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 #include <linux/uidgid.h>
  46
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/types.h>
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/fs.h>
  51 /* XATTR_{REPLACE,CREATE} */
  52 #include <linux/xattr.h>
  53
  54 #include <ldiskfs/ldiskfs.h>
  55 #include <ldiskfs/xattr.h>
  56 #include <ldiskfs/ldiskfs_extents.h>
  57 #undef ENTRY
  58 /*
  59  * struct OBD_{ALLOC,FREE}*()
  60  * OBD_FAIL_CHECK
  61  */
  62 #include <obd_support.h>
  63 /* struct ptlrpc_thread */
  64 #include <lustre_net.h>
  65 #include <lustre_fid.h>
  66 /* process_config */
  67 #include <uapi/linux/lustre/lustre_param.h>
  68
  69 #include "osd_internal.h"
  70 #include "osd_dynlocks.h"
  71
  72 /* llo_* api support */
  73 #include <md_object.h>
  74 #include <lustre_quota.h>
  75
  76 #include <lustre_linkea.h>
  77
  78 /* Maximum EA size is limited by LNET_MTU for remote objects */
  79 #define OSD_MAX_EA_SIZE 1048364
  80
  81 int ldiskfs_pdo = 1;
  82 module_param(ldiskfs_pdo, int, 0644);
  83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  84
  85 int ldiskfs_track_declares_assert;
  86 module_param(ldiskfs_track_declares_assert, int, 0644);
  87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  88
  89 /* Slab to allocate dynlocks */
  90 struct kmem_cache *dynlock_cachep;
  91
  92 /* Slab to allocate osd_it_ea */
  93 struct kmem_cache *osd_itea_cachep;
  94
  95 static struct lu_kmem_descr ldiskfs_caches[] = {
  96         {
  97                 .ckd_cache = &dynlock_cachep,
  98                 .ckd_name  = "dynlock_cache",
  99                 .ckd_size  = sizeof(struct dynlock_handle)
 100         },
 101         {
 102                 .ckd_cache = &osd_itea_cachep,
 103                 .ckd_name  = "osd_itea_cache",
 104                 .ckd_size  = sizeof(struct osd_it_ea)
 105         },
 106         {
 107                 .ckd_cache = NULL
 108         }
 109 };
 110
 111 static const char dot[] = ".";
 112 static const char dotdot[] = "..";
 113
 114 static const struct lu_object_operations      osd_lu_obj_ops;
 115 static const struct dt_object_operations      osd_obj_ops;
 116 static const struct dt_object_operations      osd_obj_otable_it_ops;
 117 static const struct dt_index_operations       osd_index_iam_ops;
 118 static const struct dt_index_operations       osd_index_ea_ops;
 119
 120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 121                           const struct lu_fid *fid);
 122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 123                                                 struct osd_device *osd);
 124
 125 int osd_trans_declare_op2rb[] = {
 126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 137 };
 138
 139 static int osd_has_index(const struct osd_object *obj)
 140 {
 141         return obj->oo_dt.do_index_ops != NULL;
 142 }
 143
 144 static int osd_object_invariant(const struct lu_object *l)
 145 {
 146         return osd_invariant(osd_obj(l));
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 153 {
 154         struct osd_thread_info *oti = osd_oti_get(env);
 155
 156         return oti->oti_w_locks > 0 && o->oo_owner == env;
 157 }
 158
 159 /*
 160  * Concurrency: doesn't access mutable data
 161  */
 162 static int osd_root_get(const struct lu_env *env,
 163                         struct dt_device *dev, struct lu_fid *f)
 164 {
 165         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 166         return 0;
 167 }
 168
 169 /*
 170  * the following set of functions are used to maintain per-thread
 171  * cache of FID->ino mapping. this mechanism is needed to resolve
 172  * FID to inode at dt_insert() which in turn stores ino in the
 173  * directory entries to keep ldiskfs compatible with ext[34].
 174  * due to locking-originated restrictions we can't lookup ino
 175  * using LU cache (deadlock is possible). lookup using OI is quite
 176  * expensive. so instead we maintain this cache and methods like
 177  * dt_create() fill it. so in the majority of cases dt_insert() is
 178  * able to find needed mapping in lockless manner.
 179  */
 180 static struct osd_idmap_cache *
 181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 182              const struct lu_fid *fid)
 183 {
 184         struct osd_thread_info *oti = osd_oti_get(env);
 185         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 186         int i;
 187
 188         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 189                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 190                         continue;
 191                 if (idc[i].oic_dev != osd)
 192                         continue;
 193
 194                 return idc + i;
 195         }
 196
 197         return NULL;
 198 }
 199
 200 static struct osd_idmap_cache *
 201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 202             const struct lu_fid *fid)
 203 {
 204         struct osd_thread_info *oti   = osd_oti_get(env);
 205         struct osd_idmap_cache *idc;
 206         int i;
 207
 208         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 209                 i = oti->oti_ins_cache_size * 2;
 210                 if (i == 0)
 211                         i = OSD_INS_CACHE_SIZE;
 212                 OBD_ALLOC(idc, sizeof(*idc) * i);
 213                 if (idc == NULL)
 214                         return ERR_PTR(-ENOMEM);
 215                 if (oti->oti_ins_cache != NULL) {
 216                         memcpy(idc, oti->oti_ins_cache,
 217                                oti->oti_ins_cache_used * sizeof(*idc));
 218                         OBD_FREE(oti->oti_ins_cache,
 219                                  oti->oti_ins_cache_used * sizeof(*idc));
 220                 }
 221                 oti->oti_ins_cache = idc;
 222                 oti->oti_ins_cache_size = i;
 223         }
 224
 225         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 226         idc->oic_fid = *fid;
 227         idc->oic_dev = osd;
 228         idc->oic_lid.oii_ino = 0;
 229         idc->oic_lid.oii_gen = 0;
 230         idc->oic_remote = 0;
 231
 232         return idc;
 233 }
 234
 235 /*
 236  * lookup mapping for the given fid in the cache, initialize a
 237  * new one if not found. the initialization checks whether the
 238  * object is local or remote. for local objects, OI is used to
 239  * learn ino/generation. the function is used when the caller
 240  * has no information about the object, e.g. at dt_insert().
 241  */
 242 static struct osd_idmap_cache *
 243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 244                      const struct lu_fid *fid)
 245 {
 246         struct osd_idmap_cache *idc;
 247         int rc;
 248
 249         idc = osd_idc_find(env, osd, fid);
 250         LASSERT(!IS_ERR(idc));
 251         if (idc != NULL)
 252                 return idc;
 253
 254         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 255                osd->od_svname, PFID(fid));
 256
 257         /* new mapping is needed */
 258         idc = osd_idc_add(env, osd, fid);
 259         if (IS_ERR(idc)) {
 260                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 261                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 262                 return idc;
 263         }
 264
 265         /* initialize it */
 266         rc = osd_remote_fid(env, osd, fid);
 267         if (unlikely(rc < 0))
 268                 return ERR_PTR(rc);
 269
 270         if (rc == 0) {
 271                 /* the object is local, lookup in OI */
 272                 /* XXX: probably cheaper to lookup in LU first? */
 273                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 274                                    &idc->oic_lid, 0);
 275                 if (unlikely(rc < 0)) {
 276                         CERROR("can't lookup: rc = %d\n", rc);
 277                         return ERR_PTR(rc);
 278                 }
 279         } else {
 280                 /* the object is remote */
 281                 idc->oic_remote = 1;
 282         }
 283
 284         return idc;
 285 }
 286
 287 /*
 288  * lookup mapping for given FID and fill it from the given object.
 289  * the object is lolcal by definition.
 290  */
 291 static int osd_idc_find_and_init(const struct lu_env *env,
 292                                  struct osd_device *osd,
 293                                  struct osd_object *obj)
 294 {
 295         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 296         struct osd_idmap_cache *idc;
 297
 298         idc = osd_idc_find(env, osd, fid);
 299         LASSERT(!IS_ERR(idc));
 300         if (idc != NULL) {
 301                 if (obj->oo_inode == NULL)
 302                         return 0;
 303                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 304                         LASSERT(idc->oic_lid.oii_ino == 0);
 305                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 306                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 307                 }
 308                 return 0;
 309         }
 310
 311         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 312                osd->od_svname, PFID(fid));
 313
 314         /* new mapping is needed */
 315         idc = osd_idc_add(env, osd, fid);
 316         if (IS_ERR(idc)) {
 317                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 318                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 319                 return PTR_ERR(idc);
 320         }
 321
 322         if (obj->oo_inode != NULL) {
 323                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 324                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 325         }
 326         return 0;
 327 }
 328
 329 /*
 330  * OSD object methods.
 331  */
 332
 333 /*
 334  * Concurrency: no concurrent access is possible that early in object
 335  * life-cycle.
 336  */
 337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 338                                           const struct lu_object_header *hdr,
 339                                           struct lu_device *d)
 340 {
 341         struct osd_object *mo;
 342
 343         OBD_ALLOC_PTR(mo);
 344         if (mo != NULL) {
 345                 struct lu_object *l;
 346                 struct lu_object_header *h;
 347                 struct osd_device *o = osd_dev(d);
 348
 349                 l = &mo->oo_dt.do_lu;
 350                 if (unlikely(o->od_in_init)) {
 351                         OBD_ALLOC_PTR(h);
 352                         if (!h) {
 353                                 OBD_FREE_PTR(mo);
 354                                 return NULL;
 355                         }
 356
 357                         lu_object_header_init(h);
 358                         lu_object_init(l, h, d);
 359                         lu_object_add_top(h, l);
 360                         mo->oo_header = h;
 361                 } else {
 362                         dt_object_init(&mo->oo_dt, NULL, d);
 363                         mo->oo_header = NULL;
 364                 }
 365
 366                 mo->oo_dt.do_ops = &osd_obj_ops;
 367                 l->lo_ops = &osd_lu_obj_ops;
 368                 init_rwsem(&mo->oo_sem);
 369                 init_rwsem(&mo->oo_ext_idx_sem);
 370                 spin_lock_init(&mo->oo_guard);
 371                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 372                 return l;
 373         }
 374         return NULL;
 375 }
 376
 377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 378                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 379 {
 380         int rc;
 381
 382         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 383                              (void *)loa, sizeof(*loa));
 384         if (rc > 0) {
 385                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 386
 387                 if (rc < sizeof(*lma))
 388                         return -EINVAL;
 389
 390                 rc = 0;
 391                 lustre_loa_swab(loa, true);
 392                 /* Check LMA compatibility */
 393                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 394                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 395                               "for fid = "DFID", ino = %lu\n",
 396                               osd_ino2name(inode),
 397                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 398                               PFID(&lma->lma_self_fid), inode->i_ino);
 399                         rc = -EOPNOTSUPP;
 400                 }
 401         } else if (rc == 0) {
 402                 rc = -ENODATA;
 403         }
 404
 405         return rc;
 406 }
 407
 408 /*
 409  * retrieve object from backend ext fs.
 410  **/
 411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 412                        struct osd_inode_id *id)
 413 {
 414         int rc;
 415         struct inode *inode = NULL;
 416
 417         /*
 418          * if we look for an inode withing a running
 419          * transaction, then we risk to deadlock
 420          * osd_dirent_check_repair() breaks this
 421          */
 422          /* LASSERT(current->journal_info == NULL); */
 423
 424         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 425         if (IS_ERR(inode)) {
 426                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 427                        id->oii_ino, PTR_ERR(inode));
 428         } else if (id->oii_gen != OSD_OII_NOGEN &&
 429                    inode->i_generation != id->oii_gen) {
 430                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 431                        "i_generation = %u\n",
 432                        id->oii_ino, id->oii_gen, inode->i_generation);
 433                 iput(inode);
 434                 inode = ERR_PTR(-ESTALE);
 435         } else if (inode->i_nlink == 0) {
 436                 /*
 437                  * due to parallel readdir and unlink,
 438                  * we can have dead inode here.
 439                  */
 440                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 441                 iput(inode);
 442                 inode = ERR_PTR(-ESTALE);
 443         } else if (is_bad_inode(inode)) {
 444                 CWARN("%s: bad inode: ino = %u\n",
 445                       osd_dev2name(dev), id->oii_ino);
 446                 iput(inode);
 447                 inode = ERR_PTR(-ENOENT);
 448         } else if ((rc = osd_attach_jinode(inode))) {
 449                 iput(inode);
 450                 inode = ERR_PTR(rc);
 451         } else {
 452                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 453                 if (id->oii_gen == OSD_OII_NOGEN)
 454                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 455
 456                 /*
 457                  * Do not update file c/mtime in ldiskfs.
 458                  * NB: we don't have any lock to protect this because we don't
 459                  * have reference on osd_object now, but contention with
 460                  * another lookup + attr_set can't happen in the tiny window
 461                  * between if (...) and set S_NOCMTIME.
 462                  */
 463                 if (!(inode->i_flags & S_NOCMTIME))
 464                         inode->i_flags |= S_NOCMTIME;
 465         }
 466         return inode;
 467 }
 468
 469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 470                           handle_t *handle, struct dentry *child,
 471                           struct inode *inode, struct htree_lock *hlock)
 472 {
 473         int rc, rc2;
 474
 475         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 476         if (rc == -ENOBUFS || rc == -ENOSPC) {
 477                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 478                 struct inode *parent = child->d_parent->d_inode;
 479                 struct lu_fid *fid = NULL;
 480
 481                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 482                 if (!rc2) {
 483                         fid = &loa->loa_lma.lma_self_fid;
 484                 } else if (rc2 == -ENODATA) {
 485                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 486                                 fid = &info->oti_fid3;
 487                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 488                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 489                                 fid = &info->oti_fid3;
 490                                 lu_igif_build(fid, parent->i_ino,
 491                                               parent->i_generation);
 492                         }
 493                 }
 494
 495                 if (fid != NULL)
 496                         /* below message is checked in sanity.sh test_129 */
 497                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
 498                               osd_name(osd), parent->i_ino, PFID(fid),
 499                               rc == -ENOSPC ? "has reached" : "is approaching");
 500                 else
 501                         /* below message is checked in sanity.sh test_129 */
 502                         CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
 503                               osd_name(osd), parent->i_ino,
 504                               rc == -ENOSPC ? "has reached" : "is approaching");
 505
 506                 /* ignore such error now */
 507                 if (rc == -ENOBUFS)
 508                         rc = 0;
 509         }
 510
 511         return rc;
 512 }
 513
 514
 515 struct inode *
 516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 517              struct osd_inode_id *id, struct lu_fid *fid)
 518 {
 519         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 520         struct inode *inode;
 521         int rc;
 522
 523         inode = osd_iget(info, dev, id);
 524         if (IS_ERR(inode))
 525                 return inode;
 526
 527         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 528         if (!rc) {
 529                 *fid = loa->loa_lma.lma_self_fid;
 530         } else if (rc == -ENODATA) {
 531                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 532                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 533                 else
 534                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 535         } else {
 536                 iput(inode);
 537                 inode = ERR_PTR(rc);
 538         }
 539         return inode;
 540 }
 541
 542 static struct inode *osd_iget_check(struct osd_thread_info *info,
 543                                     struct osd_device *dev,
 544                                     const struct lu_fid *fid,
 545                                     struct osd_inode_id *id,
 546                                     bool trusted)
 547 {
 548         struct inode *inode;
 549         int rc = 0;
 550
 551         ENTRY;
 552
 553         /*
 554          * The cached OI mapping is trustable. If we cannot locate the inode
 555          * via the cached OI mapping, then return the failure to the caller
 556          * directly without further OI checking.
 557          */
 558
 559 again:
 560         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 561         if (IS_ERR(inode)) {
 562                 rc = PTR_ERR(inode);
 563                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 564                         goto check_oi;
 565
 566                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 567                        PFID(fid), id->oii_ino, rc);
 568                 GOTO(put, rc);
 569         }
 570
 571         if (is_bad_inode(inode)) {
 572                 rc = -ENOENT;
 573                 if (!trusted)
 574                         goto check_oi;
 575
 576                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 577                        PFID(fid), id->oii_ino);
 578                 GOTO(put, rc);
 579         }
 580
 581         if (id->oii_gen != OSD_OII_NOGEN &&
 582             inode->i_generation != id->oii_gen) {
 583                 rc = -ESTALE;
 584                 if (!trusted)
 585                         goto check_oi;
 586
 587                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 588                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 589                        id->oii_ino, id->oii_gen, inode->i_generation);
 590                 GOTO(put, rc);
 591         }
 592
 593         if (inode->i_nlink == 0) {
 594                 rc = -ENOENT;
 595                 if (!trusted)
 596                         goto check_oi;
 597
 598                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 599                        PFID(fid), id->oii_ino);
 600                 GOTO(put, rc);
 601         }
 602
 603         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 604
 605 check_oi:
 606         if (rc != 0) {
 607                 __u32 saved_ino = id->oii_ino;
 608                 __u32 saved_gen = id->oii_gen;
 609
 610                 LASSERT(!trusted);
 611                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 612
 613                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 614                 /*
 615                  * XXX: There are four possible cases:
 616                  *      1. rc = 0.
 617                  *         Backup/restore caused the OI invalid.
 618                  *      2. rc = 0.
 619                  *         Someone unlinked the object but NOT removed
 620                  *         the OI mapping, such as mount target device
 621                  *         as ldiskfs, and modify something directly.
 622                  *      3. rc = -ENOENT.
 623                  *         Someone just removed the object between the
 624                  *         former oi_lookup and the iget. It is normal.
 625                  *      4. Other failure cases.
 626                  *
 627                  *      Generally, when the device is mounted, it will
 628                  *      auto check whether the system is restored from
 629                  *      file-level backup or not. We trust such detect
 630                  *      to distinguish the 1st case from the 2nd case:
 631                  *      if the OI files are consistent but may contain
 632                  *      stale OI mappings because of case 2, if iget()
 633                  *      returns -ENOENT or -ESTALE, then it should be
 634                  *      the case 2.
 635                  */
 636                 if (rc != 0)
 637                         /*
 638                          * If the OI mapping was in OI file before the
 639                          * osd_iget_check(), but now, it is disappear,
 640                          * then it must be removed by race. That is a
 641                          * normal race case.
 642                          */
 643                         GOTO(put, rc);
 644
 645                 /*
 646                  * It is the OI scrub updated the OI mapping by race.
 647                  * The new OI mapping must be valid.
 648                  */
 649                 if (saved_ino != id->oii_ino ||
 650                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 651                         if (!IS_ERR(inode))
 652                                 iput(inode);
 653
 654                         trusted = true;
 655                         goto again;
 656                 }
 657
 658                 if (IS_ERR(inode)) {
 659                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 660                             SF_INCONSISTENT)
 661                                 /*
 662                                  * It still can be the case 2, but we cannot
 663                                  * distinguish it from the case 1. So return
 664                                  * -EREMCHG to block current operation until
 665                                  *  OI scrub rebuilt the OI mappings.
 666                                  */
 667                                 rc = -EREMCHG;
 668                         else
 669                                 rc = -ENOENT;
 670
 671                         GOTO(put, rc);
 672                 }
 673
 674                 if (inode->i_generation == id->oii_gen)
 675                         rc = -ENOENT;
 676                 else
 677                         rc = -EREMCHG;
 678         } else {
 679                 if (id->oii_gen == OSD_OII_NOGEN)
 680                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 681
 682                 /*
 683                  * Do not update file c/mtime in ldiskfs.
 684                  * NB: we don't have any lock to protect this because we don't
 685                  * have reference on osd_object now, but contention with
 686                  * another lookup + attr_set can't happen in the tiny window
 687                  * between if (...) and set S_NOCMTIME.
 688                  */
 689                 if (!(inode->i_flags & S_NOCMTIME))
 690                         inode->i_flags |= S_NOCMTIME;
 691         }
 692
 693         GOTO(put, rc);
 694
 695 put:
 696         if (rc != 0) {
 697                 if (!IS_ERR(inode))
 698                         iput(inode);
 699
 700                 inode = ERR_PTR(rc);
 701         }
 702
 703         return inode;
 704 }
 705
 706 /**
 707  * \retval +v: new filter_fid does not contain self-fid
 708  * \retval 0:  filter_fid_18_23, contains self-fid
 709  * \retval -v: other failure cases
 710  */
 711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 712                  struct dentry *dentry, struct lu_fid *fid)
 713 {
 714         struct filter_fid *ff = &info->oti_ff;
 715         struct ost_id *ostid = &info->oti_ostid;
 716         int rc;
 717
 718         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 719         if (rc == sizeof(struct filter_fid_18_23)) {
 720                 struct filter_fid_18_23 *ff_old = (void *)ff;
 721
 722                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 723                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 724                 /*
 725                  * XXX: use 0 as the index for compatibility, the caller will
 726                  * handle index related issues when necessary.
 727                  */
 728                 if (!rc)
 729                         ostid_to_fid(fid, ostid, 0);
 730         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 731                 rc = 1;
 732         } else if (rc >= 0) {
 733                 rc = -EINVAL;
 734         }
 735
 736         return rc;
 737 }
 738
 739 static int osd_lma_self_repair(struct osd_thread_info *info,
 740                                struct osd_device *osd, struct inode *inode,
 741                                const struct lu_fid *fid, __u32 compat)
 742 {
 743         handle_t *jh;
 744         int rc;
 745
 746         LASSERT(current->journal_info == NULL);
 747
 748         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 749                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 750         if (IS_ERR(jh)) {
 751                 rc = PTR_ERR(jh);
 752                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 753                       osd_name(osd), rc);
 754                 return rc;
 755         }
 756
 757         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 758         if (rc != 0)
 759                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 760                       osd_name(osd), rc);
 761         ldiskfs_journal_stop(jh);
 762         return rc;
 763 }
 764
 765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 766 {
 767         struct osd_thread_info *info = osd_oti_get(env);
 768         struct osd_device *osd = osd_obj2dev(obj);
 769         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 770         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 771         struct inode *inode = obj->oo_inode;
 772         struct dentry *dentry = &info->oti_obj_dentry;
 773         struct lu_fid *fid = NULL;
 774         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 775         int rc;
 776
 777         ENTRY;
 778
 779         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 780                              (void *)loa, sizeof(*loa));
 781         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 782                 fid = &lma->lma_self_fid;
 783                 rc = osd_get_idif(info, inode, dentry, fid);
 784                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 785                         /*
 786                          * For the given OST-object, if it has neither LMA nor
 787                          * FID in XATTR_NAME_FID, then the given FID (which is
 788                          * contained in the @obj, from client RPC for locating
 789                          * the OST-object) is trusted. We use it to generate
 790                          * the LMA.
 791                          */
 792                         osd_lma_self_repair(info, osd, inode, rfid,
 793                                             LMAC_FID_ON_OST);
 794                         RETURN(0);
 795                 }
 796         }
 797
 798         if (rc < 0)
 799                 RETURN(rc);
 800
 801         if (rc > 0) {
 802                 rc = 0;
 803                 lustre_lma_swab(lma);
 804                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 805                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 806                               S_ISREG(inode->i_mode)))) {
 807                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 808                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 809                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 810                               PFID(rfid), inode->i_ino);
 811                         rc = -EOPNOTSUPP;
 812                 } else {
 813                         fid = &lma->lma_self_fid;
 814                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 815                             osd->od_is_ost)
 816                                 obj->oo_pfid_in_lma = 1;
 817                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 818                             !osd->od_is_ost)
 819                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 820                 }
 821         }
 822
 823         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 824                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 825                         struct ost_id   *oi   = &info->oti_ostid;
 826                         struct lu_fid   *fid1 = &info->oti_fid3;
 827                         __u32            idx  = fid_idif_ost_idx(rfid);
 828
 829                         /*
 830                          * For old IDIF, the OST index is not part of the IDIF,
 831                          * Means that different OSTs may have the same IDIFs.
 832                          * Under such case, we need to make some compatible
 833                          * check to make sure to trigger OI scrub properly.
 834                          */
 835                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 836                                 /* Given @rfid is new, LMA is old. */
 837                                 fid_to_ostid(fid, oi);
 838                                 ostid_to_fid(fid1, oi, idx);
 839                                 if (lu_fid_eq(fid1, rfid)) {
 840                                         if (osd->od_index_in_idif)
 841                                                 osd_lma_self_repair(info, osd,
 842                                                         inode, rfid,
 843                                                         LMAC_FID_ON_OST);
 844                                         RETURN(0);
 845                                 }
 846                         }
 847                 }
 848
 849                 rc = -EREMCHG;
 850         }
 851
 852         RETURN(rc);
 853 }
 854
 855 struct osd_check_lmv_buf {
 856         /* please keep it as first member */
 857         struct dir_context ctx;
 858         struct osd_thread_info *oclb_info;
 859         struct osd_device *oclb_dev;
 860         struct osd_idmap_cache *oclb_oic;
 861         int oclb_items;
 862         bool oclb_found;
 863 };
 864
 865 /**
 866  * It is called internally by ->iterate*() to filter out the
 867  * local slave object's FID of the striped directory.
 868  *
 869  * \retval      1 found the local slave's FID
 870  * \retval      0 continue to check next item
 871  * \retval      -ve for failure
 872  */
 873 #ifdef HAVE_FILLDIR_USE_CTX
 874 static int osd_stripe_dir_filldir(struct dir_context *buf,
 875 #else
 876 static int osd_stripe_dir_filldir(void *buf,
 877 #endif
 878                                   const char *name, int namelen,
 879                                   loff_t offset, __u64 ino, unsigned int d_type)
 880 {
 881         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 882         struct osd_thread_info *oti = oclb->oclb_info;
 883         struct lu_fid *fid = &oti->oti_fid3;
 884         struct osd_inode_id *id = &oti->oti_id3;
 885         struct osd_device *dev = oclb->oclb_dev;
 886         struct osd_idmap_cache *oic = oclb->oclb_oic;
 887         struct inode *inode;
 888
 889         oclb->oclb_items++;
 890
 891         if (name[0] == '.')
 892                 return 0;
 893
 894         fid_zero(fid);
 895         sscanf(name + 1, SFID, RFID(fid));
 896         if (!fid_is_sane(fid))
 897                 return 0;
 898
 899         if (osd_remote_fid(oti->oti_env, dev, fid))
 900                 return 0;
 901
 902         osd_id_gen(id, ino, OSD_OII_NOGEN);
 903         inode = osd_iget(oti, dev, id);
 904         if (IS_ERR(inode))
 905                 return PTR_ERR(inode);
 906
 907         iput(inode);
 908         osd_add_oi_cache(oti, dev, id, fid);
 909         oic->oic_fid = *fid;
 910         oic->oic_lid = *id;
 911         oic->oic_dev = dev;
 912         osd_oii_insert(dev, oic, true);
 913         oclb->oclb_found = true;
 914
 915         return 1;
 916 }
 917
 918 /*
 919  * When lookup item under striped directory, we need to locate the master
 920  * MDT-object of the striped directory firstly, then the client will send
 921  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 922  * and the item's name. If the system is restored from MDT file level backup,
 923  * then before the OI scrub completely built the OI files, the OI mappings of
 924  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 925  * not a problem for the master MDT-object. Because when locate the master
 926  * MDT-object, we will do name based lookup (for the striped directory itself)
 927  * firstly, during such process we can setup the correct OI mapping for the
 928  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 929  * the client will not trigger name based lookup on the MDT to locate the slave
 930  * MDT-object before locating item under the striped directory, then when
 931  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 932  * is invalid and does not know what the right OI mapping is, then the MDT has
 933  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 934  * the OI file, related OI mapping is unknown yet, please try again later. And
 935  * then client will re-try the RPC again and again until related OI mapping has
 936  * been updated. That is quite inefficient.
 937  *
 938  * To resolve above trouble, we will handle it as the following two cases:
 939  *
 940  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 941  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 942  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 943  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 944  *    directly. Please check osd_fid_lookup().
 945  *
 946  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 947  *    Under such case, during lookup the master MDT-object, we will lookup the
 948  *    slave MDT-object via readdir against the master MDT-object, because the
 949  *    slave MDT-objects information are stored as sub-directories with the name
 950  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 951  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 952  *    the correct OI mapping for the slave MDT-object.
 953  */
 954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 955                          struct inode *inode, struct osd_idmap_cache *oic)
 956 {
 957         struct lu_buf *buf = &oti->oti_big_buf;
 958         struct dentry *dentry = &oti->oti_obj_dentry;
 959         struct file *filp = &oti->oti_file;
 960         const struct file_operations *fops;
 961         struct lmv_mds_md_v1 *lmv1;
 962         struct osd_check_lmv_buf oclb = {
 963                 .ctx.actor = osd_stripe_dir_filldir,
 964                 .oclb_info = oti,
 965                 .oclb_dev = dev,
 966                 .oclb_oic = oic,
 967                 .oclb_found = false,
 968         };
 969         int rc = 0;
 970
 971         ENTRY;
 972
 973 again:
 974         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 975                              buf->lb_len);
 976         if (rc == -ERANGE) {
 977                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 978                 if (rc > 0) {
 979                         lu_buf_realloc(buf, rc);
 980                         if (buf->lb_buf == NULL)
 981                                 GOTO(out, rc = -ENOMEM);
 982
 983                         goto again;
 984                 }
 985         }
 986
 987         if (unlikely(rc == 0 || rc == -ENODATA))
 988                 GOTO(out, rc = 0);
 989
 990         if (rc < 0)
 991                 GOTO(out, rc);
 992
 993         if (unlikely(buf->lb_buf == NULL)) {
 994                 lu_buf_realloc(buf, rc);
 995                 if (buf->lb_buf == NULL)
 996                         GOTO(out, rc = -ENOMEM);
 997
 998                 goto again;
 999         }
1000
1001         lmv1 = buf->lb_buf;
1002         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1003                 GOTO(out, rc = 0);
1004
1005         fops = inode->i_fop;
1006         dentry->d_inode = inode;
1007         dentry->d_sb = inode->i_sb;
1008         filp->f_pos = 0;
1009         filp->f_path.dentry = dentry;
1010         filp->f_flags |= O_NOATIME;
1011         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012         filp->f_mapping = inode->i_mapping;
1013         filp->f_op = fops;
1014         filp->private_data = NULL;
1015         filp->f_cred = current_cred();
1016         set_file_inode(filp, inode);
1017         rc = osd_security_file_alloc(filp);
1018         if (rc)
1019                 goto out;
1020
1021         do {
1022                 oclb.oclb_items = 0;
1023                 rc = iterate_dir(filp, &oclb.ctx);
1024         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026         fops->release(inode, filp);
1027
1028 out:
1029         if (rc < 0)
1030                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031                        DFID": rc = %d\n", osd_ino2name(inode),
1032                        inode->i_ino, inode->i_generation,
1033                        PFID(&oic->oic_fid), rc);
1034         else
1035                 rc = 0;
1036
1037         RETURN(rc);
1038 }
1039
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041                           const struct lu_fid *fid,
1042                           const struct lu_object_conf *conf)
1043 {
1044         struct osd_thread_info *info;
1045         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046         struct osd_device *dev;
1047         struct osd_idmap_cache *oic;
1048         struct osd_inode_id *id;
1049         struct inode *inode = NULL;
1050         struct lustre_scrub *scrub;
1051         struct scrub_file *sf;
1052         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1053         __u32 saved_ino;
1054         __u32 saved_gen;
1055         int result = 0;
1056         int rc1 = 0;
1057         bool remote = false;
1058         bool trusted = true;
1059         bool updated = false;
1060         bool checked = false;
1061
1062         ENTRY;
1063
1064         LINVRNT(osd_invariant(obj));
1065         LASSERT(obj->oo_inode == NULL);
1066         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1067
1068         dev = osd_dev(ldev);
1069         scrub = &dev->od_scrub.os_scrub;
1070         sf = &scrub->os_file;
1071         info = osd_oti_get(env);
1072         LASSERT(info);
1073         oic = &info->oti_cache;
1074
1075         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1076                 RETURN(-ENOENT);
1077
1078         /*
1079          * For the object is created as locking anchor, or for the object to
1080          * be created on disk. No need to osd_oi_lookup() at here because FID
1081          * shouldn't never be re-used, if it's really a duplicate FID from
1082          * unexpected reason, we should be able to detect it later by calling
1083          * do_create->osd_oi_insert().
1084          */
1085         if (conf && conf->loc_flags & LOC_F_NEW)
1086                 GOTO(out, result = 0);
1087
1088         /* Search order: 1. per-thread cache. */
1089         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1090                 id = &oic->oic_lid;
1091                 goto iget;
1092         }
1093
1094         id = &info->oti_id;
1095         if (!list_empty(&scrub->os_inconsistent_items)) {
1096                 /* Search order: 2. OI scrub pending list. */
1097                 result = osd_oii_lookup(dev, fid, id);
1098                 if (!result)
1099                         goto iget;
1100         }
1101
1102         /*
1103          * The OI mapping in the OI file can be updated by the OI scrub
1104          * when we locate the inode via FID. So it may be not trustable.
1105          */
1106         trusted = false;
1107
1108         /* Search order: 3. OI files. */
1109         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110         if (result == -ENOENT) {
1111                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1114                                       sf->sf_oi_bitmap))
1115                         GOTO(out, result = 0);
1116
1117                 goto trigger;
1118         }
1119
1120         /* -ESTALE is returned if inode of OST object doesn't exist */
1121         if (result == -ESTALE &&
1122             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123                 GOTO(out, result = 0);
1124         }
1125
1126         if (result)
1127                 GOTO(out, result);
1128
1129 iget:
1130         obj->oo_inode = NULL;
1131         /* for later passes through checks, not true on first pass */
1132         if (!IS_ERR_OR_NULL(inode))
1133                 iput(inode);
1134
1135         inode = osd_iget_check(info, dev, fid, id, trusted);
1136         if (!IS_ERR(inode)) {
1137                 obj->oo_inode = inode;
1138                 result = 0;
1139                 if (remote)
1140                         goto trigger;
1141
1142                 goto check_lma;
1143         }
1144
1145         result = PTR_ERR(inode);
1146         if (result == -ENOENT || result == -ESTALE)
1147                 GOTO(out, result = 0);
1148
1149         if (result != -EREMCHG)
1150                 GOTO(out, result);
1151
1152 trigger:
1153         /*
1154          * We still have chance to get the valid inode: for the
1155          * object which is referenced by remote name entry, the
1156          * object on the local MDT will be linked under the dir
1157          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1158          *
1159          * We do not know whether the object for the given FID
1160          * is referenced by some remote name entry or not, and
1161          * especially for DNE II, a multiple-linked object may
1162          * have many name entries reside on many MDTs.
1163          *
1164          * To simplify the operation, OSD will not distinguish
1165          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166          * only happened for the RPC from other MDT during the
1167          * OI scrub, or for the client side RPC with FID only,
1168          * such as FID to path, or from old connected client.
1169          */
1170         if (!remote) {
1171                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1172                 if (!rc1) {
1173                         remote = true;
1174                         trusted = true;
1175                         flags |= SS_AUTO_PARTIAL;
1176                         flags &= ~SS_AUTO_FULL;
1177                         goto iget;
1178                 }
1179         }
1180
1181         if (thread_is_running(&scrub->os_thread)) {
1182                 if (scrub->os_partial_scan && !scrub->os_in_join)
1183                         goto join;
1184
1185                 if (IS_ERR_OR_NULL(inode) || result)
1186                         GOTO(out, result = -EINPROGRESS);
1187
1188                 LASSERT(remote);
1189                 LASSERT(obj->oo_inode == inode);
1190
1191                 osd_add_oi_cache(info, dev, id, fid);
1192                 osd_oii_insert(dev, oic, true);
1193                 goto found;
1194         }
1195
1196         if (dev->od_auto_scrub_interval == AS_NEVER) {
1197                 if (!remote)
1198                         GOTO(out, result = -EREMCHG);
1199
1200                 LASSERT(!result);
1201                 LASSERT(obj->oo_inode == inode);
1202
1203                 osd_add_oi_cache(info, dev, id, fid);
1204                 goto found;
1205         }
1206
1207 join:
1208         rc1 = osd_scrub_start(env, dev, flags);
1209         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211         if (rc1 && rc1 != -EALREADY)
1212                 GOTO(out, result = -EREMCHG);
1213
1214         if (IS_ERR_OR_NULL(inode) || result)
1215                 GOTO(out, result = -EINPROGRESS);
1216
1217         LASSERT(remote);
1218         LASSERT(obj->oo_inode == inode);
1219
1220         osd_add_oi_cache(info, dev, id, fid);
1221         osd_oii_insert(dev, oic, true);
1222         goto found;
1223
1224 check_lma:
1225         checked = true;
1226         if (unlikely(obj->oo_header))
1227                 goto found;
1228
1229         result = osd_check_lma(env, obj);
1230         if (!result)
1231                 goto found;
1232
1233         LASSERTF(id->oii_ino == inode->i_ino &&
1234                  id->oii_gen == inode->i_generation,
1235                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236                  PFID(fid), id->oii_ino, id->oii_gen,
1237                  inode->i_ino, inode->i_generation);
1238
1239         saved_ino = inode->i_ino;
1240         saved_gen = inode->i_generation;
1241
1242         if (unlikely(result == -ENODATA)) {
1243                 /*
1244                  * If the OI scrub updated the OI mapping by race, it
1245                  * must be valid. Trust the inode that has no LMA EA.
1246                  */
1247                 if (updated)
1248                         goto found;
1249
1250                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1251                 if (!result) {
1252                         /*
1253                          * The OI mapping is still there, the inode is still
1254                          * valid. It is just becaues the inode has no LMA EA.
1255                          */
1256                         if (saved_ino == id->oii_ino &&
1257                             saved_gen == id->oii_gen)
1258                                 goto found;
1259
1260                         /*
1261                          * It is the OI scrub updated the OI mapping by race.
1262                          * The new OI mapping must be valid.
1263                          */
1264                         trusted = true;
1265                         updated = true;
1266                         goto iget;
1267                 }
1268
1269                 /*
1270                  * "result == -ENOENT" means that the OI mappinghas been
1271                  * removed by race, so the inode belongs to other object.
1272                  *
1273                  * Others error can be returned  directly.
1274                  */
1275                 if (result == -ENOENT) {
1276                         LASSERT(trusted);
1277
1278                         obj->oo_inode = NULL;
1279                         result = 0;
1280                 }
1281         }
1282
1283         if (result != -EREMCHG)
1284                 GOTO(out, result);
1285
1286         LASSERT(!updated);
1287
1288         /*
1289          * if two OST objects map to the same inode, and inode mode is
1290          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291          * reserved by precreate, and not written yet, in this case, don't
1292          * set inode for the object whose FID mismatch, so that it can create
1293          * inode and not block precreate.
1294          */
1295         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297                 obj->oo_inode = NULL;
1298                 GOTO(out, result = 0);
1299         }
1300
1301         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1302         /*
1303          * "result == -ENOENT" means the cached OI mapping has been removed
1304          * from the OI file by race, above inode belongs to other object.
1305          */
1306         if (result == -ENOENT) {
1307                 LASSERT(trusted);
1308
1309                 obj->oo_inode = NULL;
1310                 GOTO(out, result = 0);
1311         }
1312
1313         if (result)
1314                 GOTO(out, result);
1315
1316         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1317                 result = -EREMCHG;
1318                 goto trigger;
1319         }
1320
1321         /*
1322          * It is the OI scrub updated the OI mapping by race.
1323          * The new OI mapping must be valid.
1324          */
1325         trusted = true;
1326         updated = true;
1327         goto iget;
1328
1329 found:
1330         if (!checked) {
1331                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1333
1334                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1335                 if (!result) {
1336                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1337                             dev->od_is_ost)
1338                                 obj->oo_pfid_in_lma = 1;
1339                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1340                             !dev->od_is_ost)
1341                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342                 } else if (result != -ENODATA) {
1343                         GOTO(out, result);
1344                 }
1345         }
1346
1347         obj->oo_compat_dot_created = 1;
1348         obj->oo_compat_dotdot_created = 1;
1349
1350         if (S_ISDIR(inode->i_mode) &&
1351             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352                 osd_check_lmv(info, dev, inode, oic);
1353
1354         result = osd_attach_jinode(inode);
1355         if (result)
1356                 GOTO(out, result);
1357
1358         if (!ldiskfs_pdo)
1359                 GOTO(out, result = 0);
1360
1361         LASSERT(!obj->oo_hl_head);
1362         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1363
1364         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1365
1366 out:
1367         if (result || !obj->oo_inode) {
1368                 if (!IS_ERR_OR_NULL(inode))
1369                         iput(inode);
1370
1371                 obj->oo_inode = NULL;
1372                 if (trusted)
1373                         fid_zero(&oic->oic_fid);
1374         }
1375
1376         LINVRNT(osd_invariant(obj));
1377         return result;
1378 }
1379
1380 /*
1381  * Concurrency: shouldn't matter.
1382  */
1383 static void osd_object_init0(struct osd_object *obj)
1384 {
1385         LASSERT(obj->oo_inode != NULL);
1386         obj->oo_dt.do_body_ops = &osd_body_ops;
1387         obj->oo_dt.do_lu.lo_header->loh_attr |=
1388                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1389 }
1390
1391 /*
1392  * Concurrency: no concurrent access is possible that early in object
1393  * life-cycle.
1394  */
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396                            const struct lu_object_conf *conf)
1397 {
1398         struct osd_object *obj = osd_obj(l);
1399         int result;
1400
1401         LINVRNT(osd_invariant(obj));
1402
1403         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1404                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1405                 l->lo_header->loh_attr |= LOHA_EXISTS;
1406                 return 0;
1407         }
1408
1409         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1410         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1411         if (result == 0 && obj->oo_inode != NULL) {
1412                 struct osd_thread_info *oti = osd_oti_get(env);
1413                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1414
1415                 osd_object_init0(obj);
1416                 if (unlikely(obj->oo_header))
1417                         return 0;
1418
1419                 result = osd_get_lma(oti, obj->oo_inode,
1420                                      &oti->oti_obj_dentry, loa);
1421                 if (!result) {
1422                         /*
1423                          * Convert LMAI flags to lustre LMA flags
1424                          * and cache it to oo_lma_flags
1425                          */
1426                         obj->oo_lma_flags =
1427                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1428                 } else if (result == -ENODATA) {
1429                         result = 0;
1430                 }
1431         }
1432
1433         LINVRNT(osd_invariant(obj));
1434         return result;
1435 }
1436
1437 /*
1438  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1439  * The left part is for value, binary mode.
1440  */
1441 struct osd_xattr_entry {
1442         struct list_head        oxe_list;
1443         size_t                  oxe_len;
1444         size_t                  oxe_namelen;
1445         bool                    oxe_exist;
1446         struct rcu_head         oxe_rcu;
1447         char                    oxe_buf[0];
1448 };
1449
1450 static int osd_oxc_get(struct osd_object *obj, const char *name,
1451                        struct lu_buf *buf)
1452 {
1453         struct osd_xattr_entry *tmp;
1454         struct osd_xattr_entry *oxe = NULL;
1455         size_t namelen = strlen(name);
1456         int rc;
1457
1458         ENTRY;
1459
1460         rcu_read_lock();
1461         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1462                 if (namelen == tmp->oxe_namelen &&
1463                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1464                         oxe = tmp;
1465                         break;
1466                 }
1467         }
1468
1469         if (oxe == NULL)
1470                 GOTO(out, rc = -ENOENT);
1471
1472         if (!oxe->oxe_exist)
1473                 GOTO(out, rc = -ENODATA);
1474
1475         /* vallen */
1476         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1477         LASSERT(rc > 0);
1478
1479         if (buf->lb_buf == NULL)
1480                 GOTO(out, rc);
1481
1482         if (buf->lb_len < rc)
1483                 GOTO(out, rc = -ERANGE);
1484
1485         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1486         EXIT;
1487 out:
1488         rcu_read_unlock();
1489
1490         return rc;
1491 }
1492
1493 static void osd_oxc_free(struct rcu_head *head)
1494 {
1495         struct osd_xattr_entry *oxe;
1496
1497         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1498         OBD_FREE(oxe, oxe->oxe_len);
1499 }
1500
1501 static void osd_oxc_add(struct osd_object *obj, const char *name,
1502                         const char *buf, int buflen)
1503 {
1504         struct osd_xattr_entry *oxe;
1505         struct osd_xattr_entry *old = NULL;
1506         struct osd_xattr_entry *tmp;
1507         size_t namelen = strlen(name);
1508         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1509
1510         OBD_ALLOC(oxe, len);
1511         if (oxe == NULL)
1512                 return;
1513
1514         INIT_LIST_HEAD(&oxe->oxe_list);
1515         oxe->oxe_len = len;
1516         oxe->oxe_namelen = namelen;
1517         memcpy(oxe->oxe_buf, name, namelen);
1518         if (buflen > 0) {
1519                 LASSERT(buf != NULL);
1520                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1521                 oxe->oxe_exist = true;
1522         } else {
1523                 oxe->oxe_exist = false;
1524         }
1525
1526         /* this should be rarely called, just remove old and add new */
1527         spin_lock(&obj->oo_guard);
1528         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1529                 if (namelen == tmp->oxe_namelen &&
1530                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1531                         old = tmp;
1532                         break;
1533                 }
1534         }
1535         if (old != NULL) {
1536                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1537                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1538         } else {
1539                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1540         }
1541         spin_unlock(&obj->oo_guard);
1542 }
1543
1544 static void osd_oxc_del(struct osd_object *obj, const char *name)
1545 {
1546         struct osd_xattr_entry *oxe;
1547         size_t namelen = strlen(name);
1548
1549         spin_lock(&obj->oo_guard);
1550         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1551                 if (namelen == oxe->oxe_namelen &&
1552                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1553                         list_del_rcu(&oxe->oxe_list);
1554                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1555                         break;
1556                 }
1557         }
1558         spin_unlock(&obj->oo_guard);
1559 }
1560
1561 static void osd_oxc_fini(struct osd_object *obj)
1562 {
1563         struct osd_xattr_entry *oxe, *next;
1564
1565         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1566                 list_del(&oxe->oxe_list);
1567                 OBD_FREE(oxe, oxe->oxe_len);
1568         }
1569 }
1570
1571 /*
1572  * Concurrency: no concurrent access is possible that late in object
1573  * life-cycle.
1574  */
1575 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1576 {
1577         struct osd_object *obj = osd_obj(l);
1578         struct lu_object_header *h = obj->oo_header;
1579
1580         LINVRNT(osd_invariant(obj));
1581
1582         osd_oxc_fini(obj);
1583         dt_object_fini(&obj->oo_dt);
1584         if (obj->oo_hl_head != NULL)
1585                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1586         OBD_FREE_PTR(obj);
1587         if (unlikely(h)) {
1588                 lu_object_header_fini(h);
1589                 OBD_FREE_PTR(h);
1590         }
1591 }
1592
1593 /*
1594  * Concurrency: no concurrent access is possible that late in object
1595  * life-cycle.
1596  */
1597 static void osd_index_fini(struct osd_object *o)
1598 {
1599         struct iam_container *bag;
1600
1601         if (o->oo_dir != NULL) {
1602                 bag = &o->oo_dir->od_container;
1603                 if (o->oo_inode != NULL) {
1604                         if (bag->ic_object == o->oo_inode)
1605                                 iam_container_fini(bag);
1606                 }
1607                 OBD_FREE_PTR(o->oo_dir);
1608                 o->oo_dir = NULL;
1609         }
1610 }
1611
1612 /*
1613  * Concurrency: no concurrent access is possible that late in object
1614  * life-cycle (for all existing callers, that is. New callers have to provide
1615  * their own locking.)
1616  */
1617 static int osd_inode_unlinked(const struct inode *inode)
1618 {
1619         return inode->i_nlink == 0;
1620 }
1621
1622 enum {
1623         OSD_TXN_OI_DELETE_CREDITS    = 20,
1624         OSD_TXN_INODE_DELETE_CREDITS = 20
1625 };
1626
1627 /*
1628  * Journal
1629  */
1630
1631 #if OSD_THANDLE_STATS
1632 /**
1633  * Set time when the handle is allocated
1634  */
1635 static void osd_th_alloced(struct osd_thandle *oth)
1636 {
1637         oth->oth_alloced = ktime_get();
1638 }
1639
1640 /**
1641  * Set time when the handle started
1642  */
1643 static void osd_th_started(struct osd_thandle *oth)
1644 {
1645         oth->oth_started = ktime_get();
1646 }
1647
1648 /**
1649  * Check whether the we deal with this handle for too long.
1650  */
1651 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1652                                 ktime_t alloced, ktime_t started,
1653                                 ktime_t closed)
1654 {
1655         ktime_t now = ktime_get();
1656
1657         LASSERT(dev != NULL);
1658
1659         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1660                             ktime_us_delta(started, alloced));
1661         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1662                             ktime_us_delta(closed, started));
1663         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1664                             ktime_us_delta(now, closed));
1665
1666         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1667                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1668                                 oth, now, alloced, started, closed);
1669                 libcfs_debug_dumpstack(NULL);
1670         }
1671 }
1672
1673 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1674 {                                                                       \
1675         ktime_t __closed = ktime_get();                                 \
1676         ktime_t __alloced = oth->oth_alloced;                           \
1677         ktime_t __started = oth->oth_started;                           \
1678                                                                         \
1679         expr;                                                           \
1680         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1681 }
1682
1683 #else /* OSD_THANDLE_STATS */
1684
1685 #define osd_th_alloced(h)                  do {} while(0)
1686 #define osd_th_started(h)                  do {} while(0)
1687 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1688
1689 #endif /* OSD_THANDLE_STATS */
1690
1691 /*
1692  * Concurrency: doesn't access mutable data.
1693  */
1694 static int osd_param_is_not_sane(const struct osd_device *dev,
1695                                  const struct thandle *th)
1696 {
1697         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1698
1699         return oh->ot_credits > osd_transaction_size(dev);
1700 }
1701
1702 /*
1703  * Concurrency: shouldn't matter.
1704  */
1705 static void osd_trans_commit_cb(struct super_block *sb,
1706                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1707 {
1708         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1709         struct thandle *th = &oh->ot_super;
1710         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1711         struct dt_txn_commit_cb *dcb, *tmp;
1712
1713         LASSERT(oh->ot_handle == NULL);
1714
1715         if (error)
1716                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1717
1718         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1719         /* call per-transaction callbacks if any */
1720         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1721                                  dcb_linkage) {
1722                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1723                          "commit callback entry: magic=%x name='%s'\n",
1724                          dcb->dcb_magic, dcb->dcb_name);
1725                 list_del_init(&dcb->dcb_linkage);
1726                 dcb->dcb_func(NULL, th, dcb, error);
1727         }
1728
1729         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1730         lu_device_put(lud);
1731         th->th_dev = NULL;
1732
1733         OBD_FREE_PTR(oh);
1734 }
1735
1736 #ifndef HAVE_SB_START_WRITE
1737 # define sb_start_write(sb) do {} while (0)
1738 # define sb_end_write(sb) do {} while (0)
1739 #endif
1740
1741 static struct thandle *osd_trans_create(const struct lu_env *env,
1742                                         struct dt_device *d)
1743 {
1744         struct osd_thread_info *oti = osd_oti_get(env);
1745         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1746         struct osd_thandle *oh;
1747         struct thandle *th;
1748
1749         ENTRY;
1750
1751         if (d->dd_rdonly) {
1752                 CERROR("%s: someone try to start transaction under "
1753                        "readonly mode, should be disabled.\n",
1754                        osd_name(osd_dt_dev(d)));
1755                 dump_stack();
1756                 RETURN(ERR_PTR(-EROFS));
1757         }
1758
1759         /* on pending IO in this thread should left from prev. request */
1760         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1761
1762         sb_start_write(osd_sb(osd_dt_dev(d)));
1763
1764         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1765         if (!oh) {
1766                 sb_end_write(osd_sb(osd_dt_dev(d)));
1767                 RETURN(ERR_PTR(-ENOMEM));
1768         }
1769
1770         oh->ot_quota_trans = &oti->oti_quota_trans;
1771         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1772         th = &oh->ot_super;
1773         th->th_dev = d;
1774         th->th_result = 0;
1775         oh->ot_credits = 0;
1776         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1777         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1778         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1779         osd_th_alloced(oh);
1780
1781         memset(oti->oti_declare_ops, 0,
1782                sizeof(oti->oti_declare_ops));
1783         memset(oti->oti_declare_ops_cred, 0,
1784                sizeof(oti->oti_declare_ops_cred));
1785         memset(oti->oti_declare_ops_used, 0,
1786                sizeof(oti->oti_declare_ops_used));
1787
1788         oti->oti_ins_cache_depth++;
1789
1790         RETURN(th);
1791 }
1792
1793 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1794 {
1795         struct osd_thread_info *oti = osd_oti_get(env);
1796         struct osd_thandle *oh;
1797
1798         oh = container_of0(th, struct osd_thandle, ot_super);
1799         LASSERT(oh != NULL);
1800
1801         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1802               oti->oti_declare_ops[OSD_OT_CREATE],
1803               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1804               oti->oti_declare_ops_used[OSD_OT_CREATE],
1805               oti->oti_declare_ops[OSD_OT_DESTROY],
1806               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1807               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1808         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1809               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1810               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1811               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1812               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1813               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1814               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1815         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1816               oti->oti_declare_ops[OSD_OT_WRITE],
1817               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1818               oti->oti_declare_ops_used[OSD_OT_WRITE],
1819               oti->oti_declare_ops[OSD_OT_PUNCH],
1820               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1821               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1822               oti->oti_declare_ops[OSD_OT_QUOTA],
1823               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1824               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1825         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1826               oti->oti_declare_ops[OSD_OT_INSERT],
1827               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1828               oti->oti_declare_ops_used[OSD_OT_INSERT],
1829               oti->oti_declare_ops[OSD_OT_DELETE],
1830               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1831               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1832         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1833               oti->oti_declare_ops[OSD_OT_REF_ADD],
1834               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1835               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1836               oti->oti_declare_ops[OSD_OT_REF_DEL],
1837               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1838               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1839 }
1840
1841 /*
1842  * Concurrency: shouldn't matter.
1843  */
1844 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1845                            struct thandle *th)
1846 {
1847         struct osd_thread_info *oti = osd_oti_get(env);
1848         struct osd_device *dev = osd_dt_dev(d);
1849         handle_t *jh;
1850         struct osd_thandle *oh;
1851         int rc;
1852
1853         ENTRY;
1854
1855         LASSERT(current->journal_info == NULL);
1856
1857         oh = container_of0(th, struct osd_thandle, ot_super);
1858         LASSERT(oh != NULL);
1859         LASSERT(oh->ot_handle == NULL);
1860
1861         rc = dt_txn_hook_start(env, d, th);
1862         if (rc != 0)
1863                 GOTO(out, rc);
1864
1865         if (unlikely(osd_param_is_not_sane(dev, th))) {
1866                 static unsigned long last_printed;
1867                 static int last_credits;
1868
1869                 /*
1870                  * don't make noise on a tiny testing systems
1871                  * actual credits misuse will be caught anyway
1872                  */
1873                 if (last_credits != oh->ot_credits &&
1874                     time_after(jiffies, last_printed +
1875                                cfs_time_seconds(60)) &&
1876                     osd_transaction_size(dev) > 512) {
1877                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1878                               oh->ot_credits, osd_transaction_size(dev));
1879                         osd_trans_dump_creds(env, th);
1880                         libcfs_debug_dumpstack(NULL);
1881                         last_credits = oh->ot_credits;
1882                         last_printed = jiffies;
1883                 }
1884                 /*
1885                  * XXX Limit the credits to 'max_transaction_buffers', and
1886                  *     let the underlying filesystem to catch the error if
1887                  *     we really need so many credits.
1888                  *
1889                  *     This should be removed when we can calculate the
1890                  *     credits precisely.
1891                  */
1892                 oh->ot_credits = osd_transaction_size(dev);
1893         } else if (ldiskfs_track_declares_assert != 0) {
1894                 /*
1895                  * reserve few credits to prevent an assertion in JBD
1896                  * our debugging mechanism will be able to detected
1897                  * overuse. this can help to debug single-update
1898                  * transactions
1899                  */
1900                 oh->ot_credits += 10;
1901                 if (unlikely(osd_param_is_not_sane(dev, th)))
1902                         oh->ot_credits = osd_transaction_size(dev);
1903         }
1904
1905         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1906                 GOTO(out, rc = -EIO);
1907
1908         /*
1909          * XXX temporary stuff. Some abstraction layer should
1910          * be used.
1911          */
1912         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1913         osd_th_started(oh);
1914         if (!IS_ERR(jh)) {
1915                 oh->ot_handle = jh;
1916                 LASSERT(oti->oti_txns == 0);
1917
1918                 lu_device_get(&d->dd_lu_dev);
1919                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1920                               "osd-tx", th);
1921                 oti->oti_txns++;
1922                 rc = 0;
1923         } else {
1924                 rc = PTR_ERR(jh);
1925         }
1926 out:
1927         RETURN(rc);
1928 }
1929
1930 static int osd_seq_exists(const struct lu_env *env,
1931                           struct osd_device *osd, u64 seq)
1932 {
1933         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1934         struct seq_server_site *ss = osd_seq_site(osd);
1935         int rc;
1936
1937         ENTRY;
1938
1939         LASSERT(ss != NULL);
1940         LASSERT(ss->ss_server_fld != NULL);
1941
1942         rc = osd_fld_lookup(env, osd, seq, range);
1943         if (rc != 0) {
1944                 if (rc != -ENOENT)
1945                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1946                                osd_name(osd), seq, rc);
1947                 RETURN(0);
1948         }
1949
1950         RETURN(ss->ss_node_id == range->lsr_index);
1951 }
1952
1953 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1954 {
1955         struct dt_txn_commit_cb *dcb;
1956         struct dt_txn_commit_cb *tmp;
1957
1958         /* call per-transaction stop callbacks if any */
1959         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1960                                  dcb_linkage) {
1961                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1962                          "commit callback entry: magic=%x name='%s'\n",
1963                          dcb->dcb_magic, dcb->dcb_name);
1964                 list_del_init(&dcb->dcb_linkage);
1965                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1966         }
1967 }
1968
1969 /*
1970  * Concurrency: shouldn't matter.
1971  */
1972 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1973                           struct thandle *th)
1974 {
1975         struct osd_thread_info *oti = osd_oti_get(env);
1976         struct osd_thandle *oh;
1977         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1978         struct osd_device *osd = osd_dt_dev(th->th_dev);
1979         struct qsd_instance *qsd = osd_def_qsd(osd);
1980         struct lquota_trans *qtrans;
1981         LIST_HEAD(truncates);
1982         int rc = 0, remove_agents = 0;
1983
1984         ENTRY;
1985
1986         oh = container_of0(th, struct osd_thandle, ot_super);
1987
1988         remove_agents = oh->ot_remove_agents;
1989
1990         qtrans = oh->ot_quota_trans;
1991         oh->ot_quota_trans = NULL;
1992
1993         /* move locks to local list, stop tx, execute truncates */
1994         list_splice(&oh->ot_trunc_locks, &truncates);
1995
1996         if (oh->ot_handle != NULL) {
1997                 int rc2;
1998
1999                 handle_t *hdl = oh->ot_handle;
2000
2001                 /*
2002                  * add commit callback
2003                  * notice we don't do this in osd_trans_start()
2004                  * as underlying transaction can change during truncate
2005                  */
2006                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2007                                              &oh->ot_jcb);
2008
2009                 LASSERT(oti->oti_txns == 1);
2010                 oti->oti_txns--;
2011
2012                 rc = dt_txn_hook_stop(env, th);
2013                 if (rc != 0)
2014                         CERROR("%s: failed in transaction hook: rc = %d\n",
2015                                osd_name(osd), rc);
2016
2017                 osd_trans_stop_cb(oh, rc);
2018                 /* hook functions might modify th_sync */
2019                 hdl->h_sync = th->th_sync;
2020
2021                 oh->ot_handle = NULL;
2022                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2023                 if (rc2 != 0)
2024                         CERROR("%s: failed to stop transaction: rc = %d\n",
2025                                osd_name(osd), rc2);
2026                 if (!rc)
2027                         rc = rc2;
2028
2029                 osd_process_truncates(&truncates);
2030         } else {
2031                 osd_trans_stop_cb(oh, th->th_result);
2032                 OBD_FREE_PTR(oh);
2033         }
2034
2035         osd_trunc_unlock_all(&truncates);
2036
2037         /* inform the quota slave device that the transaction is stopping */
2038         qsd_op_end(env, qsd, qtrans);
2039
2040         /*
2041          * as we want IO to journal and data IO be concurrent, we don't block
2042          * awaiting data IO completion in osd_do_bio(), instead we wait here
2043          * once transaction is submitted to the journal. all reqular requests
2044          * don't do direct IO (except read/write), thus this wait_event becomes
2045          * no-op for them.
2046          *
2047          * IMPORTANT: we have to wait till any IO submited by the thread is
2048          * completed otherwise iobuf may be corrupted by different request
2049          */
2050         wait_event(iobuf->dr_wait,
2051                        atomic_read(&iobuf->dr_numreqs) == 0);
2052         osd_fini_iobuf(osd, iobuf);
2053         if (!rc)
2054                 rc = iobuf->dr_error;
2055
2056         if (unlikely(remove_agents != 0))
2057                 osd_process_scheduled_agent_removals(env, osd);
2058
2059         oti->oti_ins_cache_depth--;
2060         /* reset OI cache for safety */
2061         if (oti->oti_ins_cache_depth == 0)
2062                 oti->oti_ins_cache_used = 0;
2063
2064         sb_end_write(osd_sb(osd));
2065
2066         RETURN(rc);
2067 }
2068
2069 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2070 {
2071         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2072                                                ot_super);
2073
2074         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2075         LASSERT(&dcb->dcb_func != NULL);
2076         if (dcb->dcb_flags & DCB_TRANS_STOP)
2077                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2078         else
2079                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2080
2081         return 0;
2082 }
2083
2084 /*
2085  * Called just before object is freed. Releases all resources except for
2086  * object itself (that is released by osd_object_free()).
2087  *
2088  * Concurrency: no concurrent access is possible that late in object
2089  * life-cycle.
2090  */
2091 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2092 {
2093         struct osd_object *obj = osd_obj(l);
2094         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2095         struct inode *inode = obj->oo_inode;
2096         __u64 projid;
2097         qid_t uid;
2098         qid_t gid;
2099
2100         LINVRNT(osd_invariant(obj));
2101
2102         /*
2103          * If object is unlinked remove fid->ino mapping from object index.
2104          */
2105
2106         osd_index_fini(obj);
2107
2108         if (!inode)
2109                 return;
2110
2111         uid = i_uid_read(inode);
2112         gid = i_gid_read(inode);
2113         projid = i_projid_read(inode);
2114
2115         obj->oo_inode = NULL;
2116         iput(inode);
2117
2118         /* do not rebalance quota if the caller needs to release memory
2119          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2120          * transaction and risk to deadlock - LU-12178 */
2121         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2122                 return;
2123
2124         if (!obj->oo_header && qsd) {
2125                 struct osd_thread_info *info = osd_oti_get(env);
2126                 struct lquota_id_info *qi = &info->oti_qi;
2127
2128                 /* Release granted quota to master if necessary */
2129                 qi->lqi_id.qid_uid = uid;
2130                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2131
2132                 qi->lqi_id.qid_uid = gid;
2133                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2134
2135                 qi->lqi_id.qid_uid = projid;
2136                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2137         }
2138 }
2139
2140 /*
2141  * Concurrency: ->loo_object_release() is called under site spin-lock.
2142  */
2143 static void osd_object_release(const struct lu_env *env,
2144                                struct lu_object *l)
2145 {
2146         struct osd_object *o = osd_obj(l);
2147
2148         /*
2149          * nobody should be releasing a non-destroyed object with nlink=0
2150          * the API allows this, but ldiskfs doesn't like and then report
2151          * this inode as deleted
2152          */
2153         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2154                   o->oo_inode->i_nlink == 0));
2155 }
2156
2157 /*
2158  * Concurrency: shouldn't matter.
2159  */
2160 static int osd_object_print(const struct lu_env *env, void *cookie,
2161                             lu_printer_t p, const struct lu_object *l)
2162 {
2163         struct osd_object *o = osd_obj(l);
2164         struct iam_descr *d;
2165
2166         if (o->oo_dir != NULL)
2167                 d = o->oo_dir->od_container.ic_descr;
2168         else
2169                 d = NULL;
2170         return (*p)(env, cookie,
2171                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2172                     o, o->oo_inode,
2173                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2174                     o->oo_inode ? o->oo_inode->i_generation : 0,
2175                     d ? d->id_ops->id_name : "plain");
2176 }
2177
2178 /*
2179  * Concurrency: shouldn't matter.
2180  */
2181 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2182                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2183 {
2184         struct osd_device *osd = osd_dt_dev(d);
2185         struct super_block *sb = osd_sb(osd);
2186         struct kstatfs *ksfs;
2187         __u64 reserved;
2188         int result = 0;
2189
2190         if (unlikely(osd->od_mnt == NULL))
2191                 return -EINPROGRESS;
2192
2193         /* osd_lproc.c call this without env, allocate ksfs for that case */
2194         if (unlikely(env == NULL)) {
2195                 OBD_ALLOC_PTR(ksfs);
2196                 if (ksfs == NULL)
2197                         return -ENOMEM;
2198         } else {
2199                 ksfs = &osd_oti_get(env)->oti_ksfs;
2200         }
2201
2202         result = sb->s_op->statfs(sb->s_root, ksfs);
2203         if (result)
2204                 goto out;
2205
2206         statfs_pack(sfs, ksfs);
2207         if (unlikely(sb->s_flags & SB_RDONLY))
2208                 sfs->os_state |= OS_STATE_READONLY;
2209
2210         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2211
2212         if (ldiskfs_has_feature_extents(sb))
2213                 sfs->os_maxbytes = sb->s_maxbytes;
2214         else
2215                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2216
2217         /*
2218          * Reserve some space so to avoid fragmenting the filesystem too much.
2219          * Fragmentation not only impacts performance, but can also increase
2220          * metadata overhead significantly, causing grant calculation to be
2221          * wrong.
2222          *
2223          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2224          */
2225         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2226         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2227         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2228                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2229
2230         sfs->os_blocks -= reserved;
2231         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2232         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2233
2234 out:
2235         if (unlikely(env == NULL))
2236                 OBD_FREE_PTR(ksfs);
2237         return result;
2238 }
2239
2240 /**
2241  * Estimate space needed for file creations. We assume the largest filename
2242  * which is 2^64 - 1, hence a filename of 20 chars.
2243  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2244  */
2245 #ifdef __LDISKFS_DIR_REC_LEN
2246 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2247 #else
2248 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2249 #endif
2250
2251 /*
2252  * Concurrency: doesn't access mutable data.
2253  */
2254 static void osd_conf_get(const struct lu_env *env,
2255                          const struct dt_device *dev,
2256                          struct dt_device_param *param)
2257 {
2258         struct osd_device *d = osd_dt_dev(dev);
2259         struct super_block *sb = osd_sb(d);
2260         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2261         const char *name;
2262         int ea_overhead;
2263
2264         /*
2265          * XXX should be taken from not-yet-existing fs abstraction layer.
2266          */
2267         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2268         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2269         param->ddp_symlink_max  = sb->s_blocksize;
2270         param->ddp_mount_type   = LDD_MT_LDISKFS;
2271         if (ldiskfs_has_feature_extents(sb))
2272                 param->ddp_maxbytes = sb->s_maxbytes;
2273         else
2274                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2275         /*
2276          * inode are statically allocated, so per-inode space consumption
2277          * is the space consumed by the directory entry
2278          */
2279         param->ddp_inodespace     = PER_OBJ_USAGE;
2280         /*
2281          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2282          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2283          * maximum length to not under-count the actual number of extents
2284          * needed for writing a file if there are sub-optimal block allocations.
2285          */
2286         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2287         /* worst-case extent insertion metadata overhead */
2288         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2289         param->ddp_mntopts = 0;
2290         if (test_opt(sb, XATTR_USER))
2291                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2292         if (test_opt(sb, POSIX_ACL))
2293                 param->ddp_mntopts |= MNTOPT_ACL;
2294
2295         /*
2296          * LOD might calculate the max stripe count based on max_ea_size,
2297          * so we need take account in the overhead as well,
2298          * xattr_header + magic + xattr_entry_head
2299          */
2300         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2301                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2302
2303 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2304         if (ldiskfs_has_feature_ea_inode(sb))
2305                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2306                                                                 ea_overhead;
2307         else
2308 #endif
2309                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2310
2311         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2312                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2313
2314         /*
2315          * Preferred RPC size for efficient disk IO.  4MB shows good
2316          * all-around performance for ldiskfs, but use bigalloc chunk size
2317          * by default if larger.
2318          */
2319 #if defined(LDISKFS_CLUSTER_SIZE)
2320         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2321                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2322         else
2323 #endif
2324                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2325
2326         param->ddp_t10_cksum_type = 0;
2327         if (bi) {
2328                 unsigned short interval = blk_integrity_interval(bi);
2329                 name = blk_integrity_name(bi);
2330                 /*
2331                  * Expected values:
2332                  * T10-DIF-TYPE1-CRC
2333                  * T10-DIF-TYPE3-CRC
2334                  * T10-DIF-TYPE1-IP
2335                  * T10-DIF-TYPE3-IP
2336                  */
2337                 if (strncmp(name, "T10-DIF-TYPE",
2338                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2339                         /* also skip "1/3-" at end */
2340                         const int type_off = sizeof("T10-DIF-TYPE.");
2341                         char type_number = name[type_off - 2];
2342
2343                         if (interval != 512 && interval != 4096) {
2344                                 CERROR("%s: unsupported T10PI sector size %u\n",
2345                                        d->od_svname, interval);
2346                         } else if (type_number != '1' && type_number != '3') {
2347                                 CERROR("%s: unsupported T10PI type %s\n",
2348                                        d->od_svname, name);
2349                         } else if (strcmp(name + type_off, "CRC") == 0) {
2350                                 d->od_t10_type = type_number == '1' ?
2351                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2352                                 param->ddp_t10_cksum_type = interval == 512 ?
2353                                         OBD_CKSUM_T10CRC512 :
2354                                         OBD_CKSUM_T10CRC4K;
2355                         } else if (strcmp(name + type_off, "IP") == 0) {
2356                                 d->od_t10_type = type_number == '1' ?
2357                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2358                                 param->ddp_t10_cksum_type = interval == 512 ?
2359                                         OBD_CKSUM_T10IP512 :
2360                                         OBD_CKSUM_T10IP4K;
2361                         } else {
2362                                 CERROR("%s: unsupported checksum type of "
2363                                        "T10PI type '%s'",
2364                                        d->od_svname, name);
2365                         }
2366
2367                 } else {
2368                         CERROR("%s: unsupported T10PI type '%s'",
2369                                d->od_svname, name);
2370                 }
2371         }
2372 }
2373
2374 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2375 {
2376         return osd_sb(osd_dt_dev(d));
2377 }
2378
2379 /*
2380  * Concurrency: shouldn't matter.
2381  */
2382 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2383 {
2384         int rc;
2385         struct super_block *s = osd_sb(osd_dt_dev(d));
2386         ENTRY;
2387
2388         down_read(&s->s_umount);
2389         rc = s->s_op->sync_fs(s, 1);
2390         up_read(&s->s_umount);
2391
2392         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2393                rc);
2394
2395         return rc;
2396 }
2397
2398 /**
2399  * Start commit for OSD device.
2400  *
2401  * An implementation of dt_commit_async method for OSD device.
2402  * Asychronously starts underlayng fs sync and thereby a transaction
2403  * commit.
2404  *
2405  * \param env environment
2406  * \param d dt device
2407  *
2408  * \see dt_device_operations
2409  */
2410 static int osd_commit_async(const struct lu_env *env,
2411                             struct dt_device *d)
2412 {
2413         struct super_block *s = osd_sb(osd_dt_dev(d));
2414         int rc;
2415
2416         ENTRY;
2417
2418         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2419         down_read(&s->s_umount);
2420         rc = s->s_op->sync_fs(s, 0);
2421         up_read(&s->s_umount);
2422
2423         RETURN(rc);
2424 }
2425
2426 /* Our own copy of the set readonly functions if present, or NU if not. */
2427 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2428 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2429 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2430 static int (*priv_security_file_alloc)(struct file *file);
2431
2432 int osd_security_file_alloc(struct file *file)
2433 {
2434         if (priv_security_file_alloc)
2435                 return priv_security_file_alloc(file);
2436         return 0;
2437 }
2438
2439 /*
2440  * Concurrency: shouldn't matter.
2441  */
2442 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2443 {
2444         struct super_block *sb = osd_sb(osd_dt_dev(d));
2445         struct block_device *dev = sb->s_bdev;
2446         int rc = -EOPNOTSUPP;
2447
2448         ENTRY;
2449
2450         if (priv_dev_set_rdonly) {
2451                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2452
2453                 rc = 0;
2454                 CERROR("*** setting %s read-only ***\n",
2455                        osd_dt_dev(d)->od_svname);
2456
2457                 if (sb->s_op->freeze_fs) {
2458                         rc = sb->s_op->freeze_fs(sb);
2459                         if (rc)
2460                                 goto out;
2461                 }
2462
2463                 if (jdev && (jdev != dev)) {
2464                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2465                                (long)jdev);
2466                         priv_dev_set_rdonly(jdev);
2467                 }
2468                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2469                 priv_dev_set_rdonly(dev);
2470
2471                 if (sb->s_op->unfreeze_fs)
2472                         sb->s_op->unfreeze_fs(sb);
2473         }
2474
2475 out:
2476         if (rc)
2477                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2478                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2479
2480         RETURN(rc);
2481 }
2482
2483 /**
2484  * Note: we do not count into QUOTA here.
2485  * If we mount with --data_journal we may need more.
2486  */
2487 const int osd_dto_credits_noquota[DTO_NR] = {
2488         /**
2489          * Insert.
2490          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2491          * SINGLEDATA_TRANS_BLOCKS(8)
2492          * XXX Note: maybe iam need more, since iam have more level than
2493          *           EXT3 htree.
2494          */
2495         [DTO_INDEX_INSERT]  = 16,
2496         /**
2497          * Delete
2498          * just modify a single entry, probably merge few within a block
2499          */
2500         [DTO_INDEX_DELETE]  = 1,
2501         /**
2502          * Used for OI scrub
2503          */
2504         [DTO_INDEX_UPDATE]  = 16,
2505         /**
2506          * 4(inode, inode bits, groups, GDT)
2507          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2508          */
2509         [DTO_OBJECT_CREATE] = 4,
2510         /**
2511          * 4(inode, inode bits, groups, GDT)
2512          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2513          */
2514         [DTO_OBJECT_DELETE] = 4,
2515         /**
2516          * Attr set credits (inode)
2517          */
2518         [DTO_ATTR_SET_BASE] = 1,
2519         /**
2520          * Xattr set. The same as xattr of EXT3.
2521          * DATA_TRANS_BLOCKS(14)
2522          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2523          * are also counted in. Do not know why?
2524          */
2525         [DTO_XATTR_SET]     = 14,
2526         /**
2527          * credits for inode change during write.
2528          */
2529         [DTO_WRITE_BASE]    = 3,
2530         /**
2531          * credits for single block write.
2532          */
2533         [DTO_WRITE_BLOCK]   = 14,
2534         /**
2535          * Attr set credits for chown.
2536          * This is extra credits for setattr, and it is null without quota
2537          */
2538         [DTO_ATTR_SET_CHOWN] = 0
2539 };
2540
2541 static const struct dt_device_operations osd_dt_ops = {
2542         .dt_root_get       = osd_root_get,
2543         .dt_statfs         = osd_statfs,
2544         .dt_trans_create   = osd_trans_create,
2545         .dt_trans_start    = osd_trans_start,
2546         .dt_trans_stop     = osd_trans_stop,
2547         .dt_trans_cb_add   = osd_trans_cb_add,
2548         .dt_conf_get       = osd_conf_get,
2549         .dt_mnt_sb_get     = osd_mnt_sb_get,
2550         .dt_sync           = osd_sync,
2551         .dt_ro             = osd_ro,
2552         .dt_commit_async   = osd_commit_async,
2553 };
2554
2555 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2556                           unsigned int role)
2557 {
2558         struct osd_object *obj = osd_dt_obj(dt);
2559         struct osd_thread_info *oti = osd_oti_get(env);
2560
2561         LINVRNT(osd_invariant(obj));
2562
2563         LASSERT(obj->oo_owner != env);
2564         down_read_nested(&obj->oo_sem, role);
2565
2566         LASSERT(obj->oo_owner == NULL);
2567         oti->oti_r_locks++;
2568 }
2569
2570 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2571                            unsigned int role)
2572 {
2573         struct osd_object *obj = osd_dt_obj(dt);
2574         struct osd_thread_info *oti = osd_oti_get(env);
2575
2576         LINVRNT(osd_invariant(obj));
2577
2578         LASSERT(obj->oo_owner != env);
2579         down_write_nested(&obj->oo_sem, role);
2580
2581         LASSERT(obj->oo_owner == NULL);
2582         obj->oo_owner = env;
2583         oti->oti_w_locks++;
2584 }
2585
2586 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2587 {
2588         struct osd_object *obj = osd_dt_obj(dt);
2589         struct osd_thread_info *oti = osd_oti_get(env);
2590
2591         LINVRNT(osd_invariant(obj));
2592
2593         LASSERT(oti->oti_r_locks > 0);
2594         oti->oti_r_locks--;
2595         up_read(&obj->oo_sem);
2596 }
2597
2598 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2599 {
2600         struct osd_object *obj = osd_dt_obj(dt);
2601         struct osd_thread_info *oti = osd_oti_get(env);
2602
2603         LINVRNT(osd_invariant(obj));
2604
2605         LASSERT(obj->oo_owner == env);
2606         LASSERT(oti->oti_w_locks > 0);
2607         oti->oti_w_locks--;
2608         obj->oo_owner = NULL;
2609         up_write(&obj->oo_sem);
2610 }
2611
2612 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2613 {
2614         struct osd_object *obj = osd_dt_obj(dt);
2615
2616         LINVRNT(osd_invariant(obj));
2617
2618         return obj->oo_owner == env;
2619 }
2620
2621 static void osd_inode_getattr(const struct lu_env *env,
2622                               struct inode *inode, struct lu_attr *attr)
2623 {
2624         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2625                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2626                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2627                            LA_BLKSIZE | LA_TYPE;
2628
2629         attr->la_atime = inode->i_atime.tv_sec;
2630         attr->la_mtime = inode->i_mtime.tv_sec;
2631         attr->la_ctime = inode->i_ctime.tv_sec;
2632         attr->la_mode    = inode->i_mode;
2633         attr->la_size    = i_size_read(inode);
2634         attr->la_blocks  = inode->i_blocks;
2635         attr->la_uid     = i_uid_read(inode);
2636         attr->la_gid     = i_gid_read(inode);
2637         attr->la_projid  = i_projid_read(inode);
2638         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2639         attr->la_nlink   = inode->i_nlink;
2640         attr->la_rdev    = inode->i_rdev;
2641         attr->la_blksize = 1 << inode->i_blkbits;
2642         attr->la_blkbits = inode->i_blkbits;
2643         /*
2644          * Ext4 did not transfer inherit flags from raw inode
2645          * to inode flags, and ext4 internally test raw inode
2646          * @i_flags directly. Instead of patching ext4, we do it here.
2647          */
2648         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2649                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2650 }
2651
2652 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2653                         struct lu_attr *attr)
2654 {
2655         struct osd_object *obj = osd_dt_obj(dt);
2656
2657         if (unlikely(!dt_object_exists(dt)))
2658                 return -ENOENT;
2659         if (unlikely(obj->oo_destroyed))
2660                 return -ENOENT;
2661
2662         LASSERT(!dt_object_remote(dt));
2663         LINVRNT(osd_invariant(obj));
2664
2665         spin_lock(&obj->oo_guard);
2666         osd_inode_getattr(env, obj->oo_inode, attr);
2667         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2668                 attr->la_valid |= LA_FLAGS;
2669                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2670         }
2671         spin_unlock(&obj->oo_guard);
2672
2673         return 0;
2674 }
2675
2676 static int osd_declare_attr_qid(const struct lu_env *env,
2677                                 struct osd_object *obj,
2678                                 struct osd_thandle *oh, long long bspace,
2679                                 qid_t old_id, qid_t new_id, bool enforce,
2680                                 unsigned int type, bool ignore_edquot)
2681 {
2682         int rc;
2683         struct osd_thread_info *info = osd_oti_get(env);
2684         struct lquota_id_info  *qi = &info->oti_qi;
2685
2686         qi->lqi_type = type;
2687         /* inode accounting */
2688         qi->lqi_is_blk = false;
2689
2690         /* one more inode for the new id ... */
2691         qi->lqi_id.qid_uid = new_id;
2692         qi->lqi_space      = 1;
2693         /* Reserve credits for the new id */
2694         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2695         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2696                 rc = 0;
2697         if (rc)
2698                 RETURN(rc);
2699
2700         /* and one less inode for the current id */
2701         qi->lqi_id.qid_uid = old_id;
2702         qi->lqi_space = -1;
2703         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2704         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2705                 rc = 0;
2706         if (rc)
2707                 RETURN(rc);
2708
2709         /* block accounting */
2710         qi->lqi_is_blk = true;
2711
2712         /* more blocks for the new id ... */
2713         qi->lqi_id.qid_uid = new_id;
2714         qi->lqi_space      = bspace;
2715         /*
2716          * Credits for the new uid has been reserved, re-use "obj"
2717          * to save credit reservation.
2718          */
2719         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2720         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2721                 rc = 0;
2722         if (rc)
2723                 RETURN(rc);
2724
2725         /* and finally less blocks for the current uid */
2726         qi->lqi_id.qid_uid = old_id;
2727         qi->lqi_space      = -bspace;
2728         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2729         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2730                 rc = 0;
2731
2732         RETURN(rc);
2733 }
2734
2735 static int osd_declare_attr_set(const struct lu_env *env,
2736                                 struct dt_object *dt,
2737                                 const struct lu_attr *attr,
2738                                 struct thandle *handle)
2739 {
2740         struct osd_thandle *oh;
2741         struct osd_object *obj;
2742         qid_t uid;
2743         qid_t gid;
2744         long long bspace;
2745         int rc = 0;
2746         bool enforce;
2747
2748         ENTRY;
2749
2750         LASSERT(dt != NULL);
2751         LASSERT(handle != NULL);
2752
2753         obj = osd_dt_obj(dt);
2754         LASSERT(osd_invariant(obj));
2755
2756         oh = container_of0(handle, struct osd_thandle, ot_super);
2757         LASSERT(oh->ot_handle == NULL);
2758
2759         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2760                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2761
2762         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2763                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2764
2765         if (attr == NULL || obj->oo_inode == NULL)
2766                 RETURN(rc);
2767
2768         bspace   = obj->oo_inode->i_blocks << 9;
2769         bspace   = toqb(bspace);
2770
2771         /*
2772          * Changing ownership is always preformed by super user, it should not
2773          * fail with EDQUOT unless required explicitly.
2774          *
2775          * We still need to call the osd_declare_qid() to calculate the journal
2776          * credits for updating quota accounting files and to trigger quota
2777          * space adjustment once the operation is completed.
2778          */
2779         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2780                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2781
2782                 if (!ignore_edquot)
2783                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2784                                "(the quota space is %lld)\n",
2785                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2786                                attr->la_gid, bspace);
2787
2788                 /* USERQUOTA */
2789                 uid = i_uid_read(obj->oo_inode);
2790                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2791                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2792                                           attr->la_uid, enforce, USRQUOTA,
2793                                           true);
2794                 if (rc)
2795                         RETURN(rc);
2796
2797                 gid = i_gid_read(obj->oo_inode);
2798                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2799                        attr->la_uid, gid, attr->la_gid);
2800                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2801                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2802                                           attr->la_gid, enforce, GRPQUOTA,
2803                                           ignore_edquot);
2804                 if (rc)
2805                         RETURN(rc);
2806
2807         }
2808 #ifdef HAVE_PROJECT_QUOTA
2809         if (attr->la_valid & LA_PROJID) {
2810                 __u32 projid = i_projid_read(obj->oo_inode);
2811
2812                 enforce = (attr->la_valid & LA_PROJID) &&
2813                                         (attr->la_projid != projid);
2814                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2815                                           (qid_t)projid, (qid_t)attr->la_projid,
2816                                           enforce, PRJQUOTA, true);
2817                 if (rc)
2818                         RETURN(rc);
2819         }
2820 #endif
2821         RETURN(rc);
2822 }
2823
2824 static int osd_inode_setattr(const struct lu_env *env,
2825                              struct inode *inode, const struct lu_attr *attr)
2826 {
2827         __u64 bits = attr->la_valid;
2828
2829         /* Only allow set size for regular file */
2830         if (!S_ISREG(inode->i_mode))
2831                 bits &= ~(LA_SIZE | LA_BLOCKS);
2832
2833         if (bits == 0)
2834                 return 0;
2835
2836         if (bits & LA_ATIME)
2837                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2838         if (bits & LA_CTIME)
2839                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2840         if (bits & LA_MTIME)
2841                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2842         if (bits & LA_SIZE) {
2843                 spin_lock(&inode->i_lock);
2844                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2845                 i_size_write(inode, attr->la_size);
2846                 spin_unlock(&inode->i_lock);
2847         }
2848
2849         /*
2850          * OSD should not change "i_blocks" which is used by quota.
2851          * "i_blocks" should be changed by ldiskfs only.
2852          */
2853         if (bits & LA_MODE)
2854                 inode->i_mode = (inode->i_mode & S_IFMT) |
2855                                 (attr->la_mode & ~S_IFMT);
2856         if (bits & LA_UID)
2857                 i_uid_write(inode, attr->la_uid);
2858         if (bits & LA_GID)
2859                 i_gid_write(inode, attr->la_gid);
2860         if (bits & LA_PROJID)
2861                 i_projid_write(inode, attr->la_projid);
2862         if (bits & LA_NLINK)
2863                 set_nlink(inode, attr->la_nlink);
2864         if (bits & LA_RDEV)
2865                 inode->i_rdev = attr->la_rdev;
2866
2867         if (bits & LA_FLAGS) {
2868                 /* always keep S_NOCMTIME */
2869                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2870                                  S_NOCMTIME;
2871                 /*
2872                  * Ext4 did not transfer inherit flags from
2873                  * @inode->i_flags to raw inode i_flags when writing
2874                  * flags, we do it explictly here.
2875                  */
2876                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2877                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2878                 else
2879                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2880         }
2881         return 0;
2882 }
2883
2884 #ifdef HAVE_PROJECT_QUOTA
2885 static int osd_transfer_project(struct inode *inode, __u32 projid)
2886 {
2887         struct super_block *sb = inode->i_sb;
2888         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2889         int err;
2890         kprojid_t kprojid;
2891         struct ldiskfs_iloc iloc;
2892         struct ldiskfs_inode *raw_inode;
2893         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2894
2895         if (!ldiskfs_has_feature_project(sb)) {
2896                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2897                         == LDISKFS_DEF_PROJID);
2898                 if (projid != LDISKFS_DEF_PROJID)
2899                         return -EOPNOTSUPP;
2900                 else
2901                         return 0;
2902         }
2903
2904         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2905                 return -EOPNOTSUPP;
2906
2907         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2908         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2909                 return 0;
2910
2911         err = ldiskfs_get_inode_loc(inode, &iloc);
2912         if (err)
2913                 return err;
2914
2915         raw_inode = ldiskfs_raw_inode(&iloc);
2916         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2917                 err = -EOVERFLOW;
2918                 brelse(iloc.bh);
2919                 return err;
2920         }
2921         brelse(iloc.bh);
2922
2923         dquot_initialize(inode);
2924         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2925         if (transfer_to[PRJQUOTA]) {
2926                 err = __dquot_transfer(inode, transfer_to);
2927                 dqput(transfer_to[PRJQUOTA]);
2928                 if (err)
2929                         return err;
2930         }
2931
2932         return err;
2933 }
2934 #endif
2935
2936 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2937 {
2938         int rc;
2939
2940         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2941             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2942                 struct iattr iattr;
2943
2944                 CDEBUG(D_QUOTA,
2945                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2946                        inode->i_ino, i_uid_read(inode), attr->la_uid,
2947                        i_gid_read(inode), attr->la_gid);
2948
2949                 dquot_initialize(inode);
2950                 iattr.ia_valid = 0;
2951                 if (attr->la_valid & LA_UID)
2952                         iattr.ia_valid |= ATTR_UID;
2953                 if (attr->la_valid & LA_GID)
2954                         iattr.ia_valid |= ATTR_GID;
2955                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2956                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2957
2958                 rc = dquot_transfer(inode, &iattr);
2959                 if (rc) {
2960                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2961                                "enforcement enabled on the ldiskfs "
2962                                "filesystem?\n", inode->i_sb->s_id, rc);
2963                         return rc;
2964                 }
2965         }
2966
2967         /* Handle project id transfer here properly */
2968         if (attr->la_valid & LA_PROJID &&
2969             attr->la_projid != i_projid_read(inode)) {
2970 #ifdef HAVE_PROJECT_QUOTA
2971                 rc = osd_transfer_project(inode, attr->la_projid);
2972 #else
2973                 rc = -ENOTSUPP;
2974 #endif
2975                 if (rc) {
2976                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2977                                "enforcement enabled on the ldiskfs "
2978                                "filesystem?\n", inode->i_sb->s_id, rc);
2979                         return rc;
2980                 }
2981         }
2982         return 0;
2983 }
2984
2985 static int osd_attr_set(const struct lu_env *env,
2986                         struct dt_object *dt,
2987                         const struct lu_attr *attr,
2988                         struct thandle *handle)
2989 {
2990         struct osd_object *obj = osd_dt_obj(dt);
2991         struct inode *inode;
2992         int rc;
2993
2994         if (!dt_object_exists(dt))
2995                 return -ENOENT;
2996
2997         LASSERT(handle != NULL);
2998         LASSERT(!dt_object_remote(dt));
2999         LASSERT(osd_invariant(obj));
3000
3001         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3002
3003         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3004             !osd_obj2dev(obj)->od_is_ost) {
3005                 struct osd_thread_info *oti = osd_oti_get(env);
3006                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3007                 struct lu_fid *fid1 = &oti->oti_fid;
3008                 struct osd_inode_id *id = &oti->oti_id;
3009                 struct iam_path_descr *ipd;
3010                 struct iam_container *bag;
3011                 struct osd_thandle *oh;
3012                 int rc;
3013
3014                 fid_cpu_to_be(fid1, fid0);
3015                 memset(id, 1, sizeof(*id));
3016                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3017                                   fid0)->oi_dir.od_container;
3018                 ipd = osd_idx_ipd_get(env, bag);
3019                 if (unlikely(ipd == NULL))
3020                         RETURN(-ENOMEM);
3021
3022                 oh = container_of0(handle, struct osd_thandle, ot_super);
3023                 rc = iam_update(oh->ot_handle, bag,
3024                                 (const struct iam_key *)fid1,
3025                                 (const struct iam_rec *)id, ipd);
3026                 osd_ipd_put(env, bag, ipd);
3027                 return(rc > 0 ? 0 : rc);
3028         }
3029
3030         inode = obj->oo_inode;
3031
3032         rc = osd_quota_transfer(inode, attr);
3033         if (rc)
3034                 return rc;
3035
3036         spin_lock(&obj->oo_guard);
3037         rc = osd_inode_setattr(env, inode, attr);
3038         spin_unlock(&obj->oo_guard);
3039         if (rc != 0)
3040                 GOTO(out, rc);
3041
3042         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3043
3044         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3045
3046         if (!(attr->la_valid & LA_FLAGS))
3047                 GOTO(out, rc);
3048
3049         /* Let's check if there are extra flags need to be set into LMA */
3050         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3051                 struct osd_thread_info *info = osd_oti_get(env);
3052                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3053
3054                 LASSERT(!obj->oo_pfid_in_lma);
3055
3056                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3057                                  &info->oti_ost_attrs);
3058                 if (rc)
3059                         GOTO(out, rc);
3060
3061                 lma->lma_incompat |=
3062                         lustre_to_lma_flags(attr->la_flags);
3063                 lustre_lma_swab(lma);
3064
3065                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3066
3067                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3068                                      lma, sizeof(*lma), XATTR_REPLACE);
3069                 if (rc != 0) {
3070                         struct osd_device *osd = osd_obj2dev(obj);
3071
3072                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3073                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3074                               lma->lma_incompat, rc);
3075                 } else {
3076                         obj->oo_lma_flags =
3077                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3078                 }
3079                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3080         }
3081 out:
3082
3083         return rc;
3084 }
3085
3086 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3087                                            struct osd_object *obj,
3088                                            const char *name, const int namelen)
3089 {
3090         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3091 }
3092
3093 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3094                       umode_t mode, struct dt_allocation_hint *hint,
3095                       struct thandle *th, struct lu_attr *attr)
3096 {
3097         int result;
3098         struct osd_device *osd = osd_obj2dev(obj);
3099         struct osd_thandle *oth;
3100         struct dt_object *parent = NULL;
3101         struct inode *inode;
3102         uid_t owner[2] = {0, 0};
3103
3104         if (attr->la_valid & LA_UID)
3105                 owner[0] = attr->la_uid;
3106         if (attr->la_valid & LA_GID)
3107                 owner[1] = attr->la_gid;
3108
3109         LINVRNT(osd_invariant(obj));
3110         LASSERT(obj->oo_inode == NULL);
3111         LASSERT(obj->oo_hl_head == NULL);
3112
3113         if (S_ISDIR(mode) && ldiskfs_pdo) {
3114                 obj->oo_hl_head =
3115                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3116                 if (obj->oo_hl_head == NULL)
3117                         return -ENOMEM;
3118         }
3119
3120         oth = container_of(th, struct osd_thandle, ot_super);
3121         LASSERT(oth->ot_handle->h_transaction != NULL);
3122
3123         if (hint != NULL && hint->dah_parent != NULL &&
3124             !dt_object_remote(hint->dah_parent))
3125                 parent = hint->dah_parent;
3126
3127         inode = ldiskfs_create_inode(oth->ot_handle,
3128                                      parent ? osd_dt_obj(parent)->oo_inode :
3129                                               osd_sb(osd)->s_root->d_inode,
3130                                      mode, owner);
3131         if (!IS_ERR(inode)) {
3132                 /* Do not update file c/mtime in ldiskfs. */
3133                 inode->i_flags |= S_NOCMTIME;
3134
3135                 /*
3136                  * For new created object, it must be consistent,
3137                  * and it is unnecessary to scrub against it.
3138                  */
3139                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3140
3141                 obj->oo_inode = inode;
3142                 result = 0;
3143         } else {
3144                 if (obj->oo_hl_head != NULL) {
3145                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3146                         obj->oo_hl_head = NULL;
3147                 }
3148                 result = PTR_ERR(inode);
3149         }
3150         LINVRNT(osd_invariant(obj));
3151         return result;
3152 }
3153
3154 enum {
3155         OSD_NAME_LEN = 255
3156 };
3157
3158 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3159                      struct lu_attr *attr,
3160                      struct dt_allocation_hint *hint,
3161                      struct dt_object_format *dof,
3162                      struct thandle *th)
3163 {
3164         int result;
3165         struct osd_thandle *oth;
3166         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3167
3168         LASSERT(S_ISDIR(attr->la_mode));
3169
3170         oth = container_of(th, struct osd_thandle, ot_super);
3171         LASSERT(oth->ot_handle->h_transaction != NULL);
3172         result = osd_mkfile(info, obj, mode, hint, th, attr);
3173
3174         return result;
3175 }
3176
3177 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3178                         struct lu_attr *attr,
3179                         struct dt_allocation_hint *hint,
3180                         struct dt_object_format *dof,
3181                         struct thandle *th)
3182 {
3183         int result;
3184         struct osd_thandle *oth;
3185         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3186
3187         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3188
3189         LASSERT(S_ISREG(attr->la_mode));
3190
3191         oth = container_of(th, struct osd_thandle, ot_super);
3192         LASSERT(oth->ot_handle->h_transaction != NULL);
3193
3194         result = osd_mkfile(info, obj, mode, hint, th, attr);
3195         if (result == 0) {
3196                 LASSERT(obj->oo_inode != NULL);
3197                 if (feat->dif_flags & DT_IND_VARKEY)
3198                         result = iam_lvar_create(obj->oo_inode,
3199                                                  feat->dif_keysize_max,
3200                                                  feat->dif_ptrsize,
3201                                                  feat->dif_recsize_max,
3202                                                  oth->ot_handle);
3203                 else
3204                         result = iam_lfix_create(obj->oo_inode,
3205                                                  feat->dif_keysize_max,
3206                                                  feat->dif_ptrsize,
3207                                                  feat->dif_recsize_max,
3208                                                  oth->ot_handle);
3209         }
3210         return result;
3211 }
3212
3213 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3214                      struct lu_attr *attr,
3215                      struct dt_allocation_hint *hint,
3216                      struct dt_object_format *dof,
3217                      struct thandle *th)
3218 {
3219         LASSERT(S_ISREG(attr->la_mode));
3220         return osd_mkfile(info, obj, (attr->la_mode &
3221                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3222                           attr);
3223 }
3224
3225 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3226                      struct lu_attr *attr,
3227                      struct dt_allocation_hint *hint,
3228                      struct dt_object_format *dof,
3229                      struct thandle *th)
3230 {
3231         LASSERT(S_ISLNK(attr->la_mode));
3232         return osd_mkfile(info, obj, (attr->la_mode &
3233                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3234                           attr);
3235 }
3236
3237 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3238                      struct lu_attr *attr,
3239                      struct dt_allocation_hint *hint,
3240                      struct dt_object_format *dof,
3241                      struct thandle *th)
3242 {
3243         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3244         int result;
3245
3246         LINVRNT(osd_invariant(obj));
3247         LASSERT(obj->oo_inode == NULL);
3248         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3249                 S_ISFIFO(mode) || S_ISSOCK(mode));
3250
3251         result = osd_mkfile(info, obj, mode, hint, th, attr);
3252         if (result == 0) {
3253                 LASSERT(obj->oo_inode != NULL);
3254                 /*
3255                  * This inode should be marked dirty for i_rdev.  Currently
3256                  * that is done in the osd_attr_init().
3257                  */
3258                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3259                                    attr->la_rdev);
3260         }
3261         LINVRNT(osd_invariant(obj));
3262         return result;
3263 }
3264
3265 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3266                               struct lu_attr *,
3267                               struct dt_allocation_hint *hint,
3268                               struct dt_object_format *dof,
3269                               struct thandle *);
3270
3271 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3272 {
3273         osd_obj_type_f result;
3274
3275         switch (type) {
3276         case DFT_DIR:
3277                 result = osd_mkdir;
3278                 break;
3279         case DFT_REGULAR:
3280                 result = osd_mkreg;
3281                 break;
3282         case DFT_SYM:
3283                 result = osd_mksym;
3284                 break;
3285         case DFT_NODE:
3286                 result = osd_mknod;
3287                 break;
3288         case DFT_INDEX:
3289                 result = osd_mk_index;
3290                 break;
3291
3292         default:
3293                 LBUG();
3294                 break;
3295         }
3296         return result;
3297 }
3298
3299
3300 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3301                         struct dt_object *parent, struct dt_object *child,
3302                         umode_t child_mode)
3303 {
3304         LASSERT(ah);
3305
3306         ah->dah_parent = parent;
3307         ah->dah_mode = child_mode;
3308
3309         if (parent != NULL && !dt_object_remote(parent)) {
3310                 /* will help to find FID->ino at dt_insert("..") */
3311                 struct osd_object *pobj = osd_dt_obj(parent);
3312
3313                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3314         }
3315 }
3316
3317 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3318                           struct lu_attr *attr, struct dt_object_format *dof)
3319 {
3320         struct inode *inode = obj->oo_inode;
3321         __u64 valid = attr->la_valid;
3322         int result;
3323
3324         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3325
3326         if (dof->dof_type != DFT_NODE)
3327                 attr->la_valid &= ~LA_RDEV;
3328         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3329                 attr->la_valid &= ~LA_ATIME;
3330         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3331                 attr->la_valid &= ~LA_CTIME;
3332         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3333                 attr->la_valid &= ~LA_MTIME;
3334
3335         result = osd_quota_transfer(inode, attr);
3336         if (result)
3337                 return;
3338
3339         if (attr->la_valid != 0) {
3340                 result = osd_inode_setattr(info->oti_env, inode, attr);
3341                 /*
3342                  * The osd_inode_setattr() should always succeed here.  The
3343                  * only error that could be returned is EDQUOT when we are
3344                  * trying to change the UID or GID of the inode. However, this
3345                  * should not happen since quota enforcement is no longer
3346                  * enabled on ldiskfs (lquota takes care of it).
3347                  */
3348                 LASSERTF(result == 0, "%d\n", result);
3349                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3350         }
3351
3352         attr->la_valid = valid;
3353 }
3354
3355 /**
3356  * Helper function for osd_create()
3357  *
3358  * \retval 0, on success
3359  */
3360 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3361                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3362                         struct dt_object_format *dof, struct thandle *th)
3363 {
3364         int result;
3365         __u32 umask;
3366
3367         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3368
3369         /* we drop umask so that permissions we pass are not affected */
3370         umask = current->fs->umask;
3371         current->fs->umask = 0;
3372
3373         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3374                                                   th);
3375         if (likely(obj->oo_inode != NULL)) {
3376                 LASSERT(obj->oo_inode->i_state & I_NEW);
3377
3378                 /*
3379                  * Unlock the inode before attr initialization to avoid
3380                  * unnecessary dqget operations. LU-6378
3381                  */
3382                 unlock_new_inode(obj->oo_inode);
3383         }
3384
3385         if (likely(result == 0)) {
3386                 osd_attr_init(info, obj, attr, dof);
3387                 osd_object_init0(obj);
3388         }
3389
3390         /* restore previous umask value */
3391         current->fs->umask = umask;
3392
3393         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3394
3395         return result;
3396 }
3397
3398 /**
3399  * Helper function for osd_create()
3400  *
3401  * \retval 0, on success
3402  */
3403 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3404                            const struct lu_fid *fid, struct thandle *th)
3405 {
3406         struct osd_thread_info *info = osd_oti_get(env);
3407         struct osd_inode_id    *id   = &info->oti_id;
3408         struct osd_device      *osd  = osd_obj2dev(obj);
3409         struct osd_thandle     *oh;
3410         int rc;
3411
3412         LASSERT(obj->oo_inode != NULL);
3413
3414         oh = container_of0(th, struct osd_thandle, ot_super);
3415         LASSERT(oh->ot_handle);
3416         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3417
3418         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3419         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3420                            OI_CHECK_FLD, NULL);
3421         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3422                 struct lu_fid next_fid = *fid;
3423
3424                 /* insert next object in advance, and map to the same inode */
3425                 next_fid.f_oid++;
3426                 if (next_fid.f_oid != 0) {
3427                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3428                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3429                                       OI_CHECK_FLD, NULL);
3430                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3431                 }
3432         }
3433
3434         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3435
3436         return rc;
3437 }
3438
3439 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3440                    u64 seq, struct lu_seq_range *range)
3441 {
3442         struct seq_server_site *ss = osd_seq_site(osd);
3443
3444         if (fid_seq_is_idif(seq)) {
3445                 fld_range_set_ost(range);
3446                 range->lsr_index = idif_ost_idx(seq);
3447                 return 0;
3448         }
3449
3450         if (!fid_seq_in_fldb(seq)) {
3451                 fld_range_set_mdt(range);
3452                 if (ss != NULL)
3453                         /*
3454                          * FIXME: If ss is NULL, it suppose not get lsr_index
3455                          * at all
3456                          */
3457                         range->lsr_index = ss->ss_node_id;
3458                 return 0;
3459         }
3460
3461         LASSERT(ss != NULL);
3462         fld_range_set_any(range);
3463         /* OSD will only do local fld lookup */
3464         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3465 }
3466
3467 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3468                               struct lu_attr *attr,
3469                               struct dt_allocation_hint *hint,
3470                               struct dt_object_format *dof,
3471                               struct thandle *handle)
3472 {
3473         struct osd_thandle *oh;
3474         int rc;
3475
3476         ENTRY;
3477
3478         LASSERT(handle != NULL);
3479
3480         oh = container_of0(handle, struct osd_thandle, ot_super);
3481         LASSERT(oh->ot_handle == NULL);
3482
3483         /*
3484          * EA object consumes more credits than regular object: osd_mk_index
3485          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3486          * leaf_node, could involves the block, block bitmap, groups, GDT
3487          * change for each block, so add 4 * 2 credits in that case.
3488          */
3489         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3490                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3491                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3492         /*
3493          * Reuse idle OI block may cause additional one OI block
3494          * to be changed.
3495          */
3496         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3497                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3498         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3499                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3500                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3501
3502         /* will help to find FID->ino mapping at dt_insert() */
3503         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3504                                    osd_dt_obj(dt));
3505         if (rc != 0)
3506                 RETURN(rc);
3507
3508         if (!attr)
3509                 RETURN(0);
3510
3511         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3512                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3513                                    NULL, OSD_QID_INODE);
3514         if (rc != 0)
3515                 RETURN(rc);
3516
3517         RETURN(rc);
3518 }
3519
3520 /**
3521  * Called to destroy on-disk representation of the object
3522  *
3523  * Concurrency: must be locked
3524  */
3525 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3526                                struct thandle *th)
3527 {
3528         struct osd_object *obj = osd_dt_obj(dt);
3529         struct inode *inode = obj->oo_inode;
3530         struct osd_thandle *oh;
3531         int rc;
3532
3533         ENTRY;
3534
3535         if (inode == NULL)
3536                 RETURN(-ENOENT);
3537
3538         oh = container_of0(th, struct osd_thandle, ot_super);
3539         LASSERT(oh->ot_handle == NULL);
3540
3541         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3542                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3543
3544         /* For removing agent entry */
3545         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3546                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3547
3548         /*
3549          * Recycle idle OI leaf may cause additional three OI blocks
3550          * to be changed.
3551          */
3552         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3553                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3554                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3555         /* one less inode */
3556         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3557                                    i_projid_read(inode), -1, oh, obj, NULL,
3558                                    OSD_QID_INODE);
3559         if (rc)
3560                 RETURN(rc);
3561         /* data to be truncated */
3562         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3563                                    i_projid_read(inode), 0, oh, obj, NULL,
3564                                    OSD_QID_BLK);
3565         if (rc)
3566                 RETURN(rc);
3567
3568         /*
3569          * will help to find FID->ino when this object is being
3570          * added to PENDING
3571          */
3572         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3573
3574         RETURN(rc);
3575 }
3576
3577 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3578                        struct thandle *th)
3579 {
3580         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3581         struct osd_object *obj = osd_dt_obj(dt);
3582         struct inode *inode = obj->oo_inode;
3583         struct osd_device *osd = osd_obj2dev(obj);
3584         struct osd_thandle *oh;
3585         int result;
3586
3587         ENTRY;
3588
3589         oh = container_of0(th, struct osd_thandle, ot_super);
3590         LASSERT(oh->ot_handle);
3591         LASSERT(inode);
3592         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3593
3594         if (unlikely(fid_is_acct(fid)))
3595                 RETURN(-EPERM);
3596
3597         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3598                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3599                 if (result != 0)
3600                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3601                                osd_name(osd), PFID(fid), result);
3602         }
3603
3604         if (S_ISDIR(inode->i_mode)) {
3605                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3606                         inode->i_nlink == 2);
3607
3608                 spin_lock(&obj->oo_guard);
3609                 clear_nlink(inode);
3610                 spin_unlock(&obj->oo_guard);
3611                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3612         }
3613
3614         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3615
3616         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3617
3618         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3619                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3620                                        oh->ot_handle, OI_CHECK_FLD);
3621
3622         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3623         /* XXX: add to ext3 orphan list */
3624         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3625
3626         /* not needed in the cache anymore */
3627         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3628         obj->oo_destroyed = 1;
3629
3630         RETURN(0);
3631 }
3632
3633 /**
3634  * Put the fid into lustre_mdt_attrs, and then place the structure
3635  * inode's ea. This fid should not be altered during the life time
3636  * of the inode.
3637  *
3638  * \retval +ve, on success
3639  * \retval -ve, on error
3640  *
3641  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3642  */
3643 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3644                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3645 {
3646         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3647         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3648         int rc;
3649
3650         ENTRY;
3651
3652         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3653                 RETURN(0);
3654
3655         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3656                 rc = -ENOMEM;
3657
3658         lustre_loa_init(loa, fid, compat, incompat);
3659         lustre_loa_swab(loa, false);
3660
3661         /*
3662          * For the OST device with 256 bytes inode size by default,
3663          * the PFID EA will be stored together with LMA EA to avoid
3664          * performance trouble. Otherwise the PFID EA can be stored
3665          * independently. LU-8998
3666          */
3667         if ((compat & LMAC_FID_ON_OST) &&
3668             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3669                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3670                                      sizeof(*loa), XATTR_CREATE);
3671         else
3672                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3673                                      sizeof(*lma), XATTR_CREATE);
3674         /*
3675          * LMA may already exist, but we need to check that all the
3676          * desired compat/incompat flags have been added.
3677          */
3678         if (unlikely(rc == -EEXIST)) {
3679                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3680                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3681                 if (rc < 0)
3682                         RETURN(rc);
3683
3684                 if (rc < sizeof(*lma))
3685                         RETURN(-EINVAL);
3686
3687                 lustre_loa_swab(loa, true);
3688                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3689                     ((compat == 0 && incompat == 0) ||
3690                      (!(~lma->lma_compat & compat) &&
3691                       !(~lma->lma_incompat & incompat))))
3692                         RETURN(0);
3693
3694                 lma->lma_self_fid = *fid;
3695                 lma->lma_compat |= compat;
3696                 lma->lma_incompat |= incompat;
3697                 if (rc == sizeof(*lma)) {
3698                         lustre_lma_swab(lma);
3699                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3700                                              sizeof(*lma), XATTR_REPLACE);
3701                 } else {
3702                         lustre_loa_swab(loa, false);
3703                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3704                                              sizeof(*loa), XATTR_REPLACE);
3705                 }
3706         }
3707
3708         RETURN(rc);
3709 }
3710
3711 /**
3712  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3713  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3714  * To have compatilibility with 1.8 ldiskfs driver we need to have
3715  * magic number at start of fid data.
3716  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3717  * its inmemory API.
3718  */
3719 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3720                                          const struct lu_fid *fid)
3721 {
3722         if (!fid_is_namespace_visible(fid) ||
3723             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3724                 param->edp_magic = 0;
3725                 return;
3726         }
3727
3728         param->edp_magic = LDISKFS_LUFID_MAGIC;
3729         param->edp_len =  sizeof(struct lu_fid) + 1;
3730         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3731 }
3732
3733 /**
3734  * Try to read the fid from inode ea into dt_rec.
3735  *
3736  * \param fid object fid.
3737  *
3738  * \retval 0 on success
3739  */
3740 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3741                           __u32 ino, struct lu_fid *fid,
3742                           struct osd_inode_id *id)
3743 {
3744         struct osd_thread_info *info  = osd_oti_get(env);
3745         struct inode *inode;
3746
3747         ENTRY;
3748
3749         osd_id_gen(id, ino, OSD_OII_NOGEN);
3750         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3751         if (IS_ERR(inode))
3752                 RETURN(PTR_ERR(inode));
3753
3754         iput(inode);
3755         RETURN(0);
3756 }
3757
3758 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3759                                         struct inode *dir,
3760                                         struct inode *parent_dir,
3761                                         const struct lu_fid *dot_fid,
3762                                         const struct lu_fid *dot_dot_fid,
3763                                         struct osd_thandle *oth)
3764 {
3765         struct ldiskfs_dentry_param *dot_ldp;
3766         struct ldiskfs_dentry_param *dot_dot_ldp;
3767         __u32 saved_nlink = dir->i_nlink;
3768         int rc;
3769
3770         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3771         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3772
3773         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3774         dot_ldp->edp_magic = 0;
3775
3776         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3777                                     dir, dot_ldp, dot_dot_ldp);
3778         /*
3779          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3780          * the subseqent ref_add() will increase the dir->i_nlink
3781          * as 3. That is incorrect for new created directory.
3782          *
3783          * It looks like hack, because we want to make the OSD API
3784          * to be order-independent for new created directory object
3785          * between dt_insert(..) and ref_add() operations.
3786          *
3787          * Here, we only restore the in-RAM dir-inode's nlink attr,
3788          * becuase if the nlink attr is not 2, then there will be
3789          * ref_add() called following the dt_insert(..), such call
3790          * will make both the in-RAM and on-disk dir-inode's nlink
3791          * attr to be set as 2. LU-7447
3792          */
3793         set_nlink(dir, saved_nlink);
3794         return rc;
3795 }
3796
3797 /**
3798  * Create an local agent inode for remote entry
3799  */
3800 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3801                                                   struct osd_device *osd,
3802                                                   struct osd_object *pobj,
3803                                                   const struct lu_fid *fid,
3804                                                   __u32 type,
3805                                                   struct thandle *th)
3806 {
3807         struct osd_thread_info *info = osd_oti_get(env);
3808         struct inode *local;
3809         struct osd_thandle *oh;
3810         int rc;
3811
3812         ENTRY;
3813
3814         LASSERT(th);
3815         oh = container_of(th, struct osd_thandle, ot_super);
3816         LASSERT(oh->ot_handle->h_transaction != NULL);
3817
3818         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3819                                      NULL);
3820         if (IS_ERR(local)) {
3821                 CERROR("%s: create local error %d\n", osd_name(osd),
3822                        (int)PTR_ERR(local));
3823                 RETURN(local);
3824         }
3825
3826         /*
3827          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3828          * correct gid on remote file, not agent here
3829          */
3830         local->i_gid = current_fsgid();
3831         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3832
3833         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3834          * That gives e2fsck something to look at and be happy, and allows
3835          * debugging if we need to determine where this symlink came from.
3836          */
3837         if (S_ISLNK(type)) {
3838                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3839                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3840                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3841
3842                 i_size_write(local, rc);
3843                 LDISKFS_I(local)->i_disksize = rc;
3844         }
3845         unlock_new_inode(local);
3846
3847         /* Agent inode should not have project ID */
3848 #ifdef  HAVE_PROJECT_QUOTA
3849         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3850             i_projid_read(pobj->oo_inode) != 0) {
3851                 rc = osd_transfer_project(local, 0);
3852                 if (rc) {
3853                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3854                                "quota enforcement enabled on the ldiskfs "
3855                                "filesystem?\n", local->i_sb->s_id, rc);
3856                         RETURN(ERR_PTR(rc));
3857                 }
3858         }
3859 #endif
3860         /* Set special LMA flag for local agent inode */
3861         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3862         if (rc != 0) {
3863                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3864                        osd_name(osd), PFID(fid), rc);
3865                 RETURN(ERR_PTR(rc));
3866         }
3867
3868         if (!S_ISDIR(type))
3869                 RETURN(local);
3870
3871         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3872                                          lu_object_fid(&pobj->oo_dt.do_lu),
3873                                          fid, oh);
3874         if (rc != 0) {
3875                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3876                         osd_name(osd), PFID(fid), rc);
3877                 RETURN(ERR_PTR(rc));
3878         }
3879
3880         RETURN(local);
3881 }
3882
3883 /**
3884  * when direntry is deleted, we have to take care of possible agent inode
3885  * referenced by that. unfortunately we can't do this at that point:
3886  * iget() within a running transaction leads to deadlock and we better do
3887  * not call that every delete declaration to save performance. so we put
3888  * a potention agent inode on a list and process that once the transaction
3889  * is over. Notice it's not any worse in terms of real orphans as regular
3890  * object destroy doesn't put inodes on the on-disk orphan list. this should
3891  * be addressed separately
3892  */
3893 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3894                                             struct osd_thandle *oh,
3895                                             __u32 ino)
3896 {
3897         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3898         struct osd_obj_orphan *oor;
3899
3900         OBD_ALLOC_PTR(oor);
3901         if (oor == NULL)
3902                 return -ENOMEM;
3903
3904         oor->oor_ino = ino;
3905         oor->oor_env = (struct lu_env *)env;
3906         spin_lock(&osd->od_osfs_lock);
3907         list_add(&oor->oor_list, &osd->od_orphan_list);
3908         spin_unlock(&osd->od_osfs_lock);
3909
3910         oh->ot_remove_agents = 1;
3911
3912         return 0;
3913
3914 }
3915
3916 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3917                                                 struct osd_device *osd)
3918 {
3919         struct osd_thread_info *info = osd_oti_get(env);
3920         struct osd_obj_orphan *oor, *tmp;
3921         struct osd_inode_id id;
3922         struct list_head list;
3923         struct inode *inode;
3924         struct lu_fid fid;
3925         handle_t *jh;
3926         __u32 ino;
3927
3928         INIT_LIST_HEAD(&list);
3929
3930         spin_lock(&osd->od_osfs_lock);
3931         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3932                 if (oor->oor_env == env) {
3933                         list_del(&oor->oor_list);
3934                         list_add(&oor->oor_list, &list);
3935                 }
3936         }
3937         spin_unlock(&osd->od_osfs_lock);
3938
3939         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3940
3941                 ino = oor->oor_ino;
3942
3943                 list_del(&oor->oor_list);
3944                 OBD_FREE_PTR(oor);
3945
3946                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3947                 inode = osd_iget_fid(info, osd, &id, &fid);
3948                 if (IS_ERR(inode))
3949                         continue;
3950
3951                 if (!osd_remote_fid(env, osd, &fid)) {
3952                         iput(inode);
3953                         continue;
3954                 }
3955
3956                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3957                 clear_nlink(inode);
3958                 mark_inode_dirty(inode);
3959                 ldiskfs_journal_stop(jh);
3960                 iput(inode);
3961         }
3962
3963         return 0;
3964 }
3965
3966 /**
3967  * OSD layer object create function for OST objects (b=11826).
3968  *
3969  * The FID is inserted into inode xattr here.
3970  *
3971  * \retval   0, on success
3972  * \retval -ve, on error
3973  */
3974 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3975                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3976                       struct dt_object_format *dof, struct thandle *th)
3977 {
3978         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3979         struct osd_object *obj = osd_dt_obj(dt);
3980         struct osd_thread_info *info = osd_oti_get(env);
3981         int result, on_ost = 0;
3982
3983         ENTRY;
3984
3985         if (dt_object_exists(dt))
3986                 RETURN(-EEXIST);
3987
3988         LINVRNT(osd_invariant(obj));
3989         LASSERT(!dt_object_remote(dt));
3990         LASSERT(osd_is_write_locked(env, obj));
3991         LASSERT(th != NULL);
3992
3993         if (unlikely(fid_is_acct(fid)))
3994                 /*
3995                  * Quota files can't be created from the kernel any more,
3996                  * 'tune2fs -O quota' will take care of creating them
3997                  */
3998                 RETURN(-EPERM);
3999
4000         result = __osd_create(info, obj, attr, hint, dof, th);
4001         if (result == 0) {
4002                 if (fid_is_idif(fid) &&
4003                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4004                         struct lu_fid *tfid = &info->oti_fid;
4005                         struct ost_id *oi   = &info->oti_ostid;
4006
4007                         fid_to_ostid(fid, oi);
4008                         ostid_to_fid(tfid, oi, 0);
4009                         on_ost = 1;
4010                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4011                                                 LMAC_FID_ON_OST, 0);
4012                 } else {
4013                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4014                                                fid, OI_CHECK_FLD);
4015                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4016                                                 on_ost ? LMAC_FID_ON_OST : 0,
4017                                                 0);
4018                 }
4019                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4020                         obj->oo_dt.do_body_ops = &osd_body_ops;
4021         }
4022
4023         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4024                 result = __osd_oi_insert(env, obj, fid, th);
4025
4026         /*
4027          * a small optimization - dt_insert() isn't usually applied
4028          * to OST objects, so we don't need to cache OI mapping for
4029          * OST objects
4030          */
4031         if (result == 0 && on_ost == 0) {
4032                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4033
4034                 result = osd_idc_find_and_init(env, osd, obj);
4035                 LASSERT(result == 0);
4036         }
4037
4038         LASSERT(ergo(result == 0,
4039                      dt_object_exists(dt) && !dt_object_remote(dt)));
4040         LINVRNT(osd_invariant(obj));
4041         RETURN(result);
4042 }
4043
4044 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4045                                struct thandle *handle)
4046 {
4047         struct osd_thandle *oh;
4048
4049         /* it's possible that object doesn't exist yet */
4050         LASSERT(handle != NULL);
4051
4052         oh = container_of0(handle, struct osd_thandle, ot_super);
4053         LASSERT(oh->ot_handle == NULL);
4054
4055         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4056                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4057
4058         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4059
4060         return 0;
4061 }
4062
4063 /*
4064  * Concurrency: @dt is write locked.
4065  */
4066 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4067                        struct thandle *th)
4068 {
4069         struct osd_object *obj = osd_dt_obj(dt);
4070         struct inode *inode = obj->oo_inode;
4071         struct osd_thandle *oh;
4072         int rc = 0;
4073
4074         if (!dt_object_exists(dt) || obj->oo_destroyed)
4075                 return -ENOENT;
4076
4077         LINVRNT(osd_invariant(obj));
4078         LASSERT(!dt_object_remote(dt));
4079         LASSERT(osd_is_write_locked(env, obj));
4080         LASSERT(th != NULL);
4081
4082         oh = container_of0(th, struct osd_thandle, ot_super);
4083         LASSERT(oh->ot_handle != NULL);
4084
4085         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4086
4087         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4088                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4089         /*
4090          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4091          * (65000) subdirectories by storing "1" in i_nlink if the link count
4092          * would otherwise overflow. Directory tranversal tools understand
4093          * that (st_nlink == 1) indicates that the filesystem dose not track
4094          * hard links count on the directory, and will not abort subdirectory
4095          * scanning early once (st_nlink - 2) subdirs have been found.
4096          *
4097          * This also has to properly handle the case of inodes with nlink == 0
4098          * in case they are being linked into the PENDING directory
4099          */
4100         spin_lock(&obj->oo_guard);
4101         if (unlikely(inode->i_nlink == 0))
4102                 /* inc_nlink from 0 may cause WARN_ON */
4103                 set_nlink(inode, 1);
4104         else {
4105                 ldiskfs_inc_count(oh->ot_handle, inode);
4106                 if (!S_ISDIR(inode->i_mode))
4107                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4108         }
4109         spin_unlock(&obj->oo_guard);
4110
4111         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4112         LINVRNT(osd_invariant(obj));
4113
4114         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4115
4116         return rc;
4117 }
4118
4119 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4120                                struct thandle *handle)
4121 {
4122         struct osd_thandle *oh;
4123
4124         if (!dt_object_exists(dt))
4125                 return -ENOENT;
4126
4127         LASSERT(!dt_object_remote(dt));
4128         LASSERT(handle != NULL);
4129
4130         oh = container_of0(handle, struct osd_thandle, ot_super);
4131         LASSERT(oh->ot_handle == NULL);
4132
4133         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4134                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4135
4136         return 0;
4137 }
4138
4139 /*
4140  * Concurrency: @dt is write locked.
4141  */
4142 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4143                        struct thandle *th)
4144 {
4145         struct osd_object *obj = osd_dt_obj(dt);
4146         struct inode *inode = obj->oo_inode;
4147         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4148         struct osd_thandle *oh;
4149
4150         if (!dt_object_exists(dt))
4151                 return -ENOENT;
4152
4153         LINVRNT(osd_invariant(obj));
4154         LASSERT(!dt_object_remote(dt));
4155         LASSERT(osd_is_write_locked(env, obj));
4156         LASSERT(th != NULL);
4157
4158         oh = container_of0(th, struct osd_thandle, ot_super);
4159         LASSERT(oh->ot_handle != NULL);
4160
4161         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4162
4163         spin_lock(&obj->oo_guard);
4164         /*
4165          * That can be result of upgrade from old Lustre version and
4166          * applied only to local files.  Just skip this ref_del call.
4167          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4168          */
4169         if (inode->i_nlink == 0) {
4170                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4171                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4172                              ", maybe an upgraded file? (LU-3915)\n",
4173                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4174                 spin_unlock(&obj->oo_guard);
4175                 return 0;
4176         }
4177
4178         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4179                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4180
4181         ldiskfs_dec_count(oh->ot_handle, inode);
4182         spin_unlock(&obj->oo_guard);
4183
4184         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4185         LINVRNT(osd_invariant(obj));
4186
4187         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4188
4189         return 0;
4190 }
4191
4192 /*
4193  * Concurrency: @dt is read locked.
4194  */
4195 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4196                          struct lu_buf *buf, const char *name)
4197 {
4198         struct osd_object      *obj    = osd_dt_obj(dt);
4199         struct inode           *inode  = obj->oo_inode;
4200         struct osd_thread_info *info   = osd_oti_get(env);
4201         struct dentry          *dentry = &info->oti_obj_dentry;
4202         bool cache_xattr = false;
4203         int rc;
4204
4205         LASSERT(buf);
4206
4207         /* version get is not real XATTR but uses xattr API */
4208         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4209                 dt_obj_version_t *ver = buf->lb_buf;
4210
4211                 /*
4212                  * for version we are just using xattr API but change inode
4213                  * field instead
4214                  */
4215                 if (buf->lb_len == 0)
4216                         return sizeof(dt_obj_version_t);
4217
4218                 if (buf->lb_len < sizeof(dt_obj_version_t))
4219                         return -ERANGE;
4220
4221                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4222                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4223
4224                 *ver = LDISKFS_I(inode)->i_fs_version;
4225
4226                 return sizeof(dt_obj_version_t);
4227         }
4228
4229         if (!dt_object_exists(dt))
4230                 return -ENOENT;
4231
4232         LASSERT(!dt_object_remote(dt));
4233         LASSERT(inode->i_op != NULL);
4234 #ifdef HAVE_IOP_XATTR
4235         LASSERT(inode->i_op->getxattr != NULL);
4236 #endif
4237
4238         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4239             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4240                 cache_xattr = true;
4241
4242         if (cache_xattr) {
4243                 rc = osd_oxc_get(obj, name, buf);
4244                 if (rc != -ENOENT)
4245                         return rc;
4246         }
4247
4248         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4249                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4250                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4251                 struct filter_fid *ff;
4252                 struct ost_layout *ol;
4253
4254                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4255                 if (rc)
4256                         return rc;
4257
4258                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4259
4260                 rc = sizeof(*ff);
4261                 if (buf->lb_len == 0 || !buf->lb_buf)
4262                         return rc;
4263
4264                 if (buf->lb_len < rc)
4265                         return -ERANGE;
4266
4267                 ff = buf->lb_buf;
4268                 ol = &ff->ff_layout;
4269                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4270                                                   PFID_STRIPE_IDX_BITS);
4271                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4272                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4273                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4274                 if (lma->lma_compat & LMAC_COMP_INFO) {
4275                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4276                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4277                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4278                 } else {
4279                         ol->ol_comp_start = 0;
4280                         ol->ol_comp_end = 0;
4281                         ol->ol_comp_id = 0;
4282                 }
4283         } else {
4284                 rc = __osd_xattr_get(inode, dentry, name,
4285                                      buf->lb_buf, buf->lb_len);
4286         }
4287
4288         if (cache_xattr) {
4289                 if (rc == -ENOENT || rc == -ENODATA)
4290                         osd_oxc_add(obj, name, NULL, 0);
4291                 else if (rc > 0 && buf->lb_buf != NULL)
4292                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4293         }
4294
4295         return rc;
4296 }
4297
4298 static int osd_declare_xattr_set(const struct lu_env *env,
4299                                  struct dt_object *dt,
4300                                  const struct lu_buf *buf, const char *name,
4301                                  int fl, struct thandle *handle)
4302 {
4303         struct osd_thandle *oh;
4304         int credits = 0;
4305         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4306
4307         LASSERT(handle != NULL);
4308
4309         oh = container_of0(handle, struct osd_thandle, ot_super);
4310         LASSERT(oh->ot_handle == NULL);
4311
4312         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4313                 /*
4314                  * For non-upgrading case, the LMA is set first and
4315                  * usually fit inode. But for upgrade case, the LMA
4316                  * may be in another separated EA block.
4317                  */
4318                 if (dt_object_exists(dt)) {
4319                         if (fl == LU_XATTR_REPLACE)
4320                                 credits = 1;
4321                         else
4322                                 goto upgrade;
4323                 }
4324         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4325                 credits = 1;
4326         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4327                 /* We may need to delete the old PFID EA. */
4328                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4329                 if (fl == LU_XATTR_REPLACE)
4330                         credits += 1;
4331                 else
4332                         goto upgrade;
4333         } else {
4334                 /*
4335                  * If some name entry resides on remote MDT, then will create
4336                  * agent entry under remote parent. On the other hand, if the
4337                  * remote entry will be removed, then related agent entry may
4338                  * need to be removed from the remote parent. So there may be
4339                  * kinds of cases, let's declare enough credits. The credits
4340                  * for create agent entry is enough for remove case.
4341                  */
4342                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4343                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4344                         if (dt_object_exists(dt))
4345                                 credits += 1; /* For updating LMA */
4346                 }
4347
4348 upgrade:
4349                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4350
4351                 if (buf != NULL) {
4352                         ssize_t buflen;
4353
4354                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4355                                 /*
4356                                  * learn xattr size from osd_xattr_get if
4357                                  * attribute has not been read yet
4358                                  */
4359                                 buflen = __osd_xattr_get(
4360                                     osd_dt_obj(dt)->oo_inode,
4361                                     &osd_oti_get(env)->oti_obj_dentry,
4362                                     name, NULL, 0);
4363                                 if (buflen < 0)
4364                                         buflen = 0;
4365                         } else {
4366                                 buflen = buf->lb_len;
4367                         }
4368
4369                         if (buflen > sb->s_blocksize) {
4370                                 credits += osd_calc_bkmap_credits(
4371                                     sb, NULL, 0, -1,
4372                                     (buflen + sb->s_blocksize - 1) >>
4373                                     sb->s_blocksize_bits);
4374                         }
4375                 }
4376                 /*
4377                  * xattr set may involve inode quota change, reserve credits for
4378                  * dquot_initialize()
4379                  */
4380                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4381         }
4382
4383         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4384
4385         return 0;
4386 }
4387
4388 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4389                               const struct lu_buf *buf, int fl,
4390                               struct thandle *handle)
4391 {
4392         struct osd_thread_info *info = osd_oti_get(env);
4393         struct dentry *dentry = &info->oti_obj_dentry;
4394         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4395         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4396         struct inode *inode = obj->oo_inode;
4397         struct filter_fid *ff = buf->lb_buf;
4398         struct ost_layout *ol = &ff->ff_layout;
4399         int flags = XATTR_REPLACE;
4400         int rc;
4401
4402         ENTRY;
4403
4404         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4405                 RETURN(-EINVAL);
4406
4407         rc = osd_get_lma(info, inode, dentry, loa);
4408         if (rc == -ENODATA) {
4409                 /* Usually for upgarding from old device */
4410                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4411                                 LMAC_FID_ON_OST, 0);
4412                 flags = XATTR_CREATE;
4413         } else if (rc) {
4414                 RETURN(rc);
4415         }
4416
4417         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4418                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4419                         RETURN(-EEXIST);
4420
4421                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4422                         /* Separate PFID EA from LMA */
4423                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4424                         lustre_lma_swab(lma);
4425                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4426                                              sizeof(*lma), XATTR_REPLACE);
4427                         if (!rc) {
4428                                 obj->oo_pfid_in_lma = 0;
4429                                 rc = LU_XATTR_CREATE;
4430                         }
4431
4432                         RETURN(rc);
4433                 }
4434         } else {
4435                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4436                         RETURN(fl);
4437
4438                 /*
4439                  * Old client does not send stripe information,
4440                  * then store the PFID EA on disk separatedly.
4441                  */
4442                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4443                              ol->ol_stripe_size == 0))
4444                         RETURN(fl);
4445
4446                 /* Remove old PFID EA entry firstly. */
4447                 dquot_initialize(inode);
4448                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4449                 if (rc == -ENODATA) {
4450                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4451                                 RETURN(rc);
4452                 } else if (rc) {
4453                         RETURN(rc);
4454                 }
4455         }
4456
4457         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4458         if (likely(ol->ol_stripe_size != 0)) {
4459                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4460                                              PFID_STRIPE_IDX_BITS;
4461                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4462                 lma->lma_compat |= LMAC_STRIPE_INFO;
4463                 if (ol->ol_comp_id != 0) {
4464                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4465                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4466                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4467                         lma->lma_compat |= LMAC_COMP_INFO;
4468                 }
4469         }
4470
4471         lustre_loa_swab(loa, false);
4472
4473         /* Store the PFID EA inside LMA. */
4474         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4475                              flags);
4476         if (!rc)
4477                 obj->oo_pfid_in_lma = 1;
4478
4479         RETURN(rc);
4480 }
4481
4482 /*
4483  * In DNE environment, the object (in spite of regular file or directory)
4484  * and its name entry may reside on different MDTs. Under such case, we will
4485  * create an agent entry on the MDT where the object resides. The agent entry
4486  * references the object locally, that makes the object to be visible to the
4487  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4488  * such as 'tar' can handle the object properly.
4489  *
4490  * We handle the agent entry during set linkEA that is the common interface
4491  * for both regular file and directroy, can handle kinds of cases, such as
4492  * create/link/unlink/rename, and so on.
4493  *
4494  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4495  *
4496  * XXX: There are two known issues:
4497  * 1. For one object, we will create at most one agent entry even if there
4498  *    may be more than one cross-MDTs hard links on the object. So the local
4499  *    e2fsck may claim that the object's nlink is larger than the name entries
4500  *    that reference such inode. And in further, the e2fsck will fix the nlink
4501  *    attribute to match the local references. Then it will cause the object's
4502  *    nlink attribute to be inconsistent with the global references. it is bad
4503  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4504  *    other hand, the global namespace LFSCK can repair the object's attribute
4505  *    according to the linkEA.
4506  * 2. There may be too many hard links on the object as to its linkEA overflow,
4507  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4508  *    case happened, then at this point, we do not know whether there are some
4509  *    cross-MDTs reference. But there are local references, it guarantees that
4510  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4511  */
4512 static int osd_xattr_handle_linkea(const struct lu_env *env,
4513                                    struct osd_device *osd,
4514                                    struct osd_object *obj,
4515                                    const struct lu_buf *buf,
4516                                    struct thandle *handle)
4517 {
4518         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4519         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4520         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4521         struct lu_name tmpname;
4522         struct osd_thandle *oh;
4523         int rc;
4524         bool remote = false;
4525
4526         ENTRY;
4527
4528         oh = container_of0(handle, struct osd_thandle, ot_super);
4529         LASSERT(oh->ot_handle != NULL);
4530
4531         rc = linkea_init_with_rec(&ldata);
4532         if (!rc) {
4533                 linkea_first_entry(&ldata);
4534                 while (ldata.ld_lee != NULL && !remote) {
4535                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4536                                             &tmpname, tfid);
4537                         if (osd_remote_fid(env, osd, tfid) > 0)
4538                                 remote = true;
4539                         else
4540                                 linkea_next_entry(&ldata);
4541                 }
4542         } else if (rc == -ENODATA) {
4543                 rc = 0;
4544         } else {
4545                 RETURN(rc);
4546         }
4547
4548         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4549                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4550                 if (rc)
4551                         CERROR("%s: failed to remove agent entry for "DFID
4552                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4553         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4554                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4555                 if (rc)
4556                         CERROR("%s: failed to create agent entry for "DFID
4557                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4558         }
4559
4560         RETURN(rc);
4561 }
4562
4563 /*
4564  * Concurrency: @dt is write locked.
4565  */
4566 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4567                          const struct lu_buf *buf, const char *name, int fl,
4568                          struct thandle *handle)
4569 {
4570         struct osd_object *obj = osd_dt_obj(dt);
4571         struct osd_device *osd = osd_obj2dev(obj);
4572         struct inode *inode = obj->oo_inode;
4573         struct osd_thread_info *info = osd_oti_get(env);
4574         int fs_flags = 0;
4575         int len;
4576         int rc;
4577
4578         ENTRY;
4579
4580         LASSERT(handle);
4581         LASSERT(buf);
4582
4583         /* version set is not real XATTR */
4584         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4585                 dt_obj_version_t *version = buf->lb_buf;
4586
4587                 /*
4588                  * for version we are just using xattr API but change inode
4589                  * field instead
4590                  */
4591                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4592
4593                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4594                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4595
4596                 LDISKFS_I(inode)->i_fs_version = *version;
4597                 /*
4598                  * Version is set after all inode operations are finished,
4599                  * so we should mark it dirty here
4600                  */
4601                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4602
4603                 RETURN(0);
4604         }
4605
4606         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4607                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4608
4609         len = buf->lb_len;
4610         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4611
4612         /*
4613          * For the OST device with 256 bytes inode size by default,
4614          * the PFID EA will be stored together with LMA EA to avoid
4615          * performance trouble. Otherwise the PFID EA can be stored
4616          * independently. LU-8998
4617          */
4618         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4619             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4620                 LASSERT(buf->lb_buf);
4621
4622                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4623                 if (fl <= 0)
4624                         RETURN(fl);
4625         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4626                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4627                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4628
4629                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4630                 if (rc)
4631                         RETURN(rc);
4632
4633                 lma->lma_incompat |= LMAI_STRIPED;
4634                 lustre_lma_swab(lma);
4635                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4636                                      sizeof(*lma), XATTR_REPLACE);
4637                 if (rc != 0)
4638                         RETURN(rc);
4639         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4640                 LASSERT(!osd->od_is_ost);
4641
4642                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4643                 if (rc)
4644                         RETURN(rc);
4645         }
4646
4647         if (fl & LU_XATTR_REPLACE)
4648                 fs_flags |= XATTR_REPLACE;
4649
4650         if (fl & LU_XATTR_CREATE)
4651                 fs_flags |= XATTR_CREATE;
4652
4653         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4654         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4655
4656         if (rc == 0 &&
4657             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4658              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4659                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4660
4661         return rc;
4662 }
4663
4664 /*
4665  * Concurrency: @dt is read locked.
4666  */
4667 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4668                           const struct lu_buf *buf)
4669 {
4670         struct osd_object *obj = osd_dt_obj(dt);
4671         struct inode *inode = obj->oo_inode;
4672         struct osd_thread_info *info = osd_oti_get(env);
4673         struct dentry *dentry = &info->oti_obj_dentry;
4674
4675         if (!dt_object_exists(dt))
4676                 return -ENOENT;
4677
4678         LASSERT(!dt_object_remote(dt));
4679         LASSERT(inode->i_op != NULL);
4680         LASSERT(inode->i_op->listxattr != NULL);
4681
4682         dentry->d_inode = inode;
4683         dentry->d_sb = inode->i_sb;
4684         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4685 }
4686
4687 static int osd_declare_xattr_del(const struct lu_env *env,
4688                                  struct dt_object *dt, const char *name,
4689                                  struct thandle *handle)
4690 {
4691         struct osd_thandle *oh;
4692         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4693
4694         LASSERT(!dt_object_remote(dt));
4695         LASSERT(handle != NULL);
4696
4697         oh = container_of0(handle, struct osd_thandle, ot_super);
4698         LASSERT(oh->ot_handle == NULL);
4699
4700         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4701                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4702         /*
4703          * xattr del may involve inode quota change, reserve credits for
4704          * dquot_initialize()
4705          */
4706         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4707
4708         return 0;
4709 }
4710
4711 /*
4712  * Concurrency: @dt is write locked.
4713  */
4714 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4715                          const char *name, struct thandle *handle)
4716 {
4717         struct osd_object *obj = osd_dt_obj(dt);
4718         struct inode *inode = obj->oo_inode;
4719         struct osd_thread_info *info = osd_oti_get(env);
4720         struct dentry *dentry = &info->oti_obj_dentry;
4721         int rc;
4722
4723         if (!dt_object_exists(dt))
4724                 return -ENOENT;
4725
4726         LASSERT(!dt_object_remote(dt));
4727         LASSERT(inode->i_op != NULL);
4728         LASSERT(handle != NULL);
4729 #ifdef HAVE_IOP_XATTR
4730         LASSERT(inode->i_op->removexattr != NULL);
4731 #endif
4732
4733         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4734
4735         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4736                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4737
4738                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4739                                  &info->oti_ost_attrs);
4740                 if (!rc) {
4741                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4742
4743                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4744                         lustre_lma_swab(lma);
4745                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4746                                              sizeof(*lma), XATTR_REPLACE);
4747                         if (!rc)
4748                                 obj->oo_pfid_in_lma = 0;
4749                 }
4750         } else {
4751                 dquot_initialize(inode);
4752                 dentry->d_inode = inode;
4753                 dentry->d_sb = inode->i_sb;
4754                 rc = osd_removexattr(dentry, inode, name);
4755         }
4756
4757         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4758
4759         if (rc == 0 &&
4760             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4761              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4762                 osd_oxc_del(obj, name);
4763
4764         return rc;
4765 }
4766
4767 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4768                            __u64 start, __u64 end)
4769 {
4770         struct osd_object *obj = osd_dt_obj(dt);
4771         struct inode *inode = obj->oo_inode;
4772         struct osd_thread_info *info = osd_oti_get(env);
4773         struct dentry *dentry = &info->oti_obj_dentry;
4774         struct file *file = &info->oti_file;
4775         int rc;
4776
4777         ENTRY;
4778
4779         dentry->d_inode = inode;
4780         dentry->d_sb = inode->i_sb;
4781         file->f_path.dentry = dentry;
4782         file->f_mapping = inode->i_mapping;
4783         file->f_op = inode->i_fop;
4784         set_file_inode(file, inode);
4785
4786         rc = vfs_fsync_range(file, start, end, 0);
4787
4788         RETURN(rc);
4789 }
4790
4791 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4792 {
4793         return 0;
4794 }
4795
4796 /*
4797  * Index operations.
4798  */
4799
4800 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4801                                const struct dt_index_features *feat)
4802 {
4803         struct iam_descr *descr;
4804
4805         if (osd_object_is_root(o))
4806                 return feat == &dt_directory_features;
4807
4808         LASSERT(o->oo_dir != NULL);
4809
4810         descr = o->oo_dir->od_container.ic_descr;
4811         if (feat == &dt_directory_features) {
4812                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4813                         return 1;
4814                 else
4815                         return 0;
4816         } else {
4817                 return feat->dif_keysize_min <= descr->id_key_size &&
4818                        descr->id_key_size <= feat->dif_keysize_max &&
4819                        feat->dif_recsize_min <= descr->id_rec_size &&
4820                        descr->id_rec_size <= feat->dif_recsize_max &&
4821                        !(feat->dif_flags & (DT_IND_VARKEY |
4822                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4823                        ergo(feat->dif_flags & DT_IND_UPDATE,
4824                             1 /* XXX check that object (and fs) is writable */);
4825         }
4826 }
4827
4828 static int osd_iam_container_init(const struct lu_env *env,
4829                                   struct osd_object *obj,
4830                                   struct osd_directory *dir)
4831 {
4832         struct iam_container *bag = &dir->od_container;
4833         int result;
4834
4835         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4836         if (result != 0)
4837                 return result;
4838
4839         result = iam_container_setup(bag);
4840         if (result == 0)
4841                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4842         else
4843                 iam_container_fini(bag);
4844
4845         return result;
4846 }
4847
4848
4849 /*
4850  * Concurrency: no external locking is necessary.
4851  */
4852 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4853                          const struct dt_index_features *feat)
4854 {
4855         int result;
4856         int skip_iam = 0;
4857         struct osd_object *obj = osd_dt_obj(dt);
4858
4859         LINVRNT(osd_invariant(obj));
4860
4861         if (osd_object_is_root(obj)) {
4862                 dt->do_index_ops = &osd_index_ea_ops;
4863                 result = 0;
4864         } else if (feat == &dt_directory_features) {
4865                 dt->do_index_ops = &osd_index_ea_ops;
4866                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4867                         result = 0;
4868                 else
4869                         result = -ENOTDIR;
4870                 skip_iam = 1;
4871         } else if (unlikely(feat == &dt_otable_features)) {
4872                 dt->do_index_ops = &osd_otable_ops;
4873                 return 0;
4874         } else if (unlikely(feat == &dt_acct_features)) {
4875                 dt->do_index_ops = &osd_acct_index_ops;
4876                 result = 0;
4877                 skip_iam = 1;
4878         } else if (!osd_has_index(obj)) {
4879                 struct osd_directory *dir;
4880                 struct osd_device *osd = osd_obj2dev(obj);
4881                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4882
4883                 OBD_ALLOC_PTR(dir);
4884                 if (dir) {
4885
4886                         spin_lock(&obj->oo_guard);
4887                         if (obj->oo_dir == NULL)
4888                                 obj->oo_dir = dir;
4889                         else
4890                                 /*
4891                                  * Concurrent thread allocated container data.
4892                                  */
4893                                 OBD_FREE_PTR(dir);
4894                         spin_unlock(&obj->oo_guard);
4895                         /*
4896                          * Now, that we have container data, serialize its
4897                          * initialization.
4898                          */
4899                         down_write(&obj->oo_ext_idx_sem);
4900                         /*
4901                          * recheck under lock.
4902                          */
4903
4904                         if (osd_has_index(obj)) {
4905                                 result = 0;
4906                                 goto unlock;
4907                         }
4908
4909                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4910                         if (result || feat == &dt_lfsck_namespace_features ||
4911                             feat == &dt_lfsck_layout_orphan_features ||
4912                             feat == &dt_lfsck_layout_dangling_features)
4913                                 goto unlock;
4914
4915                         result = osd_index_register(osd, fid,
4916                                                     feat->dif_keysize_max,
4917                                                     feat->dif_recsize_max);
4918                         if (result < 0)
4919                                 CWARN("%s: failed to register index "
4920                                       DFID": rc = %d\n",
4921                                       osd_name(osd), PFID(fid), result);
4922                         else if (result > 0)
4923                                 result = 0;
4924                         else
4925                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4926                                        " (%d/%d) registered\n",
4927                                        osd_name(osd), PFID(fid),
4928                                        (int)feat->dif_keysize_max,
4929                                        (int)feat->dif_recsize_max);
4930
4931 unlock:
4932                         up_write(&obj->oo_ext_idx_sem);
4933                 } else {
4934                         result = -ENOMEM;
4935                 }
4936         } else {
4937                 result = 0;
4938         }
4939
4940         if (result == 0 && skip_iam == 0) {
4941                 if (!osd_iam_index_probe(env, obj, feat))
4942                         result = -ENOTDIR;
4943         }
4944         LINVRNT(osd_invariant(obj));
4945
4946         return result;
4947 }
4948
4949 static int osd_otable_it_attr_get(const struct lu_env *env,
4950                                  struct dt_object *dt,
4951                                  struct lu_attr *attr)
4952 {
4953         attr->la_valid = 0;
4954         return 0;
4955 }
4956
4957 static const struct dt_object_operations osd_obj_ops = {
4958         .do_read_lock           = osd_read_lock,
4959         .do_write_lock          = osd_write_lock,
4960         .do_read_unlock         = osd_read_unlock,
4961         .do_write_unlock        = osd_write_unlock,
4962         .do_write_locked        = osd_write_locked,
4963         .do_attr_get            = osd_attr_get,
4964         .do_declare_attr_set    = osd_declare_attr_set,
4965         .do_attr_set            = osd_attr_set,
4966         .do_ah_init             = osd_ah_init,
4967         .do_declare_create      = osd_declare_create,
4968         .do_create              = osd_create,
4969         .do_declare_destroy     = osd_declare_destroy,
4970         .do_destroy             = osd_destroy,
4971         .do_index_try           = osd_index_try,
4972         .do_declare_ref_add     = osd_declare_ref_add,
4973         .do_ref_add             = osd_ref_add,
4974         .do_declare_ref_del     = osd_declare_ref_del,
4975         .do_ref_del             = osd_ref_del,
4976         .do_xattr_get           = osd_xattr_get,
4977         .do_declare_xattr_set   = osd_declare_xattr_set,
4978         .do_xattr_set           = osd_xattr_set,
4979         .do_declare_xattr_del   = osd_declare_xattr_del,
4980         .do_xattr_del           = osd_xattr_del,
4981         .do_xattr_list          = osd_xattr_list,
4982         .do_object_sync         = osd_object_sync,
4983         .do_invalidate          = osd_invalidate,
4984 };
4985
4986 static const struct dt_object_operations osd_obj_otable_it_ops = {
4987         .do_attr_get    = osd_otable_it_attr_get,
4988         .do_index_try   = osd_index_try,
4989 };
4990
4991 static int osd_index_declare_iam_delete(const struct lu_env *env,
4992                                         struct dt_object *dt,
4993                                         const struct dt_key *key,
4994                                         struct thandle *handle)
4995 {
4996         struct osd_thandle *oh;
4997
4998         oh = container_of0(handle, struct osd_thandle, ot_super);
4999         LASSERT(oh->ot_handle == NULL);
5000
5001         /* Recycle  may cause additional three blocks to be changed. */
5002         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5003                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5004
5005         return 0;
5006 }
5007
5008 /**
5009  *      delete a (key, value) pair from index \a dt specified by \a key
5010  *
5011  *      \param  dt      osd index object
5012  *      \param  key     key for index
5013  *      \param  rec     record reference
5014  *      \param  handle  transaction handler
5015  *
5016  *      \retval  0  success
5017  *      \retval -ve   failure
5018  */
5019 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5020                                 const struct dt_key *key,
5021                                 struct thandle *handle)
5022 {
5023         struct osd_thread_info *oti = osd_oti_get(env);
5024         struct osd_object *obj = osd_dt_obj(dt);
5025         struct osd_thandle *oh;
5026         struct iam_path_descr *ipd;
5027         struct iam_container *bag = &obj->oo_dir->od_container;
5028         int rc;
5029
5030         ENTRY;
5031
5032         if (!dt_object_exists(dt))
5033                 RETURN(-ENOENT);
5034
5035         LINVRNT(osd_invariant(obj));
5036         LASSERT(!dt_object_remote(dt));
5037         LASSERT(bag->ic_object == obj->oo_inode);
5038         LASSERT(handle != NULL);
5039
5040         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5041
5042         ipd = osd_idx_ipd_get(env, bag);
5043         if (unlikely(ipd == NULL))
5044                 RETURN(-ENOMEM);
5045
5046         oh = container_of0(handle, struct osd_thandle, ot_super);
5047         LASSERT(oh->ot_handle != NULL);
5048         LASSERT(oh->ot_handle->h_transaction != NULL);
5049
5050         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5051                 /* swab quota uid/gid provided by caller */
5052                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5053                 key = (const struct dt_key *)&oti->oti_quota_id;
5054         }
5055
5056         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5057         osd_ipd_put(env, bag, ipd);
5058         LINVRNT(osd_invariant(obj));
5059         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5060         RETURN(rc);
5061 }
5062
5063 static int osd_index_declare_ea_delete(const struct lu_env *env,
5064                                        struct dt_object *dt,
5065                                        const struct dt_key *key,
5066                                        struct thandle *handle)
5067 {
5068         struct osd_thandle *oh;
5069         struct inode *inode;
5070         int rc, credits;
5071
5072         ENTRY;
5073
5074         LASSERT(!dt_object_remote(dt));
5075         LASSERT(handle != NULL);
5076
5077         oh = container_of0(handle, struct osd_thandle, ot_super);
5078         LASSERT(oh->ot_handle == NULL);
5079
5080         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5081         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5082
5083         inode = osd_dt_obj(dt)->oo_inode;
5084         if (inode == NULL)
5085                 RETURN(-ENOENT);
5086
5087         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5088                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5089                                    NULL, OSD_QID_BLK);
5090         RETURN(rc);
5091 }
5092
5093 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5094                                           struct dt_rec *fid)
5095 {
5096         struct osd_fid_pack *rec;
5097         int rc = -ENODATA;
5098
5099         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5100                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5101                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5102                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5103                         rc = -EINVAL;
5104         }
5105         return rc;
5106 }
5107
5108 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5109                           const struct lu_fid *fid)
5110 {
5111         struct seq_server_site *ss = osd_seq_site(osd);
5112
5113         ENTRY;
5114
5115         /* FID seqs not in FLDB, must be local seq */
5116         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5117                 RETURN(0);
5118
5119         /*
5120          * If FLD is not being initialized yet, it only happens during the
5121          * initialization, likely during mgs initialization, and we assume
5122          * this is local FID.
5123          */
5124         if (ss == NULL || ss->ss_server_fld == NULL)
5125                 RETURN(0);
5126
5127         /* Only check the local FLDB here */
5128         if (osd_seq_exists(env, osd, fid_seq(fid)))
5129                 RETURN(0);
5130
5131         RETURN(1);
5132 }
5133
5134 static void osd_take_care_of_agent(const struct lu_env *env,
5135                                    struct osd_device *osd,
5136                                    struct osd_thandle *oh,
5137                                    struct ldiskfs_dir_entry_2 *de)
5138 {
5139         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5140         struct osd_idmap_cache *idc;
5141         int rc, schedule = 0;
5142
5143         LASSERT(de != NULL);
5144
5145         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5146         if (likely(rc == 0)) {
5147                 idc = osd_idc_find_or_init(env, osd, fid);
5148                 if (IS_ERR(idc) || idc->oic_remote)
5149                         schedule = 1;
5150         } else if (rc == -ENODATA) {
5151                 /*
5152                  * can't get FID, postpone to the end of the
5153                  * transaction when iget() is safe
5154                  */
5155                 schedule = 1;
5156         } else {
5157                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5158         }
5159         if (schedule)
5160                 osd_schedule_agent_inode_removal(env, oh,
5161                                                  le32_to_cpu(de->inode));
5162 }
5163
5164 /**
5165  * Index delete function for interoperability mode (b11826).
5166  * It will remove the directory entry added by osd_index_ea_insert().
5167  * This entry is needed to maintain name->fid mapping.
5168  *
5169  * \param key,  key i.e. file entry to be deleted
5170  *
5171  * \retval   0, on success
5172  * \retval -ve, on error
5173  */
5174 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5175                                const struct dt_key *key, struct thandle *handle)
5176 {
5177         struct osd_object *obj = osd_dt_obj(dt);
5178         struct inode *dir = obj->oo_inode;
5179         struct dentry *dentry;
5180         struct osd_thandle *oh;
5181         struct ldiskfs_dir_entry_2 *de = NULL;
5182         struct buffer_head *bh;
5183         struct htree_lock *hlock = NULL;
5184         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5185         int rc;
5186
5187         ENTRY;
5188
5189         if (!dt_object_exists(dt))
5190                 RETURN(-ENOENT);
5191
5192         LINVRNT(osd_invariant(obj));
5193         LASSERT(!dt_object_remote(dt));
5194         LASSERT(handle != NULL);
5195
5196         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5197
5198         oh = container_of(handle, struct osd_thandle, ot_super);
5199         LASSERT(oh->ot_handle != NULL);
5200         LASSERT(oh->ot_handle->h_transaction != NULL);
5201
5202         dquot_initialize(dir);
5203         dentry = osd_child_dentry_get(env, obj,
5204                                       (char *)key, strlen((char *)key));
5205
5206         if (obj->oo_hl_head != NULL) {
5207                 hlock = osd_oti_get(env)->oti_hlock;
5208                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5209                                    dir, LDISKFS_HLOCK_DEL);
5210         } else {
5211                 down_write(&obj->oo_ext_idx_sem);
5212         }
5213
5214         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5215         if (!IS_ERR(bh)) {
5216                 /*
5217                  * If this is not the ".." entry, it might be a remote DNE
5218                  * entry and  we need to check if the FID is for a remote
5219                  * MDT.  If the FID is  not in the directory entry (e.g.
5220                  * upgraded 1.8 filesystem without dirdata enabled) then
5221                  * we need to get the FID from the LMA. For a remote directory
5222                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5223                  * case.
5224                  *
5225                  * Delete the entry before the agent inode in order to
5226                  * simplify error handling.  At worst an error after deleting
5227                  * the entry first might leak the agent inode afterward. The
5228                  * reverse would need filesystem abort in case of error deleting
5229                  * the entry after the agent had been removed, or leave a
5230                  * dangling entry pointing at a random inode.
5231                  */
5232                 if (strcmp((char *)key, dotdot) != 0)
5233                         osd_take_care_of_agent(env, osd, oh, de);
5234                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5235                 brelse(bh);
5236         } else {
5237                 rc = PTR_ERR(bh);
5238         }
5239         if (hlock != NULL)
5240                 ldiskfs_htree_unlock(hlock);
5241         else
5242                 up_write(&obj->oo_ext_idx_sem);
5243
5244         GOTO(out, rc);
5245 out:
5246         LASSERT(osd_invariant(obj));
5247         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5248         RETURN(rc);
5249 }
5250
5251 /**
5252  *      Lookup index for \a key and copy record to \a rec.
5253  *
5254  *      \param  dt      osd index object
5255  *      \param  key     key for index
5256  *      \param  rec     record reference
5257  *
5258  *      \retval  +ve  success : exact mach
5259  *      \retval  0    return record with key not greater than \a key
5260  *      \retval -ve   failure
5261  */
5262 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5263                                 struct dt_rec *rec, const struct dt_key *key)
5264 {
5265         struct osd_object *obj = osd_dt_obj(dt);
5266         struct iam_path_descr *ipd;
5267         struct iam_container *bag = &obj->oo_dir->od_container;
5268         struct osd_thread_info *oti = osd_oti_get(env);
5269         struct iam_iterator *it = &oti->oti_idx_it;
5270         struct iam_rec *iam_rec;
5271         int rc;
5272
5273         ENTRY;
5274
5275         if (!dt_object_exists(dt))
5276                 RETURN(-ENOENT);
5277
5278         LASSERT(osd_invariant(obj));
5279         LASSERT(!dt_object_remote(dt));
5280         LASSERT(bag->ic_object == obj->oo_inode);
5281
5282         ipd = osd_idx_ipd_get(env, bag);
5283         if (IS_ERR(ipd))
5284                 RETURN(-ENOMEM);
5285
5286         /* got ipd now we can start iterator. */
5287         iam_it_init(it, bag, 0, ipd);
5288
5289         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5290                 /* swab quota uid/gid provided by caller */
5291                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5292                 key = (const struct dt_key *)&oti->oti_quota_id;
5293         }
5294
5295         rc = iam_it_get(it, (struct iam_key *)key);
5296         if (rc >= 0) {
5297                 if (S_ISDIR(obj->oo_inode->i_mode))
5298                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5299                 else
5300                         iam_rec = (struct iam_rec *)rec;
5301
5302                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5303
5304                 if (S_ISDIR(obj->oo_inode->i_mode))
5305                         osd_fid_unpack((struct lu_fid *)rec,
5306                                        (struct osd_fid_pack *)iam_rec);
5307                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5308                         osd_quota_unpack(obj, rec);
5309         }
5310
5311         iam_it_put(it);
5312         iam_it_fini(it);
5313         osd_ipd_put(env, bag, ipd);
5314
5315         LINVRNT(osd_invariant(obj));
5316
5317         RETURN(rc);
5318 }
5319
5320 static int osd_index_declare_iam_insert(const struct lu_env *env,
5321                                         struct dt_object *dt,
5322                                         const struct dt_rec *rec,
5323                                         const struct dt_key *key,
5324                                         struct thandle *handle)
5325 {
5326         struct osd_thandle *oh;
5327
5328         LASSERT(handle != NULL);
5329
5330         oh = container_of0(handle, struct osd_thandle, ot_super);
5331         LASSERT(oh->ot_handle == NULL);
5332
5333         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5334                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5335
5336         return 0;
5337 }
5338
5339 /**
5340  *      Inserts (key, value) pair in \a dt index object.
5341  *
5342  *      \param  dt      osd index object
5343  *      \param  key     key for index
5344  *      \param  rec     record reference
5345  *      \param  th      transaction handler
5346  *
5347  *      \retval  0  success
5348  *      \retval -ve failure
5349  */
5350 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5351                                 const struct dt_rec *rec,
5352                                 const struct dt_key *key, struct thandle *th)
5353 {
5354         struct osd_object *obj = osd_dt_obj(dt);
5355         struct iam_path_descr *ipd;
5356         struct osd_thandle *oh;
5357         struct iam_container *bag;
5358         struct osd_thread_info *oti = osd_oti_get(env);
5359         struct iam_rec *iam_rec;
5360         int rc;
5361
5362         ENTRY;
5363
5364         if (!dt_object_exists(dt))
5365                 RETURN(-ENOENT);
5366
5367         LINVRNT(osd_invariant(obj));
5368         LASSERT(!dt_object_remote(dt));
5369
5370         bag = &obj->oo_dir->od_container;
5371         LASSERT(bag->ic_object == obj->oo_inode);
5372         LASSERT(th != NULL);
5373
5374         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5375
5376         ipd = osd_idx_ipd_get(env, bag);
5377         if (unlikely(ipd == NULL))
5378                 RETURN(-ENOMEM);
5379
5380         oh = container_of0(th, struct osd_thandle, ot_super);
5381         LASSERT(oh->ot_handle != NULL);
5382         LASSERT(oh->ot_handle->h_transaction != NULL);
5383         if (S_ISDIR(obj->oo_inode->i_mode)) {
5384                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5385                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5386                              &oti->oti_fid);
5387         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5388                 /* pack quota uid/gid */
5389                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5390                 key = (const struct dt_key *)&oti->oti_quota_id;
5391                 /* pack quota record */
5392                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5393                 iam_rec = (struct iam_rec *)rec;
5394         } else {
5395                 iam_rec = (struct iam_rec *)rec;
5396         }
5397
5398         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5399                         iam_rec, ipd);
5400         osd_ipd_put(env, bag, ipd);
5401         LINVRNT(osd_invariant(obj));
5402         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5403         RETURN(rc);
5404 }
5405
5406 /**
5407  * Calls ldiskfs_add_entry() to add directory entry
5408  * into the directory. This is required for
5409  * interoperability mode (b11826)
5410  *
5411  * \retval   0, on success
5412  * \retval -ve, on error
5413  */
5414 static int __osd_ea_add_rec(struct osd_thread_info *info,
5415                             struct osd_object *pobj, struct inode  *cinode,
5416                             const char *name, const struct lu_fid *fid,
5417                             struct htree_lock *hlock, struct thandle *th)
5418 {
5419         struct ldiskfs_dentry_param *ldp;
5420         struct dentry *child;
5421         struct osd_thandle *oth;
5422         int rc;
5423
5424         oth = container_of(th, struct osd_thandle, ot_super);
5425         LASSERT(oth->ot_handle != NULL);
5426         LASSERT(oth->ot_handle->h_transaction != NULL);
5427         LASSERT(pobj->oo_inode);
5428
5429         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5430         if (unlikely(pobj->oo_inode ==
5431                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5432                 ldp->edp_magic = 0;
5433         else
5434                 osd_get_ldiskfs_dirent_param(ldp, fid);
5435         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5436         child->d_fsdata = (void *)ldp;
5437         dquot_initialize(pobj->oo_inode);
5438         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5439                                    child, cinode, hlock);
5440         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5441                 struct ldiskfs_dir_entry_2      *de;
5442                 struct buffer_head              *bh;
5443                 int                              rc1;
5444
5445                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5446                                             NULL, hlock);
5447                 if (!IS_ERR(bh)) {
5448                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5449                                                                bh);
5450                         if (rc1 == 0) {
5451                                 if (S_ISDIR(cinode->i_mode))
5452                                         de->file_type = LDISKFS_DIRENT_LUFID |
5453                                                         LDISKFS_FT_REG_FILE;
5454                                 else
5455                                         de->file_type = LDISKFS_DIRENT_LUFID |
5456                                                         LDISKFS_FT_DIR;
5457                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5458                                                               NULL, bh);
5459                         }
5460                         brelse(bh);
5461                 }
5462         }
5463
5464         RETURN(rc);
5465 }
5466
5467 /**
5468  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5469  * into the directory.Also sets flags into osd object to
5470  * indicate dot and dotdot are created. This is required for
5471  * interoperability mode (b11826)
5472  *
5473  * \param dir   directory for dot and dotdot fixup.
5474  * \param obj   child object for linking
5475  *
5476  * \retval   0, on success
5477  * \retval -ve, on error
5478  */
5479 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5480                               struct osd_object *dir,
5481                               struct inode *parent_dir, const char *name,
5482                               const struct lu_fid *dot_fid,
5483                               const struct lu_fid *dot_dot_fid,
5484                               struct thandle *th)
5485 {
5486         struct inode *inode = dir->oo_inode;
5487         struct osd_thandle *oth;
5488         int result = 0;
5489
5490         oth = container_of(th, struct osd_thandle, ot_super);
5491         LASSERT(oth->ot_handle->h_transaction != NULL);
5492         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5493
5494         if (strcmp(name, dot) == 0) {
5495                 if (dir->oo_compat_dot_created) {
5496                         result = -EEXIST;
5497                 } else {
5498                         LASSERT(inode->i_ino == parent_dir->i_ino);
5499                         dir->oo_compat_dot_created = 1;
5500                         result = 0;
5501                 }
5502         } else if (strcmp(name, dotdot) == 0) {
5503                 if (!dir->oo_compat_dot_created)
5504                         return -EINVAL;
5505                 /* in case of rename, dotdot is already created */
5506                 if (dir->oo_compat_dotdot_created) {
5507                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5508                                                 dot_dot_fid, NULL, th);
5509                 }
5510
5511                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5512                         struct lu_fid tfid = *dot_dot_fid;
5513
5514                         tfid.f_oid--;
5515                         result = osd_add_dot_dotdot_internal(info,
5516                                         dir->oo_inode, parent_dir, dot_fid,
5517                                         &tfid, oth);
5518                 } else {
5519                         result = osd_add_dot_dotdot_internal(info,
5520                                         dir->oo_inode, parent_dir, dot_fid,
5521                                         dot_dot_fid, oth);
5522                 }
5523
5524                 if (result == 0)
5525                         dir->oo_compat_dotdot_created = 1;
5526         }
5527
5528         return result;
5529 }
5530
5531
5532 /**
5533  * It will call the appropriate osd_add* function and return the
5534  * value, return by respective functions.
5535  */
5536 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5537                           struct inode *cinode, const char *name,
5538                           const struct lu_fid *fid, struct thandle *th)
5539 {
5540         struct osd_thread_info *info = osd_oti_get(env);
5541         struct htree_lock *hlock;
5542         int rc;
5543
5544         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5545
5546         if (name[0] == '.' && (name[1] == '\0' ||
5547                                (name[1] == '.' && name[2] == '\0'))) {
5548                 if (hlock != NULL) {
5549                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5550                                            pobj->oo_inode, 0);
5551                 } else {
5552                         down_write(&pobj->oo_ext_idx_sem);
5553                 }
5554
5555                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5556                                         lu_object_fid(&pobj->oo_dt.do_lu),
5557                                         fid, th);
5558         } else {
5559                 if (hlock != NULL) {
5560                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5561                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5562                 } else {
5563                         down_write(&pobj->oo_ext_idx_sem);
5564                 }
5565
5566                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5567                         struct lu_fid *tfid = &info->oti_fid;
5568
5569                         *tfid = *fid;
5570                         tfid->f_ver = ~0;
5571                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5572                                               tfid, hlock, th);
5573                 } else {
5574                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5575                                               hlock, th);
5576                 }
5577         }
5578         if (hlock != NULL)
5579                 ldiskfs_htree_unlock(hlock);
5580         else
5581                 up_write(&pobj->oo_ext_idx_sem);
5582
5583         return rc;
5584 }
5585
5586 static int
5587 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5588                       struct osd_idmap_cache *oic)
5589 {
5590         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5591         struct lu_fid *fid = &oic->oic_fid;
5592         struct osd_inode_id *id = &oic->oic_lid;
5593         struct inode *inode = NULL;
5594         int once = 0;
5595         bool insert;
5596         int rc;
5597
5598         ENTRY;
5599
5600         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5601                 RETURN(0);
5602
5603         if (thread_is_running(&scrub->os_thread) &&
5604             scrub->os_pos_current > id->oii_ino)
5605                 RETURN(0);
5606
5607         if (dev->od_auto_scrub_interval == AS_NEVER ||
5608             ktime_get_real_seconds() <
5609             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5610                 RETURN(0);
5611
5612 again:
5613         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5614         if (rc == -ENOENT) {
5615                 __u32 gen = id->oii_gen;
5616
5617                 insert = true;
5618                 if (inode != NULL)
5619                         goto trigger;
5620
5621                 inode = osd_iget(oti, dev, id);
5622                 /* The inode has been removed (by race maybe). */
5623                 if (IS_ERR(inode)) {
5624                         rc = PTR_ERR(inode);
5625
5626                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5627                 }
5628
5629                 /* The OI mapping is lost. */
5630                 if (gen != OSD_OII_NOGEN)
5631                         goto trigger;
5632
5633                 /*
5634                  * The inode may has been reused by others, we do not know,
5635                  * leave it to be handled by subsequent osd_fid_lookup().
5636                  */
5637                 GOTO(out, rc = 0);
5638         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5639                 GOTO(out, rc);
5640         }
5641
5642         insert = false;
5643
5644 trigger:
5645         if (thread_is_running(&scrub->os_thread)) {
5646                 if (inode == NULL) {
5647                         inode = osd_iget(oti, dev, id);
5648                         /* The inode has been removed (by race maybe). */
5649                         if (IS_ERR(inode)) {
5650                                 rc = PTR_ERR(inode);
5651
5652                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5653                         }
5654                 }
5655
5656                 rc = osd_oii_insert(dev, oic, insert);
5657                 /*
5658                  * There is race condition between osd_oi_lookup and OI scrub.
5659                  * The OI scrub finished just after osd_oi_lookup() failure.
5660                  * Under such case, it is unnecessary to trigger OI scrub again,
5661                  * but try to call osd_oi_lookup() again.
5662                  */
5663                 if (unlikely(rc == -EAGAIN))
5664                         goto again;
5665
5666                 if (!S_ISDIR(inode->i_mode))
5667                         rc = 0;
5668                 else
5669                         rc = osd_check_lmv(oti, dev, inode, oic);
5670
5671                 GOTO(out, rc);
5672         }
5673
5674         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5675                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5676                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5677                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5678                        "%s: trigger partial OI scrub for RPC inconsistency "
5679                        "checking FID "DFID": rc = %d\n",
5680                        osd_dev2name(dev), PFID(fid), rc);
5681                 if (rc == 0 || rc == -EALREADY)
5682                         goto again;
5683         }
5684
5685         GOTO(out, rc);
5686
5687 out:
5688         if (inode)
5689                 iput(inode);
5690
5691         RETURN(rc);
5692 }
5693
5694 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5695                                struct osd_device *dev,
5696                                struct osd_idmap_cache *oic,
5697                                struct lu_fid *fid, __u32 ino)
5698 {
5699         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5700         struct inode *inode;
5701         int rc;
5702
5703         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5704         inode = osd_iget(oti, dev, &oic->oic_lid);
5705         if (IS_ERR(inode)) {
5706                 fid_zero(&oic->oic_fid);
5707                 return PTR_ERR(inode);
5708         }
5709
5710         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5711         iput(inode);
5712         if (rc != 0)
5713                 fid_zero(&oic->oic_fid);
5714         else
5715                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5716         return rc;
5717 }
5718
5719 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5720                       struct osd_inode_id *id, const struct lu_fid *fid)
5721 {
5722         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5723                id->oii_ino, id->oii_gen, info);
5724         info->oti_cache.oic_lid = *id;
5725         info->oti_cache.oic_fid = *fid;
5726         info->oti_cache.oic_dev = osd;
5727 }
5728
5729 /**
5730  * Get parent FID from the linkEA.
5731  *
5732  * For a directory which parent resides on remote MDT, to satisfy the
5733  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5734  * the other hand, to make the lookup(..) on the directory can return
5735  * the real parent FID, we append the real parent FID after its ".."
5736  * name entry in the /REMOTE_PARENT_DIR.
5737  *
5738  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5739  * backup. So after the restore, we cannot get the right parent FID from
5740  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5741  * we have stored the real parent FID in the directory object's linkEA,
5742  * we can parse the linkEA for the real parent FID.
5743  *
5744  * \param[in] env       pointer to the thread context
5745  * \param[in] obj       pointer to the object to be handled
5746  * \param[out]fid       pointer to the buffer to hold the parent FID
5747  *
5748  * \retval              0 for getting the real parent FID successfully
5749  * \retval              negative error number on failure
5750  */
5751 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5752                                     struct osd_object *obj,
5753                                     struct lu_fid *fid)
5754 {
5755         struct osd_thread_info *oti = osd_oti_get(env);
5756         struct lu_buf *buf = &oti->oti_big_buf;
5757         struct dentry *dentry = &oti->oti_obj_dentry;
5758         struct inode *inode = obj->oo_inode;
5759         struct linkea_data ldata = { NULL };
5760         int rc;
5761
5762         ENTRY;
5763
5764         fid_zero(fid);
5765         if (!S_ISDIR(inode->i_mode))
5766                 RETURN(-EIO);
5767
5768 again:
5769         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5770                              buf->lb_buf, buf->lb_len);
5771         if (rc == -ERANGE) {
5772                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5773                                      NULL, 0);
5774                 if (rc > 0) {
5775                         lu_buf_realloc(buf, rc);
5776                         if (buf->lb_buf == NULL)
5777                                 RETURN(-ENOMEM);
5778
5779                         goto again;
5780                 }
5781         }
5782
5783         if (unlikely(rc == 0))
5784                 RETURN(-ENODATA);
5785
5786         if (rc < 0)
5787                 RETURN(rc);
5788
5789         if (unlikely(buf->lb_buf == NULL)) {
5790                 lu_buf_realloc(buf, rc);
5791                 if (buf->lb_buf == NULL)
5792                         RETURN(-ENOMEM);
5793
5794                 goto again;
5795         }
5796
5797         ldata.ld_buf = buf;
5798         rc = linkea_init_with_rec(&ldata);
5799         if (!rc) {
5800                 linkea_first_entry(&ldata);
5801                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5802         }
5803
5804         RETURN(rc);
5805 }
5806
5807 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5808                                     struct inode *inode,
5809                                     const struct lu_fid *pfid,
5810                                     const char *name, const int namelen)
5811 {
5812         struct osd_thread_info *oti = osd_oti_get(env);
5813         struct lu_buf *buf = &oti->oti_big_buf;
5814         struct dentry *dentry = &oti->oti_obj_dentry;
5815         struct linkea_data ldata = { NULL };
5816         struct lu_name cname = { .ln_name = name,
5817                                  .ln_namelen = namelen };
5818         int rc;
5819
5820         ENTRY;
5821
5822 again:
5823         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5824                              buf->lb_buf, buf->lb_len);
5825         if (rc == -ERANGE)
5826                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5827
5828         if (rc < 0)
5829                 RETURN(rc);
5830
5831         if (unlikely(rc == 0))
5832                 RETURN(-ENODATA);
5833
5834         if (buf->lb_len < rc) {
5835                 lu_buf_realloc(buf, rc);
5836                 if (buf->lb_buf == NULL)
5837                         RETURN(-ENOMEM);
5838
5839                 goto again;
5840         }
5841
5842         ldata.ld_buf = buf;
5843         rc = linkea_init_with_rec(&ldata);
5844         if (!rc)
5845                 rc = linkea_links_find(&ldata, &cname, pfid);
5846
5847         RETURN(rc);
5848 }
5849
5850 /**
5851  * Calls ->lookup() to find dentry. From dentry get inode and
5852  * read inode's ea to get fid. This is required for  interoperability
5853  * mode (b11826)
5854  *
5855  * \retval   0, on success
5856  * \retval -ve, on error
5857  */
5858 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5859                              struct dt_rec *rec, const struct dt_key *key)
5860 {
5861         struct inode *dir = obj->oo_inode;
5862         struct dentry *dentry;
5863         struct ldiskfs_dir_entry_2 *de;
5864         struct buffer_head *bh;
5865         struct lu_fid *fid = (struct lu_fid *)rec;
5866         struct htree_lock *hlock = NULL;
5867         int ino;
5868         int rc;
5869
5870         ENTRY;
5871
5872         LASSERT(dir->i_op != NULL);
5873         LASSERT(dir->i_op->lookup != NULL);
5874
5875         dentry = osd_child_dentry_get(env, obj,
5876                                       (char *)key, strlen((char *)key));
5877
5878         if (obj->oo_hl_head != NULL) {
5879                 hlock = osd_oti_get(env)->oti_hlock;
5880                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5881                                    dir, LDISKFS_HLOCK_LOOKUP);
5882         } else {
5883                 down_read(&obj->oo_ext_idx_sem);
5884         }
5885
5886         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5887         if (!IS_ERR(bh)) {
5888                 struct osd_thread_info *oti = osd_oti_get(env);
5889                 struct osd_inode_id *id = &oti->oti_id;
5890                 struct osd_idmap_cache *oic = &oti->oti_cache;
5891                 struct osd_device *dev = osd_obj2dev(obj);
5892
5893                 ino = le32_to_cpu(de->inode);
5894                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5895                         brelse(bh);
5896                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5897                         GOTO(out, rc);
5898                 }
5899
5900                 rc = osd_get_fid_from_dentry(de, rec);
5901
5902                 /* done with de, release bh */
5903                 brelse(bh);
5904                 if (rc != 0) {
5905                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5906                                 const char *name = (const char *)key;
5907
5908                                 /*
5909                                  * If the parent is on remote MDT, and there
5910                                  * is no FID-in-dirent, then we have to get
5911                                  * the parent FID from the linkEA.
5912                                  */
5913                                 if (likely(strlen(name) == 2 &&
5914                                            name[0] == '.' && name[1] == '.'))
5915                                         rc = osd_get_pfid_from_linkea(env, obj,
5916                                                                       fid);
5917                         } else {
5918                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5919                         }
5920                 } else {
5921                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5922                 }
5923
5924                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5925                         fid_zero(&oic->oic_fid);
5926
5927                         GOTO(out, rc);
5928                 }
5929
5930                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5931                 rc = osd_consistency_check(oti, dev, oic);
5932                 if (rc == -ENOENT)
5933                         fid_zero(&oic->oic_fid);
5934                 else
5935                         /* Other error should not affect lookup result. */
5936                         rc = 0;
5937         } else {
5938                 rc = PTR_ERR(bh);
5939         }
5940
5941         GOTO(out, rc);
5942
5943 out:
5944         if (hlock != NULL)
5945                 ldiskfs_htree_unlock(hlock);
5946         else
5947                 up_read(&obj->oo_ext_idx_sem);
5948         return rc;
5949 }
5950
5951 static int osd_index_declare_ea_insert(const struct lu_env *env,
5952                                        struct dt_object *dt,
5953                                        const struct dt_rec *rec,
5954                                        const struct dt_key *key,
5955                                        struct thandle *handle)
5956 {
5957         struct osd_thandle *oh;
5958         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5959         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5960         const struct lu_fid *fid = rec1->rec_fid;
5961         int credits, rc = 0;
5962         struct osd_idmap_cache *idc;
5963
5964         ENTRY;
5965
5966         LASSERT(!dt_object_remote(dt));
5967         LASSERT(handle != NULL);
5968         LASSERT(fid != NULL);
5969         LASSERT(rec1->rec_type != 0);
5970
5971         oh = container_of0(handle, struct osd_thandle, ot_super);
5972         LASSERT(oh->ot_handle == NULL);
5973
5974         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5975
5976         /*
5977          * we can't call iget() while a transactions is running
5978          * (this can lead to a deadlock), but we need to know
5979          * inum and object type. so we find this information at
5980          * declaration and cache in per-thread info
5981          */
5982         idc = osd_idc_find_or_init(env, osd, fid);
5983         if (IS_ERR(idc))
5984                 RETURN(PTR_ERR(idc));
5985         if (idc->oic_remote) {
5986                 /*
5987                  * a reference to remote inode is represented by an
5988                  * agent inode which we have to create
5989                  */
5990                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5991                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5992         }
5993
5994         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5995
5996         if (osd_dt_obj(dt)->oo_inode != NULL) {
5997                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5998
5999                 /*
6000                  * We ignore block quota on meta pool (MDTs), so needn't
6001                  * calculate how many blocks will be consumed by this index
6002                  * insert
6003                  */
6004                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6005                                            i_gid_read(inode),
6006                                            i_projid_read(inode), 0,
6007                                            oh, osd_dt_obj(dt), NULL,
6008                                            OSD_QID_BLK);
6009                 if (rc)
6010                         RETURN(rc);
6011
6012 #ifdef HAVE_PROJECT_QUOTA
6013                 /*
6014                  * Reserve credits for local agent inode to transfer
6015                  * to 0, quota enforcement is ignored in this case.
6016                  */
6017                 if (idc->oic_remote &&
6018                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6019                     i_projid_read(inode) != 0)
6020                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6021                                                   0, i_projid_read(inode),
6022                                                   0, false, PRJQUOTA, true);
6023 #endif
6024         }
6025
6026         RETURN(rc);
6027 }
6028
6029 /**
6030  * Index add function for interoperability mode (b11826).
6031  * It will add the directory entry.This entry is needed to
6032  * maintain name->fid mapping.
6033  *
6034  * \param key it is key i.e. file entry to be inserted
6035  * \param rec it is value of given key i.e. fid
6036  *
6037  * \retval   0, on success
6038  * \retval -ve, on error
6039  */
6040 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6041                                const struct dt_rec *rec,
6042                                const struct dt_key *key, struct thandle *th)
6043 {
6044         struct osd_object *obj = osd_dt_obj(dt);
6045         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6046         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6047         const struct lu_fid *fid = rec1->rec_fid;
6048         const char *name = (const char *)key;
6049         struct osd_thread_info *oti = osd_oti_get(env);
6050         struct inode *child_inode = NULL;
6051         struct osd_idmap_cache *idc;
6052         int rc;
6053
6054         ENTRY;
6055
6056         if (!dt_object_exists(dt))
6057                 RETURN(-ENOENT);
6058
6059         LASSERT(osd_invariant(obj));
6060         LASSERT(!dt_object_remote(dt));
6061         LASSERT(th != NULL);
6062
6063         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6064
6065         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6066
6067         idc = osd_idc_find(env, osd, fid);
6068         if (unlikely(idc == NULL)) {
6069                 idc = osd_idc_find_or_init(env, osd, fid);
6070                 if (IS_ERR(idc)) {
6071                         /*
6072                          * this dt_insert() wasn't declared properly, so
6073                          * FID is missing in OI cache. we better do not
6074                          * lookup FID in FLDB/OI and don't risk to deadlock,
6075                          * but in some special cases (lfsck testing, etc)
6076                          * it's much simpler than fixing a caller.
6077                          *
6078                          * normally this error should be placed after the first
6079                          * find, but migrate may attach source stripes to
6080                          * target, which doesn't create stripes.
6081                          */
6082                         CERROR("%s: "DFID" wasn't declared for insert\n",
6083                                osd_name(osd), PFID(fid));
6084                         dump_stack();
6085                         RETURN(PTR_ERR(idc));
6086                 }
6087         }
6088
6089         if (idc->oic_remote) {
6090                 /* Insert remote entry */
6091                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6092                         child_inode =
6093                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6094                 } else {
6095                         child_inode = osd_create_local_agent_inode(env, osd,
6096                                         obj, fid, rec1->rec_type & S_IFMT, th);
6097                         if (IS_ERR(child_inode))
6098                                 RETURN(PTR_ERR(child_inode));
6099                 }
6100         } else {
6101                 /* Insert local entry */
6102                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6103                         /* for a reason OI cache wasn't filled properly */
6104                         CERROR("%s: OIC for "DFID" isn't filled\n",
6105                                osd_name(osd), PFID(fid));
6106                         RETURN(-EINVAL);
6107                 }
6108                 child_inode = oti->oti_inode;
6109                 if (unlikely(child_inode == NULL)) {
6110                         struct ldiskfs_inode_info *lii;
6111
6112                         OBD_ALLOC_PTR(lii);
6113                         if (lii == NULL)
6114                                 RETURN(-ENOMEM);
6115                         child_inode = oti->oti_inode = &lii->vfs_inode;
6116                 }
6117                 child_inode->i_sb = osd_sb(osd);
6118                 child_inode->i_ino = idc->oic_lid.oii_ino;
6119                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6120         }
6121
6122         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6123
6124         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6125                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6126
6127         if (child_inode && child_inode != oti->oti_inode)
6128                 iput(child_inode);
6129         LASSERT(osd_invariant(obj));
6130         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6131         RETURN(rc);
6132 }
6133
6134 /**
6135  *  Initialize osd Iterator for given osd index object.
6136  *
6137  *  \param  dt      osd index object
6138  */
6139
6140 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6141                                      struct dt_object *dt,
6142                                      __u32 unused)
6143 {
6144         struct osd_it_iam *it;
6145         struct osd_object *obj = osd_dt_obj(dt);
6146         struct lu_object *lo = &dt->do_lu;
6147         struct iam_path_descr *ipd;
6148         struct iam_container *bag = &obj->oo_dir->od_container;
6149
6150         if (!dt_object_exists(dt))
6151                 return ERR_PTR(-ENOENT);
6152
6153         OBD_ALLOC_PTR(it);
6154         if (it == NULL)
6155                 return ERR_PTR(-ENOMEM);
6156
6157         ipd = osd_it_ipd_get(env, bag);
6158         if (likely(ipd != NULL)) {
6159                 it->oi_obj = obj;
6160                 it->oi_ipd = ipd;
6161                 lu_object_get(lo);
6162                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6163                 return (struct dt_it *)it;
6164         } else {
6165                 OBD_FREE_PTR(it);
6166                 return ERR_PTR(-ENOMEM);
6167         }
6168 }
6169
6170 /**
6171  * free given Iterator.
6172  */
6173 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6174 {
6175         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6176         struct osd_object *obj = it->oi_obj;
6177
6178         iam_it_fini(&it->oi_it);
6179         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6180         osd_object_put(env, obj);
6181         OBD_FREE_PTR(it);
6182 }
6183
6184 /**
6185  *  Move Iterator to record specified by \a key
6186  *
6187  *  \param  di      osd iterator
6188  *  \param  key     key for index
6189  *
6190  *  \retval +ve  di points to record with least key not larger than key
6191  *  \retval  0   di points to exact matched key
6192  *  \retval -ve  failure
6193  */
6194
6195 static int osd_it_iam_get(const struct lu_env *env,
6196                           struct dt_it *di, const struct dt_key *key)
6197 {
6198         struct osd_thread_info *oti = osd_oti_get(env);
6199         struct osd_it_iam *it = (struct osd_it_iam *)di;
6200
6201         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6202                 /* swab quota uid/gid */
6203                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6204                 key = (struct dt_key *)&oti->oti_quota_id;
6205         }
6206
6207         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6208 }
6209
6210 /**
6211  *  Release Iterator
6212  *
6213  *  \param  di      osd iterator
6214  */
6215 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6216 {
6217         struct osd_it_iam *it = (struct osd_it_iam *)di;
6218
6219         iam_it_put(&it->oi_it);
6220 }
6221
6222 /**
6223  *  Move iterator by one record
6224  *
6225  *  \param  di      osd iterator
6226  *
6227  *  \retval +1   end of container reached
6228  *  \retval  0   success
6229  *  \retval -ve  failure
6230  */
6231
6232 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6233 {
6234         struct osd_it_iam *it = (struct osd_it_iam *)di;
6235
6236         return iam_it_next(&it->oi_it);
6237 }
6238
6239 /**
6240  * Return pointer to the key under iterator.
6241  */
6242
6243 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6244                                      const struct dt_it *di)
6245 {
6246         struct osd_thread_info *oti = osd_oti_get(env);
6247         struct osd_it_iam *it = (struct osd_it_iam *)di;
6248         struct osd_object *obj = it->oi_obj;
6249         struct dt_key *key;
6250
6251         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6252
6253         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6254                 /* swab quota uid/gid */
6255                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6256                 key = (struct dt_key *)&oti->oti_quota_id;
6257         }
6258
6259         return key;
6260 }
6261
6262 /**
6263  * Return size of key under iterator (in bytes)
6264  */
6265
6266 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6267 {
6268         struct osd_it_iam *it = (struct osd_it_iam *)di;
6269
6270         return iam_it_key_size(&it->oi_it);
6271 }
6272
6273 static inline void
6274 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6275 {
6276         /* check if file type is required */
6277         if (ent->lde_attrs & LUDA_TYPE) {
6278                 struct luda_type *lt;
6279                 int align = sizeof(*lt) - 1;
6280
6281                 len = (len + align) & ~align;
6282                 lt = (struct luda_type *)(ent->lde_name + len);
6283                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6284         }
6285
6286         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6287 }
6288
6289 /**
6290  * build lu direct from backend fs dirent.
6291  */
6292
6293 static inline void
6294 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6295                    char *name, __u16 namelen, __u16 type, __u32 attr)
6296 {
6297         ent->lde_attrs = attr | LUDA_FID;
6298         fid_cpu_to_le(&ent->lde_fid, fid);
6299
6300         ent->lde_hash = cpu_to_le64(offset);
6301         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6302
6303         strncpy(ent->lde_name, name, namelen);
6304         ent->lde_name[namelen] = '\0';
6305         ent->lde_namelen = cpu_to_le16(namelen);
6306
6307         /* append lustre attributes */
6308         osd_it_append_attrs(ent, namelen, type);
6309 }
6310
6311 /**
6312  * Return pointer to the record under iterator.
6313  */
6314 static int osd_it_iam_rec(const struct lu_env *env,
6315                           const struct dt_it *di,
6316                           struct dt_rec *dtrec, __u32 attr)
6317 {
6318         struct osd_it_iam *it = (struct osd_it_iam *)di;
6319         struct osd_thread_info *info = osd_oti_get(env);
6320
6321         ENTRY;
6322
6323         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6324                 const struct osd_fid_pack *rec;
6325                 struct lu_fid *fid = &info->oti_fid;
6326                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6327                 char *name;
6328                 int namelen;
6329                 __u64 hash;
6330                 int rc;
6331
6332                 name = (char *)iam_it_key_get(&it->oi_it);
6333                 if (IS_ERR(name))
6334                         RETURN(PTR_ERR(name));
6335
6336                 namelen = iam_it_key_size(&it->oi_it);
6337
6338                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6339                 if (IS_ERR(rec))
6340                         RETURN(PTR_ERR(rec));
6341
6342                 rc = osd_fid_unpack(fid, rec);
6343                 if (rc)
6344                         RETURN(rc);
6345
6346                 hash = iam_it_store(&it->oi_it);
6347
6348                 /* IAM does not store object type in IAM index (dir) */
6349                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6350                                    0, LUDA_FID);
6351         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6352                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6353                            (struct iam_rec *)dtrec);
6354                 osd_quota_unpack(it->oi_obj, dtrec);
6355         } else {
6356                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6357                            (struct iam_rec *)dtrec);
6358         }
6359
6360         RETURN(0);
6361 }
6362
6363 /**
6364  * Returns cookie for current Iterator position.
6365  */
6366 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6367 {
6368         struct osd_it_iam *it = (struct osd_it_iam *)di;
6369
6370         return iam_it_store(&it->oi_it);
6371 }
6372
6373 /**
6374  * Restore iterator from cookie.
6375  *
6376  * \param  di      osd iterator
6377  * \param  hash    Iterator location cookie
6378  *
6379  * \retval +ve  di points to record with least key not larger than key.
6380  * \retval  0   di points to exact matched key
6381  * \retval -ve  failure
6382  */
6383
6384 static int osd_it_iam_load(const struct lu_env *env,
6385                            const struct dt_it *di, __u64 hash)
6386 {
6387         struct osd_it_iam *it = (struct osd_it_iam *)di;
6388
6389         return iam_it_load(&it->oi_it, hash);
6390 }
6391
6392 static const struct dt_index_operations osd_index_iam_ops = {
6393         .dio_lookup         = osd_index_iam_lookup,
6394         .dio_declare_insert = osd_index_declare_iam_insert,
6395         .dio_insert         = osd_index_iam_insert,
6396         .dio_declare_delete = osd_index_declare_iam_delete,
6397         .dio_delete         = osd_index_iam_delete,
6398         .dio_it     = {
6399                 .init     = osd_it_iam_init,
6400                 .fini     = osd_it_iam_fini,
6401                 .get      = osd_it_iam_get,
6402                 .put      = osd_it_iam_put,
6403                 .next     = osd_it_iam_next,
6404                 .key      = osd_it_iam_key,
6405                 .key_size = osd_it_iam_key_size,
6406                 .rec      = osd_it_iam_rec,
6407                 .store    = osd_it_iam_store,
6408                 .load     = osd_it_iam_load
6409         }
6410 };
6411
6412
6413 /**
6414  * Creates or initializes iterator context.
6415  *
6416  * \retval struct osd_it_ea, iterator structure on success
6417  *
6418  */
6419 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6420                                     struct dt_object *dt,
6421                                     __u32 attr)
6422 {
6423         struct osd_object *obj = osd_dt_obj(dt);
6424         struct osd_thread_info *info = osd_oti_get(env);
6425         struct osd_it_ea *oie;
6426         struct file *file;
6427         struct lu_object *lo = &dt->do_lu;
6428         struct dentry *obj_dentry;
6429
6430         ENTRY;
6431
6432         if (!dt_object_exists(dt) || obj->oo_destroyed)
6433                 RETURN(ERR_PTR(-ENOENT));
6434
6435         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6436         if (oie == NULL)
6437                 RETURN(ERR_PTR(-ENOMEM));
6438         obj_dentry = &oie->oie_dentry;
6439
6440         obj_dentry->d_inode = obj->oo_inode;
6441         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6442         obj_dentry->d_name.hash = 0;
6443
6444         oie->oie_rd_dirent       = 0;
6445         oie->oie_it_dirent       = 0;
6446         oie->oie_dirent          = NULL;
6447         if (unlikely(!info->oti_it_ea_buf_used)) {
6448                 oie->oie_buf = info->oti_it_ea_buf;
6449                 info->oti_it_ea_buf_used = 1;
6450         } else {
6451                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6452                 if (oie->oie_buf == NULL)
6453                         RETURN(ERR_PTR(-ENOMEM));
6454         }
6455         oie->oie_obj = obj;
6456
6457         file = &oie->oie_file;
6458
6459         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6460         if (attr & LUDA_64BITHASH)
6461                 file->f_mode    = FMODE_64BITHASH;
6462         else
6463                 file->f_mode    = FMODE_32BITHASH;
6464         file->f_path.dentry     = obj_dentry;
6465         file->f_mapping         = obj->oo_inode->i_mapping;
6466         file->f_op              = obj->oo_inode->i_fop;
6467         set_file_inode(file, obj->oo_inode);
6468
6469         lu_object_get(lo);
6470         RETURN((struct dt_it *)oie);
6471 }
6472
6473 /**
6474  * Destroy or finishes iterator context.
6475  *
6476  * \param di iterator structure to be destroyed
6477  */
6478 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6479 {
6480         struct osd_thread_info *info = osd_oti_get(env);
6481         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6482         struct osd_object *obj = oie->oie_obj;
6483         struct inode *inode = obj->oo_inode;
6484
6485         ENTRY;
6486         oie->oie_file.f_op->release(inode, &oie->oie_file);
6487         osd_object_put(env, obj);
6488         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6489                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6490         else
6491                 info->oti_it_ea_buf_used = 0;
6492         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6493         EXIT;
6494 }
6495
6496 /**
6497  * It position the iterator at given key, so that next lookup continues from
6498  * that key Or it is similar to dio_it->load() but based on a key,
6499  * rather than file position.
6500  *
6501  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6502  * to the beginning.
6503  *
6504  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6505  */
6506 static int osd_it_ea_get(const struct lu_env *env,
6507                          struct dt_it *di, const struct dt_key *key)
6508 {
6509         struct osd_it_ea *it = (struct osd_it_ea *)di;
6510
6511         ENTRY;
6512         LASSERT(((const char *)key)[0] == '\0');
6513         it->oie_file.f_pos = 0;
6514         it->oie_rd_dirent = 0;
6515         it->oie_it_dirent = 0;
6516         it->oie_dirent = NULL;
6517
6518         RETURN(+1);
6519 }
6520
6521 /**
6522  * Does nothing
6523  */
6524 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6525 {
6526 }
6527
6528 struct osd_filldir_cbs {
6529         struct dir_context ctx;
6530         struct osd_it_ea  *it;
6531 };
6532 /**
6533  * It is called internally by ->iterate*(). It fills the
6534  * iterator's in-memory data structure with required
6535  * information i.e. name, namelen, rec_size etc.
6536  *
6537  * \param buf in which information to be filled in.
6538  * \param name name of the file in given dir
6539  *
6540  * \retval 0 on success
6541  * \retval 1 on buffer full
6542  */
6543 #ifdef HAVE_FILLDIR_USE_CTX
6544 static int osd_ldiskfs_filldir(struct dir_context *buf,
6545 #else
6546 static int osd_ldiskfs_filldir(void *buf,
6547 #endif
6548                                const char *name, int namelen,
6549                                loff_t offset, __u64 ino, unsigned int d_type)
6550 {
6551         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6552         struct osd_object *obj = it->oie_obj;
6553         struct osd_it_ea_dirent *ent = it->oie_dirent;
6554         struct lu_fid *fid = &ent->oied_fid;
6555         struct osd_fid_pack *rec;
6556
6557         ENTRY;
6558
6559 /* this should never happen */
6560         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6561                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6562                 RETURN(-EIO);
6563         }
6564
6565         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6566             OSD_IT_EA_BUFSIZE)
6567                 RETURN(1);
6568
6569         /* "." is just the object itself. */
6570         if (namelen == 1 && name[0] == '.') {
6571                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6572         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6573                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6574                 if (osd_fid_unpack(fid, rec) != 0)
6575                         fid_zero(fid);
6576         } else {
6577                 fid_zero(fid);
6578         }
6579         d_type &= ~LDISKFS_DIRENT_LUFID;
6580
6581         /* NOT export local root. */
6582         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6583                 ino = obj->oo_inode->i_ino;
6584                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6585         }
6586
6587         ent->oied_ino     = ino;
6588         ent->oied_off     = offset;
6589         ent->oied_namelen = namelen;
6590         ent->oied_type    = d_type;
6591
6592         memcpy(ent->oied_name, name, namelen);
6593
6594         it->oie_rd_dirent++;
6595         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6596         RETURN(0);
6597 }
6598
6599 /**
6600  * Calls ->iterate*() to load a directory entry at a time
6601  * and stored it in iterator's in-memory data structure.
6602  *
6603  * \param di iterator's in memory structure
6604  *
6605  * \retval   0 on success
6606  * \retval -ve on error
6607  * \retval +1 reach the end of entry
6608  */
6609 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6610                                const struct dt_it *di)
6611 {
6612         struct osd_it_ea *it = (struct osd_it_ea *)di;
6613         struct osd_object *obj = it->oie_obj;
6614         struct inode *inode = obj->oo_inode;
6615         struct htree_lock *hlock = NULL;
6616         struct file *filp = &it->oie_file;
6617         int rc = 0;
6618         struct osd_filldir_cbs buf = {
6619                 .ctx.actor = osd_ldiskfs_filldir,
6620                 .it = it
6621         };
6622
6623         ENTRY;
6624         it->oie_dirent = it->oie_buf;
6625         it->oie_rd_dirent = 0;
6626
6627         if (obj->oo_hl_head != NULL) {
6628                 hlock = osd_oti_get(env)->oti_hlock;
6629                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6630                                    inode, LDISKFS_HLOCK_READDIR);
6631         } else {
6632                 down_read(&obj->oo_ext_idx_sem);
6633         }
6634
6635         filp->f_cred = current_cred();
6636         rc = osd_security_file_alloc(filp);
6637         if (rc)
6638                 RETURN(rc);
6639
6640         filp->f_flags |= O_NOATIME;
6641         filp->f_mode |= FMODE_NONOTIFY;
6642         rc = iterate_dir(filp, &buf.ctx);
6643         if (rc)
6644                 RETURN(rc);
6645
6646         if (hlock != NULL)
6647                 ldiskfs_htree_unlock(hlock);
6648         else
6649                 up_read(&obj->oo_ext_idx_sem);
6650
6651         if (it->oie_rd_dirent == 0) {
6652                 /*
6653                  * If it does not get any dirent, it means it has been reached
6654                  * to the end of the dir
6655                  */
6656                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6657                 if (rc == 0)
6658                         rc = 1;
6659         } else {
6660                 it->oie_dirent = it->oie_buf;
6661                 it->oie_it_dirent = 1;
6662         }
6663
6664         RETURN(rc);
6665 }
6666
6667 /**
6668  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6669  * to load a directory entry at a time and stored it in
6670  * iterator's in-memory data structure.
6671  *
6672  * \param di iterator's in memory structure
6673  *
6674  * \retval +ve iterator reached to end
6675  * \retval   0 iterator not reached to end
6676  * \retval -ve on error
6677  */
6678 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6679 {
6680         struct osd_it_ea *it = (struct osd_it_ea *)di;
6681         int rc;
6682
6683         ENTRY;
6684
6685         if (it->oie_it_dirent < it->oie_rd_dirent) {
6686                 it->oie_dirent =
6687                         (void *)it->oie_dirent +
6688                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6689                                        it->oie_dirent->oied_namelen);
6690                 it->oie_it_dirent++;
6691                 rc = 0;
6692         } else {
6693                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6694                         rc = 1;
6695                 else
6696                         rc = osd_ldiskfs_it_fill(env, di);
6697         }
6698
6699         RETURN(rc);
6700 }
6701
6702 /**
6703  * Returns the key at current position from iterator's in memory structure.
6704  *
6705  * \param di iterator's in memory structure
6706  *
6707  * \retval key i.e. struct dt_key on success
6708  */
6709 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6710                                     const struct dt_it *di)
6711 {
6712         struct osd_it_ea *it = (struct osd_it_ea *)di;
6713
6714         return (struct dt_key *)it->oie_dirent->oied_name;
6715 }
6716
6717 /**
6718  * Returns key's size at current position from iterator's in memory structure.
6719  *
6720  * \param di iterator's in memory structure
6721  *
6722  * \retval key_size i.e. struct dt_key on success
6723  */
6724 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6725 {
6726         struct osd_it_ea *it = (struct osd_it_ea *)di;
6727
6728         return it->oie_dirent->oied_namelen;
6729 }
6730
6731 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6732 {
6733         if (LDISKFS_DIR_REC_LEN(de) >=
6734             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6735                 return true;
6736
6737         return false;
6738 }
6739
6740 static inline bool
6741 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6742                      unsigned int blocksize, bool dotdot)
6743 {
6744         if (dotdot)
6745                 return osd_dotdot_has_space(de);
6746
6747         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6748             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6749                 return true;
6750
6751         return false;
6752 }
6753
6754 static int
6755 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6756                     handle_t *jh, struct dentry *dentry,
6757                     const struct lu_fid *fid, struct buffer_head *bh,
6758                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6759                     bool dotdot)
6760 {
6761         struct inode *dir = dentry->d_parent->d_inode;
6762         struct inode *inode = dentry->d_inode;
6763         struct osd_fid_pack *rec;
6764         struct ldiskfs_dentry_param *ldp;
6765         int namelen = dentry->d_name.len;
6766         int rc;
6767         struct osd_thread_info *info = osd_oti_get(env);
6768
6769         ENTRY;
6770
6771         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6772                 RETURN(0);
6773
6774         /* There is enough space to hold the FID-in-dirent. */
6775         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6776                 rc = ldiskfs_journal_get_write_access(jh, bh);
6777                 if (rc != 0)
6778                         RETURN(rc);
6779
6780                 de->name[namelen] = 0;
6781                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6782                 rec->fp_len = sizeof(struct lu_fid) + 1;
6783                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6784                 de->file_type |= LDISKFS_DIRENT_LUFID;
6785                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6786
6787                 RETURN(rc);
6788         }
6789
6790         LASSERT(!dotdot);
6791
6792         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6793         if (rc != 0)
6794                 RETURN(rc);
6795
6796         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6797         osd_get_ldiskfs_dirent_param(ldp, fid);
6798         dentry->d_fsdata = (void *)ldp;
6799         dquot_initialize(dir);
6800         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6801         /*
6802          * It is too bad, we cannot reinsert the name entry back.
6803          * That means we lose it!
6804          */
6805         if (rc != 0)
6806                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6807                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6808                        osd_ino2name(inode),
6809                        dir->i_ino, dir->i_generation, namelen,
6810                        dentry->d_name.name, PFID(fid), rc);
6811
6812         RETURN(rc);
6813 }
6814
6815 static int
6816 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6817                         struct osd_it_ea *it, struct lu_fid *fid,
6818                         struct osd_inode_id *id, __u32 *attr)
6819 {
6820         struct osd_thread_info *info = osd_oti_get(env);
6821         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6822         struct osd_device *dev = osd_obj2dev(obj);
6823         struct super_block *sb = osd_sb(dev);
6824         const char *devname = osd_name(dev);
6825         struct osd_it_ea_dirent *ent = it->oie_dirent;
6826         struct inode *dir = obj->oo_inode;
6827         struct htree_lock *hlock = NULL;
6828         struct buffer_head *bh = NULL;
6829         handle_t *jh = NULL;
6830         struct ldiskfs_dir_entry_2 *de;
6831         struct dentry *dentry;
6832         struct inode *inode;
6833         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6834         int credits;
6835         int rc;
6836         bool dotdot = false;
6837         bool dirty = false;
6838
6839         ENTRY;
6840
6841         if (ent->oied_name[0] == '.') {
6842                 if (ent->oied_namelen == 1)
6843                         RETURN(0);
6844
6845                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6846                         dotdot = true;
6847         }
6848
6849         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6850         inode = osd_iget(info, dev, id);
6851         if (IS_ERR(inode)) {
6852                 rc = PTR_ERR(inode);
6853                 if (rc == -ENOENT || rc == -ESTALE) {
6854                         /*
6855                          * Maybe dangling name entry, or
6856                          * corrupted directory entry.
6857                          */
6858                         *attr |= LUDA_UNKNOWN;
6859                         rc = 0;
6860                 } else {
6861                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6862                                "check_repair, dir = %lu/%u, name = %.*s, "
6863                                "ino = %llu, rc = %d\n",
6864                                devname, dir->i_ino, dir->i_generation,
6865                                ent->oied_namelen, ent->oied_name,
6866                                ent->oied_ino, rc);
6867                 }
6868
6869                 RETURN(rc);
6870         }
6871
6872         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6873                                            ent->oied_namelen);
6874         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6875         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6876                 lma = NULL;
6877         else if (rc != 0)
6878                 GOTO(out, rc);
6879
6880         /*
6881          * We need to ensure that the name entry is still valid.
6882          * Because it may be removed or renamed by other already.
6883          *
6884          * The unlink or rename operation will start journal before PDO lock,
6885          * so to avoid deadlock, here we need to start journal handle before
6886          * related PDO lock also. But because we do not know whether there
6887          * will be something to be repaired before PDO lock, we just start
6888          * journal without conditions.
6889          *
6890          * We may need to remove the name entry firstly, then insert back.
6891          * One credit is for user quota file update.
6892          * One credit is for group quota file update.
6893          * Two credits are for dirty inode.
6894          */
6895         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6896                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6897
6898         if (dev->od_dirent_journal != 0) {
6899
6900 again:
6901                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6902                 if (IS_ERR(jh)) {
6903                         rc = PTR_ERR(jh);
6904                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6905                                "check_repair, dir = %lu/%u, credits = %d, "
6906                                "name = %.*s, ino = %llu: rc = %d\n",
6907                                devname, dir->i_ino, dir->i_generation, credits,
6908                                ent->oied_namelen, ent->oied_name,
6909                                ent->oied_ino, rc);
6910
6911                         GOTO(out_inode, rc);
6912                 }
6913
6914                 if (obj->oo_hl_head != NULL) {
6915                         hlock = osd_oti_get(env)->oti_hlock;
6916                         /*
6917                          * "0" means exclusive lock for the whole directory.
6918                          * We need to prevent others access such name entry
6919                          * during the delete + insert. Neither HLOCK_ADD nor
6920                          * HLOCK_DEL cannot guarantee the atomicity.
6921                          */
6922                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6923                 } else {
6924                         down_write(&obj->oo_ext_idx_sem);
6925                 }
6926         } else {
6927                 if (obj->oo_hl_head != NULL) {
6928                         hlock = osd_oti_get(env)->oti_hlock;
6929                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6930                                            LDISKFS_HLOCK_LOOKUP);
6931                 } else {
6932                         down_read(&obj->oo_ext_idx_sem);
6933                 }
6934         }
6935
6936         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6937         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6938                 *attr |= LUDA_IGNORE;
6939
6940                 GOTO(out, rc = 0);
6941         }
6942
6943         /*
6944          * For dotdot entry, if there is not enough space to hold the
6945          * FID-in-dirent, just keep them there. It only happens when the
6946          * device upgraded from 1.8 or restored from MDT file-level backup.
6947          * For the whole directory, only dotdot entry have no FID-in-dirent
6948          * and needs to get FID from LMA when readdir, it will not affect the
6949          * performance much.
6950          */
6951         if (dotdot && !osd_dotdot_has_space(de)) {
6952                 *attr |= LUDA_UNKNOWN;
6953
6954                 GOTO(out, rc = 0);
6955         }
6956
6957         if (lma != NULL) {
6958                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6959                         GOTO(out, rc = 0);
6960
6961                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6962                         struct lu_fid *tfid = &lma->lma_self_fid;
6963
6964                         if (likely(dotdot &&
6965                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6966                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6967                                 /*
6968                                  * It must be REMOTE_PARENT_DIR and as the
6969                                  * 'dotdot' entry of remote directory
6970                                  */
6971                                 *attr |= LUDA_IGNORE;
6972                         } else {
6973                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6974                                        "parent directory, but got %.*s under "
6975                                        "dir = %lu/%u with the FID "DFID"\n",
6976                                        devname, ent->oied_namelen,
6977                                        ent->oied_name, dir->i_ino,
6978                                        dir->i_generation, PFID(tfid));
6979
6980                                 *attr |= LUDA_UNKNOWN;
6981                         }
6982
6983                         GOTO(out, rc = 0);
6984                 }
6985         }
6986
6987         if (!fid_is_zero(fid)) {
6988                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6989                                               ent->oied_namelen);
6990                 if (rc == -ENOENT ||
6991                     (rc == -ENODATA &&
6992                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6993                         /*
6994                          * linkEA does not recognize the dirent entry,
6995                          * it may because the dirent entry corruption
6996                          * and points to other's inode.
6997                          */
6998                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6999                                "recognize the dirent, dir = %lu/%u, "
7000                                " name = %.*s, ino = %llu, "
7001                                DFID": rc = %d\n", devname, dir->i_ino,
7002                                dir->i_generation, ent->oied_namelen,
7003                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7004                         *attr |= LUDA_UNKNOWN;
7005
7006                         GOTO(out, rc = 0);
7007                 }
7008
7009                 if (rc && rc != -ENODATA) {
7010                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7011                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7012                                DFID": rc = %d\n", devname, dir->i_ino,
7013                                dir->i_generation, ent->oied_namelen,
7014                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7015                         *attr |= LUDA_UNKNOWN;
7016
7017                         GOTO(out, rc = 0);
7018                 }
7019         }
7020
7021         if (lma != NULL) {
7022                 /*
7023                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7024                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7025                  */
7026                 if (*attr & LUDA_VERIFY_DRYRUN) {
7027                         *fid = lma->lma_self_fid;
7028                         *attr |= LUDA_REPAIR;
7029
7030                         GOTO(out, rc = 0);
7031                 }
7032
7033                 if (jh == NULL) {
7034                         brelse(bh);
7035                         dev->od_dirent_journal = 1;
7036                         if (hlock != NULL) {
7037                                 ldiskfs_htree_unlock(hlock);
7038                                 hlock = NULL;
7039                         } else {
7040                                 up_read(&obj->oo_ext_idx_sem);
7041                         }
7042
7043                         goto again;
7044                 }
7045
7046                 *fid = lma->lma_self_fid;
7047                 dirty = true;
7048                 /* Update or append the FID-in-dirent. */
7049                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7050                                          bh, de, hlock, dotdot);
7051                 if (rc == 0)
7052                         *attr |= LUDA_REPAIR;
7053                 else
7054                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7055                                "the dirent, dir = %lu/%u, name = %.*s, "
7056                                "ino = %llu, "DFID": rc = %d\n",
7057                                devname, dir->i_ino, dir->i_generation,
7058                                ent->oied_namelen, ent->oied_name,
7059                                ent->oied_ino, PFID(fid), rc);
7060         } else {
7061                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7062                 if (*attr & LUDA_VERIFY_DRYRUN) {
7063                         if (fid_is_sane(fid)) {
7064                                 *attr |= LUDA_REPAIR;
7065                         } else if (dev->od_index == 0) {
7066                                 lu_igif_build(fid, inode->i_ino,
7067                                               inode->i_generation);
7068                                 *attr |= LUDA_UPGRADE;
7069                         }
7070
7071                         GOTO(out, rc = 0);
7072                 }
7073
7074                 if (jh == NULL) {
7075                         brelse(bh);
7076                         dev->od_dirent_journal = 1;
7077                         if (hlock != NULL) {
7078                                 ldiskfs_htree_unlock(hlock);
7079                                 hlock = NULL;
7080                         } else {
7081                                 up_read(&obj->oo_ext_idx_sem);
7082                         }
7083
7084                         goto again;
7085                 }
7086
7087                 dirty = true;
7088                 if (unlikely(fid_is_sane(fid))) {
7089                         /*
7090                          * FID-in-dirent exists, but FID-in-LMA is lost.
7091                          * Trust the FID-in-dirent, and add FID-in-LMA.
7092                          */
7093                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7094                         if (rc == 0)
7095                                 *attr |= LUDA_REPAIR;
7096                         else
7097                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7098                                        "update dirent, dir = %lu/%u, "
7099                                        "name = %.*s, ino = %llu, "
7100                                        DFID": rc = %d\n",
7101                                        devname, dir->i_ino, dir->i_generation,
7102                                        ent->oied_namelen, ent->oied_name,
7103                                        ent->oied_ino, PFID(fid), rc);
7104                 } else if (dev->od_index == 0) {
7105                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7106                         /*
7107                          * It is probably IGIF object. Only aappend the
7108                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7109                          */
7110                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7111                                                  bh, de, hlock, dotdot);
7112                         if (rc == 0)
7113                                 *attr |= LUDA_UPGRADE;
7114                         else
7115                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7116                                        "after the dirent, dir = %lu/%u, "
7117                                        "name = %.*s, ino = %llu, "
7118                                        DFID": rc = %d\n",
7119                                        devname, dir->i_ino, dir->i_generation,
7120                                        ent->oied_namelen, ent->oied_name,
7121                                        ent->oied_ino, PFID(fid), rc);
7122                 }
7123         }
7124
7125         GOTO(out, rc);
7126
7127 out:
7128         if (!IS_ERR(bh))
7129                 brelse(bh);
7130         if (hlock != NULL) {
7131                 ldiskfs_htree_unlock(hlock);
7132         } else {
7133                 if (dev->od_dirent_journal != 0)
7134                         up_write(&obj->oo_ext_idx_sem);
7135                 else
7136                         up_read(&obj->oo_ext_idx_sem);
7137         }
7138
7139         if (jh != NULL)
7140                 ldiskfs_journal_stop(jh);
7141
7142 out_inode:
7143         iput(inode);
7144         if (rc >= 0 && !dirty)
7145                 dev->od_dirent_journal = 0;
7146
7147         return rc;
7148 }
7149
7150 /**
7151  * Returns the value at current position from iterator's in memory structure.
7152  *
7153  * \param di struct osd_it_ea, iterator's in memory structure
7154  * \param attr attr requested for dirent.
7155  * \param lde lustre dirent
7156  *
7157  * \retval   0 no error and \param lde has correct lustre dirent.
7158  * \retval -ve on error
7159  */
7160 static inline int osd_it_ea_rec(const struct lu_env *env,
7161                                 const struct dt_it *di,
7162                                 struct dt_rec *dtrec, __u32 attr)
7163 {
7164         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7165         struct osd_object      *obj   = it->oie_obj;
7166         struct osd_device      *dev   = osd_obj2dev(obj);
7167         struct osd_thread_info *oti   = osd_oti_get(env);
7168         struct osd_inode_id    *id    = &oti->oti_id;
7169         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7170         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7171         __u32 ino = it->oie_dirent->oied_ino;
7172         int rc = 0;
7173
7174         ENTRY;
7175
7176         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7177
7178         if (attr & LUDA_VERIFY) {
7179                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7180                         attr |= LUDA_IGNORE;
7181                         /*
7182                          * If the parent is on remote MDT, and there
7183                          * is no FID-in-dirent, then we have to get
7184                          * the parent FID from the linkEA.
7185                          */
7186                         if (!fid_is_sane(fid) &&
7187                             it->oie_dirent->oied_namelen == 2 &&
7188                             it->oie_dirent->oied_name[0] == '.' &&
7189                             it->oie_dirent->oied_name[1] == '.')
7190                                 osd_get_pfid_from_linkea(env, obj, fid);
7191                 } else {
7192                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7193                                                      &attr);
7194                 }
7195
7196                 if (!fid_is_sane(fid))
7197                         attr |= LUDA_UNKNOWN;
7198         } else {
7199                 attr &= ~LU_DIRENT_ATTRS_MASK;
7200                 if (!fid_is_sane(fid)) {
7201                         bool is_dotdot = false;
7202
7203                         if (it->oie_dirent->oied_namelen == 2 &&
7204                             it->oie_dirent->oied_name[0] == '.' &&
7205                             it->oie_dirent->oied_name[1] == '.')
7206                                 is_dotdot = true;
7207                         /*
7208                          * If the parent is on remote MDT, and there
7209                          * is no FID-in-dirent, then we have to get
7210                          * the parent FID from the linkEA.
7211                          */
7212                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7213                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7214                         } else {
7215                                 if (is_dotdot == false &&
7216                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7217                                         RETURN(-ENOENT);
7218
7219                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7220                         }
7221                 } else {
7222                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7223                 }
7224         }
7225
7226         /* Pack the entry anyway, at least the offset is right. */
7227         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7228                            it->oie_dirent->oied_name,
7229                            it->oie_dirent->oied_namelen,
7230                            it->oie_dirent->oied_type, attr);
7231
7232         if (rc < 0)
7233                 RETURN(rc);
7234
7235         if (osd_remote_fid(env, dev, fid))
7236                 RETURN(0);
7237
7238         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7239                 osd_add_oi_cache(oti, dev, id, fid);
7240
7241         RETURN(rc > 0 ? 0 : rc);
7242 }
7243
7244 /**
7245  * Returns the record size size at current position.
7246  *
7247  * This function will return record(lu_dirent) size in bytes.
7248  *
7249  * \param[in] env       execution environment
7250  * \param[in] di        iterator's in memory structure
7251  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7252  *                      calculate the lu_dirent size.
7253  *
7254  * \retval      record size(in bytes & in memory) of the current lu_dirent
7255  *              entry.
7256  */
7257 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7258                               __u32 attr)
7259 {
7260         struct osd_it_ea *it = (struct osd_it_ea *)di;
7261
7262         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7263 }
7264
7265 /**
7266  * Returns a cookie for current position of the iterator head, so that
7267  * user can use this cookie to load/start the iterator next time.
7268  *
7269  * \param di iterator's in memory structure
7270  *
7271  * \retval cookie for current position, on success
7272  */
7273 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7274 {
7275         struct osd_it_ea *it = (struct osd_it_ea *)di;
7276
7277         return it->oie_dirent->oied_off;
7278 }
7279
7280 /**
7281  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7282  * to load a directory entry at a time and stored it i inn,
7283  * in iterator's in-memory data structure.
7284  *
7285  * \param di struct osd_it_ea, iterator's in memory structure
7286  *
7287  * \retval +ve on success
7288  * \retval -ve on error
7289  */
7290 static int osd_it_ea_load(const struct lu_env *env,
7291                           const struct dt_it *di, __u64 hash)
7292 {
7293         struct osd_it_ea *it = (struct osd_it_ea *)di;
7294         int rc;
7295
7296         ENTRY;
7297         it->oie_file.f_pos = hash;
7298
7299         rc =  osd_ldiskfs_it_fill(env, di);
7300         if (rc > 0)
7301                 rc = -ENODATA;
7302
7303         if (rc == 0)
7304                 rc = 1;
7305
7306         RETURN(rc);
7307 }
7308
7309 /**
7310  * Index lookup function for interoperability mode (b11826).
7311  *
7312  * \param key,  key i.e. file name to be searched
7313  *
7314  * \retval +ve, on success
7315  * \retval -ve, on error
7316  */
7317 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7318                                struct dt_rec *rec, const struct dt_key *key)
7319 {
7320         struct osd_object *obj = osd_dt_obj(dt);
7321         int rc = 0;
7322
7323         ENTRY;
7324
7325         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7326         LINVRNT(osd_invariant(obj));
7327
7328         rc = osd_ea_lookup_rec(env, obj, rec, key);
7329         if (rc == 0)
7330                 rc = 1;
7331         RETURN(rc);
7332 }
7333
7334 /**
7335  * Index and Iterator operations for interoperability
7336  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7337  */
7338 static const struct dt_index_operations osd_index_ea_ops = {
7339         .dio_lookup         = osd_index_ea_lookup,
7340         .dio_declare_insert = osd_index_declare_ea_insert,
7341         .dio_insert         = osd_index_ea_insert,
7342         .dio_declare_delete = osd_index_declare_ea_delete,
7343         .dio_delete         = osd_index_ea_delete,
7344         .dio_it     = {
7345                 .init     = osd_it_ea_init,
7346                 .fini     = osd_it_ea_fini,
7347                 .get      = osd_it_ea_get,
7348                 .put      = osd_it_ea_put,
7349                 .next     = osd_it_ea_next,
7350                 .key      = osd_it_ea_key,
7351                 .key_size = osd_it_ea_key_size,
7352                 .rec      = osd_it_ea_rec,
7353                 .rec_size = osd_it_ea_rec_size,
7354                 .store    = osd_it_ea_store,
7355                 .load     = osd_it_ea_load
7356         }
7357 };
7358
7359 static void *osd_key_init(const struct lu_context *ctx,
7360                           struct lu_context_key *key)
7361 {
7362         struct osd_thread_info *info;
7363
7364         OBD_ALLOC_PTR(info);
7365         if (info == NULL)
7366                 return ERR_PTR(-ENOMEM);
7367
7368         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7369         if (info->oti_it_ea_buf == NULL)
7370                 goto out_free_info;
7371
7372         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7373
7374         info->oti_hlock = ldiskfs_htree_lock_alloc();
7375         if (info->oti_hlock == NULL)
7376                 goto out_free_ea;
7377
7378         return info;
7379
7380 out_free_ea:
7381         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7382 out_free_info:
7383         OBD_FREE_PTR(info);
7384         return ERR_PTR(-ENOMEM);
7385 }
7386
7387 static void osd_key_fini(const struct lu_context *ctx,
7388                          struct lu_context_key *key, void *data)
7389 {
7390         struct osd_thread_info *info = data;
7391         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7392         struct osd_idmap_cache *idc = info->oti_ins_cache;
7393
7394         if (info->oti_dio_pages) {
7395                 int i;
7396                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7397                         if (info->oti_dio_pages[i])
7398                                 __free_page(info->oti_dio_pages[i]);
7399                 }
7400                 OBD_FREE(info->oti_dio_pages,
7401                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7402         }
7403
7404         if (info->oti_inode != NULL)
7405                 OBD_FREE_PTR(lli);
7406         if (info->oti_hlock != NULL)
7407                 ldiskfs_htree_lock_free(info->oti_hlock);
7408         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7409         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7410         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7411         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7412         lu_buf_free(&info->oti_big_buf);
7413         if (idc != NULL) {
7414                 LASSERT(info->oti_ins_cache_size > 0);
7415                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7416                 info->oti_ins_cache = NULL;
7417                 info->oti_ins_cache_size = 0;
7418         }
7419         OBD_FREE_PTR(info);
7420 }
7421
7422 static void osd_key_exit(const struct lu_context *ctx,
7423                          struct lu_context_key *key, void *data)
7424 {
7425         struct osd_thread_info *info = data;
7426
7427         LASSERT(info->oti_r_locks == 0);
7428         LASSERT(info->oti_w_locks == 0);
7429         LASSERT(info->oti_txns    == 0);
7430 }
7431
7432 /* type constructor/destructor: osd_type_init, osd_type_fini */
7433 LU_TYPE_INIT_FINI(osd, &osd_key);
7434
7435 struct lu_context_key osd_key = {
7436         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7437         .lct_init = osd_key_init,
7438         .lct_fini = osd_key_fini,
7439         .lct_exit = osd_key_exit
7440 };
7441
7442
7443 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7444                            const char *name, struct lu_device *next)
7445 {
7446         struct osd_device *osd = osd_dev(d);
7447
7448         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7449             sizeof(osd->od_svname))
7450                 return -E2BIG;
7451         return osd_procfs_init(osd, name);
7452 }
7453
7454 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7455 {
7456         struct seq_server_site *ss = osd_seq_site(osd);
7457         int rc;
7458
7459         ENTRY;
7460
7461         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7462                 RETURN(0);
7463
7464         if (unlikely(ss == NULL))
7465                 RETURN(-ENODEV);
7466
7467         OBD_ALLOC_PTR(osd->od_cl_seq);
7468         if (osd->od_cl_seq == NULL)
7469                 RETURN(-ENOMEM);
7470
7471         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7472                              osd->od_svname, ss->ss_server_seq);
7473         if (rc != 0) {
7474                 OBD_FREE_PTR(osd->od_cl_seq);
7475                 osd->od_cl_seq = NULL;
7476                 RETURN(rc);
7477         }
7478
7479         if (ss->ss_node_id == 0) {
7480                 /*
7481                  * If the OSD on the sequence controller(MDT0), then allocate
7482                  * sequence here, otherwise allocate sequence after connected
7483                  * to MDT0 (see mdt_register_lwp_callback()).
7484                  */
7485                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7486                                    &osd->od_cl_seq->lcs_space, env);
7487         }
7488
7489         RETURN(rc);
7490 }
7491
7492 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7493 {
7494         if (osd->od_cl_seq == NULL)
7495                 return;
7496
7497         seq_client_fini(osd->od_cl_seq);
7498         OBD_FREE_PTR(osd->od_cl_seq);
7499         osd->od_cl_seq = NULL;
7500 }
7501
7502 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7503 {
7504         ENTRY;
7505
7506         /* shutdown quota slave instance associated with the device */
7507         if (o->od_quota_slave_md != NULL) {
7508                 struct qsd_instance *qsd = o->od_quota_slave_md;
7509
7510                 o->od_quota_slave_md = NULL;
7511                 qsd_fini(env, qsd);
7512         }
7513
7514         if (o->od_quota_slave_dt != NULL) {
7515                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7516
7517                 o->od_quota_slave_dt = NULL;
7518                 qsd_fini(env, qsd);
7519         }
7520
7521         osd_fid_fini(env, o);
7522         osd_scrub_cleanup(env, o);
7523
7524         RETURN(0);
7525 }
7526
7527 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7528 {
7529         ENTRY;
7530
7531         if (o->od_mnt != NULL) {
7532                 shrink_dcache_sb(osd_sb(o));
7533                 osd_sync(env, &o->od_dt_dev);
7534
7535                 mntput(o->od_mnt);
7536                 o->od_mnt = NULL;
7537         }
7538
7539         EXIT;
7540 }
7541
7542 static int osd_mount(const struct lu_env *env,
7543                      struct osd_device *o, struct lustre_cfg *cfg)
7544 {
7545         const char *name = lustre_cfg_string(cfg, 0);
7546         const char *dev = lustre_cfg_string(cfg, 1);
7547         const char *opts;
7548         unsigned long page, s_flags, lmd_flags = 0;
7549         struct page *__page;
7550         struct file_system_type *type;
7551         char *options = NULL;
7552         char *str;
7553         struct osd_thread_info *info = osd_oti_get(env);
7554         struct lu_fid *fid = &info->oti_fid;
7555         struct inode *inode;
7556         int rc = 0, force_over_512tb = 0;
7557
7558         ENTRY;
7559
7560         if (o->od_mnt != NULL)
7561                 RETURN(0);
7562
7563         if (strlen(dev) >= sizeof(o->od_mntdev))
7564                 RETURN(-E2BIG);
7565         strcpy(o->od_mntdev, dev);
7566
7567         str = lustre_cfg_string(cfg, 2);
7568         s_flags = simple_strtoul(str, NULL, 0);
7569         str = strstr(str, ":");
7570         if (str)
7571                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7572         opts = lustre_cfg_string(cfg, 3);
7573 #ifdef __BIG_ENDIAN
7574         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7575                 CERROR("%s: device %s extents feature is not guaranteed to "
7576                        "work on big-endian systems. Use \"bigendian_extents\" "
7577                        "mount option to override.\n", name, dev);
7578                 RETURN(-EINVAL);
7579         }
7580 #endif
7581 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7582         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7583                 CWARN("force_over_128tb option is deprecated. "
7584                       "Filesystems less than 512TB can be created without any "
7585                       "force options. Use force_over_512tb option for "
7586                       "filesystems greater than 512TB.\n");
7587         }
7588 #endif
7589 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7590         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7591                 CWARN("force_over_256tb option is deprecated. "
7592                       "Filesystems less than 512TB can be created without any "
7593                       "force options. Use force_over_512tb option for "
7594                       "filesystems greater than 512TB.\n");
7595         }
7596 #endif
7597
7598         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7599                 force_over_512tb = 1;
7600
7601         __page = alloc_page(GFP_KERNEL);
7602         if (__page == NULL)
7603                 GOTO(out, rc = -ENOMEM);
7604         page = (unsigned long)page_address(__page);
7605         options = (char *)page;
7606         *options = '\0';
7607         if (opts != NULL) {
7608                 /* strip out the options for back compatiblity */
7609                 static char *sout[] = {
7610                         "mballoc",
7611                         "iopen",
7612                         "noiopen",
7613                         "iopen_nopriv",
7614                         "extents",
7615                         "noextents",
7616                         /* strip out option we processed in osd */
7617                         "bigendian_extents",
7618                         "force_over_128tb",
7619                         "force_over_256tb",
7620                         "force_over_512tb",
7621                         NULL
7622                 };
7623                 strncat(options, opts, PAGE_SIZE);
7624                 for (rc = 0, str = options; sout[rc]; ) {
7625                         char *op = strstr(str, sout[rc]);
7626
7627                         if (op == NULL) {
7628                                 rc++;
7629                                 str = options;
7630                                 continue;
7631                         }
7632                         if (op == options || *(op - 1) == ',') {
7633                                 str = op + strlen(sout[rc]);
7634                                 if (*str == ',' || *str == '\0') {
7635                                         *str == ',' ? str++ : str;
7636                                         memmove(op, str, strlen(str) + 1);
7637                                 }
7638                         }
7639                         for (str = op; *str != ',' && *str != '\0'; str++)
7640                                 ;
7641                 }
7642         } else {
7643                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7644         }
7645
7646         /* Glom up mount options */
7647         if (*options != '\0')
7648                 strncat(options, ",", PAGE_SIZE);
7649         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7650
7651         type = get_fs_type("ldiskfs");
7652         if (!type) {
7653                 CERROR("%s: cannot find ldiskfs module\n", name);
7654                 GOTO(out, rc = -ENODEV);
7655         }
7656
7657         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7658         module_put(type->owner);
7659
7660         if (IS_ERR(o->od_mnt)) {
7661                 rc = PTR_ERR(o->od_mnt);
7662                 o->od_mnt = NULL;
7663                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7664                 GOTO(out, rc);
7665         }
7666
7667         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7668                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7669                                  force_over_512tb == 0) {
7670                 CERROR("%s: device %s LDISKFS does not support filesystems "
7671                        "greater than 512TB and can cause data corruption. "
7672                        "Use \"force_over_512tb\" mount option to override.\n",
7673                        name, dev);
7674                 GOTO(out_mnt, rc = -EINVAL);
7675         }
7676
7677         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7678                 if (priv_dev_set_rdonly) {
7679                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7680                         o->od_dt_dev.dd_rdonly = 1;
7681                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7682                                       name);
7683                 } else {
7684                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7685                                       name);
7686
7687                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7688                 }
7689         } else if (priv_dev_check_rdonly &&
7690                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7691                 CERROR("%s: underlying device %s is marked as "
7692                        "read-only. Setup failed\n", name, dev);
7693
7694                 GOTO(out_mnt, rc = -EROFS);
7695         }
7696
7697         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7698                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7699                 GOTO(out_mnt, rc = -EINVAL);
7700         }
7701
7702 #ifdef LDISKFS_MOUNT_DIRDATA
7703         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7704                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7705         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7706                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7707                       "enabling the dirdata feature. If you do not want to "
7708                       "downgrade to Lustre-1.x again, you can enable it via "
7709                       "'tune2fs -O dirdata device'\n", name, dev);
7710 #endif
7711         inode = osd_sb(o)->s_root->d_inode;
7712         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7713         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7714         if (rc != 0) {
7715                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7716                 GOTO(out_mnt, rc);
7717         }
7718
7719         if (lmd_flags & LMD_FLG_NOSCRUB)
7720                 o->od_auto_scrub_interval = AS_NEVER;
7721
7722         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7723                 /* do not use pagecache with flash-backed storage */
7724                 o->od_writethrough_cache = 0;
7725                 o->od_read_cache = 0;
7726         }
7727
7728         GOTO(out, rc = 0);
7729
7730 out_mnt:
7731         mntput(o->od_mnt);
7732         o->od_mnt = NULL;
7733
7734 out:
7735         if (__page)
7736                 __free_page(__page);
7737
7738         return rc;
7739 }
7740
7741 static struct lu_device *osd_device_fini(const struct lu_env *env,
7742                                          struct lu_device *d)
7743 {
7744         struct osd_device *o = osd_dev(d);
7745
7746         ENTRY;
7747
7748         osd_index_backup(env, o, false);
7749         osd_shutdown(env, o);
7750         osd_procfs_fini(o);
7751         osd_obj_map_fini(o);
7752         osd_umount(env, o);
7753
7754         RETURN(NULL);
7755 }
7756
7757 static int osd_device_init0(const struct lu_env *env,
7758                             struct osd_device *o,
7759                             struct lustre_cfg *cfg)
7760 {
7761         struct lu_device *l = osd2lu_dev(o);
7762         struct osd_thread_info *info;
7763         int rc;
7764         int cplen = 0;
7765
7766         /* if the module was re-loaded, env can loose its keys */
7767         rc = lu_env_refill((struct lu_env *)env);
7768         if (rc)
7769                 GOTO(out, rc);
7770         info = osd_oti_get(env);
7771         LASSERT(info);
7772
7773         l->ld_ops = &osd_lu_ops;
7774         o->od_dt_dev.dd_ops = &osd_dt_ops;
7775
7776         spin_lock_init(&o->od_osfs_lock);
7777         mutex_init(&o->od_otable_mutex);
7778         INIT_LIST_HEAD(&o->od_orphan_list);
7779         INIT_LIST_HEAD(&o->od_index_backup_list);
7780         INIT_LIST_HEAD(&o->od_index_restore_list);
7781         spin_lock_init(&o->od_lock);
7782         o->od_index_backup_policy = LIBP_NONE;
7783         o->od_t10_type = 0;
7784
7785         o->od_read_cache = 1;
7786         o->od_writethrough_cache = 1;
7787         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7788         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7789         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7790         o->od_auto_scrub_interval = AS_DEFAULT;
7791
7792         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7793                         sizeof(o->od_svname));
7794         if (cplen >= sizeof(o->od_svname)) {
7795                 rc = -E2BIG;
7796                 GOTO(out, rc);
7797         }
7798
7799         o->od_index_backup_stop = 0;
7800         o->od_index = -1; /* -1 means index is invalid */
7801         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7802         if (rc == LDD_F_SV_TYPE_OST)
7803                 o->od_is_ost = 1;
7804
7805         o->od_full_scrub_ratio = OFSR_DEFAULT;
7806         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7807         rc = osd_mount(env, o, cfg);
7808         if (rc != 0)
7809                 GOTO(out, rc);
7810
7811         /* Can only check block device after mount */
7812         o->od_nonrotational =
7813                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7814
7815         rc = osd_obj_map_init(env, o);
7816         if (rc != 0)
7817                 GOTO(out_mnt, rc);
7818
7819         rc = lu_site_init(&o->od_site, l);
7820         if (rc != 0)
7821                 GOTO(out_compat, rc);
7822         o->od_site.ls_bottom_dev = l;
7823
7824         rc = lu_site_init_finish(&o->od_site);
7825         if (rc != 0)
7826                 GOTO(out_site, rc);
7827
7828         INIT_LIST_HEAD(&o->od_ios_list);
7829         /* setup scrub, including OI files initialization */
7830         o->od_in_init = 1;
7831         rc = osd_scrub_setup(env, o);
7832         o->od_in_init = 0;
7833         if (rc < 0)
7834                 GOTO(out_site, rc);
7835
7836         rc = osd_procfs_init(o, o->od_svname);
7837         if (rc != 0) {
7838                 CERROR("%s: can't initialize procfs: rc = %d\n",
7839                        o->od_svname, rc);
7840                 GOTO(out_scrub, rc);
7841         }
7842
7843         LASSERT(l->ld_site->ls_linkage.next != NULL);
7844         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7845
7846         /* initialize quota slave instance */
7847         /* currently it's no need to prepare qsd_instance_md for OST */
7848         if (!o->od_is_ost) {
7849                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7850                                                 &o->od_dt_dev,
7851                                                 o->od_proc_entry, true);
7852                 if (IS_ERR(o->od_quota_slave_md)) {
7853                         rc = PTR_ERR(o->od_quota_slave_md);
7854                         o->od_quota_slave_md = NULL;
7855                         GOTO(out_procfs, rc);
7856                 }
7857         }
7858
7859         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7860                                         o->od_proc_entry, false);
7861
7862         if (IS_ERR(o->od_quota_slave_dt)) {
7863                 if (o->od_quota_slave_md != NULL) {
7864                         qsd_fini(env, o->od_quota_slave_md);
7865                         o->od_quota_slave_md = NULL;
7866                 }
7867
7868                 rc = PTR_ERR(o->od_quota_slave_dt);
7869                 o->od_quota_slave_dt = NULL;
7870                 GOTO(out_procfs, rc);
7871         }
7872
7873         RETURN(0);
7874
7875 out_procfs:
7876         osd_procfs_fini(o);
7877 out_scrub:
7878         osd_scrub_cleanup(env, o);
7879 out_site:
7880         lu_site_fini(&o->od_site);
7881 out_compat:
7882         osd_obj_map_fini(o);
7883 out_mnt:
7884         osd_umount(env, o);
7885 out:
7886         return rc;
7887 }
7888
7889 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7890                                           struct lu_device_type *t,
7891                                           struct lustre_cfg *cfg)
7892 {
7893         struct osd_device *o;
7894         int rc;
7895
7896         OBD_ALLOC_PTR(o);
7897         if (o == NULL)
7898                 return ERR_PTR(-ENOMEM);
7899
7900         rc = dt_device_init(&o->od_dt_dev, t);
7901         if (rc == 0) {
7902                 /*
7903                  * Because the ctx might be revived in dt_device_init,
7904                  * refill the env here
7905                  */
7906                 lu_env_refill((struct lu_env *)env);
7907                 rc = osd_device_init0(env, o, cfg);
7908                 if (rc)
7909                         dt_device_fini(&o->od_dt_dev);
7910         }
7911
7912         if (unlikely(rc != 0))
7913                 OBD_FREE_PTR(o);
7914
7915         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7916 }
7917
7918 static struct lu_device *osd_device_free(const struct lu_env *env,
7919                                          struct lu_device *d)
7920 {
7921         struct osd_device *o = osd_dev(d);
7922
7923         ENTRY;
7924
7925         /* XXX: make osd top device in order to release reference */
7926         d->ld_site->ls_top_dev = d;
7927         lu_site_purge(env, d->ld_site, -1);
7928         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7929                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7930                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7931         }
7932         lu_site_fini(&o->od_site);
7933         dt_device_fini(&o->od_dt_dev);
7934         OBD_FREE_PTR(o);
7935         RETURN(NULL);
7936 }
7937
7938 static int osd_process_config(const struct lu_env *env,
7939                               struct lu_device *d, struct lustre_cfg *cfg)
7940 {
7941         struct osd_device *o = osd_dev(d);
7942         ssize_t count;
7943         int rc;
7944
7945         ENTRY;
7946
7947         switch (cfg->lcfg_command) {
7948         case LCFG_SETUP:
7949                 rc = osd_mount(env, o, cfg);
7950                 break;
7951         case LCFG_CLEANUP:
7952                 /*
7953                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7954                  * that may happend if hit failure during mount process.
7955                  */
7956                 osd_index_backup(env, o, false);
7957                 lu_dev_del_linkage(d->ld_site, d);
7958                 rc = osd_shutdown(env, o);
7959                 break;
7960         case LCFG_PARAM:
7961                 LASSERT(&o->od_dt_dev);
7962                 count  = class_modify_config(cfg, PARAM_OSD,
7963                                              &o->od_dt_dev.dd_kobj);
7964                 if (count < 0)
7965                         count = class_modify_config(cfg, PARAM_OST,
7966                                                     &o->od_dt_dev.dd_kobj);
7967                 rc = count > 0 ? 0 : count;
7968                 break;
7969         case LCFG_PRE_CLEANUP:
7970                 osd_scrub_stop(o);
7971                 osd_index_backup(env, o,
7972                                  o->od_index_backup_policy != LIBP_NONE);
7973                 rc = 0;
7974                 break;
7975         default:
7976                 rc = -ENOSYS;
7977         }
7978
7979         RETURN(rc);
7980 }
7981
7982 static int osd_recovery_complete(const struct lu_env *env,
7983                                  struct lu_device *d)
7984 {
7985         struct osd_device *osd = osd_dev(d);
7986         int rc = 0;
7987
7988         ENTRY;
7989
7990         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7991                 RETURN(0);
7992
7993         /*
7994          * start qsd instance on recovery completion, this notifies the quota
7995          * slave code that we are about to process new requests now
7996          */
7997         rc = qsd_start(env, osd->od_quota_slave_dt);
7998         if (rc == 0 && osd->od_quota_slave_md != NULL)
7999                 rc = qsd_start(env, osd->od_quota_slave_md);
8000
8001         RETURN(rc);
8002 }
8003
8004 /*
8005  * we use exports to track all osd users
8006  */
8007 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8008                            struct obd_device *obd, struct obd_uuid *cluuid,
8009                            struct obd_connect_data *data, void *localdata)
8010 {
8011         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8012         struct lustre_handle conn;
8013         int rc;
8014
8015         ENTRY;
8016
8017         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8018
8019         rc = class_connect(&conn, obd, cluuid);
8020         if (rc)
8021                 RETURN(rc);
8022
8023         *exp = class_conn2export(&conn);
8024
8025         spin_lock(&osd->od_osfs_lock);
8026         osd->od_connects++;
8027         spin_unlock(&osd->od_osfs_lock);
8028
8029         RETURN(0);
8030 }
8031
8032 /*
8033  * once last export (we don't count self-export) disappeared
8034  * osd can be released
8035  */
8036 static int osd_obd_disconnect(struct obd_export *exp)
8037 {
8038         struct obd_device *obd = exp->exp_obd;
8039         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8040         int rc, release = 0;
8041
8042         ENTRY;
8043
8044         /* Only disconnect the underlying layers on the final disconnect. */
8045         spin_lock(&osd->od_osfs_lock);
8046         osd->od_connects--;
8047         if (osd->od_connects == 0)
8048                 release = 1;
8049         spin_unlock(&osd->od_osfs_lock);
8050
8051         rc = class_disconnect(exp); /* bz 9811 */
8052
8053         if (rc == 0 && release)
8054                 class_manual_cleanup(obd);
8055         RETURN(rc);
8056 }
8057
8058 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8059                        struct lu_device *dev)
8060 {
8061         struct osd_device *osd = osd_dev(dev);
8062         struct lr_server_data *lsd =
8063                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8064         int result = 0;
8065
8066         ENTRY;
8067
8068         if (osd->od_quota_slave_md != NULL) {
8069                 /* set up quota slave objects for inode */
8070                 result = qsd_prepare(env, osd->od_quota_slave_md);
8071                 if (result != 0)
8072                         RETURN(result);
8073         }
8074
8075         if (osd->od_quota_slave_dt != NULL) {
8076                 /* set up quota slave objects for block */
8077                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8078                 if (result != 0)
8079                         RETURN(result);
8080         }
8081
8082
8083         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8084 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8085                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8086                         osd->od_index_in_idif = 1;
8087                 } else {
8088                         osd->od_index_in_idif = 0;
8089                         result = osd_register_proc_index_in_idif(osd);
8090                         if (result != 0)
8091                                 RETURN(result);
8092                 }
8093 #else
8094                 osd->od_index_in_idif = 1;
8095 #endif
8096         }
8097
8098         result = osd_fid_init(env, osd);
8099
8100         RETURN(result);
8101 }
8102
8103 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8104                          struct lu_fid *fid, struct md_op_data *op_data)
8105 {
8106         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8107
8108         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8109 }
8110
8111 static const struct lu_object_operations osd_lu_obj_ops = {
8112         .loo_object_init      = osd_object_init,
8113         .loo_object_delete    = osd_object_delete,
8114         .loo_object_release   = osd_object_release,
8115         .loo_object_free      = osd_object_free,
8116         .loo_object_print     = osd_object_print,
8117         .loo_object_invariant = osd_object_invariant
8118 };
8119
8120 const struct lu_device_operations osd_lu_ops = {
8121         .ldo_object_alloc      = osd_object_alloc,
8122         .ldo_process_config    = osd_process_config,
8123         .ldo_recovery_complete = osd_recovery_complete,
8124         .ldo_prepare           = osd_prepare,
8125 };
8126
8127 static const struct lu_device_type_operations osd_device_type_ops = {
8128         .ldto_init = osd_type_init,
8129         .ldto_fini = osd_type_fini,
8130
8131         .ldto_start = osd_type_start,
8132         .ldto_stop  = osd_type_stop,
8133
8134         .ldto_device_alloc = osd_device_alloc,
8135         .ldto_device_free  = osd_device_free,
8136
8137         .ldto_device_init = osd_device_init,
8138         .ldto_device_fini = osd_device_fini
8139 };
8140
8141 static struct lu_device_type osd_device_type = {
8142         .ldt_tags     = LU_DEVICE_DT,
8143         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8144         .ldt_ops      = &osd_device_type_ops,
8145         .ldt_ctx_tags = LCT_LOCAL,
8146 };
8147
8148 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8149 {
8150         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8151         struct super_block *sb = osd_sb(osd);
8152
8153         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8154 }
8155
8156 /*
8157  * lprocfs legacy support.
8158  */
8159 static const struct obd_ops osd_obd_device_ops = {
8160         .o_owner = THIS_MODULE,
8161         .o_connect      = osd_obd_connect,
8162         .o_disconnect   = osd_obd_disconnect,
8163         .o_fid_alloc    = osd_fid_alloc,
8164         .o_health_check = osd_health_check,
8165 };
8166
8167 static ssize_t track_declares_assert_show(struct kobject *kobj,
8168                                    struct attribute *attr,
8169                                    char *buf)
8170 {
8171         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8172 }
8173
8174 static ssize_t track_declares_assert_store(struct kobject *kobj,
8175                                            struct attribute *attr,
8176                                            const char *buffer, size_t count)
8177 {
8178         bool track_declares_assert;
8179         int rc;
8180
8181         rc = kstrtobool(buffer, &track_declares_assert);
8182         if (rc)
8183                 return rc;
8184
8185         ldiskfs_track_declares_assert = track_declares_assert;
8186
8187         return count;
8188 }
8189 LUSTRE_RW_ATTR(track_declares_assert);
8190
8191 static int __init osd_init(void)
8192 {
8193         struct kobject *kobj;
8194         int rc;
8195
8196         BUILD_BUG_ON(BH_DXLock >=
8197                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8198 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8199         /* please, try to keep osd_thread_info smaller than a page */
8200         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8201 #endif
8202
8203         osd_oi_mod_init();
8204
8205         rc = lu_kmem_init(ldiskfs_caches);
8206         if (rc)
8207                 return rc;
8208
8209 #ifdef CONFIG_KALLSYMS
8210         priv_security_file_alloc =
8211                 (void *)kallsyms_lookup_name("security_file_alloc");
8212         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8213         priv_dev_check_rdonly =
8214                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8215 #endif
8216
8217         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8218                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8219         if (rc) {
8220                 lu_kmem_fini(ldiskfs_caches);
8221                 return rc;
8222         }
8223
8224         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8225         if (kobj) {
8226                 rc = sysfs_create_file(kobj,
8227                                        &lustre_attr_track_declares_assert.attr);
8228                 kobject_put(kobj);
8229                 if (rc) {
8230                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8231                         rc = 0;
8232                 }
8233         }
8234         return rc;
8235 }
8236
8237 static void __exit osd_exit(void)
8238 {
8239         struct kobject *kobj;
8240
8241         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8242         if (kobj) {
8243                 sysfs_remove_file(kobj,
8244                                   &lustre_attr_track_declares_assert.attr);
8245                 kobject_put(kobj);
8246         }
8247         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8248         lu_kmem_fini(ldiskfs_caches);
8249 }
8250
8251 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8252 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8253 MODULE_VERSION(LUSTRE_VERSION_STRING);
8254 MODULE_LICENSE("GPL");
8255
8256 module_init(osd_init);
8257 module_exit(osd_exit);