781
1782         RETURN(th);
1783 }
1784
1785 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1786 {
1787         struct osd_thread_info *oti = osd_oti_get(env);
1788         struct osd_thandle *oh;
1789
1790         oh = container_of0(th, struct osd_thandle, ot_super);
1791         LASSERT(oh != NULL);
1792
1793         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1794               oti->oti_declare_ops[OSD_OT_CREATE],
1795               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1796               oti->oti_declare_ops_used[OSD_OT_CREATE],
1797               oti->oti_declare_ops[OSD_OT_DESTROY],
1798               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1799               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1800         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1801               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1802               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1803               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1805               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1806               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1807         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1808               oti->oti_declare_ops[OSD_OT_WRITE],
1809               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1810               oti->oti_declare_ops_used[OSD_OT_WRITE],
1811               oti->oti_declare_ops[OSD_OT_PUNCH],
1812               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1813               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1814               oti->oti_declare_ops[OSD_OT_QUOTA],
1815               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1816               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1817         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_INSERT],
1819               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1820               oti->oti_declare_ops_used[OSD_OT_INSERT],
1821               oti->oti_declare_ops[OSD_OT_DELETE],
1822               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1823               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1824         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1825               oti->oti_declare_ops[OSD_OT_REF_ADD],
1826               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1827               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops[OSD_OT_REF_DEL],
1829               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1830               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1831 }
1832
1833 /*
1834  * Concurrency: shouldn't matter.
1835  */
1836 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1837                            struct thandle *th)
1838 {
1839         struct osd_thread_info *oti = osd_oti_get(env);
1840         struct osd_device *dev = osd_dt_dev(d);
1841         handle_t *jh;
1842         struct osd_thandle *oh;
1843         int rc;
1844
1845         ENTRY;
1846
1847         LASSERT(current->journal_info == NULL);
1848
1849         oh = container_of0(th, struct osd_thandle, ot_super);
1850         LASSERT(oh != NULL);
1851         LASSERT(oh->ot_handle == NULL);
1852
1853         rc = dt_txn_hook_start(env, d, th);
1854         if (rc != 0)
1855                 GOTO(out, rc);
1856
1857         if (unlikely(osd_param_is_not_sane(dev, th))) {
1858                 static unsigned long last_printed;
1859                 static int last_credits;
1860
1861                 /*
1862                  * don't make noise on a tiny testing systems
1863                  * actual credits misuse will be caught anyway
1864                  */
1865                 if (last_credits != oh->ot_credits &&
1866                     time_after(jiffies, last_printed +
1867                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1868                     osd_transaction_size(dev) > 512) {
1869                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1870                               oh->ot_credits, osd_transaction_size(dev));
1871                         osd_trans_dump_creds(env, th);
1872                         libcfs_debug_dumpstack(NULL);
1873                         last_credits = oh->ot_credits;
1874                         last_printed = jiffies;
1875                 }
1876                 /*
1877                  * XXX Limit the credits to 'max_transaction_buffers', and
1878                  *     let the underlying filesystem to catch the error if
1879                  *     we really need so many credits.
1880                  *
1881                  *     This should be removed when we can calculate the
1882                  *     credits precisely.
1883                  */
1884                 oh->ot_credits = osd_transaction_size(dev);
1885         } else if (ldiskfs_track_declares_assert != 0) {
1886                 /*
1887                  * reserve few credits to prevent an assertion in JBD
1888                  * our debugging mechanism will be able to detected
1889                  * overuse. this can help to debug single-update
1890                  * transactions
1891                  */
1892                 oh->ot_credits += 10;
1893                 if (unlikely(osd_param_is_not_sane(dev, th)))
1894                         oh->ot_credits = osd_transaction_size(dev);
1895         }
1896
1897         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1898                 GOTO(out, rc = -EIO);
1899
1900         /*
1901          * XXX temporary stuff. Some abstraction layer should
1902          * be used.
1903          */
1904         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1905         osd_th_started(oh);
1906         if (!IS_ERR(jh)) {
1907                 oh->ot_handle = jh;
1908                 LASSERT(oti->oti_txns == 0);
1909
1910                 lu_device_get(&d->dd_lu_dev);
1911                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1912                               "osd-tx", th);
1913                 oti->oti_txns++;
1914                 rc = 0;
1915         } else {
1916                 rc = PTR_ERR(jh);
1917         }
1918 out:
1919         RETURN(rc);
1920 }
1921
1922 static int osd_seq_exists(const struct lu_env *env,
1923                           struct osd_device *osd, u64 seq)
1924 {
1925         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1926         struct seq_server_site *ss = osd_seq_site(osd);
1927         int rc;
1928
1929         ENTRY;
1930
1931         LASSERT(ss != NULL);
1932         LASSERT(ss->ss_server_fld != NULL);
1933
1934         rc = osd_fld_lookup(env, osd, seq, range);
1935         if (rc != 0) {
1936                 if (rc != -ENOENT)
1937                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1938                                osd_name(osd), seq, rc);
1939                 RETURN(0);
1940         }
1941
1942         RETURN(ss->ss_node_id == range->lsr_index);
1943 }
1944
1945 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1946 {
1947         struct dt_txn_commit_cb *dcb;
1948         struct dt_txn_commit_cb *tmp;
1949
1950         /* call per-transaction stop callbacks if any */
1951         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1952                                  dcb_linkage) {
1953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1954                          "commit callback entry: magic=%x name='%s'\n",
1955                          dcb->dcb_magic, dcb->dcb_name);
1956                 list_del_init(&dcb->dcb_linkage);
1957                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1958         }
1959 }
1960
1961 /*
1962  * Concurrency: shouldn't matter.
1963  */
1964 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1965                           struct thandle *th)
1966 {
1967         struct osd_thread_info *oti = osd_oti_get(env);
1968         struct osd_thandle *oh;
1969         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1970         struct osd_device *osd = osd_dt_dev(th->th_dev);
1971         struct qsd_instance *qsd = osd_def_qsd(osd);
1972         struct lquota_trans *qtrans;
1973         struct list_head truncates = LIST_HEAD_INIT(truncates);
1974         int rc = 0, remove_agents = 0;
1975
1976         ENTRY;
1977
1978         oh = container_of0(th, struct osd_thandle, ot_super);
1979
1980         remove_agents = oh->ot_remove_agents;
1981
1982         qtrans = oh->ot_quota_trans;
1983         oh->ot_quota_trans = NULL;
1984
1985         /* move locks to local list, stop tx, execute truncates */
1986         list_splice(&oh->ot_trunc_locks, &truncates);
1987
1988         if (oh->ot_handle != NULL) {
1989                 int rc2;
1990
1991                 handle_t *hdl = oh->ot_handle;
1992
1993                 /*
1994                  * add commit callback
1995                  * notice we don't do this in osd_trans_start()
1996                  * as underlying transaction can change during truncate
1997                  */
1998                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1999                                              &oh->ot_jcb);
2000
2001                 LASSERT(oti->oti_txns == 1);
2002                 oti->oti_txns--;
2003
2004                 rc = dt_txn_hook_stop(env, th);
2005                 if (rc != 0)
2006                         CERROR("%s: failed in transaction hook: rc = %d\n",
2007                                osd_name(osd), rc);
2008
2009                 osd_trans_stop_cb(oh, rc);
2010                 /* hook functions might modify th_sync */
2011                 hdl->h_sync = th->th_sync;
2012
2013                 oh->ot_handle = NULL;
2014                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2015                 if (rc2 != 0)
2016                         CERROR("%s: failed to stop transaction: rc = %d\n",
2017                                osd_name(osd), rc2);
2018                 if (!rc)
2019                         rc = rc2;
2020
2021                 osd_process_truncates(&truncates);
2022         } else {
2023                 osd_trans_stop_cb(oh, th->th_result);
2024                 OBD_FREE_PTR(oh);
2025         }
2026
2027         osd_trunc_unlock_all(&truncates);
2028
2029         /* inform the quota slave device that the transaction is stopping */
2030         qsd_op_end(env, qsd, qtrans);
2031
2032         /*
2033          * as we want IO to journal and data IO be concurrent, we don't block
2034          * awaiting data IO completion in osd_do_bio(), instead we wait here
2035          * once transaction is submitted to the journal. all reqular requests
2036          * don't do direct IO (except read/write), thus this wait_event becomes
2037          * no-op for them.
2038          *
2039          * IMPORTANT: we have to wait till any IO submited by the thread is
2040          * completed otherwise iobuf may be corrupted by different request
2041          */
2042         wait_event(iobuf->dr_wait,
2043                        atomic_read(&iobuf->dr_numreqs) == 0);
2044         osd_fini_iobuf(osd, iobuf);
2045         if (!rc)
2046                 rc = iobuf->dr_error;
2047
2048         if (unlikely(remove_agents != 0))
2049                 osd_process_scheduled_agent_removals(env, osd);
2050
2051         oti->oti_ins_cache_depth--;
2052         /* reset OI cache for safety */
2053         if (oti->oti_ins_cache_depth == 0)
2054                 oti->oti_ins_cache_used = 0;
2055
2056         sb_end_write(osd_sb(osd));
2057
2058         RETURN(rc);
2059 }
2060
2061 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2062 {
2063         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2064                                                ot_super);
2065
2066         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2067         LASSERT(&dcb->dcb_func != NULL);
2068         if (dcb->dcb_flags & DCB_TRANS_STOP)
2069                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2070         else
2071                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2072
2073         return 0;
2074 }
2075
2076 /*
2077  * Called just before object is freed. Releases all resources except for
2078  * object itself (that is released by osd_object_free()).
2079  *
2080  * Concurrency: no concurrent access is possible that late in object
2081  * life-cycle.
2082  */
2083 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2084 {
2085         struct osd_object *obj = osd_obj(l);
2086         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2087         struct inode *inode = obj->oo_inode;
2088         __u64 projid;
2089         qid_t uid;
2090         qid_t gid;
2091
2092         LINVRNT(osd_invariant(obj));
2093
2094         /*
2095          * If object is unlinked remove fid->ino mapping from object index.
2096          */
2097
2098         osd_index_fini(obj);
2099
2100         if (!inode)
2101                 return;
2102
2103         uid = i_uid_read(inode);
2104         gid = i_gid_read(inode);
2105         projid = i_projid_read(inode);
2106
2107         obj->oo_inode = NULL;
2108         iput(inode);
2109
2110         /* do not rebalance quota if the caller needs to release memory
2111          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2112          * transaction and risk to deadlock - LU-12178 */
2113         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2114                 return;
2115
2116         if (!obj->oo_header && qsd) {
2117                 struct osd_thread_info *info = osd_oti_get(env);
2118                 struct lquota_id_info *qi = &info->oti_qi;
2119
2120                 /* Release granted quota to master if necessary */
2121                 qi->lqi_id.qid_uid = uid;
2122                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2123
2124                 qi->lqi_id.qid_uid = gid;
2125                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2126
2127                 qi->lqi_id.qid_uid = projid;
2128                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2129         }
2130 }
2131
2132 /*
2133  * Concurrency: ->loo_object_release() is called under site spin-lock.
2134  */
2135 static void osd_object_release(const struct lu_env *env,
2136                                struct lu_object *l)
2137 {
2138         struct osd_object *o = osd_obj(l);
2139
2140         /*
2141          * nobody should be releasing a non-destroyed object with nlink=0
2142          * the API allows this, but ldiskfs doesn't like and then report
2143          * this inode as deleted
2144          */
2145         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2146                   o->oo_inode->i_nlink == 0));
2147 }
2148
2149 /*
2150  * Concurrency: shouldn't matter.
2151  */
2152 static int osd_object_print(const struct lu_env *env, void *cookie,
2153                             lu_printer_t p, const struct lu_object *l)
2154 {
2155         struct osd_object *o = osd_obj(l);
2156         struct iam_descr *d;
2157
2158         if (o->oo_dir != NULL)
2159                 d = o->oo_dir->od_container.ic_descr;
2160         else
2161                 d = NULL;
2162         return (*p)(env, cookie,
2163                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2164                     o, o->oo_inode,
2165                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2166                     o->oo_inode ? o->oo_inode->i_generation : 0,
2167                     d ? d->id_ops->id_name : "plain");
2168 }
2169
2170 /*
2171  * Concurrency: shouldn't matter.
2172  */
2173 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2174                 struct obd_statfs *sfs)
2175 {
2176         struct osd_device *osd = osd_dt_dev(d);
2177         struct super_block *sb = osd_sb(osd);
2178         struct kstatfs *ksfs;
2179         __u64 reserved;
2180         int result = 0;
2181
2182         if (unlikely(osd->od_mnt == NULL))
2183                 return -EINPROGRESS;
2184
2185         /* osd_lproc.c call this without env, allocate ksfs for that case */
2186         if (unlikely(env == NULL)) {
2187                 OBD_ALLOC_PTR(ksfs);
2188                 if (ksfs == NULL)
2189                         return -ENOMEM;
2190         } else {
2191                 ksfs = &osd_oti_get(env)->oti_ksfs;
2192         }
2193
2194         result = sb->s_op->statfs(sb->s_root, ksfs);
2195         if (result)
2196                 goto out;
2197
2198         statfs_pack(sfs, ksfs);
2199         if (unlikely(sb->s_flags & MS_RDONLY))
2200                 sfs->os_state |= OS_STATE_READONLY;
2201
2202         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2203
2204         if (ldiskfs_has_feature_extents(sb))
2205                 sfs->os_maxbytes = sb->s_maxbytes;
2206         else
2207                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2208
2209         /*
2210          * Reserve some space so to avoid fragmenting the filesystem too much.
2211          * Fragmentation not only impacts performance, but can also increase
2212          * metadata overhead significantly, causing grant calculation to be
2213          * wrong.
2214          *
2215          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2216          */
2217         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2218         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2219         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2220                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2221
2222         sfs->os_blocks -= reserved;
2223         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2224         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2225
2226 out:
2227         if (unlikely(env == NULL))
2228                 OBD_FREE_PTR(ksfs);
2229         return result;
2230 }
2231
2232 /**
2233  * Estimate space needed for file creations. We assume the largest filename
2234  * which is 2^64 - 1, hence a filename of 20 chars.
2235  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2236  */
2237 #ifdef __LDISKFS_DIR_REC_LEN
2238 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2239 #else
2240 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2241 #endif
2242
2243 /*
2244  * Concurrency: doesn't access mutable data.
2245  */
2246 static void osd_conf_get(const struct lu_env *env,
2247                          const struct dt_device *dev,
2248                          struct dt_device_param *param)
2249 {
2250         struct osd_device *d = osd_dt_dev(dev);
2251         struct super_block *sb = osd_sb(d);
2252         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2253         const char *name;
2254         int ea_overhead;
2255
2256         /*
2257          * XXX should be taken from not-yet-existing fs abstraction layer.
2258          */
2259         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2260         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2261         param->ddp_symlink_max  = sb->s_blocksize;
2262         param->ddp_mount_type   = LDD_MT_LDISKFS;
2263         if (ldiskfs_has_feature_extents(sb))
2264                 param->ddp_maxbytes = sb->s_maxbytes;
2265         else
2266                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2267         /*
2268          * inode are statically allocated, so per-inode space consumption
2269          * is the space consumed by the directory entry
2270          */
2271         param->ddp_inodespace     = PER_OBJ_USAGE;
2272         /*
2273          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2274          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2275          * maximum length to not under-count the actual number of extents
2276          * needed for writing a file if there are sub-optimal block allocations.
2277          */
2278         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2279         /* worst-case extent insertion metadata overhead */
2280         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2281         param->ddp_mntopts = 0;
2282         if (test_opt(sb, XATTR_USER))
2283                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2284         if (test_opt(sb, POSIX_ACL))
2285                 param->ddp_mntopts |= MNTOPT_ACL;
2286
2287         /*
2288          * LOD might calculate the max stripe count based on max_ea_size,
2289          * so we need take account in the overhead as well,
2290          * xattr_header + magic + xattr_entry_head
2291          */
2292         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2293                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2294
2295 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2296         if (ldiskfs_has_feature_ea_inode(sb))
2297                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2298                                                                 ea_overhead;
2299         else
2300 #endif
2301                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2302
2303         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE - ea_overhead)
2304                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE - ea_overhead;
2305
2306         /*
2307          * Preferred RPC size for efficient disk IO.  4MB shows good
2308          * all-around performance for ldiskfs, but use bigalloc chunk size
2309          * by default if larger.
2310          */
2311 #if defined(LDISKFS_CLUSTER_SIZE)
2312         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2313                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2314         else
2315 #endif
2316                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2317
2318         param->ddp_t10_cksum_type = 0;
2319         if (bi) {
2320                 unsigned short interval = blk_integrity_interval(bi);
2321                 name = blk_integrity_name(bi);
2322                 /*
2323                  * Expected values:
2324                  * T10-DIF-TYPE1-CRC
2325                  * T10-DIF-TYPE3-CRC
2326                  * T10-DIF-TYPE1-IP
2327                  * T10-DIF-TYPE3-IP
2328                  */
2329                 if (strncmp(name, "T10-DIF-TYPE",
2330                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2331                         /* also skip "1/3-" at end */
2332                         const int type_off = sizeof("T10-DIF-TYPE.");
2333                         char type_number = name[type_off - 2];
2334
2335                         if (interval != 512 && interval != 4096) {
2336                                 CERROR("%s: unsupported T10PI sector size %u\n",
2337                                        d->od_svname, interval);
2338                         } else if (type_number != '1' && type_number != '3') {
2339                                 CERROR("%s: unsupported T10PI type %s\n",
2340                                        d->od_svname, name);
2341                         } else if (strcmp(name + type_off, "CRC") == 0) {
2342                                 d->od_t10_type = type_number == '1' ?
2343                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2344                                 param->ddp_t10_cksum_type = interval == 512 ?
2345                                         OBD_CKSUM_T10CRC512 :
2346                                         OBD_CKSUM_T10CRC4K;
2347                         } else if (strcmp(name + type_off, "IP") == 0) {
2348                                 d->od_t10_type = type_number == '1' ?
2349                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2350                                 param->ddp_t10_cksum_type = interval == 512 ?
2351                                         OBD_CKSUM_T10IP512 :
2352                                         OBD_CKSUM_T10IP4K;
2353                         } else {
2354                                 CERROR("%s: unsupported checksum type of "
2355                                        "T10PI type '%s'",
2356                                        d->od_svname, name);
2357                         }
2358
2359                 } else {
2360                         CERROR("%s: unsupported T10PI type '%s'",
2361                                d->od_svname, name);
2362                 }
2363         }
2364 }
2365
2366 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2367 {
2368         return osd_sb(osd_dt_dev(d));
2369 }
2370
2371 /*
2372  * Concurrency: shouldn't matter.
2373  */
2374 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2375 {
2376         int rc;
2377         struct super_block *s = osd_sb(osd_dt_dev(d));
2378         ENTRY;
2379
2380         down_read(&s->s_umount);
2381         rc = s->s_op->sync_fs(s, 1);
2382         up_read(&s->s_umount);
2383
2384         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2385                rc);
2386
2387         return rc;
2388 }
2389
2390 /**
2391  * Start commit for OSD device.
2392  *
2393  * An implementation of dt_commit_async method for OSD device.
2394  * Asychronously starts underlayng fs sync and thereby a transaction
2395  * commit.
2396  *
2397  * \param env environment
2398  * \param d dt device
2399  *
2400  * \see dt_device_operations
2401  */
2402 static int osd_commit_async(const struct lu_env *env,
2403                             struct dt_device *d)
2404 {
2405         struct super_block *s = osd_sb(osd_dt_dev(d));
2406         int rc;
2407
2408         ENTRY;
2409
2410         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2411         down_read(&s->s_umount);
2412         rc = s->s_op->sync_fs(s, 0);
2413         up_read(&s->s_umount);
2414
2415         RETURN(rc);
2416 }
2417
2418 /* Our own copy of the set readonly functions if present, or NU if not. */
2419 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2420 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2421 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2422
2423 /*
2424  * Concurrency: shouldn't matter.
2425  */
2426 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2427 {
2428         struct super_block *sb = osd_sb(osd_dt_dev(d));
2429         struct block_device *dev = sb->s_bdev;
2430         int rc = -EOPNOTSUPP;
2431
2432         ENTRY;
2433
2434         if (priv_dev_set_rdonly) {
2435                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2436
2437                 rc = 0;
2438                 CERROR("*** setting %s read-only ***\n",
2439                        osd_dt_dev(d)->od_svname);
2440
2441                 if (sb->s_op->freeze_fs) {
2442                         rc = sb->s_op->freeze_fs(sb);
2443                         if (rc)
2444                                 goto out;
2445                 }
2446
2447                 if (jdev && (jdev != dev)) {
2448                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2449                                (long)jdev);
2450                         priv_dev_set_rdonly(jdev);
2451                 }
2452                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2453                 priv_dev_set_rdonly(dev);
2454
2455                 if (sb->s_op->unfreeze_fs)
2456                         sb->s_op->unfreeze_fs(sb);
2457         }
2458
2459 out:
2460         if (rc)
2461                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2462                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2463
2464         RETURN(rc);
2465 }
2466
2467 /**
2468  * Note: we do not count into QUOTA here.
2469  * If we mount with --data_journal we may need more.
2470  */
2471 const int osd_dto_credits_noquota[DTO_NR] = {
2472         /**
2473          * Insert.
2474          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2475          * SINGLEDATA_TRANS_BLOCKS(8)
2476          * XXX Note: maybe iam need more, since iam have more level than
2477          *           EXT3 htree.
2478          */
2479         [DTO_INDEX_INSERT]  = 16,
2480         /**
2481          * Delete
2482          * just modify a single entry, probably merge few within a block
2483          */
2484         [DTO_INDEX_DELETE]  = 1,
2485         /**
2486          * Used for OI scrub
2487          */
2488         [DTO_INDEX_UPDATE]  = 16,
2489         /**
2490          * 4(inode, inode bits, groups, GDT)
2491          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2492          */
2493         [DTO_OBJECT_CREATE] = 4,
2494         /**
2495          * 4(inode, inode bits, groups, GDT)
2496          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2497          */
2498         [DTO_OBJECT_DELETE] = 4,
2499         /**
2500          * Attr set credits (inode)
2501          */
2502         [DTO_ATTR_SET_BASE] = 1,
2503         /**
2504          * Xattr set. The same as xattr of EXT3.
2505          * DATA_TRANS_BLOCKS(14)
2506          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2507          * are also counted in. Do not know why?
2508          */
2509         [DTO_XATTR_SET]     = 14,
2510         /**
2511          * credits for inode change during write.
2512          */
2513         [DTO_WRITE_BASE]    = 3,
2514         /**
2515          * credits for single block write.
2516          */
2517         [DTO_WRITE_BLOCK]   = 14,
2518         /**
2519          * Attr set credits for chown.
2520          * This is extra credits for setattr, and it is null without quota
2521          */
2522         [DTO_ATTR_SET_CHOWN] = 0
2523 };
2524
2525 static const struct dt_device_operations osd_dt_ops = {
2526         .dt_root_get       = osd_root_get,
2527         .dt_statfs         = osd_statfs,
2528         .dt_trans_create   = osd_trans_create,
2529         .dt_trans_start    = osd_trans_start,
2530         .dt_trans_stop     = osd_trans_stop,
2531         .dt_trans_cb_add   = osd_trans_cb_add,
2532         .dt_conf_get       = osd_conf_get,
2533         .dt_mnt_sb_get     = osd_mnt_sb_get,
2534         .dt_sync           = osd_sync,
2535         .dt_ro             = osd_ro,
2536         .dt_commit_async   = osd_commit_async,
2537 };
2538
2539 static void osd_read_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_read_nested(&obj->oo_sem, role);
2549
2550         LASSERT(obj->oo_owner == NULL);
2551         oti->oti_r_locks++;
2552 }
2553
2554 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2555                            unsigned int role)
2556 {
2557         struct osd_object *obj = osd_dt_obj(dt);
2558         struct osd_thread_info *oti = osd_oti_get(env);
2559
2560         LINVRNT(osd_invariant(obj));
2561
2562         LASSERT(obj->oo_owner != env);
2563         down_write_nested(&obj->oo_sem, role);
2564
2565         LASSERT(obj->oo_owner == NULL);
2566         obj->oo_owner = env;
2567         oti->oti_w_locks++;
2568 }
2569
2570 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2571 {
2572         struct osd_object *obj = osd_dt_obj(dt);
2573         struct osd_thread_info *oti = osd_oti_get(env);
2574
2575         LINVRNT(osd_invariant(obj));
2576
2577         LASSERT(oti->oti_r_locks > 0);
2578         oti->oti_r_locks--;
2579         up_read(&obj->oo_sem);
2580 }
2581
2582 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2583 {
2584         struct osd_object *obj = osd_dt_obj(dt);
2585         struct osd_thread_info *oti = osd_oti_get(env);
2586
2587         LINVRNT(osd_invariant(obj));
2588
2589         LASSERT(obj->oo_owner == env);
2590         LASSERT(oti->oti_w_locks > 0);
2591         oti->oti_w_locks--;
2592         obj->oo_owner = NULL;
2593         up_write(&obj->oo_sem);
2594 }
2595
2596 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2597 {
2598         struct osd_object *obj = osd_dt_obj(dt);
2599
2600         LINVRNT(osd_invariant(obj));
2601
2602         return obj->oo_owner == env;
2603 }
2604
2605 static struct timespec *osd_inode_time(const struct lu_env *env,
2606                                        struct inode *inode, __u64 seconds)
2607 {
2608         struct osd_thread_info *oti = osd_oti_get(env);
2609         struct timespec *t = &oti->oti_time;
2610
2611         t->tv_sec = seconds;
2612         t->tv_nsec = 0;
2613         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2614         return t;
2615 }
2616
2617 static void osd_inode_getattr(const struct lu_env *env,
2618                               struct inode *inode, struct lu_attr *attr)
2619 {
2620         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2621                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2622                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2623                            LA_BLKSIZE | LA_TYPE;
2624
2625         attr->la_atime = inode->i_atime.tv_sec;
2626         attr->la_mtime = inode->i_mtime.tv_sec;
2627         attr->la_ctime = inode->i_ctime.tv_sec;
2628         attr->la_mode    = inode->i_mode;
2629         attr->la_size    = i_size_read(inode);
2630         attr->la_blocks  = inode->i_blocks;
2631         attr->la_uid     = i_uid_read(inode);
2632         attr->la_gid     = i_gid_read(inode);
2633         attr->la_projid  = i_projid_read(inode);
2634         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2635         attr->la_nlink   = inode->i_nlink;
2636         attr->la_rdev    = inode->i_rdev;
2637         attr->la_blksize = 1 << inode->i_blkbits;
2638         attr->la_blkbits = inode->i_blkbits;
2639         /*
2640          * Ext4 did not transfer inherit flags from raw inode
2641          * to inode flags, and ext4 internally test raw inode
2642          * @i_flags directly. Instead of patching ext4, we do it here.
2643          */
2644         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2645                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2646 }
2647
2648 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2649                         struct lu_attr *attr)
2650 {
2651         struct osd_object *obj = osd_dt_obj(dt);
2652
2653         if (unlikely(!dt_object_exists(dt)))
2654                 return -ENOENT;
2655         if (unlikely(obj->oo_destroyed))
2656                 return -ENOENT;
2657
2658         LASSERT(!dt_object_remote(dt));
2659         LINVRNT(osd_invariant(obj));
2660
2661         spin_lock(&obj->oo_guard);
2662         osd_inode_getattr(env, obj->oo_inode, attr);
2663         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2664                 attr->la_valid |= LA_FLAGS;
2665                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2666         }
2667         spin_unlock(&obj->oo_guard);
2668
2669         return 0;
2670 }
2671
2672 static int osd_declare_attr_qid(const struct lu_env *env,
2673                                 struct osd_object *obj,
2674                                 struct osd_thandle *oh, long long bspace,
2675                                 qid_t old_id, qid_t new_id, bool enforce,
2676                                 unsigned int type, bool ignore_edquot)
2677 {
2678         int rc;
2679         struct osd_thread_info *info = osd_oti_get(env);
2680         struct lquota_id_info  *qi = &info->oti_qi;
2681
2682         qi->lqi_type = type;
2683         /* inode accounting */
2684         qi->lqi_is_blk = false;
2685
2686         /* one more inode for the new id ... */
2687         qi->lqi_id.qid_uid = new_id;
2688         qi->lqi_space      = 1;
2689         /* Reserve credits for the new id */
2690         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2691         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2692                 rc = 0;
2693         if (rc)
2694                 RETURN(rc);
2695
2696         /* and one less inode for the current id */
2697         qi->lqi_id.qid_uid = old_id;
2698         qi->lqi_space = -1;
2699         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2700         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2701                 rc = 0;
2702         if (rc)
2703                 RETURN(rc);
2704
2705         /* block accounting */
2706         qi->lqi_is_blk = true;
2707
2708         /* more blocks for the new id ... */
2709         qi->lqi_id.qid_uid = new_id;
2710         qi->lqi_space      = bspace;
2711         /*
2712          * Credits for the new uid has been reserved, re-use "obj"
2713          * to save credit reservation.
2714          */
2715         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2716         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2717                 rc = 0;
2718         if (rc)
2719                 RETURN(rc);
2720
2721         /* and finally less blocks for the current uid */
2722         qi->lqi_id.qid_uid = old_id;
2723         qi->lqi_space      = -bspace;
2724         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2725         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2726                 rc = 0;
2727
2728         RETURN(rc);
2729 }
2730
2731 static int osd_declare_attr_set(const struct lu_env *env,
2732                                 struct dt_object *dt,
2733                                 const struct lu_attr *attr,
2734                                 struct thandle *handle)
2735 {
2736         struct osd_thandle *oh;
2737         struct osd_object *obj;
2738         qid_t uid;
2739         qid_t gid;
2740         long long bspace;
2741         int rc = 0;
2742         bool enforce;
2743
2744         ENTRY;
2745
2746         LASSERT(dt != NULL);
2747         LASSERT(handle != NULL);
2748
2749         obj = osd_dt_obj(dt);
2750         LASSERT(osd_invariant(obj));
2751
2752         oh = container_of0(handle, struct osd_thandle, ot_super);
2753         LASSERT(oh->ot_handle == NULL);
2754
2755         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2756                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2757
2758         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2759                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2760
2761         if (attr == NULL || obj->oo_inode == NULL)
2762                 RETURN(rc);
2763
2764         bspace   = obj->oo_inode->i_blocks << 9;
2765         bspace   = toqb(bspace);
2766
2767         /*
2768          * Changing ownership is always preformed by super user, it should not
2769          * fail with EDQUOT unless required explicitly.
2770          *
2771          * We still need to call the osd_declare_qid() to calculate the journal
2772          * credits for updating quota accounting files and to trigger quota
2773          * space adjustment once the operation is completed.
2774          */
2775         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2776                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2777
2778                 if (!ignore_edquot)
2779                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2780                                "(the quota space is %lld)\n",
2781                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2782                                attr->la_gid, bspace);
2783
2784                 /* USERQUOTA */
2785                 uid = i_uid_read(obj->oo_inode);
2786                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2787                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2788                                           attr->la_uid, enforce, USRQUOTA,
2789                                           true);
2790                 if (rc)
2791                         RETURN(rc);
2792
2793                 gid = i_gid_read(obj->oo_inode);
2794                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2795                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2796                                           i_gid_read(obj->oo_inode),
2797                                           attr->la_gid, enforce, GRPQUOTA,
2798                                           ignore_edquot);
2799                 if (rc)
2800                         RETURN(rc);
2801
2802         }
2803 #ifdef HAVE_PROJECT_QUOTA
2804         if (attr->la_valid & LA_PROJID) {
2805                 __u32 projid = i_projid_read(obj->oo_inode);
2806
2807                 enforce = (attr->la_valid & LA_PROJID) &&
2808                                         (attr->la_projid != projid);
2809                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2810                                           (qid_t)projid, (qid_t)attr->la_projid,
2811                                           enforce, PRJQUOTA, true);
2812                 if (rc)
2813                         RETURN(rc);
2814         }
2815 #endif
2816         RETURN(rc);
2817 }
2818
2819 static int osd_inode_setattr(const struct lu_env *env,
2820                              struct inode *inode, const struct lu_attr *attr)
2821 {
2822         __u64 bits = attr->la_valid;
2823
2824         /* Only allow set size for regular file */
2825         if (!S_ISREG(inode->i_mode))
2826                 bits &= ~(LA_SIZE | LA_BLOCKS);
2827
2828         if (bits == 0)
2829                 return 0;
2830
2831         if (bits & LA_ATIME)
2832                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2833         if (bits & LA_CTIME)
2834                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2835         if (bits & LA_MTIME)
2836                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2837         if (bits & LA_SIZE) {
2838                 spin_lock(&inode->i_lock);
2839                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2840                 i_size_write(inode, attr->la_size);
2841                 spin_unlock(&inode->i_lock);
2842         }
2843
2844         /*
2845          * OSD should not change "i_blocks" which is used by quota.
2846          * "i_blocks" should be changed by ldiskfs only.
2847          */
2848         if (bits & LA_MODE)
2849                 inode->i_mode = (inode->i_mode & S_IFMT) |
2850                                 (attr->la_mode & ~S_IFMT);
2851         if (bits & LA_UID)
2852                 i_uid_write(inode, attr->la_uid);
2853         if (bits & LA_GID)
2854                 i_gid_write(inode, attr->la_gid);
2855         if (bits & LA_PROJID)
2856                 i_projid_write(inode, attr->la_projid);
2857         if (bits & LA_NLINK)
2858                 set_nlink(inode, attr->la_nlink);
2859         if (bits & LA_RDEV)
2860                 inode->i_rdev = attr->la_rdev;
2861
2862         if (bits & LA_FLAGS) {
2863                 /* always keep S_NOCMTIME */
2864                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2865                                  S_NOCMTIME;
2866                 /*
2867                  * Ext4 did not transfer inherit flags from
2868                  * @inode->i_flags to raw inode i_flags when writing
2869                  * flags, we do it explictly here.
2870                  */
2871                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2872                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2873                 else
2874                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2875         }
2876         return 0;
2877 }
2878
2879 #ifdef HAVE_PROJECT_QUOTA
2880 static int osd_transfer_project(struct inode *inode, __u32 projid)
2881 {
2882         struct super_block *sb = inode->i_sb;
2883         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2884         int err;
2885         kprojid_t kprojid;
2886         struct ldiskfs_iloc iloc;
2887         struct ldiskfs_inode *raw_inode;
2888         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2889
2890         if (!ldiskfs_has_feature_project(sb)) {
2891                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2892                         == LDISKFS_DEF_PROJID);
2893                 if (projid != LDISKFS_DEF_PROJID)
2894                         return -EOPNOTSUPP;
2895                 else
2896                         return 0;
2897         }
2898
2899         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2900                 return -EOPNOTSUPP;
2901
2902         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2903         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2904                 return 0;
2905
2906         err = ldiskfs_get_inode_loc(inode, &iloc);
2907         if (err)
2908                 return err;
2909
2910         raw_inode = ldiskfs_raw_inode(&iloc);
2911         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2912                 err = -EOVERFLOW;
2913                 brelse(iloc.bh);
2914                 return err;
2915         }
2916         brelse(iloc.bh);
2917
2918         dquot_initialize(inode);
2919         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2920         if (transfer_to[PRJQUOTA]) {
2921                 err = __dquot_transfer(inode, transfer_to);
2922                 dqput(transfer_to[PRJQUOTA]);
2923                 if (err)
2924                         return err;
2925         }
2926
2927         return err;
2928 }
2929 #endif
2930
2931 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2932 {
2933         int rc;
2934
2935         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2936             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2937                 struct iattr iattr;
2938
2939                 ll_vfs_dq_init(inode);
2940                 iattr.ia_valid = 0;
2941                 if (attr->la_valid & LA_UID)
2942                         iattr.ia_valid |= ATTR_UID;
2943                 if (attr->la_valid & LA_GID)
2944                         iattr.ia_valid |= ATTR_GID;
2945                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2946                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2947
2948                 rc = ll_vfs_dq_transfer(inode, &iattr);
2949                 if (rc) {
2950                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2951                                "enforcement enabled on the ldiskfs "
2952                                "filesystem?\n", inode->i_sb->s_id, rc);
2953                         return rc;
2954                 }
2955         }
2956
2957         /* Handle project id transfer here properly */
2958         if (attr->la_valid & LA_PROJID &&
2959             attr->la_projid != i_projid_read(inode)) {
2960 #ifdef HAVE_PROJECT_QUOTA
2961                 rc = osd_transfer_project(inode, attr->la_projid);
2962 #else
2963                 rc = -ENOTSUPP;
2964 #endif
2965                 if (rc) {
2966                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2967                                "enforcement enabled on the ldiskfs "
2968                                "filesystem?\n", inode->i_sb->s_id, rc);
2969                         return rc;
2970                 }
2971         }
2972         return 0;
2973 }
2974
2975 static int osd_attr_set(const struct lu_env *env,
2976                         struct dt_object *dt,
2977                         const struct lu_attr *attr,
2978                         struct thandle *handle)
2979 {
2980         struct osd_object *obj = osd_dt_obj(dt);
2981         struct inode *inode;
2982         int rc;
2983
2984         if (!dt_object_exists(dt))
2985                 return -ENOENT;
2986
2987         LASSERT(handle != NULL);
2988         LASSERT(!dt_object_remote(dt));
2989         LASSERT(osd_invariant(obj));
2990
2991         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2992
2993         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2994             !osd_obj2dev(obj)->od_is_ost) {
2995                 struct osd_thread_info *oti = osd_oti_get(env);
2996                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2997                 struct lu_fid *fid1 = &oti->oti_fid;
2998                 struct osd_inode_id *id = &oti->oti_id;
2999                 struct iam_path_descr *ipd;
3000                 struct iam_container *bag;
3001                 struct osd_thandle *oh;
3002                 int rc;
3003
3004                 fid_cpu_to_be(fid1, fid0);
3005                 memset(id, 1, sizeof(*id));
3006                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3007                                   fid0)->oi_dir.od_container;
3008                 ipd = osd_idx_ipd_get(env, bag);
3009                 if (unlikely(ipd == NULL))
3010                         RETURN(-ENOMEM);
3011
3012                 oh = container_of0(handle, struct osd_thandle, ot_super);
3013                 rc = iam_update(oh->ot_handle, bag,
3014                                 (const struct iam_key *)fid1,
3015                                 (const struct iam_rec *)id, ipd);
3016                 osd_ipd_put(env, bag, ipd);
3017                 return(rc > 0 ? 0 : rc);
3018         }
3019
3020         inode = obj->oo_inode;
3021
3022         rc = osd_quota_transfer(inode, attr);
3023         if (rc)
3024                 return rc;
3025
3026         spin_lock(&obj->oo_guard);
3027         rc = osd_inode_setattr(env, inode, attr);
3028         spin_unlock(&obj->oo_guard);
3029         if (rc != 0)
3030                 GOTO(out, rc);
3031
3032         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3033
3034         if (!(attr->la_valid & LA_FLAGS))
3035                 GOTO(out, rc);
3036
3037         /* Let's check if there are extra flags need to be set into LMA */
3038         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3039                 struct osd_thread_info *info = osd_oti_get(env);
3040                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3041
3042                 LASSERT(!obj->oo_pfid_in_lma);
3043
3044                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3045                                  &info->oti_ost_attrs);
3046                 if (rc)
3047                         GOTO(out, rc);
3048
3049                 lma->lma_incompat |=
3050                         lustre_to_lma_flags(attr->la_flags);
3051                 lustre_lma_swab(lma);
3052                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3053                                      lma, sizeof(*lma), XATTR_REPLACE);
3054                 if (rc != 0) {
3055                         struct osd_device *osd = osd_obj2dev(obj);
3056
3057                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3058                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3059                               lma->lma_incompat, rc);
3060                 } else {
3061                         obj->oo_lma_flags =
3062                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3063                 }
3064                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3065         }
3066 out:
3067         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3068
3069         return rc;
3070 }
3071
3072 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3073                                            struct osd_object *obj,
3074                                            const char *name, const int namelen)
3075 {
3076         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3077 }
3078
3079 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3080                       umode_t mode, struct dt_allocation_hint *hint,
3081                       struct thandle *th, struct lu_attr *attr)
3082 {
3083         int result;
3084         struct osd_device *osd = osd_obj2dev(obj);
3085         struct osd_thandle *oth;
3086         struct dt_object *parent = NULL;
3087         struct inode *inode;
3088         uid_t owner[2] = {0, 0};
3089
3090         if (attr->la_valid & LA_UID)
3091                 owner[0] = attr->la_uid;
3092         if (attr->la_valid & LA_GID)
3093                 owner[1] = attr->la_gid;
3094
3095         LINVRNT(osd_invariant(obj));
3096         LASSERT(obj->oo_inode == NULL);
3097         LASSERT(obj->oo_hl_head == NULL);
3098
3099         if (S_ISDIR(mode) && ldiskfs_pdo) {
3100                 obj->oo_hl_head =
3101                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3102                 if (obj->oo_hl_head == NULL)
3103                         return -ENOMEM;
3104         }
3105
3106         oth = container_of(th, struct osd_thandle, ot_super);
3107         LASSERT(oth->ot_handle->h_transaction != NULL);
3108
3109         if (hint != NULL && hint->dah_parent != NULL &&
3110             !dt_object_remote(hint->dah_parent))
3111                 parent = hint->dah_parent;
3112
3113         inode = ldiskfs_create_inode(oth->ot_handle,
3114                                      parent ? osd_dt_obj(parent)->oo_inode :
3115                                               osd_sb(osd)->s_root->d_inode,
3116                                      mode, owner);
3117         if (!IS_ERR(inode)) {
3118                 /* Do not update file c/mtime in ldiskfs. */
3119                 inode->i_flags |= S_NOCMTIME;
3120
3121                 /*
3122                  * For new created object, it must be consistent,
3123                  * and it is unnecessary to scrub against it.
3124                  */
3125                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3126
3127                 obj->oo_inode = inode;
3128                 result = 0;
3129         } else {
3130                 if (obj->oo_hl_head != NULL) {
3131                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3132                         obj->oo_hl_head = NULL;
3133                 }
3134                 result = PTR_ERR(inode);
3135         }
3136         LINVRNT(osd_invariant(obj));
3137         return result;
3138 }
3139
3140 enum {
3141         OSD_NAME_LEN = 255
3142 };
3143
3144 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3145                      struct lu_attr *attr,
3146                      struct dt_allocation_hint *hint,
3147                      struct dt_object_format *dof,
3148                      struct thandle *th)
3149 {
3150         int result;
3151         struct osd_thandle *oth;
3152         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3153
3154         LASSERT(S_ISDIR(attr->la_mode));
3155
3156         oth = container_of(th, struct osd_thandle, ot_super);
3157         LASSERT(oth->ot_handle->h_transaction != NULL);
3158         result = osd_mkfile(info, obj, mode, hint, th, attr);
3159
3160         return result;
3161 }
3162
3163 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3164                         struct lu_attr *attr,
3165                         struct dt_allocation_hint *hint,
3166                         struct dt_object_format *dof,
3167                         struct thandle *th)
3168 {
3169         int result;
3170         struct osd_thandle *oth;
3171         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3172
3173         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3174
3175         LASSERT(S_ISREG(attr->la_mode));
3176
3177         oth = container_of(th, struct osd_thandle, ot_super);
3178         LASSERT(oth->ot_handle->h_transaction != NULL);
3179
3180         result = osd_mkfile(info, obj, mode, hint, th, attr);
3181         if (result == 0) {
3182                 LASSERT(obj->oo_inode != NULL);
3183                 if (feat->dif_flags & DT_IND_VARKEY)
3184                         result = iam_lvar_create(obj->oo_inode,
3185                                                  feat->dif_keysize_max,
3186                                                  feat->dif_ptrsize,
3187                                                  feat->dif_recsize_max,
3188                                                  oth->ot_handle);
3189                 else
3190                         result = iam_lfix_create(obj->oo_inode,
3191                                                  feat->dif_keysize_max,
3192                                                  feat->dif_ptrsize,
3193                                                  feat->dif_recsize_max,
3194                                                  oth->ot_handle);
3195         }
3196         return result;
3197 }
3198
3199 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3200                      struct lu_attr *attr,
3201                      struct dt_allocation_hint *hint,
3202                      struct dt_object_format *dof,
3203                      struct thandle *th)
3204 {
3205         LASSERT(S_ISREG(attr->la_mode));
3206         return osd_mkfile(info, obj, (attr->la_mode &
3207                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3208                           attr);
3209 }
3210
3211 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3212                      struct lu_attr *attr,
3213                      struct dt_allocation_hint *hint,
3214                      struct dt_object_format *dof,
3215                      struct thandle *th)
3216 {
3217         LASSERT(S_ISLNK(attr->la_mode));
3218         return osd_mkfile(info, obj, (attr->la_mode &
3219                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3220                           attr);
3221 }
3222
3223 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3224                      struct lu_attr *attr,
3225                      struct dt_allocation_hint *hint,
3226                      struct dt_object_format *dof,
3227                      struct thandle *th)
3228 {
3229         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3230         int result;
3231
3232         LINVRNT(osd_invariant(obj));
3233         LASSERT(obj->oo_inode == NULL);
3234         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3235                 S_ISFIFO(mode) || S_ISSOCK(mode));
3236
3237         result = osd_mkfile(info, obj, mode, hint, th, attr);
3238         if (result == 0) {
3239                 LASSERT(obj->oo_inode != NULL);
3240                 /*
3241                  * This inode should be marked dirty for i_rdev.  Currently
3242                  * that is done in the osd_attr_init().
3243                  */
3244                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3245                                    attr->la_rdev);
3246         }
3247         LINVRNT(osd_invariant(obj));
3248         return result;
3249 }
3250
3251 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3252                               struct lu_attr *,
3253                               struct dt_allocation_hint *hint,
3254                               struct dt_object_format *dof,
3255                               struct thandle *);
3256
3257 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3258 {
3259         osd_obj_type_f result;
3260
3261         switch (type) {
3262         case DFT_DIR:
3263                 result = osd_mkdir;
3264                 break;
3265         case DFT_REGULAR:
3266                 result = osd_mkreg;
3267                 break;
3268         case DFT_SYM:
3269                 result = osd_mksym;
3270                 break;
3271         case DFT_NODE:
3272                 result = osd_mknod;
3273                 break;
3274         case DFT_INDEX:
3275                 result = osd_mk_index;
3276                 break;
3277
3278         default:
3279                 LBUG();
3280                 break;
3281         }
3282         return result;
3283 }
3284
3285
3286 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3287                         struct dt_object *parent, struct dt_object *child,
3288                         umode_t child_mode)
3289 {
3290         LASSERT(ah);
3291
3292         ah->dah_parent = parent;
3293         ah->dah_mode = child_mode;
3294
3295         if (parent != NULL && !dt_object_remote(parent)) {
3296                 /* will help to find FID->ino at dt_insert("..") */
3297                 struct osd_object *pobj = osd_dt_obj(parent);
3298
3299                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3300         }
3301 }
3302
3303 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3304                           struct lu_attr *attr, struct dt_object_format *dof)
3305 {
3306         struct inode *inode = obj->oo_inode;
3307         __u64 valid = attr->la_valid;
3308         int result;
3309
3310         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3311
3312         if (dof->dof_type != DFT_NODE)
3313                 attr->la_valid &= ~LA_RDEV;
3314         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3315                 attr->la_valid &= ~LA_ATIME;
3316         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3317                 attr->la_valid &= ~LA_CTIME;
3318         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3319                 attr->la_valid &= ~LA_MTIME;
3320
3321         result = osd_quota_transfer(inode, attr);
3322         if (result)
3323                 return;
3324
3325         if (attr->la_valid != 0) {
3326                 result = osd_inode_setattr(info->oti_env, inode, attr);
3327                 /*
3328                  * The osd_inode_setattr() should always succeed here.  The
3329                  * only error that could be returned is EDQUOT when we are
3330                  * trying to change the UID or GID of the inode. However, this
3331                  * should not happen since quota enforcement is no longer
3332                  * enabled on ldiskfs (lquota takes care of it).
3333                  */
3334                 LASSERTF(result == 0, "%d\n", result);
3335                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3336         }
3337
3338         attr->la_valid = valid;
3339 }
3340
3341 /**
3342  * Helper function for osd_create()
3343  *
3344  * \retval 0, on success
3345  */
3346 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3347                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3348                         struct dt_object_format *dof, struct thandle *th)
3349 {
3350         int result;
3351         __u32 umask;
3352
3353         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3354
3355         /* we drop umask so that permissions we pass are not affected */
3356         umask = current->fs->umask;
3357         current->fs->umask = 0;
3358
3359         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3360                                                   th);
3361         if (likely(obj->oo_inode != NULL)) {
3362                 LASSERT(obj->oo_inode->i_state & I_NEW);
3363
3364                 /*
3365                  * Unlock the inode before attr initialization to avoid
3366                  * unnecessary dqget operations. LU-6378
3367                  */
3368                 unlock_new_inode(obj->oo_inode);
3369         }
3370
3371         if (likely(result == 0)) {
3372                 osd_attr_init(info, obj, attr, dof);
3373                 osd_object_init0(obj);
3374         }
3375
3376         /* restore previous umask value */
3377         current->fs->umask = umask;
3378
3379         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3380
3381         return result;
3382 }
3383
3384 /**
3385  * Helper function for osd_create()
3386  *
3387  * \retval 0, on success
3388  */
3389 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3390                            const struct lu_fid *fid, struct thandle *th)
3391 {
3392         struct osd_thread_info *info = osd_oti_get(env);
3393         struct osd_inode_id    *id   = &info->oti_id;
3394         struct osd_device      *osd  = osd_obj2dev(obj);
3395         struct osd_thandle     *oh;
3396         int rc;
3397
3398         LASSERT(obj->oo_inode != NULL);
3399
3400         oh = container_of0(th, struct osd_thandle, ot_super);
3401         LASSERT(oh->ot_handle);
3402         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3403
3404         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3405         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3406                            OI_CHECK_FLD, NULL);
3407         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3408
3409         return rc;
3410 }
3411
3412 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3413                    u64 seq, struct lu_seq_range *range)
3414 {
3415         struct seq_server_site *ss = osd_seq_site(osd);
3416
3417         if (fid_seq_is_idif(seq)) {
3418                 fld_range_set_ost(range);
3419                 range->lsr_index = idif_ost_idx(seq);
3420                 return 0;
3421         }
3422
3423         if (!fid_seq_in_fldb(seq)) {
3424                 fld_range_set_mdt(range);
3425                 if (ss != NULL)
3426                         /*
3427                          * FIXME: If ss is NULL, it suppose not get lsr_index
3428                          * at all
3429                          */
3430                         range->lsr_index = ss->ss_node_id;
3431                 return 0;
3432         }
3433
3434         LASSERT(ss != NULL);
3435         fld_range_set_any(range);
3436         /* OSD will only do local fld lookup */
3437         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3438 }
3439
3440 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3441                               struct lu_attr *attr,
3442                               struct dt_allocation_hint *hint,
3443                               struct dt_object_format *dof,
3444                               struct thandle *handle)
3445 {
3446         struct osd_thandle *oh;
3447         int rc;
3448
3449         ENTRY;
3450
3451         LASSERT(handle != NULL);
3452
3453         oh = container_of0(handle, struct osd_thandle, ot_super);
3454         LASSERT(oh->ot_handle == NULL);
3455
3456         /*
3457          * EA object consumes more credits than regular object: osd_mk_index
3458          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3459          * leaf_node, could involves the block, block bitmap, groups, GDT
3460          * change for each block, so add 4 * 2 credits in that case.
3461          */
3462         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3463                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3464                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3465         /*
3466          * Reuse idle OI block may cause additional one OI block
3467          * to be changed.
3468          */
3469         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3470                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3471
3472         /* will help to find FID->ino mapping at dt_insert() */
3473         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3474                                    osd_dt_obj(dt));
3475         if (rc != 0)
3476                 RETURN(rc);
3477
3478         if (!attr)
3479                 RETURN(0);
3480
3481         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3482                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3483                                    NULL, OSD_QID_INODE);
3484         if (rc != 0)
3485                 RETURN(rc);
3486
3487         RETURN(rc);
3488 }
3489
3490 /**
3491  * Called to destroy on-disk representation of the object
3492  *
3493  * Concurrency: must be locked
3494  */
3495 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3496                                struct thandle *th)
3497 {
3498         struct osd_object *obj = osd_dt_obj(dt);
3499         struct inode *inode = obj->oo_inode;
3500         struct osd_thandle *oh;
3501         int rc;
3502
3503         ENTRY;
3504
3505         if (inode == NULL)
3506                 RETURN(-ENOENT);
3507
3508         oh = container_of0(th, struct osd_thandle, ot_super);
3509         LASSERT(oh->ot_handle == NULL);
3510
3511         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3512                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3513
3514         /* For removing agent entry */
3515         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3516                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3517
3518         /*
3519          * Recycle idle OI leaf may cause additional three OI blocks
3520          * to be changed.
3521          */
3522         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3523                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3524                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3525         /* one less inode */
3526         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3527                                    i_projid_read(inode), -1, oh, obj, NULL,
3528                                    OSD_QID_INODE);
3529         if (rc)
3530                 RETURN(rc);
3531         /* data to be truncated */
3532         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3533                                    i_projid_read(inode), 0, oh, obj, NULL,
3534                                    OSD_QID_BLK);
3535         if (rc)
3536                 RETURN(rc);
3537
3538         /*
3539          * will help to find FID->ino when this object is being
3540          * added to PENDING
3541          */
3542         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3543
3544         RETURN(rc);
3545 }
3546
3547 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3548                        struct thandle *th)
3549 {
3550         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3551         struct osd_object *obj = osd_dt_obj(dt);
3552         struct inode *inode = obj->oo_inode;
3553         struct osd_device *osd = osd_obj2dev(obj);
3554         struct osd_thandle *oh;
3555         int result;
3556
3557         ENTRY;
3558
3559         oh = container_of0(th, struct osd_thandle, ot_super);
3560         LASSERT(oh->ot_handle);
3561         LASSERT(inode);
3562         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3563
3564         if (unlikely(fid_is_acct(fid)))
3565                 RETURN(-EPERM);
3566
3567         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3568                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3569                 if (result != 0)
3570                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3571                                osd_name(osd), PFID(fid), result);
3572         }
3573
3574         if (S_ISDIR(inode->i_mode)) {
3575                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3576                         inode->i_nlink == 2);
3577
3578                 spin_lock(&obj->oo_guard);
3579                 clear_nlink(inode);
3580                 spin_unlock(&obj->oo_guard);
3581                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3582         }
3583
3584         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3585
3586         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3587
3588         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3589                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3590                                        oh->ot_handle, OI_CHECK_FLD);
3591
3592         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3593         /* XXX: add to ext3 orphan list */
3594         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3595
3596         /* not needed in the cache anymore */
3597         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3598         obj->oo_destroyed = 1;
3599
3600         RETURN(0);
3601 }
3602
3603 /**
3604  * Put the fid into lustre_mdt_attrs, and then place the structure
3605  * inode's ea. This fid should not be altered during the life time
3606  * of the inode.
3607  *
3608  * \retval +ve, on success
3609  * \retval -ve, on error
3610  *
3611  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3612  */
3613 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3614                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3615 {
3616         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3617         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3618         int rc;
3619
3620         ENTRY;
3621
3622         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3623                 RETURN(0);
3624
3625         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3626                 rc = -ENOMEM;
3627
3628         lustre_loa_init(loa, fid, compat, incompat);
3629         lustre_loa_swab(loa, false);
3630
3631         /*
3632          * For the OST device with 256 bytes inode size by default,
3633          * the PFID EA will be stored together with LMA EA to avoid
3634          * performance trouble. Otherwise the PFID EA can be stored
3635          * independently. LU-8998
3636          */
3637         if ((compat & LMAC_FID_ON_OST) &&
3638             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3639                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3640                                      sizeof(*loa), XATTR_CREATE);
3641         else
3642                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3643                                      sizeof(*lma), XATTR_CREATE);
3644         /*
3645          * LMA may already exist, but we need to check that all the
3646          * desired compat/incompat flags have been added.
3647          */
3648         if (unlikely(rc == -EEXIST)) {
3649                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3650                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3651                 if (rc < 0)
3652                         RETURN(rc);
3653
3654                 if (rc < sizeof(*lma))
3655                         RETURN(-EINVAL);
3656
3657                 lustre_loa_swab(loa, true);
3658                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3659                     ((compat == 0 && incompat == 0) ||
3660                      (!(~lma->lma_compat & compat) &&
3661                       !(~lma->lma_incompat & incompat))))
3662                         RETURN(0);
3663
3664                 lma->lma_self_fid = *fid;
3665                 lma->lma_compat |= compat;
3666                 lma->lma_incompat |= incompat;
3667                 if (rc == sizeof(*lma)) {
3668                         lustre_lma_swab(lma);
3669                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3670                                              sizeof(*lma), XATTR_REPLACE);
3671                 } else {
3672                         lustre_loa_swab(loa, false);
3673                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3674                                              sizeof(*loa), XATTR_REPLACE);
3675                 }
3676         }
3677
3678         RETURN(rc);
3679 }
3680
3681 /**
3682  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3683  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3684  * To have compatilibility with 1.8 ldiskfs driver we need to have
3685  * magic number at start of fid data.
3686  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3687  * its inmemory API.
3688  */
3689 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3690                                          const struct lu_fid *fid)
3691 {
3692         if (!fid_is_namespace_visible(fid) ||
3693             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3694                 param->edp_magic = 0;
3695                 return;
3696         }
3697
3698         param->edp_magic = LDISKFS_LUFID_MAGIC;
3699         param->edp_len =  sizeof(struct lu_fid) + 1;
3700         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3701 }
3702
3703 /**
3704  * Try to read the fid from inode ea into dt_rec.
3705  *
3706  * \param fid object fid.
3707  *
3708  * \retval 0 on success
3709  */
3710 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3711                           __u32 ino, struct lu_fid *fid,
3712                           struct osd_inode_id *id)
3713 {
3714         struct osd_thread_info *info  = osd_oti_get(env);
3715         struct inode *inode;
3716
3717         ENTRY;
3718
3719         osd_id_gen(id, ino, OSD_OII_NOGEN);
3720         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3721         if (IS_ERR(inode))
3722                 RETURN(PTR_ERR(inode));
3723
3724         iput(inode);
3725         RETURN(0);
3726 }
3727
3728 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3729                                         struct inode *dir,
3730                                         struct inode *parent_dir,
3731                                         const struct lu_fid *dot_fid,
3732                                         const struct lu_fid *dot_dot_fid,
3733                                         struct osd_thandle *oth)
3734 {
3735         struct ldiskfs_dentry_param *dot_ldp;
3736         struct ldiskfs_dentry_param *dot_dot_ldp;
3737         __u32 saved_nlink = dir->i_nlink;
3738         int rc;
3739
3740         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3741         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3742
3743         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3744         dot_ldp->edp_magic = 0;
3745
3746         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3747                                     dir, dot_ldp, dot_dot_ldp);
3748         /*
3749          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3750          * the subseqent ref_add() will increase the dir->i_nlink
3751          * as 3. That is incorrect for new created directory.
3752          *
3753          * It looks like hack, because we want to make the OSD API
3754          * to be order-independent for new created directory object
3755          * between dt_insert(..) and ref_add() operations.
3756          *
3757          * Here, we only restore the in-RAM dir-inode's nlink attr,
3758          * becuase if the nlink attr is not 2, then there will be
3759          * ref_add() called following the dt_insert(..), such call
3760          * will make both the in-RAM and on-disk dir-inode's nlink
3761          * attr to be set as 2. LU-7447
3762          */
3763         set_nlink(dir, saved_nlink);
3764         return rc;
3765 }
3766
3767 /**
3768  * Create an local agent inode for remote entry
3769  */
3770 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3771                                                   struct osd_device *osd,
3772                                                   struct osd_object *pobj,
3773                                                   const struct lu_fid *fid,
3774                                                   __u32 type,
3775                                                   struct thandle *th)
3776 {
3777         struct osd_thread_info *info = osd_oti_get(env);
3778         struct inode *local;
3779         struct osd_thandle *oh;
3780         int rc;
3781
3782         ENTRY;
3783
3784         LASSERT(th);
3785         oh = container_of(th, struct osd_thandle, ot_super);
3786         LASSERT(oh->ot_handle->h_transaction != NULL);
3787
3788         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3789                                      NULL);
3790         if (IS_ERR(local)) {
3791                 CERROR("%s: create local error %d\n", osd_name(osd),
3792                        (int)PTR_ERR(local));
3793                 RETURN(local);
3794         }
3795
3796         /*
3797          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3798          * correct gid on remote file, not agent here
3799          */
3800         local->i_gid = current_fsgid();
3801         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3802
3803         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3804          * That gives e2fsck something to look at and be happy, and allows
3805          * debugging if we need to determine where this symlink came from.
3806          */
3807         if (S_ISLNK(type)) {
3808                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3809                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3810                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3811
3812                 i_size_write(local, rc);
3813                 LDISKFS_I(local)->i_disksize = rc;
3814         }
3815         unlock_new_inode(local);
3816
3817         /* Agent inode should not have project ID */
3818 #ifdef  HAVE_PROJECT_QUOTA
3819         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3820             i_projid_read(pobj->oo_inode) != 0) {
3821                 rc = osd_transfer_project(local, 0);
3822                 if (rc) {
3823                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3824                                "quota enforcement enabled on the ldiskfs "
3825                                "filesystem?\n", local->i_sb->s_id, rc);
3826                         RETURN(ERR_PTR(rc));
3827                 }
3828         }
3829 #endif
3830         /* Set special LMA flag for local agent inode */
3831         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3832         if (rc != 0) {
3833                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3834                        osd_name(osd), PFID(fid), rc);
3835                 RETURN(ERR_PTR(rc));
3836         }
3837
3838         if (!S_ISDIR(type))
3839                 RETURN(local);
3840
3841         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3842                                          lu_object_fid(&pobj->oo_dt.do_lu),
3843                                          fid, oh);
3844         if (rc != 0) {
3845                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3846                         osd_name(osd), PFID(fid), rc);
3847                 RETURN(ERR_PTR(rc));
3848         }
3849
3850         RETURN(local);
3851 }
3852
3853 /**
3854  * when direntry is deleted, we have to take care of possible agent inode
3855  * referenced by that. unfortunately we can't do this at that point:
3856  * iget() within a running transaction leads to deadlock and we better do
3857  * not call that every delete declaration to save performance. so we put
3858  * a potention agent inode on a list and process that once the transaction
3859  * is over. Notice it's not any worse in terms of real orphans as regular
3860  * object destroy doesn't put inodes on the on-disk orphan list. this should
3861  * be addressed separately
3862  */
3863 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3864                                             struct osd_thandle *oh,
3865                                             __u32 ino)
3866 {
3867         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3868         struct osd_obj_orphan *oor;
3869
3870         OBD_ALLOC_PTR(oor);
3871         if (oor == NULL)
3872                 return -ENOMEM;
3873
3874         oor->oor_ino = ino;
3875         oor->oor_env = (struct lu_env *)env;
3876         spin_lock(&osd->od_osfs_lock);
3877         list_add(&oor->oor_list, &osd->od_orphan_list);
3878         spin_unlock(&osd->od_osfs_lock);
3879
3880         oh->ot_remove_agents = 1;
3881
3882         return 0;
3883
3884 }
3885
3886 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3887                                                 struct osd_device *osd)
3888 {
3889         struct osd_thread_info *info = osd_oti_get(env);
3890         struct osd_obj_orphan *oor, *tmp;
3891         struct osd_inode_id id;
3892         struct list_head list;
3893         struct inode *inode;
3894         struct lu_fid fid;
3895         handle_t *jh;
3896         __u32 ino;
3897
3898         INIT_LIST_HEAD(&list);
3899
3900         spin_lock(&osd->od_osfs_lock);
3901         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3902                 if (oor->oor_env == env) {
3903                         list_del(&oor->oor_list);
3904                         list_add(&oor->oor_list, &list);
3905                 }
3906         }
3907         spin_unlock(&osd->od_osfs_lock);
3908
3909         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3910
3911                 ino = oor->oor_ino;
3912
3913                 list_del(&oor->oor_list);
3914                 OBD_FREE_PTR(oor);
3915
3916                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3917                 inode = osd_iget_fid(info, osd, &id, &fid);
3918                 if (IS_ERR(inode))
3919                         continue;
3920
3921                 if (!osd_remote_fid(env, osd, &fid)) {
3922                         iput(inode);
3923                         continue;
3924                 }
3925
3926                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3927                 clear_nlink(inode);
3928                 mark_inode_dirty(inode);
3929                 ldiskfs_journal_stop(jh);
3930                 iput(inode);
3931         }
3932
3933         return 0;
3934 }
3935
3936 /**
3937  * OSD layer object create function for OST objects (b=11826).
3938  *
3939  * The FID is inserted into inode xattr here.
3940  *
3941  * \retval   0, on success
3942  * \retval -ve, on error
3943  */
3944 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3945                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3946                       struct dt_object_format *dof, struct thandle *th)
3947 {
3948         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3949         struct osd_object *obj = osd_dt_obj(dt);
3950         struct osd_thread_info *info = osd_oti_get(env);
3951         int result, on_ost = 0;
3952
3953         ENTRY;
3954
3955         if (dt_object_exists(dt))
3956                 RETURN(-EEXIST);
3957
3958         LINVRNT(osd_invariant(obj));
3959         LASSERT(!dt_object_remote(dt));
3960         LASSERT(osd_is_write_locked(env, obj));
3961         LASSERT(th != NULL);
3962
3963         if (unlikely(fid_is_acct(fid)))
3964                 /*
3965                  * Quota files can't be created from the kernel any more,
3966                  * 'tune2fs -O quota' will take care of creating them
3967                  */
3968                 RETURN(-EPERM);
3969
3970         result = __osd_create(info, obj, attr, hint, dof, th);
3971         if (result == 0) {
3972                 if (fid_is_idif(fid) &&
3973                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3974                         struct lu_fid *tfid = &info->oti_fid;
3975                         struct ost_id *oi   = &info->oti_ostid;
3976
3977                         fid_to_ostid(fid, oi);
3978                         ostid_to_fid(tfid, oi, 0);
3979                         on_ost = 1;
3980                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3981                                                 LMAC_FID_ON_OST, 0);
3982                 } else {
3983                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3984                                                fid, OI_CHECK_FLD);
3985                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3986                                                 on_ost ? LMAC_FID_ON_OST : 0,
3987                                                 0);
3988                 }
3989                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3990                         obj->oo_dt.do_body_ops = &osd_body_ops;
3991         }
3992
3993         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3994                 result = __osd_oi_insert(env, obj, fid, th);
3995
3996         /*
3997          * a small optimization - dt_insert() isn't usually applied
3998          * to OST objects, so we don't need to cache OI mapping for
3999          * OST objects
4000          */
4001         if (result == 0 && on_ost == 0) {
4002                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4003
4004                 result = osd_idc_find_and_init(env, osd, obj);
4005                 LASSERT(result == 0);
4006         }
4007
4008         LASSERT(ergo(result == 0,
4009                      dt_object_exists(dt) && !dt_object_remote(dt)));
4010         LINVRNT(osd_invariant(obj));
4011         RETURN(result);
4012 }
4013
4014 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4015                                struct thandle *handle)
4016 {
4017         struct osd_thandle *oh;
4018
4019         /* it's possible that object doesn't exist yet */
4020         LASSERT(handle != NULL);
4021
4022         oh = container_of0(handle, struct osd_thandle, ot_super);
4023         LASSERT(oh->ot_handle == NULL);
4024
4025         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4026                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4027
4028         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4029
4030         return 0;
4031 }
4032
4033 /*
4034  * Concurrency: @dt is write locked.
4035  */
4036 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4037                        struct thandle *th)
4038 {
4039         struct osd_object *obj = osd_dt_obj(dt);
4040         struct inode *inode = obj->oo_inode;
4041         struct osd_thandle *oh;
4042         int rc = 0;
4043
4044         if (!dt_object_exists(dt) || obj->oo_destroyed)
4045                 return -ENOENT;
4046
4047         LINVRNT(osd_invariant(obj));
4048         LASSERT(!dt_object_remote(dt));
4049         LASSERT(osd_is_write_locked(env, obj));
4050         LASSERT(th != NULL);
4051
4052         oh = container_of0(th, struct osd_thandle, ot_super);
4053         LASSERT(oh->ot_handle != NULL);
4054
4055         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4056
4057         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4058                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4059         /*
4060          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4061          * (65000) subdirectories by storing "1" in i_nlink if the link count
4062          * would otherwise overflow. Directory tranversal tools understand
4063          * that (st_nlink == 1) indicates that the filesystem dose not track
4064          * hard links count on the directory, and will not abort subdirectory
4065          * scanning early once (st_nlink - 2) subdirs have been found.
4066          *
4067          * This also has to properly handle the case of inodes with nlink == 0
4068          * in case they are being linked into the PENDING directory
4069          */
4070         spin_lock(&obj->oo_guard);
4071         if (unlikely(inode->i_nlink == 0))
4072                 /* inc_nlink from 0 may cause WARN_ON */
4073                 set_nlink(inode, 1);
4074         else {
4075                 ldiskfs_inc_count(oh->ot_handle, inode);
4076                 if (!S_ISDIR(inode->i_mode))
4077                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4078         }
4079         spin_unlock(&obj->oo_guard);
4080
4081         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4082         LINVRNT(osd_invariant(obj));
4083
4084         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4085
4086         return rc;
4087 }
4088
4089 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4090                                struct thandle *handle)
4091 {
4092         struct osd_thandle *oh;
4093
4094         if (!dt_object_exists(dt))
4095                 return -ENOENT;
4096
4097         LASSERT(!dt_object_remote(dt));
4098         LASSERT(handle != NULL);
4099
4100         oh = container_of0(handle, struct osd_thandle, ot_super);
4101         LASSERT(oh->ot_handle == NULL);
4102
4103         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4104                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4105
4106         return 0;
4107 }
4108
4109 /*
4110  * Concurrency: @dt is write locked.
4111  */
4112 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4113                        struct thandle *th)
4114 {
4115         struct osd_object *obj = osd_dt_obj(dt);
4116         struct inode *inode = obj->oo_inode;
4117         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4118         struct osd_thandle *oh;
4119
4120         if (!dt_object_exists(dt))
4121                 return -ENOENT;
4122
4123         LINVRNT(osd_invariant(obj));
4124         LASSERT(!dt_object_remote(dt));
4125         LASSERT(osd_is_write_locked(env, obj));
4126         LASSERT(th != NULL);
4127
4128         oh = container_of0(th, struct osd_thandle, ot_super);
4129         LASSERT(oh->ot_handle != NULL);
4130
4131         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4132
4133         spin_lock(&obj->oo_guard);
4134         /*
4135          * That can be result of upgrade from old Lustre version and
4136          * applied only to local files.  Just skip this ref_del call.
4137          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4138          */
4139         if (inode->i_nlink == 0) {
4140                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4141                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4142                              ", maybe an upgraded file? (LU-3915)\n",
4143                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4144                 spin_unlock(&obj->oo_guard);
4145                 return 0;
4146         }
4147
4148         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4149                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4150
4151         ldiskfs_dec_count(oh->ot_handle, inode);
4152         spin_unlock(&obj->oo_guard);
4153
4154         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4155         LINVRNT(osd_invariant(obj));
4156
4157         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4158
4159         return 0;
4160 }
4161
4162 /*
4163  * Concurrency: @dt is read locked.
4164  */
4165 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4166                          struct lu_buf *buf, const char *name)
4167 {
4168         struct osd_object      *obj    = osd_dt_obj(dt);
4169         struct inode           *inode  = obj->oo_inode;
4170         struct osd_thread_info *info   = osd_oti_get(env);
4171         struct dentry          *dentry = &info->oti_obj_dentry;
4172         bool cache_xattr = false;
4173         int rc;
4174
4175         LASSERT(buf);
4176
4177         /* version get is not real XATTR but uses xattr API */
4178         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4179                 dt_obj_version_t *ver = buf->lb_buf;
4180
4181                 /*
4182                  * for version we are just using xattr API but change inode
4183                  * field instead
4184                  */
4185                 if (buf->lb_len == 0)
4186                         return sizeof(dt_obj_version_t);
4187
4188                 if (buf->lb_len < sizeof(dt_obj_version_t))
4189                         return -ERANGE;
4190
4191                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4192                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4193
4194                 *ver = LDISKFS_I(inode)->i_fs_version;
4195
4196                 return sizeof(dt_obj_version_t);
4197         }
4198
4199         if (!dt_object_exists(dt))
4200                 return -ENOENT;
4201
4202         LASSERT(!dt_object_remote(dt));
4203         LASSERT(inode->i_op != NULL);
4204 #ifdef HAVE_IOP_XATTR
4205         LASSERT(inode->i_op->getxattr != NULL);
4206 #endif
4207
4208         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4209             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4210                 cache_xattr = true;
4211
4212         if (cache_xattr) {
4213                 rc = osd_oxc_get(obj, name, buf);
4214                 if (rc != -ENOENT)
4215                         return rc;
4216         }
4217
4218         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4219                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4220                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4221                 struct filter_fid *ff;
4222                 struct ost_layout *ol;
4223
4224                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4225                 if (rc)
4226                         return rc;
4227
4228                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4229
4230                 rc = sizeof(*ff);
4231                 if (buf->lb_len == 0 || !buf->lb_buf)
4232                         return rc;
4233
4234                 if (buf->lb_len < rc)
4235                         return -ERANGE;
4236
4237                 ff = buf->lb_buf;
4238                 ol = &ff->ff_layout;
4239                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4240                                                   PFID_STRIPE_IDX_BITS);
4241                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4242                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4243                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4244                 if (lma->lma_compat & LMAC_COMP_INFO) {
4245                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4246                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4247                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4248                 } else {
4249                         ol->ol_comp_start = 0;
4250                         ol->ol_comp_end = 0;
4251                         ol->ol_comp_id = 0;
4252                 }
4253         } else {
4254                 rc = __osd_xattr_get(inode, dentry, name,
4255                                      buf->lb_buf, buf->lb_len);
4256         }
4257
4258         if (cache_xattr) {
4259                 if (rc == -ENOENT || rc == -ENODATA)
4260                         osd_oxc_add(obj, name, NULL, 0);
4261                 else if (rc > 0 && buf->lb_buf != NULL)
4262                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4263         }
4264
4265         return rc;
4266 }
4267
4268 static int osd_declare_xattr_set(const struct lu_env *env,
4269                                  struct dt_object *dt,
4270                                  const struct lu_buf *buf, const char *name,
4271                                  int fl, struct thandle *handle)
4272 {
4273         struct osd_thandle *oh;
4274         int credits = 0;
4275         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4276
4277         LASSERT(handle != NULL);
4278
4279         oh = container_of0(handle, struct osd_thandle, ot_super);
4280         LASSERT(oh->ot_handle == NULL);
4281
4282         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4283                 /*
4284                  * For non-upgrading case, the LMA is set first and
4285                  * usually fit inode. But for upgrade case, the LMA
4286                  * may be in another separated EA block.
4287                  */
4288                 if (dt_object_exists(dt)) {
4289                         if (fl == LU_XATTR_REPLACE)
4290                                 credits = 1;
4291                         else
4292                                 goto upgrade;
4293                 }
4294         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4295                 credits = 1;
4296         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4297                 /* We may need to delete the old PFID EA. */
4298                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4299                 if (fl == LU_XATTR_REPLACE)
4300                         credits += 1;
4301                 else
4302                         goto upgrade;
4303         } else {
4304                 /*
4305                  * If some name entry resides on remote MDT, then will create
4306                  * agent entry under remote parent. On the other hand, if the
4307                  * remote entry will be removed, then related agent entry may
4308                  * need to be removed from the remote parent. So there may be
4309                  * kinds of cases, let's declare enough credits. The credits
4310                  * for create agent entry is enough for remove case.
4311                  */
4312                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4313                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4314                         if (dt_object_exists(dt))
4315                                 credits += 1; /* For updating LMA */
4316                 }
4317
4318 upgrade:
4319                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4320
4321                 if (buf != NULL) {
4322                         ssize_t buflen;
4323
4324                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4325                                 /*
4326                                  * learn xattr size from osd_xattr_get if
4327                                  * attribute has not been read yet
4328                                  */
4329                                 buflen = __osd_xattr_get(
4330                                     osd_dt_obj(dt)->oo_inode,
4331                                     &osd_oti_get(env)->oti_obj_dentry,
4332                                     name, NULL, 0);
4333                                 if (buflen < 0)
4334                                         buflen = 0;
4335                         } else {
4336                                 buflen = buf->lb_len;
4337                         }
4338
4339                         if (buflen > sb->s_blocksize) {
4340                                 credits += osd_calc_bkmap_credits(
4341                                     sb, NULL, 0, -1,
4342                                     (buflen + sb->s_blocksize - 1) >>
4343                                     sb->s_blocksize_bits);
4344                         }
4345                 }
4346                 /*
4347                  * xattr set may involve inode quota change, reserve credits for
4348                  * dquot_initialize()
4349                  */
4350                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4351         }
4352
4353         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4354
4355         return 0;
4356 }
4357
4358 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4359                               const struct lu_buf *buf, int fl,
4360                               struct thandle *handle)
4361 {
4362         struct osd_thread_info *info = osd_oti_get(env);
4363         struct dentry *dentry = &info->oti_obj_dentry;
4364         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4365         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4366         struct inode *inode = obj->oo_inode;
4367         struct filter_fid *ff = buf->lb_buf;
4368         struct ost_layout *ol = &ff->ff_layout;
4369         int flags = XATTR_REPLACE;
4370         int rc;
4371
4372         ENTRY;
4373
4374         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4375                 RETURN(-EINVAL);
4376
4377         rc = osd_get_lma(info, inode, dentry, loa);
4378         if (rc == -ENODATA) {
4379                 /* Usually for upgarding from old device */
4380                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4381                                 LMAC_FID_ON_OST, 0);
4382                 flags = XATTR_CREATE;
4383         } else if (rc) {
4384                 RETURN(rc);
4385         }
4386
4387         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4388                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4389                         RETURN(-EEXIST);
4390
4391                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4392                         /* Separate PFID EA from LMA */
4393                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4394                         lustre_lma_swab(lma);
4395                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4396                                              sizeof(*lma), XATTR_REPLACE);
4397                         if (!rc) {
4398                                 obj->oo_pfid_in_lma = 0;
4399                                 rc = LU_XATTR_CREATE;
4400                         }
4401
4402                         RETURN(rc);
4403                 }
4404         } else {
4405                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4406                         RETURN(fl);
4407
4408                 /*
4409                  * Old client does not send stripe information,
4410                  * then store the PFID EA on disk separatedly.
4411                  */
4412                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4413                              ol->ol_stripe_size == 0))
4414                         RETURN(fl);
4415
4416                 /* Remove old PFID EA entry firstly. */
4417                 ll_vfs_dq_init(inode);
4418                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4419                 if (rc == -ENODATA) {
4420                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4421                                 RETURN(rc);
4422                 } else if (rc) {
4423                         RETURN(rc);
4424                 }
4425         }
4426
4427         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4428         if (likely(ol->ol_stripe_size != 0)) {
4429                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4430                                              PFID_STRIPE_IDX_BITS;
4431                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4432                 lma->lma_compat |= LMAC_STRIPE_INFO;
4433                 if (ol->ol_comp_id != 0) {
4434                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4435                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4436                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4437                         lma->lma_compat |= LMAC_COMP_INFO;
4438                 }
4439         }
4440
4441         lustre_loa_swab(loa, false);
4442
4443         /* Store the PFID EA inside LMA. */
4444         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4445                              flags);
4446         if (!rc)
4447                 obj->oo_pfid_in_lma = 1;
4448
4449         RETURN(rc);
4450 }
4451
4452 /*
4453  * In DNE environment, the object (in spite of regular file or directory)
4454  * and its name entry may reside on different MDTs. Under such case, we will
4455  * create an agent entry on the MDT where the object resides. The agent entry
4456  * references the object locally, that makes the object to be visible to the
4457  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4458  * such as 'tar' can handle the object properly.
4459  *
4460  * We handle the agent entry during set linkEA that is the common interface
4461  * for both regular file and directroy, can handle kinds of cases, such as
4462  * create/link/unlink/rename, and so on.
4463  *
4464  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4465  *
4466  * XXX: There are two known issues:
4467  * 1. For one object, we will create at most one agent entry even if there
4468  *    may be more than one cross-MDTs hard links on the object. So the local
4469  *    e2fsck may claim that the object's nlink is larger than the name entries
4470  *    that reference such inode. And in further, the e2fsck will fix the nlink
4471  *    attribute to match the local references. Then it will cause the object's
4472  *    nlink attribute to be inconsistent with the global references. it is bad
4473  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4474  *    other hand, the global namespace LFSCK can repair the object's attribute
4475  *    according to the linkEA.
4476  * 2. There may be too many hard links on the object as to its linkEA overflow,
4477  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4478  *    case happened, then at this point, we do not know whether there are some
4479  *    cross-MDTs reference. But there are local references, it guarantees that
4480  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4481  */
4482 static int osd_xattr_handle_linkea(const struct lu_env *env,
4483                                    struct osd_device *osd,
4484                                    struct osd_object *obj,
4485                                    const struct lu_buf *buf,
4486                                    struct thandle *handle)
4487 {
4488         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4489         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4490         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4491         struct lu_name tmpname;
4492         struct osd_thandle *oh;
4493         int rc;
4494         bool remote = false;
4495
4496         ENTRY;
4497
4498         oh = container_of0(handle, struct osd_thandle, ot_super);
4499         LASSERT(oh->ot_handle != NULL);
4500
4501         rc = linkea_init_with_rec(&ldata);
4502         if (!rc) {
4503                 linkea_first_entry(&ldata);
4504                 while (ldata.ld_lee != NULL && !remote) {
4505                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4506                                             &tmpname, tfid);
4507                         if (osd_remote_fid(env, osd, tfid) > 0)
4508                                 remote = true;
4509                         else
4510                                 linkea_next_entry(&ldata);
4511                 }
4512         } else if (rc == -ENODATA) {
4513                 rc = 0;
4514         } else {
4515                 RETURN(rc);
4516         }
4517
4518         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4519                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4520                 if (rc)
4521                         CERROR("%s: failed to remove agent entry for "DFID
4522                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4523         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4524                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4525                 if (rc)
4526                         CERROR("%s: failed to create agent entry for "DFID
4527                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4528         }
4529
4530         RETURN(rc);
4531 }
4532
4533 /*
4534  * Concurrency: @dt is write locked.
4535  */
4536 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4537                          const struct lu_buf *buf, const char *name, int fl,
4538                          struct thandle *handle)
4539 {
4540         struct osd_object *obj = osd_dt_obj(dt);
4541         struct osd_device *osd = osd_obj2dev(obj);
4542         struct inode *inode = obj->oo_inode;
4543         struct osd_thread_info *info = osd_oti_get(env);
4544         int fs_flags = 0;
4545         int len;
4546         int rc;
4547
4548         ENTRY;
4549
4550         LASSERT(handle);
4551         LASSERT(buf);
4552
4553         /* version set is not real XATTR */
4554         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4555                 dt_obj_version_t *version = buf->lb_buf;
4556
4557                 /*
4558                  * for version we are just using xattr API but change inode
4559                  * field instead
4560                  */
4561                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4562
4563                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4564                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4565
4566                 LDISKFS_I(inode)->i_fs_version = *version;
4567                 /*
4568                  * Version is set after all inode operations are finished,
4569                  * so we should mark it dirty here
4570                  */
4571                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4572
4573                 RETURN(0);
4574         }
4575
4576         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4577                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4578
4579         len = buf->lb_len;
4580         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4581
4582         /*
4583          * For the OST device with 256 bytes inode size by default,
4584          * the PFID EA will be stored together with LMA EA to avoid
4585          * performance trouble. Otherwise the PFID EA can be stored
4586          * independently. LU-8998
4587          */
4588         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4589             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4590                 LASSERT(buf->lb_buf);
4591
4592                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4593                 if (fl <= 0)
4594                         RETURN(fl);
4595         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4596                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4597                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4598
4599                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4600                 if (rc)
4601                         RETURN(rc);
4602
4603                 lma->lma_incompat |= LMAI_STRIPED;
4604                 lustre_lma_swab(lma);
4605                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4606                                      sizeof(*lma), XATTR_REPLACE);
4607                 if (rc != 0)
4608                         RETURN(rc);
4609         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4610                 LASSERT(!osd->od_is_ost);
4611
4612                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4613                 if (rc)
4614                         RETURN(rc);
4615         }
4616
4617         if (fl & LU_XATTR_REPLACE)
4618                 fs_flags |= XATTR_REPLACE;
4619
4620         if (fl & LU_XATTR_CREATE)
4621                 fs_flags |= XATTR_CREATE;
4622
4623         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4624         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4625
4626         if (rc == 0 &&
4627             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4628              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4629                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4630
4631         return rc;
4632 }
4633
4634 /*
4635  * Concurrency: @dt is read locked.
4636  */
4637 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4638                           const struct lu_buf *buf)
4639 {
4640         struct osd_object *obj = osd_dt_obj(dt);
4641         struct inode *inode = obj->oo_inode;
4642         struct osd_thread_info *info = osd_oti_get(env);
4643         struct dentry *dentry = &info->oti_obj_dentry;
4644
4645         if (!dt_object_exists(dt))
4646                 return -ENOENT;
4647
4648         LASSERT(!dt_object_remote(dt));
4649         LASSERT(inode->i_op != NULL);
4650         LASSERT(inode->i_op->listxattr != NULL);
4651
4652         dentry->d_inode = inode;
4653         dentry->d_sb = inode->i_sb;
4654         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4655 }
4656
4657 static int osd_declare_xattr_del(const struct lu_env *env,
4658                                  struct dt_object *dt, const char *name,
4659                                  struct thandle *handle)
4660 {
4661         struct osd_thandle *oh;
4662         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4663
4664         LASSERT(!dt_object_remote(dt));
4665         LASSERT(handle != NULL);
4666
4667         oh = container_of0(handle, struct osd_thandle, ot_super);
4668         LASSERT(oh->ot_handle == NULL);
4669
4670         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4671                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4672         /*
4673          * xattr del may involve inode quota change, reserve credits for
4674          * dquot_initialize()
4675          */
4676         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4677
4678         return 0;
4679 }
4680
4681 /*
4682  * Concurrency: @dt is write locked.
4683  */
4684 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4685                          const char *name, struct thandle *handle)
4686 {
4687         struct osd_object *obj = osd_dt_obj(dt);
4688         struct inode *inode = obj->oo_inode;
4689         struct osd_thread_info *info = osd_oti_get(env);
4690         struct dentry *dentry = &info->oti_obj_dentry;
4691         int rc;
4692
4693         if (!dt_object_exists(dt))
4694                 return -ENOENT;
4695
4696         LASSERT(!dt_object_remote(dt));
4697         LASSERT(inode->i_op != NULL);
4698         LASSERT(handle != NULL);
4699 #ifdef HAVE_IOP_XATTR
4700         LASSERT(inode->i_op->removexattr != NULL);
4701 #endif
4702
4703         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4704
4705         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4706                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4707
4708                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4709                                  &info->oti_ost_attrs);
4710                 if (!rc) {
4711                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4712
4713                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4714                         lustre_lma_swab(lma);
4715                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4716                                              sizeof(*lma), XATTR_REPLACE);
4717                         if (!rc)
4718                                 obj->oo_pfid_in_lma = 0;
4719                 }
4720         } else {
4721                 ll_vfs_dq_init(inode);
4722                 dentry->d_inode = inode;
4723                 dentry->d_sb = inode->i_sb;
4724                 rc = osd_removexattr(dentry, inode, name);
4725         }
4726
4727         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4728
4729         if (rc == 0 &&
4730             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4731              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4732                 osd_oxc_del(obj, name);
4733
4734         return rc;
4735 }
4736
4737 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4738                            __u64 start, __u64 end)
4739 {
4740         struct osd_object *obj = osd_dt_obj(dt);
4741         struct inode *inode = obj->oo_inode;
4742         struct osd_thread_info *info = osd_oti_get(env);
4743         struct dentry *dentry = &info->oti_obj_dentry;
4744         struct file *file = &info->oti_file;
4745         int rc;
4746
4747         ENTRY;
4748
4749         dentry->d_inode = inode;
4750         dentry->d_sb = inode->i_sb;
4751         file->f_path.dentry = dentry;
4752         file->f_mapping = inode->i_mapping;
4753         file->f_op = inode->i_fop;
4754         set_file_inode(file, inode);
4755
4756         rc = ll_vfs_fsync_range(file, start, end, 0);
4757
4758         RETURN(rc);
4759 }
4760
4761 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4762 {
4763         return 0;
4764 }
4765
4766 /*
4767  * Index operations.
4768  */
4769
4770 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4771                                const struct dt_index_features *feat)
4772 {
4773         struct iam_descr *descr;
4774
4775         if (osd_object_is_root(o))
4776                 return feat == &dt_directory_features;
4777
4778         LASSERT(o->oo_dir != NULL);
4779
4780         descr = o->oo_dir->od_container.ic_descr;
4781         if (feat == &dt_directory_features) {
4782                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4783                         return 1;
4784                 else
4785                         return 0;
4786         } else {
4787                 return feat->dif_keysize_min <= descr->id_key_size &&
4788                        descr->id_key_size <= feat->dif_keysize_max &&
4789                        feat->dif_recsize_min <= descr->id_rec_size &&
4790                        descr->id_rec_size <= feat->dif_recsize_max &&
4791                        !(feat->dif_flags & (DT_IND_VARKEY |
4792                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4793                        ergo(feat->dif_flags & DT_IND_UPDATE,
4794                             1 /* XXX check that object (and fs) is writable */);
4795         }
4796 }
4797
4798 static int osd_iam_container_init(const struct lu_env *env,
4799                                   struct osd_object *obj,
4800                                   struct osd_directory *dir)
4801 {
4802         struct iam_container *bag = &dir->od_container;
4803         int result;
4804
4805         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4806         if (result != 0)
4807                 return result;
4808
4809         result = iam_container_setup(bag);
4810         if (result == 0)
4811                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4812         else
4813                 iam_container_fini(bag);
4814
4815         return result;
4816 }
4817
4818
4819 /*
4820  * Concurrency: no external locking is necessary.
4821  */
4822 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4823                          const struct dt_index_features *feat)
4824 {
4825         int result;
4826         int skip_iam = 0;
4827         struct osd_object *obj = osd_dt_obj(dt);
4828
4829         LINVRNT(osd_invariant(obj));
4830
4831         if (osd_object_is_root(obj)) {
4832                 dt->do_index_ops = &osd_index_ea_ops;
4833                 result = 0;
4834         } else if (feat == &dt_directory_features) {
4835                 dt->do_index_ops = &osd_index_ea_ops;
4836                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4837                         result = 0;
4838                 else
4839                         result = -ENOTDIR;
4840                 skip_iam = 1;
4841         } else if (unlikely(feat == &dt_otable_features)) {
4842                 dt->do_index_ops = &osd_otable_ops;
4843                 return 0;
4844         } else if (unlikely(feat == &dt_acct_features)) {
4845                 dt->do_index_ops = &osd_acct_index_ops;
4846                 result = 0;
4847                 skip_iam = 1;
4848         } else if (!osd_has_index(obj)) {
4849                 struct osd_directory *dir;
4850                 struct osd_device *osd = osd_obj2dev(obj);
4851                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4852
4853                 OBD_ALLOC_PTR(dir);
4854                 if (dir) {
4855
4856                         spin_lock(&obj->oo_guard);
4857                         if (obj->oo_dir == NULL)
4858                                 obj->oo_dir = dir;
4859                         else
4860                                 /*
4861                                  * Concurrent thread allocated container data.
4862                                  */
4863                                 OBD_FREE_PTR(dir);
4864                         spin_unlock(&obj->oo_guard);
4865                         /*
4866                          * Now, that we have container data, serialize its
4867                          * initialization.
4868                          */
4869                         down_write(&obj->oo_ext_idx_sem);
4870                         /*
4871                          * recheck under lock.
4872                          */
4873
4874                         if (osd_has_index(obj)) {
4875                                 result = 0;
4876                                 goto unlock;
4877                         }
4878
4879                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4880                         if (result || feat == &dt_lfsck_namespace_features ||
4881                             feat == &dt_lfsck_layout_orphan_features ||
4882                             feat == &dt_lfsck_layout_dangling_features)
4883                                 goto unlock;
4884
4885                         result = osd_index_register(osd, fid,
4886                                                     feat->dif_keysize_max,
4887                                                     feat->dif_recsize_max);
4888                         if (result < 0)
4889                                 CWARN("%s: failed to register index "
4890                                       DFID": rc = %d\n",
4891                                       osd_name(osd), PFID(fid), result);
4892                         else if (result > 0)
4893                                 result = 0;
4894                         else
4895                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4896                                        " (%d/%d) registered\n",
4897                                        osd_name(osd), PFID(fid),
4898                                        (int)feat->dif_keysize_max,
4899                                        (int)feat->dif_recsize_max);
4900
4901 unlock:
4902                         up_write(&obj->oo_ext_idx_sem);
4903                 } else {
4904                         result = -ENOMEM;
4905                 }
4906         } else {
4907                 result = 0;
4908         }
4909
4910         if (result == 0 && skip_iam == 0) {
4911                 if (!osd_iam_index_probe(env, obj, feat))
4912                         result = -ENOTDIR;
4913         }
4914         LINVRNT(osd_invariant(obj));
4915
4916         return result;
4917 }
4918
4919 static int osd_otable_it_attr_get(const struct lu_env *env,
4920                                  struct dt_object *dt,
4921                                  struct lu_attr *attr)
4922 {
4923         attr->la_valid = 0;
4924         return 0;
4925 }
4926
4927 static const struct dt_object_operations osd_obj_ops = {
4928         .do_read_lock           = osd_read_lock,
4929         .do_write_lock          = osd_write_lock,
4930         .do_read_unlock         = osd_read_unlock,
4931         .do_write_unlock        = osd_write_unlock,
4932         .do_write_locked        = osd_write_locked,
4933         .do_attr_get            = osd_attr_get,
4934         .do_declare_attr_set    = osd_declare_attr_set,
4935         .do_attr_set            = osd_attr_set,
4936         .do_ah_init             = osd_ah_init,
4937         .do_declare_create      = osd_declare_create,
4938         .do_create              = osd_create,
4939         .do_declare_destroy     = osd_declare_destroy,
4940         .do_destroy             = osd_destroy,
4941         .do_index_try           = osd_index_try,
4942         .do_declare_ref_add     = osd_declare_ref_add,
4943         .do_ref_add             = osd_ref_add,
4944         .do_declare_ref_del     = osd_declare_ref_del,
4945         .do_ref_del             = osd_ref_del,
4946         .do_xattr_get           = osd_xattr_get,
4947         .do_declare_xattr_set   = osd_declare_xattr_set,
4948         .do_xattr_set           = osd_xattr_set,
4949         .do_declare_xattr_del   = osd_declare_xattr_del,
4950         .do_xattr_del           = osd_xattr_del,
4951         .do_xattr_list          = osd_xattr_list,
4952         .do_object_sync         = osd_object_sync,
4953         .do_invalidate          = osd_invalidate,
4954 };
4955
4956 static const struct dt_object_operations osd_obj_otable_it_ops = {
4957         .do_attr_get    = osd_otable_it_attr_get,
4958         .do_index_try   = osd_index_try,
4959 };
4960
4961 static int osd_index_declare_iam_delete(const struct lu_env *env,
4962                                         struct dt_object *dt,
4963                                         const struct dt_key *key,
4964                                         struct thandle *handle)
4965 {
4966         struct osd_thandle *oh;
4967
4968         oh = container_of0(handle, struct osd_thandle, ot_super);
4969         LASSERT(oh->ot_handle == NULL);
4970
4971         /* Recycle  may cause additional three blocks to be changed. */
4972         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4973                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4974
4975         return 0;
4976 }
4977
4978 /**
4979  *      delete a (key, value) pair from index \a dt specified by \a key
4980  *
4981  *      \param  dt      osd index object
4982  *      \param  key     key for index
4983  *      \param  rec     record reference
4984  *      \param  handle  transaction handler
4985  *
4986  *      \retval  0  success
4987  *      \retval -ve   failure
4988  */
4989 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4990                                 const struct dt_key *key,
4991                                 struct thandle *handle)
4992 {
4993         struct osd_thread_info *oti = osd_oti_get(env);
4994         struct osd_object *obj = osd_dt_obj(dt);
4995         struct osd_thandle *oh;
4996         struct iam_path_descr *ipd;
4997         struct iam_container *bag = &obj->oo_dir->od_container;
4998         int rc;
4999
5000         ENTRY;
5001
5002         if (!dt_object_exists(dt))
5003                 RETURN(-ENOENT);
5004
5005         LINVRNT(osd_invariant(obj));
5006         LASSERT(!dt_object_remote(dt));
5007         LASSERT(bag->ic_object == obj->oo_inode);
5008         LASSERT(handle != NULL);
5009
5010         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5011
5012         ipd = osd_idx_ipd_get(env, bag);
5013         if (unlikely(ipd == NULL))
5014                 RETURN(-ENOMEM);
5015
5016         oh = container_of0(handle, struct osd_thandle, ot_super);
5017         LASSERT(oh->ot_handle != NULL);
5018         LASSERT(oh->ot_handle->h_transaction != NULL);
5019
5020         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5021                 /* swab quota uid/gid provided by caller */
5022                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5023                 key = (const struct dt_key *)&oti->oti_quota_id;
5024         }
5025
5026         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5027         osd_ipd_put(env, bag, ipd);
5028         LINVRNT(osd_invariant(obj));
5029         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5030         RETURN(rc);
5031 }
5032
5033 static int osd_index_declare_ea_delete(const struct lu_env *env,
5034                                        struct dt_object *dt,
5035                                        const struct dt_key *key,
5036                                        struct thandle *handle)
5037 {
5038         struct osd_thandle *oh;
5039         struct inode *inode;
5040         int rc, credits;
5041
5042         ENTRY;
5043
5044         LASSERT(!dt_object_remote(dt));
5045         LASSERT(handle != NULL);
5046
5047         oh = container_of0(handle, struct osd_thandle, ot_super);
5048         LASSERT(oh->ot_handle == NULL);
5049
5050         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5051         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5052
5053         inode = osd_dt_obj(dt)->oo_inode;
5054         if (inode == NULL)
5055                 RETURN(-ENOENT);
5056
5057         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5058                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5059                                    NULL, OSD_QID_BLK);
5060         RETURN(rc);
5061 }
5062
5063 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5064                                           struct dt_rec *fid)
5065 {
5066         struct osd_fid_pack *rec;
5067         int rc = -ENODATA;
5068
5069         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5070                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5071                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5072                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5073                         rc = -EINVAL;
5074         }
5075         return rc;
5076 }
5077
5078 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5079                           const struct lu_fid *fid)
5080 {
5081         struct seq_server_site *ss = osd_seq_site(osd);
5082
5083         ENTRY;
5084
5085         /* FID seqs not in FLDB, must be local seq */
5086         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5087                 RETURN(0);
5088
5089         /*
5090          * If FLD is not being initialized yet, it only happens during the
5091          * initialization, likely during mgs initialization, and we assume
5092          * this is local FID.
5093          */
5094         if (ss == NULL || ss->ss_server_fld == NULL)
5095                 RETURN(0);
5096
5097         /* Only check the local FLDB here */
5098         if (osd_seq_exists(env, osd, fid_seq(fid)))
5099                 RETURN(0);
5100
5101         RETURN(1);
5102 }
5103
5104 static void osd_take_care_of_agent(const struct lu_env *env,
5105                                    struct osd_device *osd,
5106                                    struct osd_thandle *oh,
5107                                    struct ldiskfs_dir_entry_2 *de)
5108 {
5109         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5110         struct osd_idmap_cache *idc;
5111         int rc, schedule = 0;
5112
5113         LASSERT(de != NULL);
5114
5115         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5116         if (likely(rc == 0)) {
5117                 idc = osd_idc_find_or_init(env, osd, fid);
5118                 if (IS_ERR(idc) || idc->oic_remote)
5119                         schedule = 1;
5120         } else if (rc == -ENODATA) {
5121                 /*
5122                  * can't get FID, postpone to the end of the
5123                  * transaction when iget() is safe
5124                  */
5125                 schedule = 1;
5126         } else {
5127                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5128         }
5129         if (schedule)
5130                 osd_schedule_agent_inode_removal(env, oh,
5131                                                  le32_to_cpu(de->inode));
5132 }
5133
5134 /**
5135  * Index delete function for interoperability mode (b11826).
5136  * It will remove the directory entry added by osd_index_ea_insert().
5137  * This entry is needed to maintain name->fid mapping.
5138  *
5139  * \param key,  key i.e. file entry to be deleted
5140  *
5141  * \retval   0, on success
5142  * \retval -ve, on error
5143  */
5144 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5145                                const struct dt_key *key, struct thandle *handle)
5146 {
5147         struct osd_object *obj = osd_dt_obj(dt);
5148         struct inode *dir = obj->oo_inode;
5149         struct dentry *dentry;
5150         struct osd_thandle *oh;
5151         struct ldiskfs_dir_entry_2 *de = NULL;
5152         struct buffer_head *bh;
5153         struct htree_lock *hlock = NULL;
5154         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5155         int rc;
5156
5157         ENTRY;
5158
5159         if (!dt_object_exists(dt))
5160                 RETURN(-ENOENT);
5161
5162         LINVRNT(osd_invariant(obj));
5163         LASSERT(!dt_object_remote(dt));
5164         LASSERT(handle != NULL);
5165
5166         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5167
5168         oh = container_of(handle, struct osd_thandle, ot_super);
5169         LASSERT(oh->ot_handle != NULL);
5170         LASSERT(oh->ot_handle->h_transaction != NULL);
5171
5172         ll_vfs_dq_init(dir);
5173         dentry = osd_child_dentry_get(env, obj,
5174                                       (char *)key, strlen((char *)key));
5175
5176         if (obj->oo_hl_head != NULL) {
5177                 hlock = osd_oti_get(env)->oti_hlock;
5178                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5179                                    dir, LDISKFS_HLOCK_DEL);
5180         } else {
5181                 down_write(&obj->oo_ext_idx_sem);
5182         }
5183
5184         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5185         if (!IS_ERR(bh)) {
5186                 /*
5187                  * If this is not the ".." entry, it might be a remote DNE
5188                  * entry and  we need to check if the FID is for a remote
5189                  * MDT.  If the FID is  not in the directory entry (e.g.
5190                  * upgraded 1.8 filesystem without dirdata enabled) then
5191                  * we need to get the FID from the LMA. For a remote directory
5192                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5193                  * case.
5194                  *
5195                  * Delete the entry before the agent inode in order to
5196                  * simplify error handling.  At worst an error after deleting
5197                  * the entry first might leak the agent inode afterward. The
5198                  * reverse would need filesystem abort in case of error deleting
5199                  * the entry after the agent had been removed, or leave a
5200                  * dangling entry pointing at a random inode.
5201                  */
5202                 if (strcmp((char *)key, dotdot) != 0)
5203                         osd_take_care_of_agent(env, osd, oh, de);
5204                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5205                 brelse(bh);
5206         } else {
5207                 rc = PTR_ERR(bh);
5208         }
5209         if (hlock != NULL)
5210                 ldiskfs_htree_unlock(hlock);
5211         else
5212                 up_write(&obj->oo_ext_idx_sem);
5213
5214         GOTO(out, rc);
5215 out:
5216         LASSERT(osd_invariant(obj));
5217         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5218         RETURN(rc);
5219 }
5220
5221 /**
5222  *      Lookup index for \a key and copy record to \a rec.
5223  *
5224  *      \param  dt      osd index object
5225  *      \param  key     key for index
5226  *      \param  rec     record reference
5227  *
5228  *      \retval  +ve  success : exact mach
5229  *      \retval  0    return record with key not greater than \a key
5230  *      \retval -ve   failure
5231  */
5232 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5233                                 struct dt_rec *rec, const struct dt_key *key)
5234 {
5235         struct osd_object *obj = osd_dt_obj(dt);
5236         struct iam_path_descr *ipd;
5237         struct iam_container *bag = &obj->oo_dir->od_container;
5238         struct osd_thread_info *oti = osd_oti_get(env);
5239         struct iam_iterator *it = &oti->oti_idx_it;
5240         struct iam_rec *iam_rec;
5241         int rc;
5242
5243         ENTRY;
5244
5245         if (!dt_object_exists(dt))
5246                 RETURN(-ENOENT);
5247
5248         LASSERT(osd_invariant(obj));
5249         LASSERT(!dt_object_remote(dt));
5250         LASSERT(bag->ic_object == obj->oo_inode);
5251
5252         ipd = osd_idx_ipd_get(env, bag);
5253         if (IS_ERR(ipd))
5254                 RETURN(-ENOMEM);
5255
5256         /* got ipd now we can start iterator. */
5257         iam_it_init(it, bag, 0, ipd);
5258
5259         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5260                 /* swab quota uid/gid provided by caller */
5261                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5262                 key = (const struct dt_key *)&oti->oti_quota_id;
5263         }
5264
5265         rc = iam_it_get(it, (struct iam_key *)key);
5266         if (rc >= 0) {
5267                 if (S_ISDIR(obj->oo_inode->i_mode))
5268                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5269                 else
5270                         iam_rec = (struct iam_rec *)rec;
5271
5272                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5273
5274                 if (S_ISDIR(obj->oo_inode->i_mode))
5275                         osd_fid_unpack((struct lu_fid *)rec,
5276                                        (struct osd_fid_pack *)iam_rec);
5277                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5278                         osd_quota_unpack(obj, rec);
5279         }
5280
5281         iam_it_put(it);
5282         iam_it_fini(it);
5283         osd_ipd_put(env, bag, ipd);
5284
5285         LINVRNT(osd_invariant(obj));
5286
5287         RETURN(rc);
5288 }
5289
5290 static int osd_index_declare_iam_insert(const struct lu_env *env,
5291                                         struct dt_object *dt,
5292                                         const struct dt_rec *rec,
5293                                         const struct dt_key *key,
5294                                         struct thandle *handle)
5295 {
5296         struct osd_thandle *oh;
5297
5298         LASSERT(handle != NULL);
5299
5300         oh = container_of0(handle, struct osd_thandle, ot_super);
5301         LASSERT(oh->ot_handle == NULL);
5302
5303         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5304                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5305
5306         return 0;
5307 }
5308
5309 /**
5310  *      Inserts (key, value) pair in \a dt index object.
5311  *
5312  *      \param  dt      osd index object
5313  *      \param  key     key for index
5314  *      \param  rec     record reference
5315  *      \param  th      transaction handler
5316  *
5317  *      \retval  0  success
5318  *      \retval -ve failure
5319  */
5320 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5321                                 const struct dt_rec *rec,
5322                                 const struct dt_key *key, struct thandle *th)
5323 {
5324         struct osd_object *obj = osd_dt_obj(dt);
5325         struct iam_path_descr *ipd;
5326         struct osd_thandle *oh;
5327         struct iam_container *bag;
5328         struct osd_thread_info *oti = osd_oti_get(env);
5329         struct iam_rec *iam_rec;
5330         int rc;
5331
5332         ENTRY;
5333
5334         if (!dt_object_exists(dt))
5335                 RETURN(-ENOENT);
5336
5337         LINVRNT(osd_invariant(obj));
5338         LASSERT(!dt_object_remote(dt));
5339
5340         bag = &obj->oo_dir->od_container;
5341         LASSERT(bag->ic_object == obj->oo_inode);
5342         LASSERT(th != NULL);
5343
5344         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5345
5346         ipd = osd_idx_ipd_get(env, bag);
5347         if (unlikely(ipd == NULL))
5348                 RETURN(-ENOMEM);
5349
5350         oh = container_of0(th, struct osd_thandle, ot_super);
5351         LASSERT(oh->ot_handle != NULL);
5352         LASSERT(oh->ot_handle->h_transaction != NULL);
5353         if (S_ISDIR(obj->oo_inode->i_mode)) {
5354                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5355                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5356                              &oti->oti_fid);
5357         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5358                 /* pack quota uid/gid */
5359                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5360                 key = (const struct dt_key *)&oti->oti_quota_id;
5361                 /* pack quota record */
5362                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5363                 iam_rec = (struct iam_rec *)rec;
5364         } else {
5365                 iam_rec = (struct iam_rec *)rec;
5366         }
5367
5368         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5369                         iam_rec, ipd);
5370         osd_ipd_put(env, bag, ipd);
5371         LINVRNT(osd_invariant(obj));
5372         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5373         RETURN(rc);
5374 }
5375
5376 /**
5377  * Calls ldiskfs_add_entry() to add directory entry
5378  * into the directory. This is required for
5379  * interoperability mode (b11826)
5380  *
5381  * \retval   0, on success
5382  * \retval -ve, on error
5383  */
5384 static int __osd_ea_add_rec(struct osd_thread_info *info,
5385                             struct osd_object *pobj, struct inode  *cinode,
5386                             const char *name, const struct lu_fid *fid,
5387                             struct htree_lock *hlock, struct thandle *th)
5388 {
5389         struct ldiskfs_dentry_param *ldp;
5390         struct dentry *child;
5391         struct osd_thandle *oth;
5392         int rc;
5393
5394         oth = container_of(th, struct osd_thandle, ot_super);
5395         LASSERT(oth->ot_handle != NULL);
5396         LASSERT(oth->ot_handle->h_transaction != NULL);
5397         LASSERT(pobj->oo_inode);
5398
5399         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5400         if (unlikely(pobj->oo_inode ==
5401                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5402                 ldp->edp_magic = 0;
5403         else
5404                 osd_get_ldiskfs_dirent_param(ldp, fid);
5405         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5406         child->d_fsdata = (void *)ldp;
5407         ll_vfs_dq_init(pobj->oo_inode);
5408         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5409                                    child, cinode, hlock);
5410         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5411                 struct ldiskfs_dir_entry_2      *de;
5412                 struct buffer_head              *bh;
5413                 int                              rc1;
5414
5415                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5416                                             NULL, hlock);
5417                 if (!IS_ERR(bh)) {
5418                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5419                                                                bh);
5420                         if (rc1 == 0) {
5421                                 if (S_ISDIR(cinode->i_mode))
5422                                         de->file_type = LDISKFS_DIRENT_LUFID |
5423                                                         LDISKFS_FT_REG_FILE;
5424                                 else
5425                                         de->file_type = LDISKFS_DIRENT_LUFID |
5426                                                         LDISKFS_FT_DIR;
5427                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5428                                                               NULL, bh);
5429                         }
5430                         brelse(bh);
5431                 }
5432         }
5433
5434         RETURN(rc);
5435 }
5436
5437 /**
5438  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5439  * into the directory.Also sets flags into osd object to
5440  * indicate dot and dotdot are created. This is required for
5441  * interoperability mode (b11826)
5442  *
5443  * \param dir   directory for dot and dotdot fixup.
5444  * \param obj   child object for linking
5445  *
5446  * \retval   0, on success
5447  * \retval -ve, on error
5448  */
5449 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5450                               struct osd_object *dir,
5451                               struct inode *parent_dir, const char *name,
5452                               const struct lu_fid *dot_fid,
5453                               const struct lu_fid *dot_dot_fid,
5454                               struct thandle *th)
5455 {
5456         struct inode *inode = dir->oo_inode;
5457         struct osd_thandle *oth;
5458         int result = 0;
5459
5460         oth = container_of(th, struct osd_thandle, ot_super);
5461         LASSERT(oth->ot_handle->h_transaction != NULL);
5462         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5463
5464         if (strcmp(name, dot) == 0) {
5465                 if (dir->oo_compat_dot_created) {
5466                         result = -EEXIST;
5467                 } else {
5468                         LASSERT(inode->i_ino == parent_dir->i_ino);
5469                         dir->oo_compat_dot_created = 1;
5470                         result = 0;
5471                 }
5472         } else if (strcmp(name, dotdot) == 0) {
5473                 if (!dir->oo_compat_dot_created)
5474                         return -EINVAL;
5475                 /* in case of rename, dotdot is already created */
5476                 if (dir->oo_compat_dotdot_created) {
5477                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5478                                                 dot_dot_fid, NULL, th);
5479                 }
5480
5481                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5482                         struct lu_fid tfid = *dot_dot_fid;
5483
5484                         tfid.f_oid--;
5485                         result = osd_add_dot_dotdot_internal(info,
5486                                         dir->oo_inode, parent_dir, dot_fid,
5487                                         &tfid, oth);
5488                 } else {
5489                         result = osd_add_dot_dotdot_internal(info,
5490                                         dir->oo_inode, parent_dir, dot_fid,
5491                                         dot_dot_fid, oth);
5492                 }
5493
5494                 if (result == 0)
5495                         dir->oo_compat_dotdot_created = 1;
5496         }
5497
5498         return result;
5499 }
5500
5501
5502 /**
5503  * It will call the appropriate osd_add* function and return the
5504  * value, return by respective functions.
5505  */
5506 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5507                           struct inode *cinode, const char *name,
5508                           const struct lu_fid *fid, struct thandle *th)
5509 {
5510         struct osd_thread_info *info = osd_oti_get(env);
5511         struct htree_lock *hlock;
5512         int rc;
5513
5514         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5515
5516         if (name[0] == '.' && (name[1] == '\0' ||
5517                                (name[1] == '.' && name[2] == '\0'))) {
5518                 if (hlock != NULL) {
5519                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5520                                            pobj->oo_inode, 0);
5521                 } else {
5522                         down_write(&pobj->oo_ext_idx_sem);
5523                 }
5524
5525                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5526                                         lu_object_fid(&pobj->oo_dt.do_lu),
5527                                         fid, th);
5528         } else {
5529                 if (hlock != NULL) {
5530                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5531                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5532                 } else {
5533                         down_write(&pobj->oo_ext_idx_sem);
5534                 }
5535
5536                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5537                         struct lu_fid *tfid = &info->oti_fid;
5538
5539                         *tfid = *fid;
5540                         tfid->f_ver = ~0;
5541                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5542                                               tfid, hlock, th);
5543                 } else {
5544                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5545                                               hlock, th);
5546                 }
5547         }
5548         if (hlock != NULL)
5549                 ldiskfs_htree_unlock(hlock);
5550         else
5551                 up_write(&pobj->oo_ext_idx_sem);
5552
5553         return rc;
5554 }
5555
5556 static int
5557 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5558                       struct osd_idmap_cache *oic)
5559 {
5560         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5561         struct lu_fid *fid = &oic->oic_fid;
5562         struct osd_inode_id *id = &oic->oic_lid;
5563         struct inode *inode = NULL;
5564         int once = 0;
5565         bool insert;
5566         int rc;
5567
5568         ENTRY;
5569
5570         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5571                 RETURN(0);
5572
5573         if (thread_is_running(&scrub->os_thread) &&
5574             scrub->os_pos_current > id->oii_ino)
5575                 RETURN(0);
5576
5577         if (dev->od_auto_scrub_interval == AS_NEVER ||
5578             ktime_get_real_seconds() <
5579             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5580                 RETURN(0);
5581
5582 again:
5583         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5584         if (rc == -ENOENT) {
5585                 __u32 gen = id->oii_gen;
5586
5587                 insert = true;
5588                 if (inode != NULL)
5589                         goto trigger;
5590
5591                 inode = osd_iget(oti, dev, id);
5592                 /* The inode has been removed (by race maybe). */
5593                 if (IS_ERR(inode)) {
5594                         rc = PTR_ERR(inode);
5595
5596                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5597                 }
5598
5599                 /* The OI mapping is lost. */
5600                 if (gen != OSD_OII_NOGEN)
5601                         goto trigger;
5602
5603                 /*
5604                  * The inode may has been reused by others, we do not know,
5605                  * leave it to be handled by subsequent osd_fid_lookup().
5606                  */
5607                 GOTO(out, rc = 0);
5608         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5609                 GOTO(out, rc);
5610         }
5611
5612         insert = false;
5613
5614 trigger:
5615         if (thread_is_running(&scrub->os_thread)) {
5616                 if (inode == NULL) {
5617                         inode = osd_iget(oti, dev, id);
5618                         /* The inode has been removed (by race maybe). */
5619                         if (IS_ERR(inode)) {
5620                                 rc = PTR_ERR(inode);
5621
5622                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5623                         }
5624                 }
5625
5626                 rc = osd_oii_insert(dev, oic, insert);
5627                 /*
5628                  * There is race condition between osd_oi_lookup and OI scrub.
5629                  * The OI scrub finished just after osd_oi_lookup() failure.
5630                  * Under such case, it is unnecessary to trigger OI scrub again,
5631                  * but try to call osd_oi_lookup() again.
5632                  */
5633                 if (unlikely(rc == -EAGAIN))
5634                         goto again;
5635
5636                 if (!S_ISDIR(inode->i_mode))
5637                         rc = 0;
5638                 else
5639                         rc = osd_check_lmv(oti, dev, inode, oic);
5640
5641                 GOTO(out, rc);
5642         }
5643
5644         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5645                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5646                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5647                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5648                        "%s: trigger partial OI scrub for RPC inconsistency "
5649                        "checking FID "DFID": rc = %d\n",
5650                        osd_dev2name(dev), PFID(fid), rc);
5651                 if (rc == 0 || rc == -EALREADY)
5652                         goto again;
5653         }
5654
5655         GOTO(out, rc);
5656
5657 out:
5658         if (inode)
5659                 iput(inode);
5660
5661         RETURN(rc);
5662 }
5663
5664 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5665                                struct osd_device *dev,
5666                                struct osd_idmap_cache *oic,
5667                                struct lu_fid *fid, __u32 ino)
5668 {
5669         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5670         struct inode *inode;
5671         int rc;
5672
5673         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5674         inode = osd_iget(oti, dev, &oic->oic_lid);
5675         if (IS_ERR(inode)) {
5676                 fid_zero(&oic->oic_fid);
5677                 return PTR_ERR(inode);
5678         }
5679
5680         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5681         iput(inode);
5682         if (rc != 0)
5683                 fid_zero(&oic->oic_fid);
5684         else
5685                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5686         return rc;
5687 }
5688
5689 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5690                       struct osd_inode_id *id, const struct lu_fid *fid)
5691 {
5692         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5693                id->oii_ino, id->oii_gen, info);
5694         info->oti_cache.oic_lid = *id;
5695         info->oti_cache.oic_fid = *fid;
5696         info->oti_cache.oic_dev = osd;
5697 }
5698
5699 /**
5700  * Get parent FID from the linkEA.
5701  *
5702  * For a directory which parent resides on remote MDT, to satisfy the
5703  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5704  * the other hand, to make the lookup(..) on the directory can return
5705  * the real parent FID, we append the real parent FID after its ".."
5706  * name entry in the /REMOTE_PARENT_DIR.
5707  *
5708  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5709  * backup. So after the restore, we cannot get the right parent FID from
5710  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5711  * we have stored the real parent FID in the directory object's linkEA,
5712  * we can parse the linkEA for the real parent FID.
5713  *
5714  * \param[in] env       pointer to the thread context
5715  * \param[in] obj       pointer to the object to be handled
5716  * \param[out]fid       pointer to the buffer to hold the parent FID
5717  *
5718  * \retval              0 for getting the real parent FID successfully
5719  * \retval              negative error number on failure
5720  */
5721 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5722                                     struct osd_object *obj,
5723                                     struct lu_fid *fid)
5724 {
5725         struct osd_thread_info *oti = osd_oti_get(env);
5726         struct lu_buf *buf = &oti->oti_big_buf;
5727         struct dentry *dentry = &oti->oti_obj_dentry;
5728         struct inode *inode = obj->oo_inode;
5729         struct linkea_data ldata = { NULL };
5730         int rc;
5731
5732         ENTRY;
5733
5734         fid_zero(fid);
5735         if (!S_ISDIR(inode->i_mode))
5736                 RETURN(-EIO);
5737
5738 again:
5739         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5740                              buf->lb_buf, buf->lb_len);
5741         if (rc == -ERANGE) {
5742                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5743                                      NULL, 0);
5744                 if (rc > 0) {
5745                         lu_buf_realloc(buf, rc);
5746                         if (buf->lb_buf == NULL)
5747                                 RETURN(-ENOMEM);
5748
5749                         goto again;
5750                 }
5751         }
5752
5753         if (unlikely(rc == 0))
5754                 RETURN(-ENODATA);
5755
5756         if (rc < 0)
5757                 RETURN(rc);
5758
5759         if (unlikely(buf->lb_buf == NULL)) {
5760                 lu_buf_realloc(buf, rc);
5761                 if (buf->lb_buf == NULL)
5762                         RETURN(-ENOMEM);
5763
5764                 goto again;
5765         }
5766
5767         ldata.ld_buf = buf;
5768         rc = linkea_init_with_rec(&ldata);
5769         if (!rc) {
5770                 linkea_first_entry(&ldata);
5771                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5772         }
5773
5774         RETURN(rc);
5775 }
5776
5777 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5778                                     struct inode *inode,
5779                                     const struct lu_fid *pfid,
5780                                     const char *name, const int namelen)
5781 {
5782         struct osd_thread_info *oti = osd_oti_get(env);
5783         struct lu_buf *buf = &oti->oti_big_buf;
5784         struct dentry *dentry = &oti->oti_obj_dentry;
5785         struct linkea_data ldata = { NULL };
5786         struct lu_name cname = { .ln_name = name,
5787                                  .ln_namelen = namelen };
5788         int rc;
5789
5790         ENTRY;
5791
5792 again:
5793         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5794                              buf->lb_buf, buf->lb_len);
5795         if (rc == -ERANGE)
5796                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5797
5798         if (rc < 0)
5799                 RETURN(rc);
5800
5801         if (unlikely(rc == 0))
5802                 RETURN(-ENODATA);
5803
5804         if (buf->lb_len < rc) {
5805                 lu_buf_realloc(buf, rc);
5806                 if (buf->lb_buf == NULL)
5807                         RETURN(-ENOMEM);
5808
5809                 goto again;
5810         }
5811
5812         ldata.ld_buf = buf;
5813         rc = linkea_init_with_rec(&ldata);
5814         if (!rc)
5815                 rc = linkea_links_find(&ldata, &cname, pfid);
5816
5817         RETURN(rc);
5818 }
5819
5820 /**
5821  * Calls ->lookup() to find dentry. From dentry get inode and
5822  * read inode's ea to get fid. This is required for  interoperability
5823  * mode (b11826)
5824  *
5825  * \retval   0, on success
5826  * \retval -ve, on error
5827  */
5828 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5829                              struct dt_rec *rec, const struct dt_key *key)
5830 {
5831         struct inode *dir = obj->oo_inode;
5832         struct dentry *dentry;
5833         struct ldiskfs_dir_entry_2 *de;
5834         struct buffer_head *bh;
5835         struct lu_fid *fid = (struct lu_fid *)rec;
5836         struct htree_lock *hlock = NULL;
5837         int ino;
5838         int rc;
5839
5840         ENTRY;
5841
5842         LASSERT(dir->i_op != NULL);
5843         LASSERT(dir->i_op->lookup != NULL);
5844
5845         dentry = osd_child_dentry_get(env, obj,
5846                                       (char *)key, strlen((char *)key));
5847
5848         if (obj->oo_hl_head != NULL) {
5849                 hlock = osd_oti_get(env)->oti_hlock;
5850                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5851                                    dir, LDISKFS_HLOCK_LOOKUP);
5852         } else {
5853                 down_read(&obj->oo_ext_idx_sem);
5854         }
5855
5856         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5857         if (!IS_ERR(bh)) {
5858                 struct osd_thread_info *oti = osd_oti_get(env);
5859                 struct osd_inode_id *id = &oti->oti_id;
5860                 struct osd_idmap_cache *oic = &oti->oti_cache;
5861                 struct osd_device *dev = osd_obj2dev(obj);
5862
5863                 ino = le32_to_cpu(de->inode);
5864                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5865                         brelse(bh);
5866                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5867                         GOTO(out, rc);
5868                 }
5869
5870                 rc = osd_get_fid_from_dentry(de, rec);
5871
5872                 /* done with de, release bh */
5873                 brelse(bh);
5874                 if (rc != 0) {
5875                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5876                                 const char *name = (const char *)key;
5877
5878                                 /*
5879                                  * If the parent is on remote MDT, and there
5880                                  * is no FID-in-dirent, then we have to get
5881                                  * the parent FID from the linkEA.
5882                                  */
5883                                 if (likely(strlen(name) == 2 &&
5884                                            name[0] == '.' && name[1] == '.'))
5885                                         rc = osd_get_pfid_from_linkea(env, obj,
5886                                                                       fid);
5887                         } else {
5888                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5889                         }
5890                 } else {
5891                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5892                 }
5893
5894                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5895                         fid_zero(&oic->oic_fid);
5896
5897                         GOTO(out, rc);
5898                 }
5899
5900                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5901                 rc = osd_consistency_check(oti, dev, oic);
5902                 if (rc == -ENOENT)
5903                         fid_zero(&oic->oic_fid);
5904                 else
5905                         /* Other error should not affect lookup result. */
5906                         rc = 0;
5907         } else {
5908                 rc = PTR_ERR(bh);
5909         }
5910
5911         GOTO(out, rc);
5912
5913 out:
5914         if (hlock != NULL)
5915                 ldiskfs_htree_unlock(hlock);
5916         else
5917                 up_read(&obj->oo_ext_idx_sem);
5918         return rc;
5919 }
5920
5921 static int osd_index_declare_ea_insert(const struct lu_env *env,
5922                                        struct dt_object *dt,
5923                                        const struct dt_rec *rec,
5924                                        const struct dt_key *key,
5925                                        struct thandle *handle)
5926 {
5927         struct osd_thandle *oh;
5928         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5929         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5930         const struct lu_fid *fid = rec1->rec_fid;
5931         int credits, rc = 0;
5932         struct osd_idmap_cache *idc;
5933
5934         ENTRY;
5935
5936         LASSERT(!dt_object_remote(dt));
5937         LASSERT(handle != NULL);
5938         LASSERT(fid != NULL);
5939         LASSERT(rec1->rec_type != 0);
5940
5941         oh = container_of0(handle, struct osd_thandle, ot_super);
5942         LASSERT(oh->ot_handle == NULL);
5943
5944         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5945
5946         /*
5947          * we can't call iget() while a transactions is running
5948          * (this can lead to a deadlock), but we need to know
5949          * inum and object type. so we find this information at
5950          * declaration and cache in per-thread info
5951          */
5952         idc = osd_idc_find_or_init(env, osd, fid);
5953         if (IS_ERR(idc))
5954                 RETURN(PTR_ERR(idc));
5955         if (idc->oic_remote) {
5956                 /*
5957                  * a reference to remote inode is represented by an
5958                  * agent inode which we have to create
5959                  */
5960                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5961                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5962         }
5963
5964         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5965
5966         if (osd_dt_obj(dt)->oo_inode != NULL) {
5967                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5968
5969                 /*
5970                  * We ignore block quota on meta pool (MDTs), so needn't
5971                  * calculate how many blocks will be consumed by this index
5972                  * insert
5973                  */
5974                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5975                                            i_gid_read(inode),
5976                                            i_projid_read(inode), 0,
5977                                            oh, osd_dt_obj(dt), NULL,
5978                                            OSD_QID_BLK);
5979                 if (rc)
5980                         RETURN(rc);
5981
5982 #ifdef HAVE_PROJECT_QUOTA
5983                 /*
5984                  * Reserve credits for local agent inode to transfer
5985                  * to 0, quota enforcement is ignored in this case.
5986                  */
5987                 if (idc->oic_remote &&
5988                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5989                     i_projid_read(inode) != 0)
5990                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5991                                                   0, i_projid_read(inode),
5992                                                   0, false, PRJQUOTA, true);
5993 #endif
5994         }
5995
5996         RETURN(rc);
5997 }
5998
5999 /**
6000  * Index add function for interoperability mode (b11826).
6001  * It will add the directory entry.This entry is needed to
6002  * maintain name->fid mapping.
6003  *
6004  * \param key it is key i.e. file entry to be inserted
6005  * \param rec it is value of given key i.e. fid
6006  *
6007  * \retval   0, on success
6008  * \retval -ve, on error
6009  */
6010 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6011                                const struct dt_rec *rec,
6012                                const struct dt_key *key, struct thandle *th)
6013 {
6014         struct osd_object *obj = osd_dt_obj(dt);
6015         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6016         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6017         const struct lu_fid *fid = rec1->rec_fid;
6018         const char *name = (const char *)key;
6019         struct osd_thread_info *oti = osd_oti_get(env);
6020         struct inode *child_inode = NULL;
6021         struct osd_idmap_cache *idc;
6022         int rc;
6023
6024         ENTRY;
6025
6026         if (!dt_object_exists(dt))
6027                 RETURN(-ENOENT);
6028
6029         LASSERT(osd_invariant(obj));
6030         LASSERT(!dt_object_remote(dt));
6031         LASSERT(th != NULL);
6032
6033         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6034
6035         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6036
6037         idc = osd_idc_find(env, osd, fid);
6038         if (unlikely(idc == NULL)) {
6039                 idc = osd_idc_find_or_init(env, osd, fid);
6040                 if (IS_ERR(idc)) {
6041                         /*
6042                          * this dt_insert() wasn't declared properly, so
6043                          * FID is missing in OI cache. we better do not
6044                          * lookup FID in FLDB/OI and don't risk to deadlock,
6045                          * but in some special cases (lfsck testing, etc)
6046                          * it's much simpler than fixing a caller.
6047                          *
6048                          * normally this error should be placed after the first
6049                          * find, but migrate may attach source stripes to
6050                          * target, which doesn't create stripes.
6051                          */
6052                         CERROR("%s: "DFID" wasn't declared for insert\n",
6053                                osd_name(osd), PFID(fid));
6054                         dump_stack();
6055                         RETURN(PTR_ERR(idc));
6056                 }
6057         }
6058
6059         if (idc->oic_remote) {
6060                 /* Insert remote entry */
6061                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6062                         child_inode =
6063                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6064                 } else {
6065                         child_inode = osd_create_local_agent_inode(env, osd,
6066                                         obj, fid, rec1->rec_type & S_IFMT, th);
6067                         if (IS_ERR(child_inode))
6068                                 RETURN(PTR_ERR(child_inode));
6069                 }
6070         } else {
6071                 /* Insert local entry */
6072                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6073                         /* for a reason OI cache wasn't filled properly */
6074                         CERROR("%s: OIC for "DFID" isn't filled\n",
6075                                osd_name(osd), PFID(fid));
6076                         RETURN(-EINVAL);
6077                 }
6078                 child_inode = oti->oti_inode;
6079                 if (unlikely(child_inode == NULL)) {
6080                         struct ldiskfs_inode_info *lii;
6081
6082                         OBD_ALLOC_PTR(lii);
6083                         if (lii == NULL)
6084                                 RETURN(-ENOMEM);
6085                         child_inode = oti->oti_inode = &lii->vfs_inode;
6086                 }
6087                 child_inode->i_sb = osd_sb(osd);
6088                 child_inode->i_ino = idc->oic_lid.oii_ino;
6089                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6090         }
6091
6092         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6093
6094         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6095                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6096
6097         if (child_inode && child_inode != oti->oti_inode)
6098                 iput(child_inode);
6099         LASSERT(osd_invariant(obj));
6100         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6101         RETURN(rc);
6102 }
6103
6104 /**
6105  *  Initialize osd Iterator for given osd index object.
6106  *
6107  *  \param  dt      osd index object
6108  */
6109
6110 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6111                                      struct dt_object *dt,
6112                                      __u32 unused)
6113 {
6114         struct osd_it_iam *it;
6115         struct osd_object *obj = osd_dt_obj(dt);
6116         struct lu_object *lo = &dt->do_lu;
6117         struct iam_path_descr *ipd;
6118         struct iam_container *bag = &obj->oo_dir->od_container;
6119
6120         if (!dt_object_exists(dt))
6121                 return ERR_PTR(-ENOENT);
6122
6123         OBD_ALLOC_PTR(it);
6124         if (it == NULL)
6125                 return ERR_PTR(-ENOMEM);
6126
6127         ipd = osd_it_ipd_get(env, bag);
6128         if (likely(ipd != NULL)) {
6129                 it->oi_obj = obj;
6130                 it->oi_ipd = ipd;
6131                 lu_object_get(lo);
6132                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6133                 return (struct dt_it *)it;
6134         } else {
6135                 OBD_FREE_PTR(it);
6136                 return ERR_PTR(-ENOMEM);
6137         }
6138 }
6139
6140 /**
6141  * free given Iterator.
6142  */
6143 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6144 {
6145         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6146         struct osd_object *obj = it->oi_obj;
6147
6148         iam_it_fini(&it->oi_it);
6149         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6150         osd_object_put(env, obj);
6151         OBD_FREE_PTR(it);
6152 }
6153
6154 /**
6155  *  Move Iterator to record specified by \a key
6156  *
6157  *  \param  di      osd iterator
6158  *  \param  key     key for index
6159  *
6160  *  \retval +ve  di points to record with least key not larger than key
6161  *  \retval  0   di points to exact matched key
6162  *  \retval -ve  failure
6163  */
6164
6165 static int osd_it_iam_get(const struct lu_env *env,
6166                           struct dt_it *di, const struct dt_key *key)
6167 {
6168         struct osd_thread_info *oti = osd_oti_get(env);
6169         struct osd_it_iam *it = (struct osd_it_iam *)di;
6170
6171         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6172                 /* swab quota uid/gid */
6173                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6174                 key = (struct dt_key *)&oti->oti_quota_id;
6175         }
6176
6177         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6178 }
6179
6180 /**
6181  *  Release Iterator
6182  *
6183  *  \param  di      osd iterator
6184  */
6185 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6186 {
6187         struct osd_it_iam *it = (struct osd_it_iam *)di;
6188
6189         iam_it_put(&it->oi_it);
6190 }
6191
6192 /**
6193  *  Move iterator by one record
6194  *
6195  *  \param  di      osd iterator
6196  *
6197  *  \retval +1   end of container reached
6198  *  \retval  0   success
6199  *  \retval -ve  failure
6200  */
6201
6202 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6203 {
6204         struct osd_it_iam *it = (struct osd_it_iam *)di;
6205
6206         return iam_it_next(&it->oi_it);
6207 }
6208
6209 /**
6210  * Return pointer to the key under iterator.
6211  */
6212
6213 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6214                                      const struct dt_it *di)
6215 {
6216         struct osd_thread_info *oti = osd_oti_get(env);
6217         struct osd_it_iam *it = (struct osd_it_iam *)di;
6218         struct osd_object *obj = it->oi_obj;
6219         struct dt_key *key;
6220
6221         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6222
6223         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6224                 /* swab quota uid/gid */
6225                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6226                 key = (struct dt_key *)&oti->oti_quota_id;
6227         }
6228
6229         return key;
6230 }
6231
6232 /**
6233  * Return size of key under iterator (in bytes)
6234  */
6235
6236 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6237 {
6238         struct osd_it_iam *it = (struct osd_it_iam *)di;
6239
6240         return iam_it_key_size(&it->oi_it);
6241 }
6242
6243 static inline void
6244 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6245 {
6246         /* check if file type is required */
6247         if (ent->lde_attrs & LUDA_TYPE) {
6248                 struct luda_type *lt;
6249                 int align = sizeof(*lt) - 1;
6250
6251                 len = (len + align) & ~align;
6252                 lt = (struct luda_type *)(ent->lde_name + len);
6253                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6254         }
6255
6256         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6257 }
6258
6259 /**
6260  * build lu direct from backend fs dirent.
6261  */
6262
6263 static inline void
6264 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6265                    char *name, __u16 namelen, __u16 type, __u32 attr)
6266 {
6267         ent->lde_attrs = attr | LUDA_FID;
6268         fid_cpu_to_le(&ent->lde_fid, fid);
6269
6270         ent->lde_hash = cpu_to_le64(offset);
6271         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6272
6273         strncpy(ent->lde_name, name, namelen);
6274         ent->lde_name[namelen] = '\0';
6275         ent->lde_namelen = cpu_to_le16(namelen);
6276
6277         /* append lustre attributes */
6278         osd_it_append_attrs(ent, namelen, type);
6279 }
6280
6281 /**
6282  * Return pointer to the record under iterator.
6283  */
6284 static int osd_it_iam_rec(const struct lu_env *env,
6285                           const struct dt_it *di,
6286                           struct dt_rec *dtrec, __u32 attr)
6287 {
6288         struct osd_it_iam *it = (struct osd_it_iam *)di;
6289         struct osd_thread_info *info = osd_oti_get(env);
6290
6291         ENTRY;
6292
6293         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6294                 const struct osd_fid_pack *rec;
6295                 struct lu_fid *fid = &info->oti_fid;
6296                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6297                 char *name;
6298                 int namelen;
6299                 __u64 hash;
6300                 int rc;
6301
6302                 name = (char *)iam_it_key_get(&it->oi_it);
6303                 if (IS_ERR(name))
6304                         RETURN(PTR_ERR(name));
6305
6306                 namelen = iam_it_key_size(&it->oi_it);
6307
6308                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6309                 if (IS_ERR(rec))
6310                         RETURN(PTR_ERR(rec));
6311
6312                 rc = osd_fid_unpack(fid, rec);
6313                 if (rc)
6314                         RETURN(rc);
6315
6316                 hash = iam_it_store(&it->oi_it);
6317
6318                 /* IAM does not store object type in IAM index (dir) */
6319                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6320                                    0, LUDA_FID);
6321         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6322                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6323                            (struct iam_rec *)dtrec);
6324                 osd_quota_unpack(it->oi_obj, dtrec);
6325         } else {
6326                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6327                            (struct iam_rec *)dtrec);
6328         }
6329
6330         RETURN(0);
6331 }
6332
6333 /**
6334  * Returns cookie for current Iterator position.
6335  */
6336 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6337 {
6338         struct osd_it_iam *it = (struct osd_it_iam *)di;
6339
6340         return iam_it_store(&it->oi_it);
6341 }
6342
6343 /**
6344  * Restore iterator from cookie.
6345  *
6346  * \param  di      osd iterator
6347  * \param  hash    Iterator location cookie
6348  *
6349  * \retval +ve  di points to record with least key not larger than key.
6350  * \retval  0   di points to exact matched key
6351  * \retval -ve  failure
6352  */
6353
6354 static int osd_it_iam_load(const struct lu_env *env,
6355                            const struct dt_it *di, __u64 hash)
6356 {
6357         struct osd_it_iam *it = (struct osd_it_iam *)di;
6358
6359         return iam_it_load(&it->oi_it, hash);
6360 }
6361
6362 static const struct dt_index_operations osd_index_iam_ops = {
6363         .dio_lookup         = osd_index_iam_lookup,
6364         .dio_declare_insert = osd_index_declare_iam_insert,
6365         .dio_insert         = osd_index_iam_insert,
6366         .dio_declare_delete = osd_index_declare_iam_delete,
6367         .dio_delete         = osd_index_iam_delete,
6368         .dio_it     = {
6369                 .init     = osd_it_iam_init,
6370                 .fini     = osd_it_iam_fini,
6371                 .get      = osd_it_iam_get,
6372                 .put      = osd_it_iam_put,
6373                 .next     = osd_it_iam_next,
6374                 .key      = osd_it_iam_key,
6375                 .key_size = osd_it_iam_key_size,
6376                 .rec      = osd_it_iam_rec,
6377                 .store    = osd_it_iam_store,
6378                 .load     = osd_it_iam_load
6379         }
6380 };
6381
6382
6383 /**
6384  * Creates or initializes iterator context.
6385  *
6386  * \retval struct osd_it_ea, iterator structure on success
6387  *
6388  */
6389 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6390                                     struct dt_object *dt,
6391                                     __u32 attr)
6392 {
6393         struct osd_object *obj = osd_dt_obj(dt);
6394         struct osd_thread_info *info = osd_oti_get(env);
6395         struct osd_it_ea *oie;
6396         struct file *file;
6397         struct lu_object *lo = &dt->do_lu;
6398         struct dentry *obj_dentry;
6399
6400         ENTRY;
6401
6402         if (!dt_object_exists(dt) || obj->oo_destroyed)
6403                 RETURN(ERR_PTR(-ENOENT));
6404
6405         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6406         if (oie == NULL)
6407                 RETURN(ERR_PTR(-ENOMEM));
6408         obj_dentry = &oie->oie_dentry;
6409
6410         obj_dentry->d_inode = obj->oo_inode;
6411         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6412         obj_dentry->d_name.hash = 0;
6413
6414         oie->oie_rd_dirent       = 0;
6415         oie->oie_it_dirent       = 0;
6416         oie->oie_dirent          = NULL;
6417         if (unlikely(!info->oti_it_ea_buf_used)) {
6418                 oie->oie_buf = info->oti_it_ea_buf;
6419                 info->oti_it_ea_buf_used = 1;
6420         } else {
6421                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6422                 if (oie->oie_buf == NULL)
6423                         RETURN(ERR_PTR(-ENOMEM));
6424         }
6425         oie->oie_obj = obj;
6426
6427         file = &oie->oie_file;
6428
6429         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6430         if (attr & LUDA_64BITHASH)
6431                 file->f_mode    = FMODE_64BITHASH;
6432         else
6433                 file->f_mode    = FMODE_32BITHASH;
6434         file->f_path.dentry     = obj_dentry;
6435         file->f_mapping         = obj->oo_inode->i_mapping;
6436         file->f_op              = obj->oo_inode->i_fop;
6437         set_file_inode(file, obj->oo_inode);
6438
6439         lu_object_get(lo);
6440         RETURN((struct dt_it *)oie);
6441 }
6442
6443 /**
6444  * Destroy or finishes iterator context.
6445  *
6446  * \param di iterator structure to be destroyed
6447  */
6448 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6449 {
6450         struct osd_thread_info *info = osd_oti_get(env);
6451         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6452         struct osd_object *obj = oie->oie_obj;
6453         struct inode *inode = obj->oo_inode;
6454
6455         ENTRY;
6456         oie->oie_file.f_op->release(inode, &oie->oie_file);
6457         osd_object_put(env, obj);
6458         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6459                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6460         else
6461                 info->oti_it_ea_buf_used = 0;
6462         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6463         EXIT;
6464 }
6465
6466 /**
6467  * It position the iterator at given key, so that next lookup continues from
6468  * that key Or it is similar to dio_it->load() but based on a key,
6469  * rather than file position.
6470  *
6471  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6472  * to the beginning.
6473  *
6474  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6475  */
6476 static int osd_it_ea_get(const struct lu_env *env,
6477                          struct dt_it *di, const struct dt_key *key)
6478 {
6479         struct osd_it_ea *it = (struct osd_it_ea *)di;
6480
6481         ENTRY;
6482         LASSERT(((const char *)key)[0] == '\0');
6483         it->oie_file.f_pos = 0;
6484         it->oie_rd_dirent = 0;
6485         it->oie_it_dirent = 0;
6486         it->oie_dirent = NULL;
6487
6488         RETURN(+1);
6489 }
6490
6491 /**
6492  * Does nothing
6493  */
6494 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6495 {
6496 }
6497
6498 struct osd_filldir_cbs {
6499 #ifdef HAVE_DIR_CONTEXT
6500         struct dir_context ctx;
6501 #endif
6502         struct osd_it_ea  *it;
6503 };
6504 /**
6505  * It is called internally by ->readdir(). It fills the
6506  * iterator's in-memory data structure with required
6507  * information i.e. name, namelen, rec_size etc.
6508  *
6509  * \param buf in which information to be filled in.
6510  * \param name name of the file in given dir
6511  *
6512  * \retval 0 on success
6513  * \retval 1 on buffer full
6514  */
6515 #ifdef HAVE_FILLDIR_USE_CTX
6516 static int osd_ldiskfs_filldir(struct dir_context *buf,
6517 #else
6518 static int osd_ldiskfs_filldir(void *buf,
6519 #endif
6520                                const char *name, int namelen,
6521                                loff_t offset, __u64 ino, unsigned int d_type)
6522 {
6523         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6524         struct osd_object *obj = it->oie_obj;
6525         struct osd_it_ea_dirent *ent = it->oie_dirent;
6526         struct lu_fid *fid = &ent->oied_fid;
6527         struct osd_fid_pack *rec;
6528
6529         ENTRY;
6530
6531 /* this should never happen */
6532         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6533                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6534                 RETURN(-EIO);
6535         }
6536
6537         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6538             OSD_IT_EA_BUFSIZE)
6539                 RETURN(1);
6540
6541         /* "." is just the object itself. */
6542         if (namelen == 1 && name[0] == '.') {
6543                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6544         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6545                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6546                 if (osd_fid_unpack(fid, rec) != 0)
6547                         fid_zero(fid);
6548         } else {
6549                 fid_zero(fid);
6550         }
6551         d_type &= ~LDISKFS_DIRENT_LUFID;
6552
6553         /* NOT export local root. */
6554         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6555                 ino = obj->oo_inode->i_ino;
6556                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6557         }
6558
6559         ent->oied_ino     = ino;
6560         ent->oied_off     = offset;
6561         ent->oied_namelen = namelen;
6562         ent->oied_type    = d_type;
6563
6564         memcpy(ent->oied_name, name, namelen);
6565
6566         it->oie_rd_dirent++;
6567         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6568         RETURN(0);
6569 }
6570
6571 /**
6572  * Calls ->readdir() to load a directory entry at a time
6573  * and stored it in iterator's in-memory data structure.
6574  *
6575  * \param di iterator's in memory structure
6576  *
6577  * \retval   0 on success
6578  * \retval -ve on error
6579  * \retval +1 reach the end of entry
6580  */
6581 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6582                                const struct dt_it *di)
6583 {
6584         struct osd_it_ea *it = (struct osd_it_ea *)di;
6585         struct osd_object *obj = it->oie_obj;
6586         struct inode *inode = obj->oo_inode;
6587         struct htree_lock *hlock = NULL;
6588         struct file *filp = &it->oie_file;
6589         int rc = 0;
6590         struct osd_filldir_cbs buf = {
6591 #ifdef HAVE_DIR_CONTEXT
6592                 .ctx.actor = osd_ldiskfs_filldir,
6593 #endif
6594                 .it = it
6595         };
6596
6597         ENTRY;
6598         it->oie_dirent = it->oie_buf;
6599         it->oie_rd_dirent = 0;
6600
6601         if (obj->oo_hl_head != NULL) {
6602                 hlock = osd_oti_get(env)->oti_hlock;
6603                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6604                                    inode, LDISKFS_HLOCK_READDIR);
6605         } else {
6606                 down_read(&obj->oo_ext_idx_sem);
6607         }
6608
6609 #ifdef HAVE_DIR_CONTEXT
6610         buf.ctx.pos = filp->f_pos;
6611 #ifdef HAVE_ITERATE_SHARED
6612         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6613 #else
6614         rc = inode->i_fop->iterate(filp, &buf.ctx);
6615 #endif
6616         filp->f_pos = buf.ctx.pos;
6617 #else
6618         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6619 #endif
6620
6621         if (hlock != NULL)
6622                 ldiskfs_htree_unlock(hlock);
6623         else
6624                 up_read(&obj->oo_ext_idx_sem);
6625
6626         if (it->oie_rd_dirent == 0) {
6627                 /*
6628                  * If it does not get any dirent, it means it has been reached
6629                  * to the end of the dir
6630                  */
6631                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6632                 if (rc == 0)
6633                         rc = 1;
6634         } else {
6635                 it->oie_dirent = it->oie_buf;
6636                 it->oie_it_dirent = 1;
6637         }
6638
6639         RETURN(rc);
6640 }
6641
6642 /**
6643  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6644  * to load a directory entry at a time and stored it in
6645  * iterator's in-memory data structure.
6646  *
6647  * \param di iterator's in memory structure
6648  *
6649  * \retval +ve iterator reached to end
6650  * \retval   0 iterator not reached to end
6651  * \retval -ve on error
6652  */
6653 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6654 {
6655         struct osd_it_ea *it = (struct osd_it_ea *)di;
6656         int rc;
6657
6658         ENTRY;
6659
6660         if (it->oie_it_dirent < it->oie_rd_dirent) {
6661                 it->oie_dirent =
6662                         (void *)it->oie_dirent +
6663                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6664                                        it->oie_dirent->oied_namelen);
6665                 it->oie_it_dirent++;
6666                 rc = 0;
6667         } else {
6668                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6669                         rc = 1;
6670                 else
6671                         rc = osd_ldiskfs_it_fill(env, di);
6672         }
6673
6674         RETURN(rc);
6675 }
6676
6677 /**
6678  * Returns the key at current position from iterator's in memory structure.
6679  *
6680  * \param di iterator's in memory structure
6681  *
6682  * \retval key i.e. struct dt_key on success
6683  */
6684 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6685                                     const struct dt_it *di)
6686 {
6687         struct osd_it_ea *it = (struct osd_it_ea *)di;
6688
6689         return (struct dt_key *)it->oie_dirent->oied_name;
6690 }
6691
6692 /**
6693  * Returns key's size at current position from iterator's in memory structure.
6694  *
6695  * \param di iterator's in memory structure
6696  *
6697  * \retval key_size i.e. struct dt_key on success
6698  */
6699 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6700 {
6701         struct osd_it_ea *it = (struct osd_it_ea *)di;
6702
6703         return it->oie_dirent->oied_namelen;
6704 }
6705
6706 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6707 {
6708         if (LDISKFS_DIR_REC_LEN(de) >=
6709             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6710                 return true;
6711
6712         return false;
6713 }
6714
6715 static inline bool
6716 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6717                      unsigned int blocksize, bool dotdot)
6718 {
6719         if (dotdot)
6720                 return osd_dotdot_has_space(de);
6721
6722         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6723             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6724                 return true;
6725
6726         return false;
6727 }
6728
6729 static int
6730 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6731                     handle_t *jh, struct dentry *dentry,
6732                     const struct lu_fid *fid, struct buffer_head *bh,
6733                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6734                     bool dotdot)
6735 {
6736         struct inode *dir = dentry->d_parent->d_inode;
6737         struct inode *inode = dentry->d_inode;
6738         struct osd_fid_pack *rec;
6739         struct ldiskfs_dentry_param *ldp;
6740         int namelen = dentry->d_name.len;
6741         int rc;
6742         struct osd_thread_info *info = osd_oti_get(env);
6743
6744         ENTRY;
6745
6746         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6747                 RETURN(0);
6748
6749         /* There is enough space to hold the FID-in-dirent. */
6750         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6751                 rc = ldiskfs_journal_get_write_access(jh, bh);
6752                 if (rc != 0)
6753                         RETURN(rc);
6754
6755                 de->name[namelen] = 0;
6756                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6757                 rec->fp_len = sizeof(struct lu_fid) + 1;
6758                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6759                 de->file_type |= LDISKFS_DIRENT_LUFID;
6760                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6761
6762                 RETURN(rc);
6763         }
6764
6765         LASSERT(!dotdot);
6766
6767         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6768         if (rc != 0)
6769                 RETURN(rc);
6770
6771         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6772         osd_get_ldiskfs_dirent_param(ldp, fid);
6773         dentry->d_fsdata = (void *)ldp;
6774         ll_vfs_dq_init(dir);
6775         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6776         /*
6777          * It is too bad, we cannot reinsert the name entry back.
6778          * That means we lose it!
6779          */
6780         if (rc != 0)
6781                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6782                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6783                        osd_ino2name(inode),
6784                        dir->i_ino, dir->i_generation, namelen,
6785                        dentry->d_name.name, PFID(fid), rc);
6786
6787         RETURN(rc);
6788 }
6789
6790 static int
6791 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6792                         struct osd_it_ea *it, struct lu_fid *fid,
6793                         struct osd_inode_id *id, __u32 *attr)
6794 {
6795         struct osd_thread_info *info = osd_oti_get(env);
6796         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6797         struct osd_device *dev = osd_obj2dev(obj);
6798         struct super_block *sb = osd_sb(dev);
6799         const char *devname = osd_name(dev);
6800         struct osd_it_ea_dirent *ent = it->oie_dirent;
6801         struct inode *dir = obj->oo_inode;
6802         struct htree_lock *hlock = NULL;
6803         struct buffer_head *bh = NULL;
6804         handle_t *jh = NULL;
6805         struct ldiskfs_dir_entry_2 *de;
6806         struct dentry *dentry;
6807         struct inode *inode;
6808         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6809         int credits;
6810         int rc;
6811         bool dotdot = false;
6812         bool dirty = false;
6813
6814         ENTRY;
6815
6816         if (ent->oied_name[0] == '.') {
6817                 if (ent->oied_namelen == 1)
6818                         RETURN(0);
6819
6820                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6821                         dotdot = true;
6822         }
6823
6824         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6825         inode = osd_iget(info, dev, id);
6826         if (IS_ERR(inode)) {
6827                 rc = PTR_ERR(inode);
6828                 if (rc == -ENOENT || rc == -ESTALE) {
6829                         /*
6830                          * Maybe dangling name entry, or
6831                          * corrupted directory entry.
6832                          */
6833                         *attr |= LUDA_UNKNOWN;
6834                         rc = 0;
6835                 } else {
6836                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6837                                "check_repair, dir = %lu/%u, name = %.*s, "
6838                                "ino = %llu, rc = %d\n",
6839                                devname, dir->i_ino, dir->i_generation,
6840                                ent->oied_namelen, ent->oied_name,
6841                                ent->oied_ino, rc);
6842                 }
6843
6844                 RETURN(rc);
6845         }
6846
6847         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6848                                            ent->oied_namelen);
6849         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6850         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6851                 lma = NULL;
6852         else if (rc != 0)
6853                 GOTO(out, rc);
6854
6855         /*
6856          * We need to ensure that the name entry is still valid.
6857          * Because it may be removed or renamed by other already.
6858          *
6859          * The unlink or rename operation will start journal before PDO lock,
6860          * so to avoid deadlock, here we need to start journal handle before
6861          * related PDO lock also. But because we do not know whether there
6862          * will be something to be repaired before PDO lock, we just start
6863          * journal without conditions.
6864          *
6865          * We may need to remove the name entry firstly, then insert back.
6866          * One credit is for user quota file update.
6867          * One credit is for group quota file update.
6868          * Two credits are for dirty inode.
6869          */
6870         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6871                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6872
6873         if (dev->od_dirent_journal != 0) {
6874
6875 again:
6876                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6877                 if (IS_ERR(jh)) {
6878                         rc = PTR_ERR(jh);
6879                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6880                                "check_repair, dir = %lu/%u, credits = %d, "
6881                                "name = %.*s, ino = %llu: rc = %d\n",
6882                                devname, dir->i_ino, dir->i_generation, credits,
6883                                ent->oied_namelen, ent->oied_name,
6884                                ent->oied_ino, rc);
6885
6886                         GOTO(out_inode, rc);
6887                 }
6888
6889                 if (obj->oo_hl_head != NULL) {
6890                         hlock = osd_oti_get(env)->oti_hlock;
6891                         /*
6892                          * "0" means exclusive lock for the whole directory.
6893                          * We need to prevent others access such name entry
6894                          * during the delete + insert. Neither HLOCK_ADD nor
6895                          * HLOCK_DEL cannot guarantee the atomicity.
6896                          */
6897                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6898                 } else {
6899                         down_write(&obj->oo_ext_idx_sem);
6900                 }
6901         } else {
6902                 if (obj->oo_hl_head != NULL) {
6903                         hlock = osd_oti_get(env)->oti_hlock;
6904                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6905                                            LDISKFS_HLOCK_LOOKUP);
6906                 } else {
6907                         down_read(&obj->oo_ext_idx_sem);
6908                 }
6909         }
6910
6911         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6912         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6913                 *attr |= LUDA_IGNORE;
6914
6915                 GOTO(out, rc = 0);
6916         }
6917
6918         /*
6919          * For dotdot entry, if there is not enough space to hold the
6920          * FID-in-dirent, just keep them there. It only happens when the
6921          * device upgraded from 1.8 or restored from MDT file-level backup.
6922          * For the whole directory, only dotdot entry have no FID-in-dirent
6923          * and needs to get FID from LMA when readdir, it will not affect the
6924          * performance much.
6925          */
6926         if (dotdot && !osd_dotdot_has_space(de)) {
6927                 *attr |= LUDA_UNKNOWN;
6928
6929                 GOTO(out, rc = 0);
6930         }
6931
6932         if (lma != NULL) {
6933                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6934                         GOTO(out, rc = 0);
6935
6936                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6937                         struct lu_fid *tfid = &lma->lma_self_fid;
6938
6939                         if (likely(dotdot &&
6940                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6941                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6942                                 /*
6943                                  * It must be REMOTE_PARENT_DIR and as the
6944                                  * 'dotdot' entry of remote directory
6945                                  */
6946                                 *attr |= LUDA_IGNORE;
6947                         } else {
6948                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6949                                        "parent directory, but got %.*s under "
6950                                        "dir = %lu/%u with the FID "DFID"\n",
6951                                        devname, ent->oied_namelen,
6952                                        ent->oied_name, dir->i_ino,
6953                                        dir->i_generation, PFID(tfid));
6954
6955                                 *attr |= LUDA_UNKNOWN;
6956                         }
6957
6958                         GOTO(out, rc = 0);
6959                 }
6960         }
6961
6962         if (!fid_is_zero(fid)) {
6963                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6964                                               ent->oied_namelen);
6965                 if (rc == -ENOENT ||
6966                     (rc == -ENODATA &&
6967                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6968                         /*
6969                          * linkEA does not recognize the dirent entry,
6970                          * it may because the dirent entry corruption
6971                          * and points to other's inode.
6972                          */
6973                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6974                                "recognize the dirent, dir = %lu/%u, "
6975                                " name = %.*s, ino = %llu, "
6976                                DFID": rc = %d\n", devname, dir->i_ino,
6977                                dir->i_generation, ent->oied_namelen,
6978                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6979                         *attr |= LUDA_UNKNOWN;
6980
6981                         GOTO(out, rc = 0);
6982                 }
6983
6984                 if (rc && rc != -ENODATA) {
6985                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6986                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6987                                DFID": rc = %d\n", devname, dir->i_ino,
6988                                dir->i_generation, ent->oied_namelen,
6989                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6990                         *attr |= LUDA_UNKNOWN;
6991
6992                         GOTO(out, rc = 0);
6993                 }
6994         }
6995
6996         if (lma != NULL) {
6997                 /*
6998                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6999                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7000                  */
7001                 if (*attr & LUDA_VERIFY_DRYRUN) {
7002                         *fid = lma->lma_self_fid;
7003                         *attr |= LUDA_REPAIR;
7004
7005                         GOTO(out, rc = 0);
7006                 }
7007
7008                 if (jh == NULL) {
7009                         brelse(bh);
7010                         dev->od_dirent_journal = 1;
7011                         if (hlock != NULL) {
7012                                 ldiskfs_htree_unlock(hlock);
7013                                 hlock = NULL;
7014                         } else {
7015                                 up_read(&obj->oo_ext_idx_sem);
7016                         }
7017
7018                         goto again;
7019                 }
7020
7021                 *fid = lma->lma_self_fid;
7022                 dirty = true;
7023                 /* Update or append the FID-in-dirent. */
7024                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7025                                          bh, de, hlock, dotdot);
7026                 if (rc == 0)
7027                         *attr |= LUDA_REPAIR;
7028                 else
7029                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7030                                "the dirent, dir = %lu/%u, name = %.*s, "
7031                                "ino = %llu, "DFID": rc = %d\n",
7032                                devname, dir->i_ino, dir->i_generation,
7033                                ent->oied_namelen, ent->oied_name,
7034                                ent->oied_ino, PFID(fid), rc);
7035         } else {
7036                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7037                 if (*attr & LUDA_VERIFY_DRYRUN) {
7038                         if (fid_is_sane(fid)) {
7039                                 *attr |= LUDA_REPAIR;
7040                         } else if (dev->od_index == 0) {
7041                                 lu_igif_build(fid, inode->i_ino,
7042                                               inode->i_generation);
7043                                 *attr |= LUDA_UPGRADE;
7044                         }
7045
7046                         GOTO(out, rc = 0);
7047                 }
7048
7049                 if (jh == NULL) {
7050                         brelse(bh);
7051                         dev->od_dirent_journal = 1;
7052                         if (hlock != NULL) {
7053                                 ldiskfs_htree_unlock(hlock);
7054                                 hlock = NULL;
7055                         } else {
7056                                 up_read(&obj->oo_ext_idx_sem);
7057                         }
7058
7059                         goto again;
7060                 }
7061
7062                 dirty = true;
7063                 if (unlikely(fid_is_sane(fid))) {
7064                         /*
7065                          * FID-in-dirent exists, but FID-in-LMA is lost.
7066                          * Trust the FID-in-dirent, and add FID-in-LMA.
7067                          */
7068                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7069                         if (rc == 0)
7070                                 *attr |= LUDA_REPAIR;
7071                         else
7072                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7073                                        "update dirent, dir = %lu/%u, "
7074                                        "name = %.*s, ino = %llu, "
7075                                        DFID": rc = %d\n",
7076                                        devname, dir->i_ino, dir->i_generation,
7077                                        ent->oied_namelen, ent->oied_name,
7078                                        ent->oied_ino, PFID(fid), rc);
7079                 } else if (dev->od_index == 0) {
7080                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7081                         /*
7082                          * It is probably IGIF object. Only aappend the
7083                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7084                          */
7085                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7086                                                  bh, de, hlock, dotdot);
7087                         if (rc == 0)
7088                                 *attr |= LUDA_UPGRADE;
7089                         else
7090                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7091                                        "after the dirent, dir = %lu/%u, "
7092                                        "name = %.*s, ino = %llu, "
7093                                        DFID": rc = %d\n",
7094                                        devname, dir->i_ino, dir->i_generation,
7095                                        ent->oied_namelen, ent->oied_name,
7096                                        ent->oied_ino, PFID(fid), rc);
7097                 }
7098         }
7099
7100         GOTO(out, rc);
7101
7102 out:
7103         if (!IS_ERR(bh))
7104                 brelse(bh);
7105         if (hlock != NULL) {
7106                 ldiskfs_htree_unlock(hlock);
7107         } else {
7108                 if (dev->od_dirent_journal != 0)
7109                         up_write(&obj->oo_ext_idx_sem);
7110                 else
7111                         up_read(&obj->oo_ext_idx_sem);
7112         }
7113
7114         if (jh != NULL)
7115                 ldiskfs_journal_stop(jh);
7116
7117 out_inode:
7118         iput(inode);
7119         if (rc >= 0 && !dirty)
7120                 dev->od_dirent_journal = 0;
7121
7122         return rc;
7123 }
7124
7125 /**
7126  * Returns the value at current position from iterator's in memory structure.
7127  *
7128  * \param di struct osd_it_ea, iterator's in memory structure
7129  * \param attr attr requested for dirent.
7130  * \param lde lustre dirent
7131  *
7132  * \retval   0 no error and \param lde has correct lustre dirent.
7133  * \retval -ve on error
7134  */
7135 static inline int osd_it_ea_rec(const struct lu_env *env,
7136                                 const struct dt_it *di,
7137                                 struct dt_rec *dtrec, __u32 attr)
7138 {
7139         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7140         struct osd_object      *obj   = it->oie_obj;
7141         struct osd_device      *dev   = osd_obj2dev(obj);
7142         struct osd_thread_info *oti   = osd_oti_get(env);
7143         struct osd_inode_id    *id    = &oti->oti_id;
7144         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7145         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7146         __u32 ino = it->oie_dirent->oied_ino;
7147         int rc = 0;
7148
7149         ENTRY;
7150
7151         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7152
7153         if (attr & LUDA_VERIFY) {
7154                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7155                         attr |= LUDA_IGNORE;
7156                         /*
7157                          * If the parent is on remote MDT, and there
7158                          * is no FID-in-dirent, then we have to get
7159                          * the parent FID from the linkEA.
7160                          */
7161                         if (!fid_is_sane(fid) &&
7162                             it->oie_dirent->oied_namelen == 2 &&
7163                             it->oie_dirent->oied_name[0] == '.' &&
7164                             it->oie_dirent->oied_name[1] == '.')
7165                                 osd_get_pfid_from_linkea(env, obj, fid);
7166                 } else {
7167                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7168                                                      &attr);
7169                 }
7170
7171                 if (!fid_is_sane(fid))
7172                         attr |= LUDA_UNKNOWN;
7173         } else {
7174                 attr &= ~LU_DIRENT_ATTRS_MASK;
7175                 if (!fid_is_sane(fid)) {
7176                         bool is_dotdot = false;
7177
7178                         if (it->oie_dirent->oied_namelen == 2 &&
7179                             it->oie_dirent->oied_name[0] == '.' &&
7180                             it->oie_dirent->oied_name[1] == '.')
7181                                 is_dotdot = true;
7182                         /*
7183                          * If the parent is on remote MDT, and there
7184                          * is no FID-in-dirent, then we have to get
7185                          * the parent FID from the linkEA.
7186                          */
7187                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7188                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7189                         } else {
7190                                 if (is_dotdot == false &&
7191                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7192                                         RETURN(-ENOENT);
7193
7194                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7195                         }
7196                 } else {
7197                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7198                 }
7199         }
7200
7201         /* Pack the entry anyway, at least the offset is right. */
7202         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7203                            it->oie_dirent->oied_name,
7204                            it->oie_dirent->oied_namelen,
7205                            it->oie_dirent->oied_type, attr);
7206
7207         if (rc < 0)
7208                 RETURN(rc);
7209
7210         if (osd_remote_fid(env, dev, fid))
7211                 RETURN(0);
7212
7213         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7214                 osd_add_oi_cache(oti, dev, id, fid);
7215
7216         RETURN(rc > 0 ? 0 : rc);
7217 }
7218
7219 /**
7220  * Returns the record size size at current position.
7221  *
7222  * This function will return record(lu_dirent) size in bytes.
7223  *
7224  * \param[in] env       execution environment
7225  * \param[in] di        iterator's in memory structure
7226  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7227  *                      calculate the lu_dirent size.
7228  *
7229  * \retval      record size(in bytes & in memory) of the current lu_dirent
7230  *              entry.
7231  */
7232 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7233                               __u32 attr)
7234 {
7235         struct osd_it_ea *it = (struct osd_it_ea *)di;
7236
7237         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7238 }
7239
7240 /**
7241  * Returns a cookie for current position of the iterator head, so that
7242  * user can use this cookie to load/start the iterator next time.
7243  *
7244  * \param di iterator's in memory structure
7245  *
7246  * \retval cookie for current position, on success
7247  */
7248 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7249 {
7250         struct osd_it_ea *it = (struct osd_it_ea *)di;
7251
7252         return it->oie_dirent->oied_off;
7253 }
7254
7255 /**
7256  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7257  * to load a directory entry at a time and stored it i inn,
7258  * in iterator's in-memory data structure.
7259  *
7260  * \param di struct osd_it_ea, iterator's in memory structure
7261  *
7262  * \retval +ve on success
7263  * \retval -ve on error
7264  */
7265 static int osd_it_ea_load(const struct lu_env *env,
7266                           const struct dt_it *di, __u64 hash)
7267 {
7268         struct osd_it_ea *it = (struct osd_it_ea *)di;
7269         int rc;
7270
7271         ENTRY;
7272         it->oie_file.f_pos = hash;
7273
7274         rc =  osd_ldiskfs_it_fill(env, di);
7275         if (rc > 0)
7276                 rc = -ENODATA;
7277
7278         if (rc == 0)
7279                 rc = 1;
7280
7281         RETURN(rc);
7282 }
7283
7284 /**
7285  * Index lookup function for interoperability mode (b11826).
7286  *
7287  * \param key,  key i.e. file name to be searched
7288  *
7289  * \retval +ve, on success
7290  * \retval -ve, on error
7291  */
7292 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7293                                struct dt_rec *rec, const struct dt_key *key)
7294 {
7295         struct osd_object *obj = osd_dt_obj(dt);
7296         int rc = 0;
7297
7298         ENTRY;
7299
7300         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7301         LINVRNT(osd_invariant(obj));
7302
7303         rc = osd_ea_lookup_rec(env, obj, rec, key);
7304         if (rc == 0)
7305                 rc = 1;
7306         RETURN(rc);
7307 }
7308
7309 /**
7310  * Index and Iterator operations for interoperability
7311  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7312  */
7313 static const struct dt_index_operations osd_index_ea_ops = {
7314         .dio_lookup         = osd_index_ea_lookup,
7315         .dio_declare_insert = osd_index_declare_ea_insert,
7316         .dio_insert         = osd_index_ea_insert,
7317         .dio_declare_delete = osd_index_declare_ea_delete,
7318         .dio_delete         = osd_index_ea_delete,
7319         .dio_it     = {
7320                 .init     = osd_it_ea_init,
7321                 .fini     = osd_it_ea_fini,
7322                 .get      = osd_it_ea_get,
7323                 .put      = osd_it_ea_put,
7324                 .next     = osd_it_ea_next,
7325                 .key      = osd_it_ea_key,
7326                 .key_size = osd_it_ea_key_size,
7327                 .rec      = osd_it_ea_rec,
7328                 .rec_size = osd_it_ea_rec_size,
7329                 .store    = osd_it_ea_store,
7330                 .load     = osd_it_ea_load
7331         }
7332 };
7333
7334 static void *osd_key_init(const struct lu_context *ctx,
7335                           struct lu_context_key *key)
7336 {
7337         struct osd_thread_info *info;
7338
7339         OBD_ALLOC_PTR(info);
7340         if (info == NULL)
7341                 return ERR_PTR(-ENOMEM);
7342
7343         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7344         if (info->oti_it_ea_buf == NULL)
7345                 goto out_free_info;
7346
7347         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7348
7349         info->oti_hlock = ldiskfs_htree_lock_alloc();
7350         if (info->oti_hlock == NULL)
7351                 goto out_free_ea;
7352
7353         return info;
7354
7355 out_free_ea:
7356         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7357 out_free_info:
7358         OBD_FREE_PTR(info);
7359         return ERR_PTR(-ENOMEM);
7360 }
7361
7362 static void osd_key_fini(const struct lu_context *ctx,
7363                          struct lu_context_key *key, void *data)
7364 {
7365         struct osd_thread_info *info = data;
7366         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7367         struct osd_idmap_cache *idc = info->oti_ins_cache;
7368
7369         if (info->oti_dio_pages) {
7370                 int i;
7371                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7372                         if (info->oti_dio_pages[i])
7373                                 __free_page(info->oti_dio_pages[i]);
7374                 }
7375                 OBD_FREE(info->oti_dio_pages,
7376                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7377         }
7378
7379         if (info->oti_inode != NULL)
7380                 OBD_FREE_PTR(lli);
7381         if (info->oti_hlock != NULL)
7382                 ldiskfs_htree_lock_free(info->oti_hlock);
7383         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7384         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7385         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7386         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7387         lu_buf_free(&info->oti_big_buf);
7388         if (idc != NULL) {
7389                 LASSERT(info->oti_ins_cache_size > 0);
7390                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7391                 info->oti_ins_cache = NULL;
7392                 info->oti_ins_cache_size = 0;
7393         }
7394         OBD_FREE_PTR(info);
7395 }
7396
7397 static void osd_key_exit(const struct lu_context *ctx,
7398                          struct lu_context_key *key, void *data)
7399 {
7400         struct osd_thread_info *info = data;
7401
7402         LASSERT(info->oti_r_locks == 0);
7403         LASSERT(info->oti_w_locks == 0);
7404         LASSERT(info->oti_txns    == 0);
7405 }
7406
7407 /* type constructor/destructor: osd_type_init, osd_type_fini */
7408 LU_TYPE_INIT_FINI(osd, &osd_key);
7409
7410 struct lu_context_key osd_key = {
7411         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7412         .lct_init = osd_key_init,
7413         .lct_fini = osd_key_fini,
7414         .lct_exit = osd_key_exit
7415 };
7416
7417
7418 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7419                            const char *name, struct lu_device *next)
7420 {
7421         struct osd_device *osd = osd_dev(d);
7422
7423         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7424             sizeof(osd->od_svname))
7425                 return -E2BIG;
7426         return osd_procfs_init(osd, name);
7427 }
7428
7429 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7430 {
7431         struct seq_server_site *ss = osd_seq_site(osd);
7432         int rc;
7433
7434         ENTRY;
7435
7436         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7437                 RETURN(0);
7438
7439         if (unlikely(ss == NULL))
7440                 RETURN(-ENODEV);
7441
7442         OBD_ALLOC_PTR(osd->od_cl_seq);
7443         if (osd->od_cl_seq == NULL)
7444                 RETURN(-ENOMEM);
7445
7446         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7447                              osd->od_svname, ss->ss_server_seq);
7448         if (rc != 0) {
7449                 OBD_FREE_PTR(osd->od_cl_seq);
7450                 osd->od_cl_seq = NULL;
7451                 RETURN(rc);
7452         }
7453
7454         if (ss->ss_node_id == 0) {
7455                 /*
7456                  * If the OSD on the sequence controller(MDT0), then allocate
7457                  * sequence here, otherwise allocate sequence after connected
7458                  * to MDT0 (see mdt_register_lwp_callback()).
7459                  */
7460                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7461                                    &osd->od_cl_seq->lcs_space, env);
7462         }
7463
7464         RETURN(rc);
7465 }
7466
7467 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7468 {
7469         if (osd->od_cl_seq == NULL)
7470                 return;
7471
7472         seq_client_fini(osd->od_cl_seq);
7473         OBD_FREE_PTR(osd->od_cl_seq);
7474         osd->od_cl_seq = NULL;
7475 }
7476
7477 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7478 {
7479         ENTRY;
7480
7481         /* shutdown quota slave instance associated with the device */
7482         if (o->od_quota_slave_md != NULL) {
7483                 struct qsd_instance *qsd = o->od_quota_slave_md;
7484
7485                 o->od_quota_slave_md = NULL;
7486                 qsd_fini(env, qsd);
7487         }
7488
7489         if (o->od_quota_slave_dt != NULL) {
7490                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7491
7492                 o->od_quota_slave_dt = NULL;
7493                 qsd_fini(env, qsd);
7494         }
7495
7496         osd_fid_fini(env, o);
7497         osd_scrub_cleanup(env, o);
7498
7499         RETURN(0);
7500 }
7501
7502 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7503 {
7504         ENTRY;
7505
7506         if (o->od_mnt != NULL) {
7507                 shrink_dcache_sb(osd_sb(o));
7508                 osd_sync(env, &o->od_dt_dev);
7509
7510                 mntput(o->od_mnt);
7511                 o->od_mnt = NULL;
7512         }
7513
7514         EXIT;
7515 }
7516
7517 static int osd_mount(const struct lu_env *env,
7518                      struct osd_device *o, struct lustre_cfg *cfg)
7519 {
7520         const char *name = lustre_cfg_string(cfg, 0);
7521         const char *dev = lustre_cfg_string(cfg, 1);
7522         const char *opts;
7523         unsigned long page, s_flags, lmd_flags = 0;
7524         struct page *__page;
7525         struct file_system_type *type;
7526         char *options = NULL;
7527         char *str;
7528         struct osd_thread_info *info = osd_oti_get(env);
7529         struct lu_fid *fid = &info->oti_fid;
7530         struct inode *inode;
7531         int rc = 0, force_over_512tb = 0;
7532
7533         ENTRY;
7534
7535         if (o->od_mnt != NULL)
7536                 RETURN(0);
7537
7538         if (strlen(dev) >= sizeof(o->od_mntdev))
7539                 RETURN(-E2BIG);
7540         strcpy(o->od_mntdev, dev);
7541
7542         str = lustre_cfg_string(cfg, 2);
7543         s_flags = simple_strtoul(str, NULL, 0);
7544         str = strstr(str, ":");
7545         if (str)
7546                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7547         opts = lustre_cfg_string(cfg, 3);
7548 #ifdef __BIG_ENDIAN
7549         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7550                 CERROR("%s: device %s extents feature is not guaranteed to "
7551                        "work on big-endian systems. Use \"bigendian_extents\" "
7552                        "mount option to override.\n", name, dev);
7553                 RETURN(-EINVAL);
7554         }
7555 #endif
7556 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7557         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7558                 CWARN("force_over_128tb option is deprecated. "
7559                       "Filesystems less than 512TB can be created without any "
7560                       "force options. Use force_over_512tb option for "
7561                       "filesystems greater than 512TB.\n");
7562         }
7563 #endif
7564 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7565         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7566                 CWARN("force_over_256tb option is deprecated. "
7567                       "Filesystems less than 512TB can be created without any "
7568                       "force options. Use force_over_512tb option for "
7569                       "filesystems greater than 512TB.\n");
7570         }
7571 #endif
7572
7573         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7574                 force_over_512tb = 1;
7575
7576         __page = alloc_page(GFP_KERNEL);
7577         if (__page == NULL)
7578                 GOTO(out, rc = -ENOMEM);
7579         page = (unsigned long)page_address(__page);
7580         options = (char *)page;
7581         *options = '\0';
7582         if (opts != NULL) {
7583                 /* strip out the options for back compatiblity */
7584                 static char *sout[] = {
7585                         "mballoc",
7586                         "iopen",
7587                         "noiopen",
7588                         "iopen_nopriv",
7589                         "extents",
7590                         "noextents",
7591                         /* strip out option we processed in osd */
7592                         "bigendian_extents",
7593                         "force_over_128tb",
7594                         "force_over_256tb",
7595                         "force_over_512tb",
7596                         NULL
7597                 };
7598                 strncat(options, opts, PAGE_SIZE);
7599                 for (rc = 0, str = options; sout[rc]; ) {
7600                         char *op = strstr(str, sout[rc]);
7601
7602                         if (op == NULL) {
7603                                 rc++;
7604                                 str = options;
7605                                 continue;
7606                         }
7607                         if (op == options || *(op - 1) == ',') {
7608                                 str = op + strlen(sout[rc]);
7609                                 if (*str == ',' || *str == '\0') {
7610                                         *str == ',' ? str++ : str;
7611                                         memmove(op, str, strlen(str) + 1);
7612                                 }
7613                         }
7614                         for (str = op; *str != ',' && *str != '\0'; str++)
7615                                 ;
7616                 }
7617         } else {
7618                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7619         }
7620
7621         /* Glom up mount options */
7622         if (*options != '\0')
7623                 strncat(options, ",", PAGE_SIZE);
7624         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7625
7626         type = get_fs_type("ldiskfs");
7627         if (!type) {
7628                 CERROR("%s: cannot find ldiskfs module\n", name);
7629                 GOTO(out, rc = -ENODEV);
7630         }
7631
7632         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7633         module_put(type->owner);
7634
7635         if (IS_ERR(o->od_mnt)) {
7636                 rc = PTR_ERR(o->od_mnt);
7637                 o->od_mnt = NULL;
7638                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7639                 GOTO(out, rc);
7640         }
7641
7642         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7643                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7644                                  force_over_512tb == 0) {
7645                 CERROR("%s: device %s LDISKFS does not support filesystems "
7646                        "greater than 512TB and can cause data corruption. "
7647                        "Use \"force_over_512tb\" mount option to override.\n",
7648                        name, dev);
7649                 GOTO(out_mnt, rc = -EINVAL);
7650         }
7651
7652         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7653                 if (priv_dev_set_rdonly) {
7654                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7655                         o->od_dt_dev.dd_rdonly = 1;
7656                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7657                                       name);
7658                 } else {
7659                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7660                                       name);
7661
7662                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7663                 }
7664         } else if (priv_dev_check_rdonly &&
7665                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7666                 CERROR("%s: underlying device %s is marked as "
7667                        "read-only. Setup failed\n", name, dev);
7668
7669                 GOTO(out_mnt, rc = -EROFS);
7670         }
7671
7672         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7673                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7674                 GOTO(out_mnt, rc = -EINVAL);
7675         }
7676
7677 #ifdef LDISKFS_MOUNT_DIRDATA
7678         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7679                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7680         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7681                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7682                       "enabling the dirdata feature. If you do not want to "
7683                       "downgrade to Lustre-1.x again, you can enable it via "
7684                       "'tune2fs -O dirdata device'\n", name, dev);
7685 #endif
7686         inode = osd_sb(o)->s_root->d_inode;
7687         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7688         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7689         if (rc != 0) {
7690                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7691                 GOTO(out_mnt, rc);
7692         }
7693
7694         if (lmd_flags & LMD_FLG_NOSCRUB)
7695                 o->od_auto_scrub_interval = AS_NEVER;
7696
7697         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7698                 /* do not use pagecache with flash-backed storage */
7699                 o->od_writethrough_cache = 0;
7700                 o->od_read_cache = 0;
7701         }
7702
7703         GOTO(out, rc = 0);
7704
7705 out_mnt:
7706         mntput(o->od_mnt);
7707         o->od_mnt = NULL;
7708
7709 out:
7710         if (__page)
7711                 __free_page(__page);
7712
7713         return rc;
7714 }
7715
7716 static struct lu_device *osd_device_fini(const struct lu_env *env,
7717                                          struct lu_device *d)
7718 {
7719         struct osd_device *o = osd_dev(d);
7720
7721         ENTRY;
7722
7723         osd_index_backup(env, o, false);
7724         osd_shutdown(env, o);
7725         osd_procfs_fini(o);
7726         osd_obj_map_fini(o);
7727         osd_umount(env, o);
7728
7729         RETURN(NULL);
7730 }
7731
7732 static int osd_device_init0(const struct lu_env *env,
7733                             struct osd_device *o,
7734                             struct lustre_cfg *cfg)
7735 {
7736         struct lu_device *l = osd2lu_dev(o);
7737         struct osd_thread_info *info;
7738         int rc;
7739         int cplen = 0;
7740
7741         /* if the module was re-loaded, env can loose its keys */
7742         rc = lu_env_refill((struct lu_env *)env);
7743         if (rc)
7744                 GOTO(out, rc);
7745         info = osd_oti_get(env);
7746         LASSERT(info);
7747
7748         l->ld_ops = &osd_lu_ops;
7749         o->od_dt_dev.dd_ops = &osd_dt_ops;
7750
7751         spin_lock_init(&o->od_osfs_lock);
7752         mutex_init(&o->od_otable_mutex);
7753         INIT_LIST_HEAD(&o->od_orphan_list);
7754         INIT_LIST_HEAD(&o->od_index_backup_list);
7755         INIT_LIST_HEAD(&o->od_index_restore_list);
7756         spin_lock_init(&o->od_lock);
7757         o->od_index_backup_policy = LIBP_NONE;
7758         o->od_t10_type = 0;
7759
7760         o->od_read_cache = 1;
7761         o->od_writethrough_cache = 1;
7762         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7763
7764         o->od_auto_scrub_interval = AS_DEFAULT;
7765
7766         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7767                         sizeof(o->od_svname));
7768         if (cplen >= sizeof(o->od_svname)) {
7769                 rc = -E2BIG;
7770                 GOTO(out, rc);
7771         }
7772
7773         o->od_index_backup_stop = 0;
7774         o->od_index = -1; /* -1 means index is invalid */
7775         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7776         if (rc == LDD_F_SV_TYPE_OST)
7777                 o->od_is_ost = 1;
7778
7779         o->od_full_scrub_ratio = OFSR_DEFAULT;
7780         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7781         rc = osd_mount(env, o, cfg);
7782         if (rc != 0)
7783                 GOTO(out, rc);
7784
7785         /* Can only check block device after mount */
7786         o->od_nonrotational =
7787                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7788
7789         rc = osd_obj_map_init(env, o);
7790         if (rc != 0)
7791                 GOTO(out_mnt, rc);
7792
7793         rc = lu_site_init(&o->od_site, l);
7794         if (rc != 0)
7795                 GOTO(out_compat, rc);
7796         o->od_site.ls_bottom_dev = l;
7797
7798         rc = lu_site_init_finish(&o->od_site);
7799         if (rc != 0)
7800                 GOTO(out_site, rc);
7801
7802         INIT_LIST_HEAD(&o->od_ios_list);
7803         /* setup scrub, including OI files initialization */
7804         o->od_in_init = 1;
7805         rc = osd_scrub_setup(env, o);
7806         o->od_in_init = 0;
7807         if (rc < 0)
7808                 GOTO(out_site, rc);
7809
7810         rc = osd_procfs_init(o, o->od_svname);
7811         if (rc != 0) {
7812                 CERROR("%s: can't initialize procfs: rc = %d\n",
7813                        o->od_svname, rc);
7814                 GOTO(out_scrub, rc);
7815         }
7816
7817         LASSERT(l->ld_site->ls_linkage.next != NULL);
7818         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7819
7820         /* initialize quota slave instance */
7821         /* currently it's no need to prepare qsd_instance_md for OST */
7822         if (!o->od_is_ost) {
7823                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7824                                                 &o->od_dt_dev,
7825                                                 o->od_proc_entry, true);
7826                 if (IS_ERR(o->od_quota_slave_md)) {
7827                         rc = PTR_ERR(o->od_quota_slave_md);
7828                         o->od_quota_slave_md = NULL;
7829                         GOTO(out_procfs, rc);
7830                 }
7831         }
7832
7833         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7834                                         o->od_proc_entry, false);
7835
7836         if (IS_ERR(o->od_quota_slave_dt)) {
7837                 if (o->od_quota_slave_md != NULL) {
7838                         qsd_fini(env, o->od_quota_slave_md);
7839                         o->od_quota_slave_md = NULL;
7840                 }
7841
7842                 rc = PTR_ERR(o->od_quota_slave_dt);
7843                 o->od_quota_slave_dt = NULL;
7844                 GOTO(out_procfs, rc);
7845         }
7846
7847         RETURN(0);
7848
7849 out_procfs:
7850         osd_procfs_fini(o);
7851 out_scrub:
7852         osd_scrub_cleanup(env, o);
7853 out_site:
7854         lu_site_fini(&o->od_site);
7855 out_compat:
7856         osd_obj_map_fini(o);
7857 out_mnt:
7858         osd_umount(env, o);
7859 out:
7860         return rc;
7861 }
7862
7863 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7864                                           struct lu_device_type *t,
7865                                           struct lustre_cfg *cfg)
7866 {
7867         struct osd_device *o;
7868         int rc;
7869
7870         OBD_ALLOC_PTR(o);
7871         if (o == NULL)
7872                 return ERR_PTR(-ENOMEM);
7873
7874         rc = dt_device_init(&o->od_dt_dev, t);
7875         if (rc == 0) {
7876                 /*
7877                  * Because the ctx might be revived in dt_device_init,
7878                  * refill the env here
7879                  */
7880                 lu_env_refill((struct lu_env *)env);
7881                 rc = osd_device_init0(env, o, cfg);
7882                 if (rc)
7883                         dt_device_fini(&o->od_dt_dev);
7884         }
7885
7886         if (unlikely(rc != 0))
7887                 OBD_FREE_PTR(o);
7888
7889         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7890 }
7891
7892 static struct lu_device *osd_device_free(const struct lu_env *env,
7893                                          struct lu_device *d)
7894 {
7895         struct osd_device *o = osd_dev(d);
7896
7897         ENTRY;
7898
7899         /* XXX: make osd top device in order to release reference */
7900         d->ld_site->ls_top_dev = d;
7901         lu_site_purge(env, d->ld_site, -1);
7902         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7903                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7904                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7905         }
7906         lu_site_fini(&o->od_site);
7907         dt_device_fini(&o->od_dt_dev);
7908         OBD_FREE_PTR(o);
7909         RETURN(NULL);
7910 }
7911
7912 static int osd_process_config(const struct lu_env *env,
7913                               struct lu_device *d, struct lustre_cfg *cfg)
7914 {
7915         struct osd_device *o = osd_dev(d);
7916         ssize_t count;
7917         int rc;
7918
7919         ENTRY;
7920
7921         switch (cfg->lcfg_command) {
7922         case LCFG_SETUP:
7923                 rc = osd_mount(env, o, cfg);
7924                 break;
7925         case LCFG_CLEANUP:
7926                 /*
7927                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7928                  * that may happend if hit failure during mount process.
7929                  */
7930                 osd_index_backup(env, o, false);
7931                 lu_dev_del_linkage(d->ld_site, d);
7932                 rc = osd_shutdown(env, o);
7933                 break;
7934         case LCFG_PARAM:
7935                 LASSERT(&o->od_dt_dev);
7936                 count  = class_modify_config(cfg, PARAM_OSD,
7937                                              &o->od_dt_dev.dd_kobj);
7938                 if (count < 0)
7939                         count = class_modify_config(cfg, PARAM_OST,
7940                                                     &o->od_dt_dev.dd_kobj);
7941                 rc = count > 0 ? 0 : count;
7942                 break;
7943         case LCFG_PRE_CLEANUP:
7944                 osd_scrub_stop(o);
7945                 osd_index_backup(env, o,
7946                                  o->od_index_backup_policy != LIBP_NONE);
7947                 rc = 0;
7948                 break;
7949         default:
7950                 rc = -ENOSYS;
7951         }
7952
7953         RETURN(rc);
7954 }
7955
7956 static int osd_recovery_complete(const struct lu_env *env,
7957                                  struct lu_device *d)
7958 {
7959         struct osd_device *osd = osd_dev(d);
7960         int rc = 0;
7961
7962         ENTRY;
7963
7964         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7965                 RETURN(0);
7966
7967         /*
7968          * start qsd instance on recovery completion, this notifies the quota
7969          * slave code that we are about to process new requests now
7970          */
7971         rc = qsd_start(env, osd->od_quota_slave_dt);
7972         if (rc == 0 && osd->od_quota_slave_md != NULL)
7973                 rc = qsd_start(env, osd->od_quota_slave_md);
7974
7975         RETURN(rc);
7976 }
7977
7978 /*
7979  * we use exports to track all osd users
7980  */
7981 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7982                            struct obd_device *obd, struct obd_uuid *cluuid,
7983                            struct obd_connect_data *data, void *localdata)
7984 {
7985         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7986         struct lustre_handle conn;
7987         int rc;
7988
7989         ENTRY;
7990
7991         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7992
7993         rc = class_connect(&conn, obd, cluuid);
7994         if (rc)
7995                 RETURN(rc);
7996
7997         *exp = class_conn2export(&conn);
7998
7999         spin_lock(&osd->od_osfs_lock);
8000         osd->od_connects++;
8001         spin_unlock(&osd->od_osfs_lock);
8002
8003         RETURN(0);
8004 }
8005
8006 /*
8007  * once last export (we don't count self-export) disappeared
8008  * osd can be released
8009  */
8010 static int osd_obd_disconnect(struct obd_export *exp)
8011 {
8012         struct obd_device *obd = exp->exp_obd;
8013         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8014         int rc, release = 0;
8015
8016         ENTRY;
8017
8018         /* Only disconnect the underlying layers on the final disconnect. */
8019         spin_lock(&osd->od_osfs_lock);
8020         osd->od_connects--;
8021         if (osd->od_connects == 0)
8022                 release = 1;
8023         spin_unlock(&osd->od_osfs_lock);
8024
8025         rc = class_disconnect(exp); /* bz 9811 */
8026
8027         if (rc == 0 && release)
8028                 class_manual_cleanup(obd);
8029         RETURN(rc);
8030 }
8031
8032 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8033                        struct lu_device *dev)
8034 {
8035         struct osd_device *osd = osd_dev(dev);
8036         struct lr_server_data *lsd =
8037                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8038         int result = 0;
8039
8040         ENTRY;
8041
8042         if (osd->od_quota_slave_md != NULL) {
8043                 /* set up quota slave objects for inode */
8044                 result = qsd_prepare(env, osd->od_quota_slave_md);
8045                 if (result != 0)
8046                         RETURN(result);
8047         }
8048
8049         if (osd->od_quota_slave_dt != NULL) {
8050                 /* set up quota slave objects for block */
8051                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8052                 if (result != 0)
8053                         RETURN(result);
8054         }
8055
8056
8057         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8058 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8059                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8060                         osd->od_index_in_idif = 1;
8061                 } else {
8062                         osd->od_index_in_idif = 0;
8063                         result = osd_register_proc_index_in_idif(osd);
8064                         if (result != 0)
8065                                 RETURN(result);
8066                 }
8067 #else
8068                 osd->od_index_in_idif = 1;
8069 #endif
8070         }
8071
8072         result = osd_fid_init(env, osd);
8073
8074         RETURN(result);
8075 }
8076
8077 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8078                          struct lu_fid *fid, struct md_op_data *op_data)
8079 {
8080         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8081
8082         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8083 }
8084
8085 static const struct lu_object_operations osd_lu_obj_ops = {
8086         .loo_object_init      = osd_object_init,
8087         .loo_object_delete    = osd_object_delete,
8088         .loo_object_release   = osd_object_release,
8089         .loo_object_free      = osd_object_free,
8090         .loo_object_print     = osd_object_print,
8091         .loo_object_invariant = osd_object_invariant
8092 };
8093
8094 const struct lu_device_operations osd_lu_ops = {
8095         .ldo_object_alloc      = osd_object_alloc,
8096         .ldo_process_config    = osd_process_config,
8097         .ldo_recovery_complete = osd_recovery_complete,
8098         .ldo_prepare           = osd_prepare,
8099 };
8100
8101 static const struct lu_device_type_operations osd_device_type_ops = {
8102         .ldto_init = osd_type_init,
8103         .ldto_fini = osd_type_fini,
8104
8105         .ldto_start = osd_type_start,
8106         .ldto_stop  = osd_type_stop,
8107
8108         .ldto_device_alloc = osd_device_alloc,
8109         .ldto_device_free  = osd_device_free,
8110
8111         .ldto_device_init = osd_device_init,
8112         .ldto_device_fini = osd_device_fini
8113 };
8114
8115 static struct lu_device_type osd_device_type = {
8116         .ldt_tags     = LU_DEVICE_DT,
8117         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8118         .ldt_ops      = &osd_device_type_ops,
8119         .ldt_ctx_tags = LCT_LOCAL,
8120 };
8121
8122 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8123 {
8124         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8125         struct super_block *sb = osd_sb(osd);
8126
8127         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8128 }
8129
8130 /*
8131  * lprocfs legacy support.
8132  */
8133 static struct obd_ops osd_obd_device_ops = {
8134         .o_owner = THIS_MODULE,
8135         .o_connect      = osd_obd_connect,
8136         .o_disconnect   = osd_obd_disconnect,
8137         .o_fid_alloc    = osd_fid_alloc,
8138         .o_health_check = osd_health_check,
8139 };
8140
8141 static ssize_t track_declares_assert_show(struct kobject *kobj,
8142                                    struct attribute *attr,
8143                                    char *buf)
8144 {
8145         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8146 }
8147
8148 static ssize_t track_declares_assert_store(struct kobject *kobj,
8149                                            struct attribute *attr,
8150                                            const char *buffer, size_t count)
8151 {
8152         bool track_declares_assert;
8153         int rc;
8154
8155         rc = kstrtobool(buffer, &track_declares_assert);
8156         if (rc)
8157                 return rc;
8158
8159         ldiskfs_track_declares_assert = track_declares_assert;
8160
8161         return count;
8162 }
8163 LUSTRE_RW_ATTR(track_declares_assert);
8164
8165 static int __init osd_init(void)
8166 {
8167         struct kobject *kobj;
8168         int rc;
8169
8170         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8171 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8172         /* please, try to keep osd_thread_info smaller than a page */
8173         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8174 #endif
8175
8176         osd_oi_mod_init();
8177
8178         rc = lu_kmem_init(ldiskfs_caches);
8179         if (rc)
8180                 return rc;
8181
8182 #ifdef CONFIG_KALLSYMS
8183         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8184         priv_dev_check_rdonly =
8185                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8186 #endif
8187
8188         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8189                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8190         if (rc) {
8191                 lu_kmem_fini(ldiskfs_caches);
8192                 return rc;
8193         }
8194
8195         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8196         if (kobj) {
8197                 rc = sysfs_create_file(kobj,
8198                                        &lustre_attr_track_declares_assert.attr);
8199                 kobject_put(kobj);
8200                 if (rc) {
8201                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8202                         rc = 0;
8203                 }
8204         }
8205         return rc;
8206 }
8207
8208 static void __exit osd_exit(void)
8209 {
8210         struct kobject *kobj;
8211
8212         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8213         if (kobj) {
8214                 sysfs_remove_file(kobj,
8215                                   &lustre_attr_track_declares_assert.attr);
8216                 kobject_put(kobj);
8217         }
8218         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8219         lu_kmem_fini(ldiskfs_caches);
8220 }
8221
8222 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8223 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8224 MODULE_VERSION(LUSTRE_VERSION_STRING);
8225 MODULE_LICENSE("GPL");
8226
8227 module_init(osd_init);
8228 module_exit(osd_exit);