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