lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/fs_struct.h>
  43 #include <linux/kallsyms.h>
  44 #include <linux/module.h>
  45 #include <linux/user_namespace.h>
  46 #include <linux/uidgid.h>
  47
  48 /* prerequisite for linux/xattr.h */
  49 #include <linux/types.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/fs.h>
  52 /* XATTR_{REPLACE,CREATE} */
  53 #include <linux/xattr.h>
  54
  55 #include <ldiskfs/ldiskfs.h>
  56 #include <ldiskfs/xattr.h>
  57 #include <ldiskfs/ldiskfs_extents.h>
  58 #undef ENTRY
  59 /*
  60  * struct OBD_{ALLOC,FREE}*()
  61  * OBD_FAIL_CHECK
  62  */
  63 #include <obd_support.h>
  64 /* struct ptlrpc_thread */
  65 #include <lustre_net.h>
  66 #include <lustre_fid.h>
  67 /* process_config */
  68 #include <uapi/linux/lustre/lustre_param.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_dynlocks.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75 #include <lustre_quota.h>
  76
  77 #include <lustre_linkea.h>
  78
  79 /* Maximum EA size is limited by LNET_MTU for remote objects */
  80 #define OSD_MAX_EA_SIZE 1048364
  81
  82 int ldiskfs_pdo = 1;
  83 module_param(ldiskfs_pdo, int, 0644);
  84 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  85
  86 int ldiskfs_track_declares_assert;
  87 module_param(ldiskfs_track_declares_assert, int, 0644);
  88 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  89
  90 /* Slab to allocate dynlocks */
  91 struct kmem_cache *dynlock_cachep;
  92
  93 /* Slab to allocate osd_it_ea */
  94 struct kmem_cache *osd_itea_cachep;
  95
  96 static struct lu_kmem_descr ldiskfs_caches[] = {
  97         {
  98                 .ckd_cache = &dynlock_cachep,
  99                 .ckd_name  = "dynlock_cache",
 100                 .ckd_size  = sizeof(struct dynlock_handle)
 101         },
 102         {
 103                 .ckd_cache = &osd_itea_cachep,
 104                 .ckd_name  = "osd_itea_cache",
 105                 .ckd_size  = sizeof(struct osd_it_ea)
 106         },
 107         {
 108                 .ckd_cache = NULL
 109         }
 110 };
 111
 112 static const char dot[] = ".";
 113 static const char dotdot[] = "..";
 114
 115 static const struct lu_object_operations      osd_lu_obj_ops;
 116 static const struct dt_object_operations      osd_obj_ops;
 117 static const struct dt_object_operations      osd_obj_otable_it_ops;
 118 static const struct dt_index_operations       osd_index_iam_ops;
 119 static const struct dt_index_operations       osd_index_ea_ops;
 120
 121 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 122                           const struct lu_fid *fid);
 123 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 124                                                 struct osd_device *osd);
 125
 126 int osd_trans_declare_op2rb[] = {
 127         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 128         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 129         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 130         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 131         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 132         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 133         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 134         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 135         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 136         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 137         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 138 };
 139
 140 static int osd_has_index(const struct osd_object *obj)
 141 {
 142         return obj->oo_dt.do_index_ops != NULL;
 143 }
 144
 145 static int osd_object_invariant(const struct lu_object *l)
 146 {
 147         return osd_invariant(osd_obj(l));
 148 }
 149
 150 /*
 151  * Concurrency: doesn't matter
 152  */
 153 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 154 {
 155         struct osd_thread_info *oti = osd_oti_get(env);
 156
 157         return oti->oti_w_locks > 0 && o->oo_owner == env;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't access mutable data
 162  */
 163 static int osd_root_get(const struct lu_env *env,
 164                         struct dt_device *dev, struct lu_fid *f)
 165 {
 166         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 167         return 0;
 168 }
 169
 170 /*
 171  * the following set of functions are used to maintain per-thread
 172  * cache of FID->ino mapping. this mechanism is needed to resolve
 173  * FID to inode at dt_insert() which in turn stores ino in the
 174  * directory entries to keep ldiskfs compatible with ext[34].
 175  * due to locking-originated restrictions we can't lookup ino
 176  * using LU cache (deadlock is possible). lookup using OI is quite
 177  * expensive. so instead we maintain this cache and methods like
 178  * dt_create() fill it. so in the majority of cases dt_insert() is
 179  * able to find needed mapping in lockless manner.
 180  */
 181 static struct osd_idmap_cache *
 182 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 183              const struct lu_fid *fid)
 184 {
 185         struct osd_thread_info *oti = osd_oti_get(env);
 186         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 187         int i;
 188
 189         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 190                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 191                         continue;
 192                 if (idc[i].oic_dev != osd)
 193                         continue;
 194
 195                 return idc + i;
 196         }
 197
 198         return NULL;
 199 }
 200
 201 static struct osd_idmap_cache *
 202 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 203             const struct lu_fid *fid)
 204 {
 205         struct osd_thread_info *oti   = osd_oti_get(env);
 206         struct osd_idmap_cache *idc;
 207         int i;
 208
 209         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 210                 i = oti->oti_ins_cache_size * 2;
 211                 if (i == 0)
 212                         i = OSD_INS_CACHE_SIZE;
 213                 OBD_ALLOC_PTR_ARRAY(idc, i);
 214                 if (idc == NULL)
 215                         return ERR_PTR(-ENOMEM);
 216                 if (oti->oti_ins_cache != NULL) {
 217                         memcpy(idc, oti->oti_ins_cache,
 218                                oti->oti_ins_cache_used * sizeof(*idc));
 219                         OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
 220                                            oti->oti_ins_cache_used);
 221                 }
 222                 oti->oti_ins_cache = idc;
 223                 oti->oti_ins_cache_size = i;
 224         }
 225
 226         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 227         idc->oic_fid = *fid;
 228         idc->oic_dev = osd;
 229         idc->oic_lid.oii_ino = 0;
 230         idc->oic_lid.oii_gen = 0;
 231         idc->oic_remote = 0;
 232
 233         return idc;
 234 }
 235
 236 /*
 237  * lookup mapping for the given fid in the cache, initialize a
 238  * new one if not found. the initialization checks whether the
 239  * object is local or remote. for local objects, OI is used to
 240  * learn ino/generation. the function is used when the caller
 241  * has no information about the object, e.g. at dt_insert().
 242  */
 243 static struct osd_idmap_cache *
 244 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 245                      const struct lu_fid *fid)
 246 {
 247         struct osd_idmap_cache *idc;
 248         int rc;
 249
 250         idc = osd_idc_find(env, osd, fid);
 251         LASSERT(!IS_ERR(idc));
 252         if (idc != NULL)
 253                 return idc;
 254
 255         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 256                osd->od_svname, PFID(fid));
 257
 258         /* new mapping is needed */
 259         idc = osd_idc_add(env, osd, fid);
 260         if (IS_ERR(idc)) {
 261                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 262                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 263                 return idc;
 264         }
 265
 266         /* initialize it */
 267         rc = osd_remote_fid(env, osd, fid);
 268         if (unlikely(rc < 0))
 269                 return ERR_PTR(rc);
 270
 271         if (rc == 0) {
 272                 /* the object is local, lookup in OI */
 273                 /* XXX: probably cheaper to lookup in LU first? */
 274                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 275                                    &idc->oic_lid, 0);
 276                 if (unlikely(rc < 0)) {
 277                         CERROR("can't lookup: rc = %d\n", rc);
 278                         return ERR_PTR(rc);
 279                 }
 280         } else {
 281                 /* the object is remote */
 282                 idc->oic_remote = 1;
 283         }
 284
 285         return idc;
 286 }
 287
 288 /*
 289  * lookup mapping for given FID and fill it from the given object.
 290  * the object is lolcal by definition.
 291  */
 292 static int osd_idc_find_and_init(const struct lu_env *env,
 293                                  struct osd_device *osd,
 294                                  struct osd_object *obj)
 295 {
 296         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 297         struct osd_idmap_cache *idc;
 298
 299         idc = osd_idc_find(env, osd, fid);
 300         LASSERT(!IS_ERR(idc));
 301         if (idc != NULL) {
 302                 if (obj->oo_inode == NULL)
 303                         return 0;
 304                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 305                         LASSERT(idc->oic_lid.oii_ino == 0);
 306                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 307                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 308                 }
 309                 return 0;
 310         }
 311
 312         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 313                osd->od_svname, PFID(fid));
 314
 315         /* new mapping is needed */
 316         idc = osd_idc_add(env, osd, fid);
 317         if (IS_ERR(idc)) {
 318                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 319                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 320                 return PTR_ERR(idc);
 321         }
 322
 323         if (obj->oo_inode != NULL) {
 324                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 325                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 326         }
 327         return 0;
 328 }
 329
 330 /*
 331  * OSD object methods.
 332  */
 333
 334 /*
 335  * Concurrency: no concurrent access is possible that early in object
 336  * life-cycle.
 337  */
 338 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 339                                           const struct lu_object_header *hdr,
 340                                           struct lu_device *d)
 341 {
 342         struct osd_object *mo;
 343
 344         OBD_ALLOC_PTR(mo);
 345         if (mo != NULL) {
 346                 struct lu_object *l;
 347                 struct lu_object_header *h;
 348                 struct osd_device *o = osd_dev(d);
 349
 350                 l = &mo->oo_dt.do_lu;
 351                 if (unlikely(o->od_in_init)) {
 352                         OBD_ALLOC_PTR(h);
 353                         if (!h) {
 354                                 OBD_FREE_PTR(mo);
 355                                 return NULL;
 356                         }
 357
 358                         lu_object_header_init(h);
 359                         lu_object_init(l, h, d);
 360                         lu_object_add_top(h, l);
 361                         mo->oo_header = h;
 362                 } else {
 363                         dt_object_init(&mo->oo_dt, NULL, d);
 364                         mo->oo_header = NULL;
 365                 }
 366
 367                 mo->oo_dt.do_ops = &osd_obj_ops;
 368                 l->lo_ops = &osd_lu_obj_ops;
 369                 init_rwsem(&mo->oo_sem);
 370                 init_rwsem(&mo->oo_ext_idx_sem);
 371                 spin_lock_init(&mo->oo_guard);
 372                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 373                 return l;
 374         }
 375         return NULL;
 376 }
 377
 378 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 379                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 380 {
 381         int rc;
 382
 383         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 384                              (void *)loa, sizeof(*loa));
 385         if (rc > 0) {
 386                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 387
 388                 if (rc < sizeof(*lma))
 389                         return -EINVAL;
 390
 391                 rc = 0;
 392                 lustre_loa_swab(loa, true);
 393                 /* Check LMA compatibility */
 394                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 395                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 396                               "for fid = "DFID", ino = %lu\n",
 397                               osd_ino2name(inode),
 398                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 399                               PFID(&lma->lma_self_fid), inode->i_ino);
 400                         rc = -EOPNOTSUPP;
 401                 }
 402         } else if (rc == 0) {
 403                 rc = -ENODATA;
 404         }
 405
 406         return rc;
 407 }
 408
 409 /*
 410  * retrieve object from backend ext fs.
 411  **/
 412 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 413                        struct osd_inode_id *id)
 414 {
 415         int rc;
 416         struct inode *inode = NULL;
 417
 418         /*
 419          * if we look for an inode withing a running
 420          * transaction, then we risk to deadlock
 421          * osd_dirent_check_repair() breaks this
 422          */
 423          /* LASSERT(current->journal_info == NULL); */
 424
 425         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 426         if (IS_ERR(inode)) {
 427                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 428                        id->oii_ino, PTR_ERR(inode));
 429         } else if (id->oii_gen != OSD_OII_NOGEN &&
 430                    inode->i_generation != id->oii_gen) {
 431                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 432                        "i_generation = %u\n",
 433                        id->oii_ino, id->oii_gen, inode->i_generation);
 434                 iput(inode);
 435                 inode = ERR_PTR(-ESTALE);
 436         } else if (inode->i_nlink == 0) {
 437                 /*
 438                  * due to parallel readdir and unlink,
 439                  * we can have dead inode here.
 440                  */
 441                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 442                 iput(inode);
 443                 inode = ERR_PTR(-ESTALE);
 444         } else if (is_bad_inode(inode)) {
 445                 CWARN("%s: bad inode: ino = %u\n",
 446                       osd_dev2name(dev), id->oii_ino);
 447                 iput(inode);
 448                 inode = ERR_PTR(-ENOENT);
 449         } else if ((rc = osd_attach_jinode(inode))) {
 450                 iput(inode);
 451                 inode = ERR_PTR(rc);
 452         } else {
 453                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 454                 if (id->oii_gen == OSD_OII_NOGEN)
 455                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 456
 457                 /*
 458                  * Do not update file c/mtime in ldiskfs.
 459                  * NB: we don't have any lock to protect this because we don't
 460                  * have reference on osd_object now, but contention with
 461                  * another lookup + attr_set can't happen in the tiny window
 462                  * between if (...) and set S_NOCMTIME.
 463                  */
 464                 if (!(inode->i_flags & S_NOCMTIME))
 465                         inode->i_flags |= S_NOCMTIME;
 466         }
 467         return inode;
 468 }
 469
 470 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 471                           handle_t *handle, struct dentry *child,
 472                           struct inode *inode, struct htree_lock *hlock)
 473 {
 474         int rc, rc2;
 475
 476         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 477         if (rc == -ENOBUFS || rc == -ENOSPC) {
 478                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 479                 struct inode *parent = child->d_parent->d_inode;
 480                 struct lu_fid *fid = NULL;
 481                 char fidstr[FID_LEN + 1] = "unknown";
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
 499
 500                 /* below message is checked in sanity.sh test_129 */
 501                 if (rc == -ENOSPC) {
 502                         CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
 503                               osd_name(osd), parent->i_ino, fidstr);
 504                 } else {
 505                         rc = 0; /* ignore such error now */
 506                         CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
 507                               osd_name(osd), parent->i_ino, fidstr);
 508                 }
 509
 510         }
 511
 512         return rc;
 513 }
 514
 515
 516 struct inode *
 517 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 518              struct osd_inode_id *id, struct lu_fid *fid)
 519 {
 520         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 521         struct inode *inode;
 522         int rc;
 523
 524         inode = osd_iget(info, dev, id);
 525         if (IS_ERR(inode))
 526                 return inode;
 527
 528         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 529         if (!rc) {
 530                 *fid = loa->loa_lma.lma_self_fid;
 531         } else if (rc == -ENODATA) {
 532                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 533                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 534                 else
 535                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 536         } else {
 537                 iput(inode);
 538                 inode = ERR_PTR(rc);
 539         }
 540         return inode;
 541 }
 542
 543 static struct inode *osd_iget_check(struct osd_thread_info *info,
 544                                     struct osd_device *dev,
 545                                     const struct lu_fid *fid,
 546                                     struct osd_inode_id *id,
 547                                     bool trusted)
 548 {
 549         struct inode *inode;
 550         int rc = 0;
 551
 552         ENTRY;
 553
 554         /*
 555          * The cached OI mapping is trustable. If we cannot locate the inode
 556          * via the cached OI mapping, then return the failure to the caller
 557          * directly without further OI checking.
 558          */
 559
 560 again:
 561         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 562         if (IS_ERR(inode)) {
 563                 rc = PTR_ERR(inode);
 564                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 565                         goto check_oi;
 566
 567                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 568                        PFID(fid), id->oii_ino, rc);
 569                 GOTO(put, rc);
 570         }
 571
 572         if (is_bad_inode(inode)) {
 573                 rc = -ENOENT;
 574                 if (!trusted)
 575                         goto check_oi;
 576
 577                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 578                        PFID(fid), id->oii_ino);
 579                 GOTO(put, rc);
 580         }
 581
 582         if (id->oii_gen != OSD_OII_NOGEN &&
 583             inode->i_generation != id->oii_gen) {
 584                 rc = -ESTALE;
 585                 if (!trusted)
 586                         goto check_oi;
 587
 588                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 589                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 590                        id->oii_ino, id->oii_gen, inode->i_generation);
 591                 GOTO(put, rc);
 592         }
 593
 594         if (inode->i_nlink == 0) {
 595                 rc = -ENOENT;
 596                 if (!trusted)
 597                         goto check_oi;
 598
 599                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 600                        PFID(fid), id->oii_ino);
 601                 GOTO(put, rc);
 602         }
 603
 604         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 605
 606 check_oi:
 607         if (rc != 0) {
 608                 __u32 saved_ino = id->oii_ino;
 609                 __u32 saved_gen = id->oii_gen;
 610
 611                 LASSERT(!trusted);
 612                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 613
 614                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 615                 /*
 616                  * XXX: There are four possible cases:
 617                  *      1. rc = 0.
 618                  *         Backup/restore caused the OI invalid.
 619                  *      2. rc = 0.
 620                  *         Someone unlinked the object but NOT removed
 621                  *         the OI mapping, such as mount target device
 622                  *         as ldiskfs, and modify something directly.
 623                  *      3. rc = -ENOENT.
 624                  *         Someone just removed the object between the
 625                  *         former oi_lookup and the iget. It is normal.
 626                  *      4. Other failure cases.
 627                  *
 628                  *      Generally, when the device is mounted, it will
 629                  *      auto check whether the system is restored from
 630                  *      file-level backup or not. We trust such detect
 631                  *      to distinguish the 1st case from the 2nd case:
 632                  *      if the OI files are consistent but may contain
 633                  *      stale OI mappings because of case 2, if iget()
 634                  *      returns -ENOENT or -ESTALE, then it should be
 635                  *      the case 2.
 636                  */
 637                 if (rc != 0)
 638                         /*
 639                          * If the OI mapping was in OI file before the
 640                          * osd_iget_check(), but now, it is disappear,
 641                          * then it must be removed by race. That is a
 642                          * normal race case.
 643                          */
 644                         GOTO(put, rc);
 645
 646                 /*
 647                  * It is the OI scrub updated the OI mapping by race.
 648                  * The new OI mapping must be valid.
 649                  */
 650                 if (saved_ino != id->oii_ino ||
 651                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 652                         if (!IS_ERR(inode))
 653                                 iput(inode);
 654
 655                         trusted = true;
 656                         goto again;
 657                 }
 658
 659                 if (IS_ERR(inode)) {
 660                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 661                             SF_INCONSISTENT)
 662                                 /*
 663                                  * It still can be the case 2, but we cannot
 664                                  * distinguish it from the case 1. So return
 665                                  * -EREMCHG to block current operation until
 666                                  *  OI scrub rebuilt the OI mappings.
 667                                  */
 668                                 rc = -EREMCHG;
 669                         else
 670                                 rc = -ENOENT;
 671
 672                         GOTO(put, rc);
 673                 }
 674
 675                 if (inode->i_generation == id->oii_gen)
 676                         rc = -ENOENT;
 677                 else
 678                         rc = -EREMCHG;
 679         } else {
 680                 if (id->oii_gen == OSD_OII_NOGEN)
 681                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 682
 683                 /*
 684                  * Do not update file c/mtime in ldiskfs.
 685                  * NB: we don't have any lock to protect this because we don't
 686                  * have reference on osd_object now, but contention with
 687                  * another lookup + attr_set can't happen in the tiny window
 688                  * between if (...) and set S_NOCMTIME.
 689                  */
 690                 if (!(inode->i_flags & S_NOCMTIME))
 691                         inode->i_flags |= S_NOCMTIME;
 692         }
 693
 694         GOTO(put, rc);
 695
 696 put:
 697         if (rc != 0) {
 698                 if (!IS_ERR(inode))
 699                         iput(inode);
 700
 701                 inode = ERR_PTR(rc);
 702         }
 703
 704         return inode;
 705 }
 706
 707 /**
 708  * \retval +v: new filter_fid does not contain self-fid
 709  * \retval 0:  filter_fid_18_23, contains self-fid
 710  * \retval -v: other failure cases
 711  */
 712 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 713                  struct dentry *dentry, struct lu_fid *fid)
 714 {
 715         struct filter_fid *ff = &info->oti_ff;
 716         struct ost_id *ostid = &info->oti_ostid;
 717         int rc;
 718
 719         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 720         if (rc == sizeof(struct filter_fid_18_23)) {
 721                 struct filter_fid_18_23 *ff_old = (void *)ff;
 722
 723                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 724                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 725                 /*
 726                  * XXX: use 0 as the index for compatibility, the caller will
 727                  * handle index related issues when necessary.
 728                  */
 729                 if (!rc)
 730                         ostid_to_fid(fid, ostid, 0);
 731         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 732                 rc = 1;
 733         } else if (rc >= 0) {
 734                 rc = -EINVAL;
 735         }
 736
 737         return rc;
 738 }
 739
 740 static int osd_lma_self_repair(struct osd_thread_info *info,
 741                                struct osd_device *osd, struct inode *inode,
 742                                const struct lu_fid *fid, __u32 compat)
 743 {
 744         handle_t *jh;
 745         int rc;
 746
 747         LASSERT(current->journal_info == NULL);
 748
 749         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 750                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 751         if (IS_ERR(jh)) {
 752                 rc = PTR_ERR(jh);
 753                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 754                       osd_name(osd), rc);
 755                 return rc;
 756         }
 757
 758         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 759         if (rc != 0)
 760                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 761                       osd_name(osd), rc);
 762         ldiskfs_journal_stop(jh);
 763         return rc;
 764 }
 765
 766 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 767 {
 768         struct osd_thread_info *info = osd_oti_get(env);
 769         struct osd_device *osd = osd_obj2dev(obj);
 770         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 771         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 772         struct inode *inode = obj->oo_inode;
 773         struct dentry *dentry = &info->oti_obj_dentry;
 774         struct lu_fid *fid = NULL;
 775         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 776         int rc;
 777
 778         ENTRY;
 779
 780         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 781                              (void *)loa, sizeof(*loa));
 782         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 783                 fid = &lma->lma_self_fid;
 784                 rc = osd_get_idif(info, inode, dentry, fid);
 785                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 786                         /*
 787                          * For the given OST-object, if it has neither LMA nor
 788                          * FID in XATTR_NAME_FID, then the given FID (which is
 789                          * contained in the @obj, from client RPC for locating
 790                          * the OST-object) is trusted. We use it to generate
 791                          * the LMA.
 792                          */
 793                         osd_lma_self_repair(info, osd, inode, rfid,
 794                                             LMAC_FID_ON_OST);
 795                         RETURN(0);
 796                 }
 797         }
 798
 799         if (rc < 0)
 800                 RETURN(rc);
 801
 802         if (rc > 0) {
 803                 rc = 0;
 804                 lustre_lma_swab(lma);
 805                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 806                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 807                               S_ISREG(inode->i_mode)))) {
 808                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 809                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 810                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 811                               PFID(rfid), inode->i_ino);
 812                         rc = -EOPNOTSUPP;
 813                 } else {
 814                         fid = &lma->lma_self_fid;
 815                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 816                             osd->od_is_ost)
 817                                 obj->oo_pfid_in_lma = 1;
 818                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 819                             !osd->od_is_ost)
 820                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 821                 }
 822         }
 823
 824         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 825                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 826                         struct ost_id   *oi   = &info->oti_ostid;
 827                         struct lu_fid   *fid1 = &info->oti_fid3;
 828                         __u32            idx  = fid_idif_ost_idx(rfid);
 829
 830                         /*
 831                          * For old IDIF, the OST index is not part of the IDIF,
 832                          * Means that different OSTs may have the same IDIFs.
 833                          * Under such case, we need to make some compatible
 834                          * check to make sure to trigger OI scrub properly.
 835                          */
 836                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 837                                 /* Given @rfid is new, LMA is old. */
 838                                 fid_to_ostid(fid, oi);
 839                                 ostid_to_fid(fid1, oi, idx);
 840                                 if (lu_fid_eq(fid1, rfid)) {
 841                                         if (osd->od_index_in_idif)
 842                                                 osd_lma_self_repair(info, osd,
 843                                                         inode, rfid,
 844                                                         LMAC_FID_ON_OST);
 845                                         RETURN(0);
 846                                 }
 847                         }
 848                 }
 849
 850                 rc = -EREMCHG;
 851         }
 852
 853         RETURN(rc);
 854 }
 855
 856 struct osd_check_lmv_buf {
 857         /* please keep it as first member */
 858         struct dir_context ctx;
 859         struct osd_thread_info *oclb_info;
 860         struct osd_device *oclb_dev;
 861         struct osd_idmap_cache *oclb_oic;
 862         int oclb_items;
 863         bool oclb_found;
 864 };
 865
 866 /**
 867  * It is called internally by ->iterate*() to filter out the
 868  * local slave object's FID of the striped directory.
 869  *
 870  * \retval      1 found the local slave's FID
 871  * \retval      0 continue to check next item
 872  * \retval      -ve for failure
 873  */
 874 #ifdef HAVE_FILLDIR_USE_CTX
 875 static int osd_stripe_dir_filldir(struct dir_context *buf,
 876 #else
 877 static int osd_stripe_dir_filldir(void *buf,
 878 #endif
 879                                   const char *name, int namelen,
 880                                   loff_t offset, __u64 ino, unsigned int d_type)
 881 {
 882         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 883         struct osd_thread_info *oti = oclb->oclb_info;
 884         struct lu_fid *fid = &oti->oti_fid3;
 885         struct osd_inode_id *id = &oti->oti_id3;
 886         struct osd_device *dev = oclb->oclb_dev;
 887         struct osd_idmap_cache *oic = oclb->oclb_oic;
 888         struct inode *inode;
 889
 890         oclb->oclb_items++;
 891
 892         if (name[0] == '.')
 893                 return 0;
 894
 895         fid_zero(fid);
 896         sscanf(name + 1, SFID, RFID(fid));
 897         if (!fid_is_sane(fid))
 898                 return 0;
 899
 900         if (osd_remote_fid(oti->oti_env, dev, fid))
 901                 return 0;
 902
 903         osd_id_gen(id, ino, OSD_OII_NOGEN);
 904         inode = osd_iget(oti, dev, id);
 905         if (IS_ERR(inode))
 906                 return PTR_ERR(inode);
 907
 908         iput(inode);
 909         osd_add_oi_cache(oti, dev, id, fid);
 910         oic->oic_fid = *fid;
 911         oic->oic_lid = *id;
 912         oic->oic_dev = dev;
 913         osd_oii_insert(dev, oic, true);
 914         oclb->oclb_found = true;
 915
 916         return 1;
 917 }
 918
 919 /*
 920  * When lookup item under striped directory, we need to locate the master
 921  * MDT-object of the striped directory firstly, then the client will send
 922  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 923  * and the item's name. If the system is restored from MDT file level backup,
 924  * then before the OI scrub completely built the OI files, the OI mappings of
 925  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 926  * not a problem for the master MDT-object. Because when locate the master
 927  * MDT-object, we will do name based lookup (for the striped directory itself)
 928  * firstly, during such process we can setup the correct OI mapping for the
 929  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 930  * the client will not trigger name based lookup on the MDT to locate the slave
 931  * MDT-object before locating item under the striped directory, then when
 932  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 933  * is invalid and does not know what the right OI mapping is, then the MDT has
 934  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 935  * the OI file, related OI mapping is unknown yet, please try again later. And
 936  * then client will re-try the RPC again and again until related OI mapping has
 937  * been updated. That is quite inefficient.
 938  *
 939  * To resolve above trouble, we will handle it as the following two cases:
 940  *
 941  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 942  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 943  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 944  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 945  *    directly. Please check osd_fid_lookup().
 946  *
 947  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 948  *    Under such case, during lookup the master MDT-object, we will lookup the
 949  *    slave MDT-object via readdir against the master MDT-object, because the
 950  *    slave MDT-objects information are stored as sub-directories with the name
 951  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 952  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 953  *    the correct OI mapping for the slave MDT-object.
 954  */
 955 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 956                          struct inode *inode, struct osd_idmap_cache *oic)
 957 {
 958         struct lu_buf *buf = &oti->oti_big_buf;
 959         struct dentry *dentry = &oti->oti_obj_dentry;
 960         struct file *filp;
 961         struct lmv_mds_md_v1 *lmv1;
 962         struct osd_check_lmv_buf oclb = {
 963                 .ctx.actor = osd_stripe_dir_filldir,
 964                 .oclb_info = oti,
 965                 .oclb_dev = dev,
 966                 .oclb_oic = oic,
 967                 .oclb_found = false,
 968         };
 969         int rc = 0;
 970
 971         ENTRY;
 972
 973 again:
 974         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 975                              buf->lb_len);
 976         if (rc == -ERANGE) {
 977                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 978                 if (rc > 0) {
 979                         lu_buf_realloc(buf, rc);
 980                         if (buf->lb_buf == NULL)
 981                                 GOTO(out, rc = -ENOMEM);
 982
 983                         goto again;
 984                 }
 985         }
 986
 987         if (unlikely(rc == 0 || rc == -ENODATA))
 988                 GOTO(out, rc = 0);
 989
 990         if (rc < 0)
 991                 GOTO(out, rc);
 992
 993         if (unlikely(buf->lb_buf == NULL)) {
 994                 lu_buf_realloc(buf, rc);
 995                 if (buf->lb_buf == NULL)
 996                         GOTO(out, rc = -ENOMEM);
 997
 998                 goto again;
 999         }
1000
1001         lmv1 = buf->lb_buf;
1002         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1003                 GOTO(out, rc = 0);
1004
1005         filp = osd_quasi_file(oti->oti_env, inode);
1006         rc = osd_security_file_alloc(filp);
1007         if (rc)
1008                 goto out;
1009
1010         do {
1011                 oclb.oclb_items = 0;
1012                 rc = iterate_dir(filp, &oclb.ctx);
1013         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1014                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1015         inode->i_fop->release(inode, filp);
1016
1017 out:
1018         if (rc < 0)
1019                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1020                        DFID": rc = %d\n", osd_ino2name(inode),
1021                        inode->i_ino, inode->i_generation,
1022                        PFID(&oic->oic_fid), rc);
1023         else
1024                 rc = 0;
1025
1026         RETURN(rc);
1027 }
1028
1029 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1030                           const struct lu_fid *fid,
1031                           const struct lu_object_conf *conf)
1032 {
1033         struct osd_thread_info *info;
1034         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1035         struct osd_device *dev;
1036         struct osd_idmap_cache *oic;
1037         struct osd_inode_id *id;
1038         struct inode *inode = NULL;
1039         struct lustre_scrub *scrub;
1040         struct scrub_file *sf;
1041         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1042         __u32 saved_ino;
1043         __u32 saved_gen;
1044         int result = 0;
1045         int rc1 = 0;
1046         bool remote = false;
1047         bool trusted = true;
1048         bool updated = false;
1049         bool checked = false;
1050
1051         ENTRY;
1052
1053         LINVRNT(osd_invariant(obj));
1054         LASSERT(obj->oo_inode == NULL);
1055         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1056
1057         dev = osd_dev(ldev);
1058         scrub = &dev->od_scrub.os_scrub;
1059         sf = &scrub->os_file;
1060         info = osd_oti_get(env);
1061         LASSERT(info);
1062         oic = &info->oti_cache;
1063
1064         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1065                 RETURN(-ENOENT);
1066
1067         /*
1068          * For the object is created as locking anchor, or for the object to
1069          * be created on disk. No need to osd_oi_lookup() at here because FID
1070          * shouldn't never be re-used, if it's really a duplicate FID from
1071          * unexpected reason, we should be able to detect it later by calling
1072          * do_create->osd_oi_insert().
1073          */
1074         if (conf && conf->loc_flags & LOC_F_NEW)
1075                 GOTO(out, result = 0);
1076
1077         /* Search order: 1. per-thread cache. */
1078         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1079                 id = &oic->oic_lid;
1080                 goto iget;
1081         }
1082
1083         id = &info->oti_id;
1084         if (!list_empty(&scrub->os_inconsistent_items)) {
1085                 /* Search order: 2. OI scrub pending list. */
1086                 result = osd_oii_lookup(dev, fid, id);
1087                 if (!result)
1088                         goto iget;
1089         }
1090
1091         /*
1092          * The OI mapping in the OI file can be updated by the OI scrub
1093          * when we locate the inode via FID. So it may be not trustable.
1094          */
1095         trusted = false;
1096
1097         /* Search order: 3. OI files. */
1098         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1099         if (result == -ENOENT) {
1100                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1101                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1102                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1103                                       sf->sf_oi_bitmap))
1104                         GOTO(out, result = 0);
1105
1106                 goto trigger;
1107         }
1108
1109         /* -ESTALE is returned if inode of OST object doesn't exist */
1110         if (result == -ESTALE &&
1111             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1112                 GOTO(out, result = 0);
1113         }
1114
1115         if (result)
1116                 GOTO(out, result);
1117
1118 iget:
1119         obj->oo_inode = NULL;
1120         /* for later passes through checks, not true on first pass */
1121         if (!IS_ERR_OR_NULL(inode))
1122                 iput(inode);
1123
1124         inode = osd_iget_check(info, dev, fid, id, trusted);
1125         if (!IS_ERR(inode)) {
1126                 obj->oo_inode = inode;
1127                 result = 0;
1128                 if (remote)
1129                         goto trigger;
1130
1131                 goto check_lma;
1132         }
1133
1134         result = PTR_ERR(inode);
1135         if (result == -ENOENT || result == -ESTALE)
1136                 GOTO(out, result = 0);
1137
1138         if (result != -EREMCHG)
1139                 GOTO(out, result);
1140
1141 trigger:
1142         /*
1143          * We still have chance to get the valid inode: for the
1144          * object which is referenced by remote name entry, the
1145          * object on the local MDT will be linked under the dir
1146          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1147          *
1148          * We do not know whether the object for the given FID
1149          * is referenced by some remote name entry or not, and
1150          * especially for DNE II, a multiple-linked object may
1151          * have many name entries reside on many MDTs.
1152          *
1153          * To simplify the operation, OSD will not distinguish
1154          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1155          * only happened for the RPC from other MDT during the
1156          * OI scrub, or for the client side RPC with FID only,
1157          * such as FID to path, or from old connected client.
1158          */
1159         if (!remote) {
1160                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1161                 if (!rc1) {
1162                         remote = true;
1163                         trusted = true;
1164                         flags |= SS_AUTO_PARTIAL;
1165                         flags &= ~SS_AUTO_FULL;
1166                         goto iget;
1167                 }
1168         }
1169
1170         if (scrub->os_running) {
1171                 if (scrub->os_partial_scan && !scrub->os_in_join)
1172                         goto join;
1173
1174                 osd_add_oi_cache(info, dev, id, fid);
1175                 if (IS_ERR_OR_NULL(inode) || result) {
1176                         osd_oii_insert(dev, oic, result == -ENOENT);
1177                         GOTO(out, result = -EINPROGRESS);
1178                 }
1179
1180                 LASSERT(remote);
1181                 LASSERT(obj->oo_inode == inode);
1182
1183                 osd_oii_insert(dev, oic, true);
1184                 goto found;
1185         }
1186
1187         if (dev->od_auto_scrub_interval == AS_NEVER) {
1188                 if (!remote)
1189                         GOTO(out, result = -EREMCHG);
1190
1191                 LASSERT(!result);
1192                 LASSERT(obj->oo_inode == inode);
1193
1194                 osd_add_oi_cache(info, dev, id, fid);
1195                 goto found;
1196         }
1197
1198 join:
1199         rc1 = osd_scrub_start(env, dev, flags);
1200         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1201                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1202         if (rc1 && rc1 != -EALREADY)
1203                 GOTO(out, result = -EREMCHG);
1204
1205         osd_add_oi_cache(info, dev, id, fid);
1206         if (IS_ERR_OR_NULL(inode) || result) {
1207                 osd_oii_insert(dev, oic, result == -ENOENT);
1208                 GOTO(out, result = -EINPROGRESS);
1209         }
1210
1211         LASSERT(remote);
1212         LASSERT(obj->oo_inode == inode);
1213
1214         osd_oii_insert(dev, oic, true);
1215         goto found;
1216
1217 check_lma:
1218         checked = true;
1219         if (unlikely(obj->oo_header))
1220                 goto found;
1221
1222         result = osd_check_lma(env, obj);
1223         if (!result)
1224                 goto found;
1225
1226         LASSERTF(id->oii_ino == inode->i_ino &&
1227                  id->oii_gen == inode->i_generation,
1228                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1229                  PFID(fid), id->oii_ino, id->oii_gen,
1230                  inode->i_ino, inode->i_generation);
1231
1232         saved_ino = inode->i_ino;
1233         saved_gen = inode->i_generation;
1234
1235         if (unlikely(result == -ENODATA)) {
1236                 /*
1237                  * If the OI scrub updated the OI mapping by race, it
1238                  * must be valid. Trust the inode that has no LMA EA.
1239                  */
1240                 if (updated)
1241                         goto found;
1242
1243                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1244                 if (!result) {
1245                         /*
1246                          * The OI mapping is still there, the inode is still
1247                          * valid. It is just becaues the inode has no LMA EA.
1248                          */
1249                         if (saved_ino == id->oii_ino &&
1250                             saved_gen == id->oii_gen)
1251                                 goto found;
1252
1253                         /*
1254                          * It is the OI scrub updated the OI mapping by race.
1255                          * The new OI mapping must be valid.
1256                          */
1257                         trusted = true;
1258                         updated = true;
1259                         goto iget;
1260                 }
1261
1262                 /*
1263                  * "result == -ENOENT" means that the OI mappinghas been
1264                  * removed by race, so the inode belongs to other object.
1265                  *
1266                  * Others error can be returned  directly.
1267                  */
1268                 if (result == -ENOENT) {
1269                         LASSERT(trusted);
1270
1271                         obj->oo_inode = NULL;
1272                         result = 0;
1273                 }
1274         }
1275
1276         if (result != -EREMCHG)
1277                 GOTO(out, result);
1278
1279         LASSERT(!updated);
1280
1281         /*
1282          * if two OST objects map to the same inode, and inode mode is
1283          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1284          * reserved by precreate, and not written yet, in this case, don't
1285          * set inode for the object whose FID mismatch, so that it can create
1286          * inode and not block precreate.
1287          */
1288         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1289             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1290                 obj->oo_inode = NULL;
1291                 GOTO(out, result = 0);
1292         }
1293
1294         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1295         /*
1296          * "result == -ENOENT" means the cached OI mapping has been removed
1297          * from the OI file by race, above inode belongs to other object.
1298          */
1299         if (result == -ENOENT) {
1300                 LASSERT(trusted);
1301
1302                 obj->oo_inode = NULL;
1303                 GOTO(out, result = 0);
1304         }
1305
1306         if (result)
1307                 GOTO(out, result);
1308
1309         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1310                 result = -EREMCHG;
1311                 goto trigger;
1312         }
1313
1314         /*
1315          * It is the OI scrub updated the OI mapping by race.
1316          * The new OI mapping must be valid.
1317          */
1318         trusted = true;
1319         updated = true;
1320         goto iget;
1321
1322 found:
1323         if (!checked) {
1324                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1325                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1326
1327                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1328                 if (!result) {
1329                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1330                             dev->od_is_ost)
1331                                 obj->oo_pfid_in_lma = 1;
1332                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1333                             !dev->od_is_ost)
1334                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1335                 } else if (result != -ENODATA) {
1336                         GOTO(out, result);
1337                 }
1338         }
1339
1340         obj->oo_compat_dot_created = 1;
1341         obj->oo_compat_dotdot_created = 1;
1342
1343         if (S_ISDIR(inode->i_mode) &&
1344             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1345                 osd_check_lmv(info, dev, inode, oic);
1346
1347         result = osd_attach_jinode(inode);
1348         if (result)
1349                 GOTO(out, result);
1350
1351         if (!ldiskfs_pdo)
1352                 GOTO(out, result = 0);
1353
1354         LASSERT(!obj->oo_hl_head);
1355         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1356
1357         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1358
1359 out:
1360         if (result || !obj->oo_inode) {
1361                 if (!IS_ERR_OR_NULL(inode))
1362                         iput(inode);
1363
1364                 obj->oo_inode = NULL;
1365                 if (trusted)
1366                         fid_zero(&oic->oic_fid);
1367         }
1368
1369         LINVRNT(osd_invariant(obj));
1370         return result;
1371 }
1372
1373 /*
1374  * Concurrency: shouldn't matter.
1375  */
1376 static void osd_object_init0(struct osd_object *obj)
1377 {
1378         LASSERT(obj->oo_inode != NULL);
1379         obj->oo_dt.do_body_ops = &osd_body_ops;
1380         obj->oo_dt.do_lu.lo_header->loh_attr |=
1381                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1382 }
1383
1384 /*
1385  * Concurrency: no concurrent access is possible that early in object
1386  * life-cycle.
1387  */
1388 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1389                            const struct lu_object_conf *conf)
1390 {
1391         struct osd_object *obj = osd_obj(l);
1392         int result;
1393
1394         LINVRNT(osd_invariant(obj));
1395
1396         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1397             cfs_fail_val == 2) {
1398                 struct osd_thread_info *info = osd_oti_get(env);
1399                 struct osd_idmap_cache *oic = &info->oti_cache;
1400                 /* invalidate thread cache */
1401                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1402         }
1403         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1404                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1405                 l->lo_header->loh_attr |= LOHA_EXISTS;
1406                 return 0;
1407         }
1408
1409         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1410         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1411         if (result == 0 && obj->oo_inode != NULL) {
1412                 struct osd_thread_info *oti = osd_oti_get(env);
1413                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1414
1415                 osd_object_init0(obj);
1416                 if (unlikely(obj->oo_header))
1417                         return 0;
1418
1419                 result = osd_get_lma(oti, obj->oo_inode,
1420                                      &oti->oti_obj_dentry, loa);
1421                 if (!result) {
1422                         /*
1423                          * Convert LMAI flags to lustre LMA flags
1424                          * and cache it to oo_lma_flags
1425                          */
1426                         obj->oo_lma_flags =
1427                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1428                 } else if (result == -ENODATA) {
1429                         result = 0;
1430                 }
1431         }
1432         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1433
1434         LINVRNT(osd_invariant(obj));
1435         return result;
1436 }
1437
1438 /*
1439  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1440  * The left part is for value, binary mode.
1441  */
1442 struct osd_xattr_entry {
1443         struct list_head        oxe_list;
1444         size_t                  oxe_len;
1445         size_t                  oxe_namelen;
1446         bool                    oxe_exist;
1447         struct rcu_head         oxe_rcu;
1448         char                    oxe_buf[0];
1449 };
1450
1451 static int osd_oxc_get(struct osd_object *obj, const char *name,
1452                        struct lu_buf *buf)
1453 {
1454         struct osd_xattr_entry *tmp;
1455         struct osd_xattr_entry *oxe = NULL;
1456         size_t namelen = strlen(name);
1457         int rc;
1458
1459         rcu_read_lock();
1460         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1461                 if (namelen == tmp->oxe_namelen &&
1462                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1463                         oxe = tmp;
1464                         break;
1465                 }
1466         }
1467
1468         if (oxe == NULL)
1469                 GOTO(out, rc = -ENOENT);
1470
1471         if (!oxe->oxe_exist)
1472                 GOTO(out, rc = -ENODATA);
1473
1474         /* vallen */
1475         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1476         LASSERT(rc > 0);
1477
1478         if (buf->lb_buf == NULL)
1479                 GOTO(out, rc);
1480
1481         if (buf->lb_len < rc)
1482                 GOTO(out, rc = -ERANGE);
1483
1484         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1485 out:
1486         rcu_read_unlock();
1487
1488         return rc;
1489 }
1490
1491 static void osd_oxc_free(struct rcu_head *head)
1492 {
1493         struct osd_xattr_entry *oxe;
1494
1495         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1496         OBD_FREE(oxe, oxe->oxe_len);
1497 }
1498
1499 static void osd_oxc_add(struct osd_object *obj, const char *name,
1500                         const char *buf, int buflen)
1501 {
1502         struct osd_xattr_entry *oxe;
1503         struct osd_xattr_entry *old = NULL;
1504         struct osd_xattr_entry *tmp;
1505         size_t namelen = strlen(name);
1506         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1507
1508         OBD_ALLOC(oxe, len);
1509         if (oxe == NULL)
1510                 return;
1511
1512         INIT_LIST_HEAD(&oxe->oxe_list);
1513         oxe->oxe_len = len;
1514         oxe->oxe_namelen = namelen;
1515         memcpy(oxe->oxe_buf, name, namelen);
1516         if (buflen > 0) {
1517                 LASSERT(buf != NULL);
1518                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1519                 oxe->oxe_exist = true;
1520         } else {
1521                 oxe->oxe_exist = false;
1522         }
1523
1524         /* this should be rarely called, just remove old and add new */
1525         spin_lock(&obj->oo_guard);
1526         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1527                 if (namelen == tmp->oxe_namelen &&
1528                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1529                         old = tmp;
1530                         break;
1531                 }
1532         }
1533         if (old != NULL) {
1534                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1535                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1536         } else {
1537                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1538         }
1539         spin_unlock(&obj->oo_guard);
1540 }
1541
1542 static void osd_oxc_del(struct osd_object *obj, const char *name)
1543 {
1544         struct osd_xattr_entry *oxe;
1545         size_t namelen = strlen(name);
1546
1547         spin_lock(&obj->oo_guard);
1548         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1549                 if (namelen == oxe->oxe_namelen &&
1550                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1551                         list_del_rcu(&oxe->oxe_list);
1552                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1553                         break;
1554                 }
1555         }
1556         spin_unlock(&obj->oo_guard);
1557 }
1558
1559 static void osd_oxc_fini(struct osd_object *obj)
1560 {
1561         struct osd_xattr_entry *oxe, *next;
1562
1563         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1564                 list_del(&oxe->oxe_list);
1565                 OBD_FREE(oxe, oxe->oxe_len);
1566         }
1567 }
1568
1569 /*
1570  * Concurrency: no concurrent access is possible that late in object
1571  * life-cycle.
1572  */
1573 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1574 {
1575         struct osd_object *obj = osd_obj(l);
1576         struct lu_object_header *h = obj->oo_header;
1577
1578         LINVRNT(osd_invariant(obj));
1579
1580         osd_oxc_fini(obj);
1581         dt_object_fini(&obj->oo_dt);
1582         if (obj->oo_hl_head != NULL)
1583                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1584         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1585         OBD_FREE_PTR(obj);
1586         if (unlikely(h))
1587                 lu_object_header_free(h);
1588 }
1589
1590 /*
1591  * Concurrency: no concurrent access is possible that late in object
1592  * life-cycle.
1593  */
1594 static void osd_index_fini(struct osd_object *o)
1595 {
1596         struct iam_container *bag;
1597
1598         if (o->oo_dir != NULL) {
1599                 bag = &o->oo_dir->od_container;
1600                 if (o->oo_inode != NULL) {
1601                         if (bag->ic_object == o->oo_inode)
1602                                 iam_container_fini(bag);
1603                 }
1604                 OBD_FREE_PTR(o->oo_dir);
1605                 o->oo_dir = NULL;
1606         }
1607 }
1608
1609 enum {
1610         OSD_TXN_OI_DELETE_CREDITS    = 20,
1611         OSD_TXN_INODE_DELETE_CREDITS = 20
1612 };
1613
1614 /*
1615  * Journal
1616  */
1617
1618 #if OSD_THANDLE_STATS
1619 /**
1620  * Set time when the handle is allocated
1621  */
1622 static void osd_th_alloced(struct osd_thandle *oth)
1623 {
1624         oth->oth_alloced = ktime_get();
1625 }
1626
1627 /**
1628  * Set time when the handle started
1629  */
1630 static void osd_th_started(struct osd_thandle *oth)
1631 {
1632         oth->oth_started = ktime_get();
1633 }
1634
1635 /**
1636  * Check whether the we deal with this handle for too long.
1637  */
1638 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1639                                 ktime_t alloced, ktime_t started,
1640                                 ktime_t closed)
1641 {
1642         ktime_t now = ktime_get();
1643
1644         LASSERT(dev != NULL);
1645
1646         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1647                             ktime_us_delta(started, alloced));
1648         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1649                             ktime_us_delta(closed, started));
1650         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1651                             ktime_us_delta(now, closed));
1652
1653         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1654                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1655                                 oth, now, alloced, started, closed);
1656                 libcfs_debug_dumpstack(NULL);
1657         }
1658 }
1659
1660 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1661 {                                                                       \
1662         ktime_t __closed = ktime_get();                                 \
1663         ktime_t __alloced = oth->oth_alloced;                           \
1664         ktime_t __started = oth->oth_started;                           \
1665                                                                         \
1666         expr;                                                           \
1667         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1668 }
1669
1670 #else /* OSD_THANDLE_STATS */
1671
1672 #define osd_th_alloced(h)                  do {} while(0)
1673 #define osd_th_started(h)                  do {} while(0)
1674 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1675
1676 #endif /* OSD_THANDLE_STATS */
1677
1678 /*
1679  * Concurrency: doesn't access mutable data.
1680  */
1681 static int osd_param_is_not_sane(const struct osd_device *dev,
1682                                  const struct thandle *th)
1683 {
1684         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1685
1686         return oh->ot_credits > osd_transaction_size(dev);
1687 }
1688
1689 /*
1690  * Concurrency: shouldn't matter.
1691  */
1692 static void osd_trans_commit_cb(struct super_block *sb,
1693                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1694 {
1695         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1696         struct thandle *th = &oh->ot_super;
1697         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1698         struct dt_txn_commit_cb *dcb, *tmp;
1699
1700         LASSERT(oh->ot_handle == NULL);
1701
1702         if (error)
1703                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1704
1705         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1706         /* call per-transaction callbacks if any */
1707         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1708                                  dcb_linkage) {
1709                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1710                          "commit callback entry: magic=%x name='%s'\n",
1711                          dcb->dcb_magic, dcb->dcb_name);
1712                 list_del_init(&dcb->dcb_linkage);
1713                 dcb->dcb_func(NULL, th, dcb, error);
1714         }
1715
1716         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1717         lu_device_put(lud);
1718         th->th_dev = NULL;
1719
1720         OBD_FREE_PTR(oh);
1721 }
1722
1723 static struct thandle *osd_trans_create(const struct lu_env *env,
1724                                         struct dt_device *d)
1725 {
1726         struct osd_thread_info *oti = osd_oti_get(env);
1727         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1728         struct osd_thandle *oh;
1729         struct thandle *th;
1730
1731         ENTRY;
1732
1733         if (d->dd_rdonly) {
1734                 CERROR("%s: someone try to start transaction under "
1735                        "readonly mode, should be disabled.\n",
1736                        osd_name(osd_dt_dev(d)));
1737                 dump_stack();
1738                 RETURN(ERR_PTR(-EROFS));
1739         }
1740
1741         /* on pending IO in this thread should left from prev. request */
1742         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1743
1744         sb_start_write(osd_sb(osd_dt_dev(d)));
1745
1746         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1747         if (!oh) {
1748                 sb_end_write(osd_sb(osd_dt_dev(d)));
1749                 RETURN(ERR_PTR(-ENOMEM));
1750         }
1751
1752         oh->ot_quota_trans = &oti->oti_quota_trans;
1753         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1754         th = &oh->ot_super;
1755         th->th_dev = d;
1756         th->th_result = 0;
1757         oh->ot_credits = 0;
1758         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1759         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1760         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1761         osd_th_alloced(oh);
1762
1763         memset(oti->oti_declare_ops, 0,
1764                sizeof(oti->oti_declare_ops));
1765         memset(oti->oti_declare_ops_cred, 0,
1766                sizeof(oti->oti_declare_ops_cred));
1767         memset(oti->oti_declare_ops_used, 0,
1768                sizeof(oti->oti_declare_ops_used));
1769
1770         oti->oti_ins_cache_depth++;
1771
1772         RETURN(th);
1773 }
1774
1775 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1776 {
1777         struct osd_thread_info *oti = osd_oti_get(env);
1778         struct osd_thandle *oh;
1779
1780         oh = container_of(th, struct osd_thandle, ot_super);
1781         LASSERT(oh != NULL);
1782
1783         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1784               oti->oti_declare_ops[OSD_OT_CREATE],
1785               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1786               oti->oti_declare_ops_used[OSD_OT_CREATE],
1787               oti->oti_declare_ops[OSD_OT_DESTROY],
1788               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1789               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1790         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1791               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1792               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1793               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1794               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1795               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1796               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1797         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1798               oti->oti_declare_ops[OSD_OT_WRITE],
1799               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1800               oti->oti_declare_ops_used[OSD_OT_WRITE],
1801               oti->oti_declare_ops[OSD_OT_PUNCH],
1802               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1803               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1804               oti->oti_declare_ops[OSD_OT_QUOTA],
1805               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1806               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1807         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1808               oti->oti_declare_ops[OSD_OT_INSERT],
1809               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1810               oti->oti_declare_ops_used[OSD_OT_INSERT],
1811               oti->oti_declare_ops[OSD_OT_DELETE],
1812               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1813               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1814         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1815               oti->oti_declare_ops[OSD_OT_REF_ADD],
1816               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1817               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1818               oti->oti_declare_ops[OSD_OT_REF_DEL],
1819               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1820               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1821 }
1822
1823 /*
1824  * Concurrency: shouldn't matter.
1825  */
1826 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1827                            struct thandle *th)
1828 {
1829         struct osd_thread_info *oti = osd_oti_get(env);
1830         struct osd_device *dev = osd_dt_dev(d);
1831         handle_t *jh;
1832         struct osd_thandle *oh;
1833         int rc;
1834
1835         ENTRY;
1836
1837         LASSERT(current->journal_info == NULL);
1838
1839         oh = container_of(th, struct osd_thandle, ot_super);
1840         LASSERT(oh != NULL);
1841         LASSERT(oh->ot_handle == NULL);
1842
1843         rc = dt_txn_hook_start(env, d, th);
1844         if (rc != 0)
1845                 GOTO(out, rc);
1846
1847         if (unlikely(osd_param_is_not_sane(dev, th))) {
1848                 static unsigned long last_printed;
1849                 static int last_credits;
1850
1851                 /*
1852                  * don't make noise on a tiny testing systems
1853                  * actual credits misuse will be caught anyway
1854                  */
1855                 if (last_credits != oh->ot_credits &&
1856                     time_after(jiffies, last_printed +
1857                                cfs_time_seconds(60)) &&
1858                     osd_transaction_size(dev) > 512) {
1859                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1860                               oh->ot_credits, osd_transaction_size(dev));
1861                         osd_trans_dump_creds(env, th);
1862                         libcfs_debug_dumpstack(NULL);
1863                         last_credits = oh->ot_credits;
1864                         last_printed = jiffies;
1865                 }
1866                 /*
1867                  * XXX Limit the credits to 'max_transaction_buffers', and
1868                  *     let the underlying filesystem to catch the error if
1869                  *     we really need so many credits.
1870                  *
1871                  *     This should be removed when we can calculate the
1872                  *     credits precisely.
1873                  */
1874                 oh->ot_credits = osd_transaction_size(dev);
1875         } else if (ldiskfs_track_declares_assert != 0) {
1876                 /*
1877                  * reserve few credits to prevent an assertion in JBD
1878                  * our debugging mechanism will be able to detected
1879                  * overuse. this can help to debug single-update
1880                  * transactions
1881                  */
1882                 oh->ot_credits += 10;
1883                 if (unlikely(osd_param_is_not_sane(dev, th)))
1884                         oh->ot_credits = osd_transaction_size(dev);
1885         }
1886
1887         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1888                 GOTO(out, rc = -EIO);
1889
1890         /*
1891          * XXX temporary stuff. Some abstraction layer should
1892          * be used.
1893          */
1894         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1895         osd_th_started(oh);
1896         if (!IS_ERR(jh)) {
1897                 oh->ot_handle = jh;
1898                 LASSERT(oti->oti_txns == 0);
1899
1900                 lu_device_get(&d->dd_lu_dev);
1901                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1902                               "osd-tx", th);
1903                 oti->oti_txns++;
1904                 rc = 0;
1905         } else {
1906                 rc = PTR_ERR(jh);
1907         }
1908 out:
1909         RETURN(rc);
1910 }
1911
1912 static int osd_seq_exists(const struct lu_env *env,
1913                           struct osd_device *osd, u64 seq)
1914 {
1915         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1916         struct seq_server_site *ss = osd_seq_site(osd);
1917         int rc;
1918
1919         ENTRY;
1920
1921         LASSERT(ss != NULL);
1922         LASSERT(ss->ss_server_fld != NULL);
1923
1924         rc = osd_fld_lookup(env, osd, seq, range);
1925         if (rc != 0) {
1926                 if (rc != -ENOENT)
1927                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1928                                osd_name(osd), seq, rc);
1929                 RETURN(0);
1930         }
1931
1932         RETURN(ss->ss_node_id == range->lsr_index);
1933 }
1934
1935 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1936 {
1937         struct dt_txn_commit_cb *dcb;
1938         struct dt_txn_commit_cb *tmp;
1939
1940         /* call per-transaction stop callbacks if any */
1941         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1942                                  dcb_linkage) {
1943                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1944                          "commit callback entry: magic=%x name='%s'\n",
1945                          dcb->dcb_magic, dcb->dcb_name);
1946                 list_del_init(&dcb->dcb_linkage);
1947                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1948         }
1949 }
1950
1951 /*
1952  * Concurrency: shouldn't matter.
1953  */
1954 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1955                           struct thandle *th)
1956 {
1957         struct osd_thread_info *oti = osd_oti_get(env);
1958         struct osd_thandle *oh;
1959         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1960         struct osd_device *osd = osd_dt_dev(th->th_dev);
1961         struct qsd_instance *qsd = osd_def_qsd(osd);
1962         struct lquota_trans *qtrans;
1963         LIST_HEAD(truncates);
1964         int rc = 0, remove_agents = 0;
1965
1966         ENTRY;
1967
1968         oh = container_of(th, struct osd_thandle, ot_super);
1969
1970         remove_agents = oh->ot_remove_agents;
1971
1972         qtrans = oh->ot_quota_trans;
1973         oh->ot_quota_trans = NULL;
1974
1975         /* move locks to local list, stop tx, execute truncates */
1976         list_splice(&oh->ot_trunc_locks, &truncates);
1977
1978         if (oh->ot_handle != NULL) {
1979                 int rc2;
1980
1981                 handle_t *hdl = oh->ot_handle;
1982
1983                 /*
1984                  * add commit callback
1985                  * notice we don't do this in osd_trans_start()
1986                  * as underlying transaction can change during truncate
1987                  */
1988                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1989                                              &oh->ot_jcb);
1990
1991                 LASSERT(oti->oti_txns == 1);
1992                 oti->oti_txns--;
1993
1994                 rc = dt_txn_hook_stop(env, th);
1995                 if (rc != 0)
1996                         CERROR("%s: failed in transaction hook: rc = %d\n",
1997                                osd_name(osd), rc);
1998
1999                 osd_trans_stop_cb(oh, rc);
2000                 /* hook functions might modify th_sync */
2001                 hdl->h_sync = th->th_sync;
2002
2003                 oh->ot_handle = NULL;
2004                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2005                 if (rc2 != 0)
2006                         CERROR("%s: failed to stop transaction: rc = %d\n",
2007                                osd_name(osd), rc2);
2008                 if (!rc)
2009                         rc = rc2;
2010
2011                 osd_process_truncates(&truncates);
2012         } else {
2013                 osd_trans_stop_cb(oh, th->th_result);
2014                 OBD_FREE_PTR(oh);
2015         }
2016
2017         osd_trunc_unlock_all(env, &truncates);
2018
2019         /* inform the quota slave device that the transaction is stopping */
2020         qsd_op_end(env, qsd, qtrans);
2021
2022         /*
2023          * as we want IO to journal and data IO be concurrent, we don't block
2024          * awaiting data IO completion in osd_do_bio(), instead we wait here
2025          * once transaction is submitted to the journal. all reqular requests
2026          * don't do direct IO (except read/write), thus this wait_event becomes
2027          * no-op for them.
2028          *
2029          * IMPORTANT: we have to wait till any IO submited by the thread is
2030          * completed otherwise iobuf may be corrupted by different request
2031          */
2032         wait_event(iobuf->dr_wait,
2033                        atomic_read(&iobuf->dr_numreqs) == 0);
2034         osd_fini_iobuf(osd, iobuf);
2035         if (!rc)
2036                 rc = iobuf->dr_error;
2037
2038         if (unlikely(remove_agents != 0))
2039                 osd_process_scheduled_agent_removals(env, osd);
2040
2041         oti->oti_ins_cache_depth--;
2042         /* reset OI cache for safety */
2043         if (oti->oti_ins_cache_depth == 0)
2044                 oti->oti_ins_cache_used = 0;
2045
2046         sb_end_write(osd_sb(osd));
2047
2048         RETURN(rc);
2049 }
2050
2051 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2052 {
2053         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2054                                               ot_super);
2055
2056         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2057         LASSERT(&dcb->dcb_func != NULL);
2058         if (dcb->dcb_flags & DCB_TRANS_STOP)
2059                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2060         else
2061                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2062
2063         return 0;
2064 }
2065
2066 /*
2067  * Called just before object is freed. Releases all resources except for
2068  * object itself (that is released by osd_object_free()).
2069  *
2070  * Concurrency: no concurrent access is possible that late in object
2071  * life-cycle.
2072  */
2073 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2074 {
2075         struct osd_object *obj = osd_obj(l);
2076         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2077         struct inode *inode = obj->oo_inode;
2078         __u64 projid;
2079         qid_t uid;
2080         qid_t gid;
2081
2082         LINVRNT(osd_invariant(obj));
2083
2084         /*
2085          * If object is unlinked remove fid->ino mapping from object index.
2086          */
2087
2088         osd_index_fini(obj);
2089
2090         if (!inode)
2091                 return;
2092
2093         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2094                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2095
2096         uid = i_uid_read(inode);
2097         gid = i_gid_read(inode);
2098         projid = i_projid_read(inode);
2099
2100         obj->oo_inode = NULL;
2101         iput(inode);
2102
2103         /* do not rebalance quota if the caller needs to release memory
2104          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2105          * transaction and risk to deadlock - LU-12178 */
2106         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2107                 return;
2108
2109         if (!obj->oo_header && qsd) {
2110                 struct osd_thread_info *info = osd_oti_get(env);
2111                 struct lquota_id_info *qi = &info->oti_qi;
2112
2113                 /* Release granted quota to master if necessary */
2114                 qi->lqi_id.qid_uid = uid;
2115                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2116
2117                 qi->lqi_id.qid_uid = gid;
2118                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2119
2120                 qi->lqi_id.qid_uid = projid;
2121                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2122         }
2123 }
2124
2125 /*
2126  * Concurrency: ->loo_object_release() is called under site spin-lock.
2127  */
2128 static void osd_object_release(const struct lu_env *env,
2129                                struct lu_object *l)
2130 {
2131         struct osd_object *o = osd_obj(l);
2132
2133         /*
2134          * nobody should be releasing a non-destroyed object with nlink=0
2135          * the API allows this, but ldiskfs doesn't like and then report
2136          * this inode as deleted
2137          */
2138         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2139                   o->oo_inode->i_nlink == 0));
2140 }
2141
2142 /*
2143  * Concurrency: shouldn't matter.
2144  */
2145 static int osd_object_print(const struct lu_env *env, void *cookie,
2146                             lu_printer_t p, const struct lu_object *l)
2147 {
2148         struct osd_object *o = osd_obj(l);
2149         struct iam_descr *d;
2150
2151         if (o->oo_dir != NULL)
2152                 d = o->oo_dir->od_container.ic_descr;
2153         else
2154                 d = NULL;
2155         return (*p)(env, cookie,
2156                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2157                     o, o->oo_inode,
2158                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2159                     o->oo_inode ? o->oo_inode->i_generation : 0,
2160                     d ? d->id_ops->id_name : "plain");
2161 }
2162
2163 /*
2164  * Concurrency: shouldn't matter.
2165  */
2166 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2167                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2168 {
2169         struct osd_device *osd = osd_dt_dev(d);
2170         struct super_block *sb = osd_sb(osd);
2171         struct kstatfs *ksfs;
2172         __u64 reserved;
2173         int result = 0;
2174
2175         if (unlikely(osd->od_mnt == NULL))
2176                 return -EINPROGRESS;
2177
2178         /* osd_lproc.c call this without env, allocate ksfs for that case */
2179         if (unlikely(env == NULL)) {
2180                 OBD_ALLOC_PTR(ksfs);
2181                 if (ksfs == NULL)
2182                         return -ENOMEM;
2183         } else {
2184                 ksfs = &osd_oti_get(env)->oti_ksfs;
2185         }
2186
2187         result = sb->s_op->statfs(sb->s_root, ksfs);
2188         if (result)
2189                 goto out;
2190
2191         statfs_pack(sfs, ksfs);
2192         if (unlikely(sb->s_flags & SB_RDONLY))
2193                 sfs->os_state |= OS_STATFS_READONLY;
2194
2195         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2196
2197         if (ldiskfs_has_feature_extents(sb))
2198                 sfs->os_maxbytes = sb->s_maxbytes;
2199         else
2200                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2201
2202         /*
2203          * Reserve some space so to avoid fragmenting the filesystem too much.
2204          * Fragmentation not only impacts performance, but can also increase
2205          * metadata overhead significantly, causing grant calculation to be
2206          * wrong.
2207          *
2208          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2209          */
2210         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2211         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2212         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2213                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2214
2215         sfs->os_blocks -= reserved;
2216         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2217         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2218
2219 out:
2220         if (unlikely(env == NULL))
2221                 OBD_FREE_PTR(ksfs);
2222         return result;
2223 }
2224
2225 /**
2226  * Estimate space needed for file creations. We assume the largest filename
2227  * which is 2^64 - 1, hence a filename of 20 chars.
2228  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2229  */
2230 #ifdef __LDISKFS_DIR_REC_LEN
2231 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2232 #else
2233 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2234 #endif
2235
2236 /*
2237  * Concurrency: doesn't access mutable data.
2238  */
2239 static void osd_conf_get(const struct lu_env *env,
2240                          const struct dt_device *dev,
2241                          struct dt_device_param *param)
2242 {
2243         struct osd_device *d = osd_dt_dev(dev);
2244         struct super_block *sb = osd_sb(d);
2245         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2246         const char *name;
2247         int ea_overhead;
2248
2249         /*
2250          * XXX should be taken from not-yet-existing fs abstraction layer.
2251          */
2252         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2253         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2254         param->ddp_symlink_max  = sb->s_blocksize;
2255         param->ddp_mount_type   = LDD_MT_LDISKFS;
2256         if (ldiskfs_has_feature_extents(sb))
2257                 param->ddp_maxbytes = sb->s_maxbytes;
2258         else
2259                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2260         /*
2261          * inode are statically allocated, so per-inode space consumption
2262          * is the space consumed by the directory entry
2263          */
2264         param->ddp_inodespace     = PER_OBJ_USAGE;
2265         /*
2266          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2267          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2268          * maximum length to not under-count the actual number of extents
2269          * needed for writing a file if there are sub-optimal block allocations.
2270          */
2271         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2272         /* worst-case extent insertion metadata overhead */
2273         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2274         param->ddp_mntopts = 0;
2275         if (test_opt(sb, XATTR_USER))
2276                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2277         if (test_opt(sb, POSIX_ACL))
2278                 param->ddp_mntopts |= MNTOPT_ACL;
2279
2280         /*
2281          * LOD might calculate the max stripe count based on max_ea_size,
2282          * so we need take account in the overhead as well,
2283          * xattr_header + magic + xattr_entry_head
2284          */
2285         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2286                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2287
2288 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2289         if (ldiskfs_has_feature_ea_inode(sb))
2290                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2291                                                                 ea_overhead;
2292         else
2293 #endif
2294                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2295
2296         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2297                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2298
2299         /*
2300          * Preferred RPC size for efficient disk IO.  4MB shows good
2301          * all-around performance for ldiskfs, but use bigalloc chunk size
2302          * by default if larger.
2303          */
2304 #if defined(LDISKFS_CLUSTER_SIZE)
2305         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2306                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2307         else
2308 #endif
2309                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2310
2311         param->ddp_t10_cksum_type = 0;
2312         if (bi) {
2313                 unsigned short interval = blk_integrity_interval(bi);
2314                 name = blk_integrity_name(bi);
2315                 /*
2316                  * Expected values:
2317                  * T10-DIF-TYPE1-CRC
2318                  * T10-DIF-TYPE3-CRC
2319                  * T10-DIF-TYPE1-IP
2320                  * T10-DIF-TYPE3-IP
2321                  */
2322                 if (strncmp(name, "T10-DIF-TYPE",
2323                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2324                         /* also skip "1/3-" at end */
2325                         const int type_off = sizeof("T10-DIF-TYPE.");
2326                         char type_number = name[type_off - 2];
2327
2328                         if (interval != 512 && interval != 4096) {
2329                                 CERROR("%s: unsupported T10PI sector size %u\n",
2330                                        d->od_svname, interval);
2331                         } else if (type_number != '1' && type_number != '3') {
2332                                 CERROR("%s: unsupported T10PI type %s\n",
2333                                        d->od_svname, name);
2334                         } else if (strcmp(name + type_off, "CRC") == 0) {
2335                                 d->od_t10_type = type_number == '1' ?
2336                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2337                                 param->ddp_t10_cksum_type = interval == 512 ?
2338                                         OBD_CKSUM_T10CRC512 :
2339                                         OBD_CKSUM_T10CRC4K;
2340                         } else if (strcmp(name + type_off, "IP") == 0) {
2341                                 d->od_t10_type = type_number == '1' ?
2342                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2343                                 param->ddp_t10_cksum_type = interval == 512 ?
2344                                         OBD_CKSUM_T10IP512 :
2345                                         OBD_CKSUM_T10IP4K;
2346                         } else {
2347                                 CERROR("%s: unsupported checksum type of "
2348                                        "T10PI type '%s'",
2349                                        d->od_svname, name);
2350                         }
2351
2352                 } else {
2353                         CERROR("%s: unsupported T10PI type '%s'",
2354                                d->od_svname, name);
2355                 }
2356         }
2357
2358         param->ddp_has_lseek_data_hole = true;
2359 }
2360
2361 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2362 {
2363         return osd_sb(osd_dt_dev(d));
2364 }
2365
2366 /*
2367  * Concurrency: shouldn't matter.
2368  */
2369 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2370 {
2371         int rc;
2372         struct super_block *s = osd_sb(osd_dt_dev(d));
2373         ENTRY;
2374
2375         down_read(&s->s_umount);
2376         rc = s->s_op->sync_fs(s, 1);
2377         up_read(&s->s_umount);
2378
2379         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2380                rc);
2381
2382         return rc;
2383 }
2384
2385 /**
2386  * Start commit for OSD device.
2387  *
2388  * An implementation of dt_commit_async method for OSD device.
2389  * Asychronously starts underlayng fs sync and thereby a transaction
2390  * commit.
2391  *
2392  * \param env environment
2393  * \param d dt device
2394  *
2395  * \see dt_device_operations
2396  */
2397 static int osd_commit_async(const struct lu_env *env,
2398                             struct dt_device *d)
2399 {
2400         struct super_block *s = osd_sb(osd_dt_dev(d));
2401         int rc;
2402
2403         ENTRY;
2404
2405         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2406         down_read(&s->s_umount);
2407         rc = s->s_op->sync_fs(s, 0);
2408         up_read(&s->s_umount);
2409
2410         RETURN(rc);
2411 }
2412
2413 static int (*priv_security_file_alloc)(struct file *file);
2414
2415 int osd_security_file_alloc(struct file *file)
2416 {
2417         if (priv_security_file_alloc)
2418                 return priv_security_file_alloc(file);
2419         return 0;
2420 }
2421
2422 /*
2423  * Concurrency: shouldn't matter.
2424  */
2425 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2426 {
2427         struct super_block *sb = osd_sb(osd_dt_dev(d));
2428         struct block_device *dev = sb->s_bdev;
2429         int rc = -EOPNOTSUPP;
2430
2431         ENTRY;
2432
2433         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2434                osd_dt_dev(d)->od_svname, (long)dev, rc);
2435
2436         RETURN(rc);
2437 }
2438
2439 /**
2440  * Note: we do not count into QUOTA here.
2441  * If we mount with --data_journal we may need more.
2442  */
2443 const int osd_dto_credits_noquota[DTO_NR] = {
2444         /**
2445          * Insert.
2446          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2447          * SINGLEDATA_TRANS_BLOCKS(8)
2448          * XXX Note: maybe iam need more, since iam have more level than
2449          *           EXT3 htree.
2450          */
2451         [DTO_INDEX_INSERT]  = 16,
2452         /**
2453          * Delete
2454          * just modify a single entry, probably merge few within a block
2455          */
2456         [DTO_INDEX_DELETE]  = 1,
2457         /**
2458          * Used for OI scrub
2459          */
2460         [DTO_INDEX_UPDATE]  = 16,
2461         /**
2462          * 4(inode, inode bits, groups, GDT)
2463          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2464          */
2465         [DTO_OBJECT_CREATE] = 4,
2466         /**
2467          * 4(inode, inode bits, groups, GDT)
2468          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2469          */
2470         [DTO_OBJECT_DELETE] = 4,
2471         /**
2472          * Attr set credits (inode)
2473          */
2474         [DTO_ATTR_SET_BASE] = 1,
2475         /**
2476          * Xattr set. The same as xattr of EXT3.
2477          * DATA_TRANS_BLOCKS(14)
2478          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2479          * are also counted in. Do not know why?
2480          */
2481         [DTO_XATTR_SET]     = 14,
2482         /**
2483          * credits for inode change during write.
2484          */
2485         [DTO_WRITE_BASE]    = 3,
2486         /**
2487          * credits for single block write.
2488          */
2489         [DTO_WRITE_BLOCK]   = 14,
2490         /**
2491          * Attr set credits for chown.
2492          * This is extra credits for setattr, and it is null without quota
2493          */
2494         [DTO_ATTR_SET_CHOWN] = 0
2495 };
2496
2497 static const struct dt_device_operations osd_dt_ops = {
2498         .dt_root_get       = osd_root_get,
2499         .dt_statfs         = osd_statfs,
2500         .dt_trans_create   = osd_trans_create,
2501         .dt_trans_start    = osd_trans_start,
2502         .dt_trans_stop     = osd_trans_stop,
2503         .dt_trans_cb_add   = osd_trans_cb_add,
2504         .dt_conf_get       = osd_conf_get,
2505         .dt_mnt_sb_get     = osd_mnt_sb_get,
2506         .dt_sync           = osd_sync,
2507         .dt_ro             = osd_ro,
2508         .dt_commit_async   = osd_commit_async,
2509 };
2510
2511 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2512                           unsigned int role)
2513 {
2514         struct osd_object *obj = osd_dt_obj(dt);
2515         struct osd_thread_info *oti = osd_oti_get(env);
2516
2517         LINVRNT(osd_invariant(obj));
2518
2519         LASSERT(obj->oo_owner != env);
2520         down_read_nested(&obj->oo_sem, role);
2521
2522         LASSERT(obj->oo_owner == NULL);
2523         oti->oti_r_locks++;
2524 }
2525
2526 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2527                            unsigned int role)
2528 {
2529         struct osd_object *obj = osd_dt_obj(dt);
2530         struct osd_thread_info *oti = osd_oti_get(env);
2531
2532         LINVRNT(osd_invariant(obj));
2533
2534         LASSERT(obj->oo_owner != env);
2535         down_write_nested(&obj->oo_sem, role);
2536
2537         LASSERT(obj->oo_owner == NULL);
2538         obj->oo_owner = env;
2539         oti->oti_w_locks++;
2540 }
2541
2542 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2543 {
2544         struct osd_object *obj = osd_dt_obj(dt);
2545         struct osd_thread_info *oti = osd_oti_get(env);
2546
2547         LINVRNT(osd_invariant(obj));
2548
2549         LASSERT(oti->oti_r_locks > 0);
2550         oti->oti_r_locks--;
2551         up_read(&obj->oo_sem);
2552 }
2553
2554 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2555 {
2556         struct osd_object *obj = osd_dt_obj(dt);
2557         struct osd_thread_info *oti = osd_oti_get(env);
2558
2559         LINVRNT(osd_invariant(obj));
2560
2561         LASSERT(obj->oo_owner == env);
2562         LASSERT(oti->oti_w_locks > 0);
2563         oti->oti_w_locks--;
2564         obj->oo_owner = NULL;
2565         up_write(&obj->oo_sem);
2566 }
2567
2568 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2569 {
2570         struct osd_object *obj = osd_dt_obj(dt);
2571
2572         LINVRNT(osd_invariant(obj));
2573
2574         return obj->oo_owner == env;
2575 }
2576
2577 static void osd_inode_getattr(const struct lu_env *env,
2578                               struct inode *inode, struct lu_attr *attr)
2579 {
2580         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2581                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2582                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2583                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2584
2585         attr->la_atime = inode->i_atime.tv_sec;
2586         attr->la_mtime = inode->i_mtime.tv_sec;
2587         attr->la_ctime = inode->i_ctime.tv_sec;
2588         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2589         attr->la_mode    = inode->i_mode;
2590         attr->la_size    = i_size_read(inode);
2591         attr->la_blocks  = inode->i_blocks;
2592         attr->la_uid     = i_uid_read(inode);
2593         attr->la_gid     = i_gid_read(inode);
2594         attr->la_projid  = i_projid_read(inode);
2595         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2596         attr->la_nlink   = inode->i_nlink;
2597         attr->la_rdev    = inode->i_rdev;
2598         attr->la_blksize = 1 << inode->i_blkbits;
2599         attr->la_blkbits = inode->i_blkbits;
2600         /*
2601          * Ext4 did not transfer inherit flags from raw inode
2602          * to inode flags, and ext4 internally test raw inode
2603          * @i_flags directly. Instead of patching ext4, we do it here.
2604          */
2605         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2606                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2607 }
2608
2609 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2610                             u64 *count)
2611 {
2612         struct osd_object *obj = osd_dt_obj(dt);
2613         const struct dt_it_ops *iops;
2614         struct dt_it *it;
2615         int rc;
2616
2617         ENTRY;
2618
2619         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2620         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2621
2622         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2623                 *count = obj->oo_dirent_count;
2624                 RETURN(0);
2625         }
2626
2627         /* directory not initialized yet */
2628         if (!dt->do_index_ops) {
2629                 *count = 0;
2630                 RETURN(0);
2631         }
2632
2633         iops = &dt->do_index_ops->dio_it;
2634         it = iops->init(env, dt, LUDA_64BITHASH);
2635         if (IS_ERR(it))
2636                 RETURN(PTR_ERR(it));
2637
2638         rc = iops->load(env, it, 0);
2639         if (rc < 0) {
2640                 if (rc == -ENODATA) {
2641                         rc = 0;
2642                         *count = 0;
2643                 }
2644                 GOTO(out, rc);
2645         }
2646         if (rc > 0)
2647                 rc = iops->next(env, it);
2648
2649         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2650                 if (rc == -ESTALE)
2651                         continue;
2652
2653                 if (iops->key_size(env, it) == 0)
2654                         continue;
2655
2656                 (*count)++;
2657         }
2658         if (rc == 1) {
2659                 obj->oo_dirent_count = *count;
2660                 rc = 0;
2661         }
2662 out:
2663         iops->put(env, it);
2664         iops->fini(env, it);
2665
2666         RETURN(rc);
2667 }
2668
2669 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2670                         struct lu_attr *attr)
2671 {
2672         struct osd_object *obj = osd_dt_obj(dt);
2673         int rc = 0;
2674
2675         if (unlikely(!dt_object_exists(dt)))
2676                 return -ENOENT;
2677         if (unlikely(obj->oo_destroyed))
2678                 return -ENOENT;
2679
2680         LASSERT(!dt_object_remote(dt));
2681         LINVRNT(osd_invariant(obj));
2682
2683         spin_lock(&obj->oo_guard);
2684         osd_inode_getattr(env, obj->oo_inode, attr);
2685         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2686                 attr->la_valid |= LA_FLAGS;
2687                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2688         }
2689         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2690                 attr->la_valid |= LA_FLAGS;
2691                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2692         }
2693         spin_unlock(&obj->oo_guard);
2694
2695         if (S_ISDIR(obj->oo_inode->i_mode) &&
2696             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2697                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2698
2699         return rc;
2700 }
2701
2702 static int osd_declare_attr_qid(const struct lu_env *env,
2703                                 struct osd_object *obj,
2704                                 struct osd_thandle *oh, long long bspace,
2705                                 qid_t old_id, qid_t new_id, bool enforce,
2706                                 unsigned int type, bool ignore_edquot)
2707 {
2708         int rc;
2709         struct osd_thread_info *info = osd_oti_get(env);
2710         struct lquota_id_info  *qi = &info->oti_qi;
2711
2712         qi->lqi_type = type;
2713         /* inode accounting */
2714         qi->lqi_is_blk = false;
2715
2716         /* one more inode for the new id ... */
2717         qi->lqi_id.qid_uid = new_id;
2718         qi->lqi_space      = 1;
2719         /* Reserve credits for the new id */
2720         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2721         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2722                 rc = 0;
2723         if (rc)
2724                 RETURN(rc);
2725
2726         /* and one less inode for the current id */
2727         qi->lqi_id.qid_uid = old_id;
2728         qi->lqi_space = -1;
2729         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2730         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2731                 rc = 0;
2732         if (rc)
2733                 RETURN(rc);
2734
2735         /* block accounting */
2736         qi->lqi_is_blk = true;
2737
2738         /* more blocks for the new id ... */
2739         qi->lqi_id.qid_uid = new_id;
2740         qi->lqi_space      = bspace;
2741         /*
2742          * Credits for the new uid has been reserved, re-use "obj"
2743          * to save credit reservation.
2744          */
2745         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2746         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2747                 rc = 0;
2748         if (rc)
2749                 RETURN(rc);
2750
2751         /* and finally less blocks for the current uid */
2752         qi->lqi_id.qid_uid = old_id;
2753         qi->lqi_space      = -bspace;
2754         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2755         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2756                 rc = 0;
2757
2758         RETURN(rc);
2759 }
2760
2761 static int osd_declare_attr_set(const struct lu_env *env,
2762                                 struct dt_object *dt,
2763                                 const struct lu_attr *attr,
2764                                 struct thandle *handle)
2765 {
2766         struct osd_thandle *oh;
2767         struct osd_object *obj;
2768         qid_t uid;
2769         qid_t gid;
2770         long long bspace;
2771         int rc = 0;
2772         bool enforce;
2773
2774         ENTRY;
2775
2776         LASSERT(dt != NULL);
2777         LASSERT(handle != NULL);
2778
2779         obj = osd_dt_obj(dt);
2780         LASSERT(osd_invariant(obj));
2781
2782         oh = container_of(handle, struct osd_thandle, ot_super);
2783         LASSERT(oh->ot_handle == NULL);
2784
2785         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2786                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2787
2788         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2789                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2790
2791         if (attr == NULL || obj->oo_inode == NULL)
2792                 RETURN(rc);
2793
2794         bspace   = obj->oo_inode->i_blocks << 9;
2795         bspace   = toqb(bspace);
2796
2797         /*
2798          * Changing ownership is always preformed by super user, it should not
2799          * fail with EDQUOT unless required explicitly.
2800          *
2801          * We still need to call the osd_declare_qid() to calculate the journal
2802          * credits for updating quota accounting files and to trigger quota
2803          * space adjustment once the operation is completed.
2804          */
2805         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2806                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2807
2808                 if (!ignore_edquot)
2809                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2810                                "(the quota space is %lld)\n",
2811                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2812                                attr->la_gid, bspace);
2813
2814                 /* USERQUOTA */
2815                 uid = i_uid_read(obj->oo_inode);
2816                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2817                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2818                                           attr->la_uid, enforce, USRQUOTA,
2819                                           true);
2820                 if (rc)
2821                         RETURN(rc);
2822
2823                 gid = i_gid_read(obj->oo_inode);
2824                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2825                        attr->la_uid, gid, attr->la_gid);
2826                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2827                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2828                                           attr->la_gid, enforce, GRPQUOTA,
2829                                           ignore_edquot);
2830                 if (rc)
2831                         RETURN(rc);
2832
2833         }
2834 #ifdef HAVE_PROJECT_QUOTA
2835         if (attr->la_valid & LA_PROJID) {
2836                 __u32 projid = i_projid_read(obj->oo_inode);
2837
2838                 enforce = (attr->la_valid & LA_PROJID) &&
2839                                         (attr->la_projid != projid);
2840                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2841                                           (qid_t)projid, (qid_t)attr->la_projid,
2842                                           enforce, PRJQUOTA, true);
2843                 if (rc)
2844                         RETURN(rc);
2845         }
2846 #endif
2847         RETURN(rc);
2848 }
2849
2850 static int osd_inode_setattr(const struct lu_env *env,
2851                              struct inode *inode, const struct lu_attr *attr)
2852 {
2853         __u64 bits = attr->la_valid;
2854
2855         /* Only allow set size for regular file */
2856         if (!S_ISREG(inode->i_mode))
2857                 bits &= ~(LA_SIZE | LA_BLOCKS);
2858
2859         if (bits == 0)
2860                 return 0;
2861
2862         if (bits & LA_ATIME)
2863                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2864         if (bits & LA_CTIME)
2865                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2866         if (bits & LA_MTIME)
2867                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2868         if (bits & LA_SIZE) {
2869                 spin_lock(&inode->i_lock);
2870                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2871                 i_size_write(inode, attr->la_size);
2872                 spin_unlock(&inode->i_lock);
2873         }
2874
2875         /*
2876          * OSD should not change "i_blocks" which is used by quota.
2877          * "i_blocks" should be changed by ldiskfs only.
2878          */
2879         if (bits & LA_MODE)
2880                 inode->i_mode = (inode->i_mode & S_IFMT) |
2881                                 (attr->la_mode & ~S_IFMT);
2882         if (bits & LA_UID)
2883                 i_uid_write(inode, attr->la_uid);
2884         if (bits & LA_GID)
2885                 i_gid_write(inode, attr->la_gid);
2886         if (bits & LA_PROJID)
2887                 i_projid_write(inode, attr->la_projid);
2888         if (bits & LA_NLINK)
2889                 set_nlink(inode, attr->la_nlink);
2890         if (bits & LA_RDEV)
2891                 inode->i_rdev = attr->la_rdev;
2892
2893         if (bits & LA_FLAGS) {
2894                 /* always keep S_NOCMTIME */
2895                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2896                                  S_NOCMTIME;
2897 #if defined(S_ENCRYPTED)
2898                 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2899                  * aware of encryption status. It is just stored into LMA
2900                  * so that it can be forwared to client side.
2901                  */
2902                 inode->i_flags &= ~S_ENCRYPTED;
2903 #endif
2904                 /*
2905                  * Ext4 did not transfer inherit flags from
2906                  * @inode->i_flags to raw inode i_flags when writing
2907                  * flags, we do it explictly here.
2908                  */
2909                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2910                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2911                 else
2912                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2913         }
2914         return 0;
2915 }
2916
2917 #ifdef HAVE_PROJECT_QUOTA
2918 static int osd_transfer_project(struct inode *inode, __u32 projid,
2919                                 struct thandle *handle)
2920 {
2921         struct super_block *sb = inode->i_sb;
2922         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2923         int err;
2924         kprojid_t kprojid;
2925         struct ldiskfs_iloc iloc;
2926         struct ldiskfs_inode *raw_inode;
2927         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2928
2929         if (!ldiskfs_has_feature_project(sb)) {
2930                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2931                         == LDISKFS_DEF_PROJID);
2932                 if (projid != LDISKFS_DEF_PROJID)
2933                         return -EOPNOTSUPP;
2934                 else
2935                         return 0;
2936         }
2937
2938         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2939                 return -EOPNOTSUPP;
2940
2941         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2942         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2943                 return 0;
2944
2945         err = ldiskfs_get_inode_loc(inode, &iloc);
2946         if (err)
2947                 return err;
2948
2949         raw_inode = ldiskfs_raw_inode(&iloc);
2950         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2951                 struct osd_thandle *oh = container_of(handle,
2952                                                       struct osd_thandle,
2953                                                       ot_super);
2954                 /**
2955                  * try to expand inode size automatically.
2956                  */
2957                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
2958                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2959                         err = -EOVERFLOW;
2960                         brelse(iloc.bh);
2961                         return err;
2962                 }
2963         }
2964         brelse(iloc.bh);
2965
2966         dquot_initialize(inode);
2967         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2968         if (transfer_to[PRJQUOTA]) {
2969                 err = __dquot_transfer(inode, transfer_to);
2970                 dqput(transfer_to[PRJQUOTA]);
2971                 if (err)
2972                         return err;
2973         }
2974
2975         return err;
2976 }
2977 #endif
2978
2979 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
2980                               struct thandle *handle)
2981 {
2982         int rc;
2983
2984         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2985             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2986                 struct iattr iattr;
2987
2988                 CDEBUG(D_QUOTA,
2989                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2990                        inode->i_ino, i_uid_read(inode), attr->la_uid,
2991                        i_gid_read(inode), attr->la_gid);
2992
2993                 dquot_initialize(inode);
2994                 iattr.ia_valid = 0;
2995                 if (attr->la_valid & LA_UID)
2996                         iattr.ia_valid |= ATTR_UID;
2997                 if (attr->la_valid & LA_GID)
2998                         iattr.ia_valid |= ATTR_GID;
2999                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3000                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3001
3002                 rc = dquot_transfer(inode, &iattr);
3003                 if (rc) {
3004                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
3005                                "enforcement enabled on the ldiskfs "
3006                                "filesystem?\n", inode->i_sb->s_id, rc);
3007                         return rc;
3008                 }
3009         }
3010
3011         /* Handle project id transfer here properly */
3012         if (attr->la_valid & LA_PROJID &&
3013             attr->la_projid != i_projid_read(inode)) {
3014 #ifdef HAVE_PROJECT_QUOTA
3015                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3016 #else
3017                 rc = -ENOTSUPP;
3018 #endif
3019                 if (rc) {
3020                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3021                                "enforcement enabled on the ldiskfs "
3022                                "filesystem?\n", inode->i_sb->s_id, rc);
3023                         return rc;
3024                 }
3025         }
3026         return 0;
3027 }
3028
3029 static int osd_attr_set(const struct lu_env *env,
3030                         struct dt_object *dt,
3031                         const struct lu_attr *attr,
3032                         struct thandle *handle)
3033 {
3034         struct osd_object *obj = osd_dt_obj(dt);
3035         struct inode *inode;
3036         int rc;
3037
3038         if (!dt_object_exists(dt))
3039                 return -ENOENT;
3040
3041         LASSERT(handle != NULL);
3042         LASSERT(!dt_object_remote(dt));
3043         LASSERT(osd_invariant(obj));
3044
3045         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3046
3047         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3048             !osd_obj2dev(obj)->od_is_ost) {
3049                 struct osd_thread_info *oti = osd_oti_get(env);
3050                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3051                 struct lu_fid *fid1 = &oti->oti_fid;
3052                 struct osd_inode_id *id = &oti->oti_id;
3053                 struct iam_path_descr *ipd;
3054                 struct iam_container *bag;
3055                 struct osd_thandle *oh;
3056                 int rc;
3057
3058                 fid_cpu_to_be(fid1, fid0);
3059                 memset(id, 1, sizeof(*id));
3060                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3061                                   fid0)->oi_dir.od_container;
3062                 ipd = osd_idx_ipd_get(env, bag);
3063                 if (unlikely(ipd == NULL))
3064                         RETURN(-ENOMEM);
3065
3066                 oh = container_of(handle, struct osd_thandle, ot_super);
3067                 rc = iam_update(oh->ot_handle, bag,
3068                                 (const struct iam_key *)fid1,
3069                                 (const struct iam_rec *)id, ipd);
3070                 osd_ipd_put(env, bag, ipd);
3071                 return(rc > 0 ? 0 : rc);
3072         }
3073
3074         inode = obj->oo_inode;
3075
3076         rc = osd_quota_transfer(inode, attr, handle);
3077         if (rc)
3078                 return rc;
3079
3080         spin_lock(&obj->oo_guard);
3081         rc = osd_inode_setattr(env, inode, attr);
3082         spin_unlock(&obj->oo_guard);
3083         if (rc != 0)
3084                 GOTO(out, rc);
3085
3086         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3087
3088         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3089
3090         if (!(attr->la_valid & LA_FLAGS))
3091                 GOTO(out, rc);
3092
3093         /* Let's check if there are extra flags need to be set into LMA */
3094         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3095                 struct osd_thread_info *info = osd_oti_get(env);
3096                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3097
3098                 LASSERT(!obj->oo_pfid_in_lma);
3099
3100                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3101                                  &info->oti_ost_attrs);
3102                 if (rc)
3103                         GOTO(out, rc);
3104
3105                 lma->lma_incompat |=
3106                         lustre_to_lma_flags(attr->la_flags);
3107                 lustre_lma_swab(lma);
3108
3109                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3110
3111                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3112                                      lma, sizeof(*lma), XATTR_REPLACE);
3113                 if (rc != 0) {
3114                         struct osd_device *osd = osd_obj2dev(obj);
3115
3116                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3117                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3118                               lma->lma_incompat, rc);
3119                 } else {
3120                         obj->oo_lma_flags =
3121                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3122                 }
3123                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3124         }
3125 out:
3126
3127         return rc;
3128 }
3129
3130 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3131                                            struct osd_object *obj,
3132                                            const char *name, const int namelen)
3133 {
3134         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3135 }
3136
3137 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3138                       umode_t mode, struct dt_allocation_hint *hint,
3139                       struct thandle *th, struct lu_attr *attr)
3140 {
3141         int result;
3142         struct osd_device *osd = osd_obj2dev(obj);
3143         struct osd_thandle *oth;
3144         struct dt_object *parent = NULL;
3145         struct inode *inode;
3146         uid_t owner[2] = {0, 0};
3147
3148         if (attr->la_valid & LA_UID)
3149                 owner[0] = attr->la_uid;
3150         if (attr->la_valid & LA_GID)
3151                 owner[1] = attr->la_gid;
3152
3153         LINVRNT(osd_invariant(obj));
3154         LASSERT(obj->oo_inode == NULL);
3155         LASSERT(obj->oo_hl_head == NULL);
3156
3157         if (S_ISDIR(mode) && ldiskfs_pdo) {
3158                 obj->oo_hl_head =
3159                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3160                 if (obj->oo_hl_head == NULL)
3161                         return -ENOMEM;
3162         }
3163
3164         oth = container_of(th, struct osd_thandle, ot_super);
3165         LASSERT(oth->ot_handle->h_transaction != NULL);
3166
3167         if (hint != NULL && hint->dah_parent != NULL &&
3168             !dt_object_remote(hint->dah_parent))
3169                 parent = hint->dah_parent;
3170
3171         inode = ldiskfs_create_inode(oth->ot_handle,
3172                                      parent ? osd_dt_obj(parent)->oo_inode :
3173                                               osd_sb(osd)->s_root->d_inode,
3174                                      mode, owner);
3175         if (!IS_ERR(inode)) {
3176                 /* Do not update file c/mtime in ldiskfs. */
3177                 inode->i_flags |= S_NOCMTIME;
3178
3179                 /*
3180                  * For new created object, it must be consistent,
3181                  * and it is unnecessary to scrub against it.
3182                  */
3183                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3184
3185                 obj->oo_inode = inode;
3186                 result = 0;
3187         } else {
3188                 if (obj->oo_hl_head != NULL) {
3189                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3190                         obj->oo_hl_head = NULL;
3191                 }
3192                 result = PTR_ERR(inode);
3193         }
3194         LINVRNT(osd_invariant(obj));
3195         return result;
3196 }
3197
3198 enum {
3199         OSD_NAME_LEN = 255
3200 };
3201
3202 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3203                      struct lu_attr *attr,
3204                      struct dt_allocation_hint *hint,
3205                      struct dt_object_format *dof,
3206                      struct thandle *th)
3207 {
3208         int result;
3209         struct osd_thandle *oth;
3210         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3211
3212         LASSERT(S_ISDIR(attr->la_mode));
3213
3214         oth = container_of(th, struct osd_thandle, ot_super);
3215         LASSERT(oth->ot_handle->h_transaction != NULL);
3216         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3217                 obj->oo_dirent_count = 0;
3218         result = osd_mkfile(info, obj, mode, hint, th, attr);
3219
3220         return result;
3221 }
3222
3223 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3224                         struct lu_attr *attr,
3225                         struct dt_allocation_hint *hint,
3226                         struct dt_object_format *dof,
3227                         struct thandle *th)
3228 {
3229         int result;
3230         struct osd_thandle *oth;
3231         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3232
3233         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3234
3235         LASSERT(S_ISREG(attr->la_mode));
3236
3237         oth = container_of(th, struct osd_thandle, ot_super);
3238         LASSERT(oth->ot_handle->h_transaction != NULL);
3239
3240         result = osd_mkfile(info, obj, mode, hint, th, attr);
3241         if (result == 0) {
3242                 LASSERT(obj->oo_inode != NULL);
3243                 if (feat->dif_flags & DT_IND_VARKEY)
3244                         result = iam_lvar_create(obj->oo_inode,
3245                                                  feat->dif_keysize_max,
3246                                                  feat->dif_ptrsize,
3247                                                  feat->dif_recsize_max,
3248                                                  oth->ot_handle);
3249                 else
3250                         result = iam_lfix_create(obj->oo_inode,
3251                                                  feat->dif_keysize_max,
3252                                                  feat->dif_ptrsize,
3253                                                  feat->dif_recsize_max,
3254                                                  oth->ot_handle);
3255         }
3256         return result;
3257 }
3258
3259 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3260                      struct lu_attr *attr,
3261                      struct dt_allocation_hint *hint,
3262                      struct dt_object_format *dof,
3263                      struct thandle *th)
3264 {
3265         LASSERT(S_ISREG(attr->la_mode));
3266         return osd_mkfile(info, obj, (attr->la_mode &
3267                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3268                           attr);
3269 }
3270
3271 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3272                      struct lu_attr *attr,
3273                      struct dt_allocation_hint *hint,
3274                      struct dt_object_format *dof,
3275                      struct thandle *th)
3276 {
3277         LASSERT(S_ISLNK(attr->la_mode));
3278         return osd_mkfile(info, obj, (attr->la_mode &
3279                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3280                           attr);
3281 }
3282
3283 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3284                      struct lu_attr *attr,
3285                      struct dt_allocation_hint *hint,
3286                      struct dt_object_format *dof,
3287                      struct thandle *th)
3288 {
3289         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3290         int result;
3291
3292         LINVRNT(osd_invariant(obj));
3293         LASSERT(obj->oo_inode == NULL);
3294         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3295                 S_ISFIFO(mode) || S_ISSOCK(mode));
3296
3297         result = osd_mkfile(info, obj, mode, hint, th, attr);
3298         if (result == 0) {
3299                 LASSERT(obj->oo_inode != NULL);
3300                 /*
3301                  * This inode should be marked dirty for i_rdev.  Currently
3302                  * that is done in the osd_attr_init().
3303                  */
3304                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3305                                    attr->la_rdev);
3306         }
3307         LINVRNT(osd_invariant(obj));
3308         return result;
3309 }
3310
3311 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3312                               struct lu_attr *,
3313                               struct dt_allocation_hint *hint,
3314                               struct dt_object_format *dof,
3315                               struct thandle *);
3316
3317 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3318 {
3319         osd_obj_type_f result;
3320
3321         switch (type) {
3322         case DFT_DIR:
3323                 result = osd_mkdir;
3324                 break;
3325         case DFT_REGULAR:
3326                 result = osd_mkreg;
3327                 break;
3328         case DFT_SYM:
3329                 result = osd_mksym;
3330                 break;
3331         case DFT_NODE:
3332                 result = osd_mknod;
3333                 break;
3334         case DFT_INDEX:
3335                 result = osd_mk_index;
3336                 break;
3337
3338         default:
3339                 LBUG();
3340                 break;
3341         }
3342         return result;
3343 }
3344
3345
3346 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3347                         struct dt_object *parent, struct dt_object *child,
3348                         umode_t child_mode)
3349 {
3350         LASSERT(ah);
3351
3352         ah->dah_parent = parent;
3353         ah->dah_mode = child_mode;
3354
3355         if (parent != NULL && !dt_object_remote(parent)) {
3356                 /* will help to find FID->ino at dt_insert("..") */
3357                 struct osd_object *pobj = osd_dt_obj(parent);
3358
3359                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3360         }
3361 }
3362
3363 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3364                           struct lu_attr *attr, struct dt_object_format *dof,
3365                           struct thandle *handle)
3366 {
3367         struct inode *inode = obj->oo_inode;
3368         __u64 valid = attr->la_valid;
3369         int result;
3370
3371         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3372
3373         if (dof->dof_type != DFT_NODE)
3374                 attr->la_valid &= ~LA_RDEV;
3375         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3376                 attr->la_valid &= ~LA_ATIME;
3377         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3378                 attr->la_valid &= ~LA_CTIME;
3379         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3380                 attr->la_valid &= ~LA_MTIME;
3381
3382         result = osd_quota_transfer(inode, attr, handle);
3383         if (result)
3384                 return;
3385
3386         if (attr->la_valid != 0) {
3387                 result = osd_inode_setattr(info->oti_env, inode, attr);
3388                 /*
3389                  * The osd_inode_setattr() should always succeed here.  The
3390                  * only error that could be returned is EDQUOT when we are
3391                  * trying to change the UID or GID of the inode. However, this
3392                  * should not happen since quota enforcement is no longer
3393                  * enabled on ldiskfs (lquota takes care of it).
3394                  */
3395                 LASSERTF(result == 0, "%d\n", result);
3396                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3397         }
3398
3399         attr->la_valid = valid;
3400 }
3401
3402 /**
3403  * Helper function for osd_create()
3404  *
3405  * \retval 0, on success
3406  */
3407 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3408                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3409                         struct dt_object_format *dof, struct thandle *th)
3410 {
3411         int result;
3412         __u32 umask;
3413
3414         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3415
3416         /* we drop umask so that permissions we pass are not affected */
3417         umask = current->fs->umask;
3418         current->fs->umask = 0;
3419
3420         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3421                                                   th);
3422         if (likely(obj->oo_inode != NULL)) {
3423                 LASSERT(obj->oo_inode->i_state & I_NEW);
3424
3425                 /*
3426                  * Unlock the inode before attr initialization to avoid
3427                  * unnecessary dqget operations. LU-6378
3428                  */
3429                 unlock_new_inode(obj->oo_inode);
3430         }
3431
3432         if (likely(result == 0)) {
3433                 osd_attr_init(info, obj, attr, dof, th);
3434                 osd_object_init0(obj);
3435         }
3436
3437         /* restore previous umask value */
3438         current->fs->umask = umask;
3439
3440         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3441
3442         return result;
3443 }
3444
3445 /**
3446  * Helper function for osd_create()
3447  *
3448  * \retval 0, on success
3449  */
3450 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3451                            const struct lu_fid *fid, struct thandle *th)
3452 {
3453         struct osd_thread_info *info = osd_oti_get(env);
3454         struct osd_inode_id    *id   = &info->oti_id;
3455         struct osd_device      *osd  = osd_obj2dev(obj);
3456         struct osd_thandle     *oh;
3457         int rc;
3458
3459         LASSERT(obj->oo_inode != NULL);
3460
3461         oh = container_of(th, struct osd_thandle, ot_super);
3462         LASSERT(oh->ot_handle);
3463         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3464
3465         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3466         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3467                            OI_CHECK_FLD, NULL);
3468         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3469                 struct lu_fid next_fid = *fid;
3470
3471                 /* insert next object in advance, and map to the same inode */
3472                 next_fid.f_oid++;
3473                 if (next_fid.f_oid != 0) {
3474                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3475                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3476                                       OI_CHECK_FLD, NULL);
3477                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3478                 }
3479         }
3480
3481         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3482
3483         return rc;
3484 }
3485
3486 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3487                    u64 seq, struct lu_seq_range *range)
3488 {
3489         struct seq_server_site *ss = osd_seq_site(osd);
3490
3491         if (fid_seq_is_idif(seq)) {
3492                 fld_range_set_ost(range);
3493                 range->lsr_index = idif_ost_idx(seq);
3494                 return 0;
3495         }
3496
3497         if (!fid_seq_in_fldb(seq)) {
3498                 fld_range_set_mdt(range);
3499                 if (ss != NULL)
3500                         /*
3501                          * FIXME: If ss is NULL, it suppose not get lsr_index
3502                          * at all
3503                          */
3504                         range->lsr_index = ss->ss_node_id;
3505                 return 0;
3506         }
3507
3508         LASSERT(ss != NULL);
3509         fld_range_set_any(range);
3510         /* OSD will only do local fld lookup */
3511         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3512 }
3513
3514 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3515                               struct lu_attr *attr,
3516                               struct dt_allocation_hint *hint,
3517                               struct dt_object_format *dof,
3518                               struct thandle *handle)
3519 {
3520         struct osd_thandle *oh;
3521         int rc;
3522
3523         ENTRY;
3524
3525         LASSERT(handle != NULL);
3526
3527         oh = container_of(handle, struct osd_thandle, ot_super);
3528         LASSERT(oh->ot_handle == NULL);
3529
3530         /*
3531          * EA object consumes more credits than regular object: osd_mk_index
3532          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3533          * leaf_node, could involves the block, block bitmap, groups, GDT
3534          * change for each block, so add 4 * 2 credits in that case.
3535          */
3536         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3537                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3538                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3539         /*
3540          * Reuse idle OI block may cause additional one OI block
3541          * to be changed.
3542          */
3543         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3544                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3545         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3546                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3547                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3548
3549         /* will help to find FID->ino mapping at dt_insert() */
3550         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3551                                    osd_dt_obj(dt));
3552         if (rc != 0)
3553                 RETURN(rc);
3554
3555         if (!attr)
3556                 RETURN(0);
3557
3558         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3559                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3560                                    NULL, OSD_QID_INODE);
3561         if (rc != 0)
3562                 RETURN(rc);
3563
3564         RETURN(rc);
3565 }
3566
3567 /**
3568  * Called to destroy on-disk representation of the object
3569  *
3570  * Concurrency: must be locked
3571  */
3572 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3573                                struct thandle *th)
3574 {
3575         struct osd_object *obj = osd_dt_obj(dt);
3576         struct inode *inode = obj->oo_inode;
3577         struct osd_thandle *oh;
3578         int rc;
3579
3580         ENTRY;
3581
3582         if (inode == NULL)
3583                 RETURN(-ENOENT);
3584
3585         oh = container_of(th, struct osd_thandle, ot_super);
3586         LASSERT(oh->ot_handle == NULL);
3587
3588         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3589                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3590
3591         /* For removing agent entry */
3592         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3593                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3594
3595         /*
3596          * Recycle idle OI leaf may cause additional three OI blocks
3597          * to be changed.
3598          */
3599         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3600                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3601                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3602         /* one less inode */
3603         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3604                                    i_projid_read(inode), -1, oh, obj, NULL,
3605                                    OSD_QID_INODE);
3606         if (rc)
3607                 RETURN(rc);
3608         /* data to be truncated */
3609         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3610                                    i_projid_read(inode), 0, oh, obj, NULL,
3611                                    OSD_QID_BLK);
3612         if (rc)
3613                 RETURN(rc);
3614
3615         /*
3616          * will help to find FID->ino when this object is being
3617          * added to PENDING
3618          */
3619         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3620
3621         RETURN(rc);
3622 }
3623
3624 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3625                        struct thandle *th)
3626 {
3627         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3628         struct osd_object *obj = osd_dt_obj(dt);
3629         struct inode *inode = obj->oo_inode;
3630         struct osd_device *osd = osd_obj2dev(obj);
3631         struct osd_thandle *oh;
3632         int result;
3633
3634         ENTRY;
3635
3636         oh = container_of(th, struct osd_thandle, ot_super);
3637         LASSERT(oh->ot_handle);
3638         LASSERT(inode);
3639         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3640
3641         if (unlikely(fid_is_acct(fid)))
3642                 RETURN(-EPERM);
3643
3644         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3645                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3646                 if (result != 0)
3647                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3648                                osd_name(osd), PFID(fid), result);
3649         }
3650
3651         if (S_ISDIR(inode->i_mode)) {
3652                 if (inode->i_nlink > 2)
3653                         CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3654                                osd_name(osd), PFID(fid), inode->i_ino,
3655                                inode->i_nlink);
3656
3657                 spin_lock(&obj->oo_guard);
3658                 clear_nlink(inode);
3659                 spin_unlock(&obj->oo_guard);
3660                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3661         }
3662
3663         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3664
3665         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3666
3667         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3668                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3669                                        oh->ot_handle, OI_CHECK_FLD);
3670
3671         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3672         /* XXX: add to ext3 orphan list */
3673         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3674
3675         /* not needed in the cache anymore */
3676         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3677         obj->oo_destroyed = 1;
3678
3679         RETURN(0);
3680 }
3681
3682 /**
3683  * Put the fid into lustre_mdt_attrs, and then place the structure
3684  * inode's ea. This fid should not be altered during the life time
3685  * of the inode.
3686  *
3687  * \retval +ve, on success
3688  * \retval -ve, on error
3689  *
3690  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3691  */
3692 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3693                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3694 {
3695         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3696         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3697         int rc;
3698
3699         ENTRY;
3700
3701         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3702                 RETURN(0);
3703
3704         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3705                 rc = -ENOMEM;
3706
3707         lustre_loa_init(loa, fid, compat, incompat);
3708         lustre_loa_swab(loa, false);
3709
3710         /*
3711          * For the OST device with 256 bytes inode size by default,
3712          * the PFID EA will be stored together with LMA EA to avoid
3713          * performance trouble. Otherwise the PFID EA can be stored
3714          * independently. LU-8998
3715          */
3716         if ((compat & LMAC_FID_ON_OST) &&
3717             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3718                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3719                                      sizeof(*loa), XATTR_CREATE);
3720         else
3721                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3722                                      sizeof(*lma), XATTR_CREATE);
3723         /*
3724          * LMA may already exist, but we need to check that all the
3725          * desired compat/incompat flags have been added.
3726          */
3727         if (unlikely(rc == -EEXIST)) {
3728                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3729                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3730                 if (rc < 0)
3731                         RETURN(rc);
3732
3733                 if (rc < sizeof(*lma))
3734                         RETURN(-EINVAL);
3735
3736                 lustre_loa_swab(loa, true);
3737                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3738                     ((compat == 0 && incompat == 0) ||
3739                      (!(~lma->lma_compat & compat) &&
3740                       !(~lma->lma_incompat & incompat))))
3741                         RETURN(0);
3742
3743                 lma->lma_self_fid = *fid;
3744                 lma->lma_compat |= compat;
3745                 lma->lma_incompat |= incompat;
3746                 if (rc == sizeof(*lma)) {
3747                         lustre_lma_swab(lma);
3748                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3749                                              sizeof(*lma), XATTR_REPLACE);
3750                 } else {
3751                         lustre_loa_swab(loa, false);
3752                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3753                                              sizeof(*loa), XATTR_REPLACE);
3754                 }
3755         }
3756
3757         RETURN(rc);
3758 }
3759
3760 /**
3761  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3762  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3763  * To have compatilibility with 1.8 ldiskfs driver we need to have
3764  * magic number at start of fid data.
3765  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3766  * its inmemory API.
3767  */
3768 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3769                                          const struct lu_fid *fid)
3770 {
3771         if (!fid_is_namespace_visible(fid) ||
3772             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3773                 param->edp_magic = 0;
3774                 return;
3775         }
3776
3777         param->edp_magic = LDISKFS_LUFID_MAGIC;
3778         param->edp_len =  sizeof(struct lu_fid) + 1;
3779         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3780 }
3781
3782 /**
3783  * Try to read the fid from inode ea into dt_rec.
3784  *
3785  * \param fid object fid.
3786  *
3787  * \retval 0 on success
3788  */
3789 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3790                           __u32 ino, struct lu_fid *fid,
3791                           struct osd_inode_id *id)
3792 {
3793         struct osd_thread_info *info  = osd_oti_get(env);
3794         struct inode *inode;
3795
3796         ENTRY;
3797
3798         osd_id_gen(id, ino, OSD_OII_NOGEN);
3799         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3800         if (IS_ERR(inode))
3801                 RETURN(PTR_ERR(inode));
3802
3803         iput(inode);
3804         RETURN(0);
3805 }
3806
3807 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3808                                         struct inode *dir,
3809                                         struct inode *parent_dir,
3810                                         const struct lu_fid *dot_fid,
3811                                         const struct lu_fid *dot_dot_fid,
3812                                         struct osd_thandle *oth)
3813 {
3814         struct ldiskfs_dentry_param *dot_ldp;
3815         struct ldiskfs_dentry_param *dot_dot_ldp;
3816         __u32 saved_nlink = dir->i_nlink;
3817         int rc;
3818
3819         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3820         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3821
3822         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3823         dot_ldp->edp_magic = 0;
3824
3825         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3826                                     dir, dot_ldp, dot_dot_ldp);
3827         /*
3828          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3829          * the subseqent ref_add() will increase the dir->i_nlink
3830          * as 3. That is incorrect for new created directory.
3831          *
3832          * It looks like hack, because we want to make the OSD API
3833          * to be order-independent for new created directory object
3834          * between dt_insert(..) and ref_add() operations.
3835          *
3836          * Here, we only restore the in-RAM dir-inode's nlink attr,
3837          * becuase if the nlink attr is not 2, then there will be
3838          * ref_add() called following the dt_insert(..), such call
3839          * will make both the in-RAM and on-disk dir-inode's nlink
3840          * attr to be set as 2. LU-7447
3841          */
3842         set_nlink(dir, saved_nlink);
3843         return rc;
3844 }
3845
3846 /**
3847  * Create an local agent inode for remote entry
3848  */
3849 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3850                                                   struct osd_device *osd,
3851                                                   struct osd_object *pobj,
3852                                                   const struct lu_fid *fid,
3853                                                   __u32 type,
3854                                                   struct thandle *th)
3855 {
3856         struct osd_thread_info *info = osd_oti_get(env);
3857         struct inode *local;
3858         struct osd_thandle *oh;
3859         uid_t own[2] = {0, 0};
3860         int rc;
3861
3862         ENTRY;
3863
3864         LASSERT(th);
3865         oh = container_of(th, struct osd_thandle, ot_super);
3866         LASSERT(oh->ot_handle->h_transaction != NULL);
3867
3868         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type, own);
3869         if (IS_ERR(local)) {
3870                 CERROR("%s: create local error %d\n", osd_name(osd),
3871                        (int)PTR_ERR(local));
3872                 RETURN(local);
3873         }
3874
3875         /*
3876          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3877          * correct gid on remote file, not agent here
3878          */
3879         local->i_gid = current_fsgid();
3880         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3881
3882         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3883          * That gives e2fsck something to look at and be happy, and allows
3884          * debugging if we need to determine where this symlink came from.
3885          */
3886         if (S_ISLNK(type)) {
3887                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3888                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3889                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3890
3891                 i_size_write(local, rc);
3892                 LDISKFS_I(local)->i_disksize = rc;
3893         }
3894         unlock_new_inode(local);
3895
3896         /* Agent inode should not have project ID */
3897 #ifdef  HAVE_PROJECT_QUOTA
3898         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3899             i_projid_read(pobj->oo_inode) != 0) {
3900                 rc = osd_transfer_project(local, 0, th);
3901                 if (rc) {
3902                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3903                                "quota enforcement enabled on the ldiskfs "
3904                                "filesystem?\n", local->i_sb->s_id, rc);
3905                         RETURN(ERR_PTR(rc));
3906                 }
3907         }
3908 #endif
3909         /* Set special LMA flag for local agent inode */
3910         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3911         if (rc != 0) {
3912                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3913                        osd_name(osd), PFID(fid), rc);
3914                 RETURN(ERR_PTR(rc));
3915         }
3916
3917         if (!S_ISDIR(type))
3918                 RETURN(local);
3919
3920         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3921                                          lu_object_fid(&pobj->oo_dt.do_lu),
3922                                          fid, oh);
3923         if (rc != 0) {
3924                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3925                         osd_name(osd), PFID(fid), rc);
3926                 RETURN(ERR_PTR(rc));
3927         }
3928
3929         RETURN(local);
3930 }
3931
3932 /**
3933  * when direntry is deleted, we have to take care of possible agent inode
3934  * referenced by that. unfortunately we can't do this at that point:
3935  * iget() within a running transaction leads to deadlock and we better do
3936  * not call that every delete declaration to save performance. so we put
3937  * a potention agent inode on a list and process that once the transaction
3938  * is over. Notice it's not any worse in terms of real orphans as regular
3939  * object destroy doesn't put inodes on the on-disk orphan list. this should
3940  * be addressed separately
3941  */
3942 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3943                                             struct osd_thandle *oh,
3944                                             __u32 ino)
3945 {
3946         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3947         struct osd_obj_orphan *oor;
3948
3949         OBD_ALLOC_PTR(oor);
3950         if (oor == NULL)
3951                 return -ENOMEM;
3952
3953         oor->oor_ino = ino;
3954         oor->oor_env = (struct lu_env *)env;
3955         spin_lock(&osd->od_osfs_lock);
3956         list_add(&oor->oor_list, &osd->od_orphan_list);
3957         spin_unlock(&osd->od_osfs_lock);
3958
3959         oh->ot_remove_agents = 1;
3960
3961         return 0;
3962
3963 }
3964
3965 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3966                                                 struct osd_device *osd)
3967 {
3968         struct osd_thread_info *info = osd_oti_get(env);
3969         struct osd_obj_orphan *oor, *tmp;
3970         struct osd_inode_id id;
3971         LIST_HEAD(list);
3972         struct inode *inode;
3973         struct lu_fid fid;
3974         handle_t *jh;
3975         __u32 ino;
3976
3977         spin_lock(&osd->od_osfs_lock);
3978         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3979                 if (oor->oor_env == env)
3980                         list_move(&oor->oor_list, &list);
3981         }
3982         spin_unlock(&osd->od_osfs_lock);
3983
3984         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3985
3986                 ino = oor->oor_ino;
3987
3988                 list_del(&oor->oor_list);
3989                 OBD_FREE_PTR(oor);
3990
3991                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3992                 inode = osd_iget_fid(info, osd, &id, &fid);
3993                 if (IS_ERR(inode))
3994                         continue;
3995
3996                 if (!osd_remote_fid(env, osd, &fid)) {
3997                         iput(inode);
3998                         continue;
3999                 }
4000
4001                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4002                 clear_nlink(inode);
4003                 mark_inode_dirty(inode);
4004                 ldiskfs_journal_stop(jh);
4005                 iput(inode);
4006         }
4007
4008         return 0;
4009 }
4010
4011 /**
4012  * OSD layer object create function for OST objects (b=11826).
4013  *
4014  * The FID is inserted into inode xattr here.
4015  *
4016  * \retval   0, on success
4017  * \retval -ve, on error
4018  */
4019 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4020                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4021                       struct dt_object_format *dof, struct thandle *th)
4022 {
4023         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4024         struct osd_object *obj = osd_dt_obj(dt);
4025         struct osd_thread_info *info = osd_oti_get(env);
4026         int result, on_ost = 0;
4027
4028         ENTRY;
4029
4030         if (dt_object_exists(dt))
4031                 RETURN(-EEXIST);
4032
4033         LINVRNT(osd_invariant(obj));
4034         LASSERT(!dt_object_remote(dt));
4035         LASSERT(osd_is_write_locked(env, obj));
4036         LASSERT(th != NULL);
4037
4038         if (unlikely(fid_is_acct(fid)))
4039                 /*
4040                  * Quota files can't be created from the kernel any more,
4041                  * 'tune2fs -O quota' will take care of creating them
4042                  */
4043                 RETURN(-EPERM);
4044
4045         result = __osd_create(info, obj, attr, hint, dof, th);
4046         if (result == 0) {
4047                 if (fid_is_idif(fid) &&
4048                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4049                         struct lu_fid *tfid = &info->oti_fid;
4050                         struct ost_id *oi   = &info->oti_ostid;
4051
4052                         fid_to_ostid(fid, oi);
4053                         ostid_to_fid(tfid, oi, 0);
4054                         on_ost = 1;
4055                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4056                                                 LMAC_FID_ON_OST, 0);
4057                 } else {
4058                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4059                                                fid, OI_CHECK_FLD);
4060                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4061                                                 on_ost ? LMAC_FID_ON_OST : 0,
4062                                                 0);
4063                 }
4064                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4065                         obj->oo_dt.do_body_ops = &osd_body_ops;
4066         }
4067
4068         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4069                 result = __osd_oi_insert(env, obj, fid, th);
4070
4071         /*
4072          * a small optimization - dt_insert() isn't usually applied
4073          * to OST objects, so we don't need to cache OI mapping for
4074          * OST objects
4075          */
4076         if (result == 0 && on_ost == 0) {
4077                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4078
4079                 result = osd_idc_find_and_init(env, osd, obj);
4080                 LASSERT(result == 0);
4081         }
4082
4083         LASSERT(ergo(result == 0,
4084                      dt_object_exists(dt) && !dt_object_remote(dt)));
4085         LINVRNT(osd_invariant(obj));
4086         RETURN(result);
4087 }
4088
4089 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4090                                struct thandle *handle)
4091 {
4092         struct osd_thandle *oh;
4093
4094         /* it's possible that object doesn't exist yet */
4095         LASSERT(handle != NULL);
4096
4097         oh = container_of(handle, struct osd_thandle, ot_super);
4098         LASSERT(oh->ot_handle == NULL);
4099
4100         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4101                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4102
4103         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4104
4105         return 0;
4106 }
4107
4108 /*
4109  * Concurrency: @dt is write locked.
4110  */
4111 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4112                        struct thandle *th)
4113 {
4114         struct osd_object *obj = osd_dt_obj(dt);
4115         struct inode *inode = obj->oo_inode;
4116         struct osd_thandle *oh;
4117         int rc = 0;
4118
4119         if (!dt_object_exists(dt) || obj->oo_destroyed)
4120                 return -ENOENT;
4121
4122         LINVRNT(osd_invariant(obj));
4123         LASSERT(!dt_object_remote(dt));
4124         LASSERT(osd_is_write_locked(env, obj));
4125         LASSERT(th != NULL);
4126
4127         oh = container_of(th, struct osd_thandle, ot_super);
4128         LASSERT(oh->ot_handle != NULL);
4129
4130         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4131
4132         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4133                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4134         /*
4135          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4136          * (65000) subdirectories by storing "1" in i_nlink if the link count
4137          * would otherwise overflow. Directory tranversal tools understand
4138          * that (st_nlink == 1) indicates that the filesystem dose not track
4139          * hard links count on the directory, and will not abort subdirectory
4140          * scanning early once (st_nlink - 2) subdirs have been found.
4141          *
4142          * This also has to properly handle the case of inodes with nlink == 0
4143          * in case they are being linked into the PENDING directory
4144          */
4145         spin_lock(&obj->oo_guard);
4146         if (unlikely(inode->i_nlink == 0))
4147                 /* inc_nlink from 0 may cause WARN_ON */
4148                 set_nlink(inode, 1);
4149         else {
4150                 ldiskfs_inc_count(oh->ot_handle, inode);
4151                 if (!S_ISDIR(inode->i_mode))
4152                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4153         }
4154         spin_unlock(&obj->oo_guard);
4155
4156         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4157         LINVRNT(osd_invariant(obj));
4158
4159         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4160
4161         return rc;
4162 }
4163
4164 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4165                                struct thandle *handle)
4166 {
4167         struct osd_thandle *oh;
4168
4169         if (!dt_object_exists(dt))
4170                 return -ENOENT;
4171
4172         LASSERT(!dt_object_remote(dt));
4173         LASSERT(handle != NULL);
4174
4175         oh = container_of(handle, struct osd_thandle, ot_super);
4176         LASSERT(oh->ot_handle == NULL);
4177
4178         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4179                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4180
4181         return 0;
4182 }
4183
4184 /*
4185  * Concurrency: @dt is write locked.
4186  */
4187 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4188                        struct thandle *th)
4189 {
4190         struct osd_object *obj = osd_dt_obj(dt);
4191         struct inode *inode = obj->oo_inode;
4192         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4193         struct osd_thandle *oh;
4194
4195         if (!dt_object_exists(dt))
4196                 return -ENOENT;
4197
4198         LINVRNT(osd_invariant(obj));
4199         LASSERT(!dt_object_remote(dt));
4200         LASSERT(osd_is_write_locked(env, obj));
4201         LASSERT(th != NULL);
4202
4203         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4204                 return -EIO;
4205
4206         oh = container_of(th, struct osd_thandle, ot_super);
4207         LASSERT(oh->ot_handle != NULL);
4208
4209         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4210
4211         spin_lock(&obj->oo_guard);
4212         /*
4213          * That can be result of upgrade from old Lustre version and
4214          * applied only to local files.  Just skip this ref_del call.
4215          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4216          */
4217         if (inode->i_nlink == 0) {
4218                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4219                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4220                              ", maybe an upgraded file? (LU-3915)\n",
4221                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4222                 spin_unlock(&obj->oo_guard);
4223                 return 0;
4224         }
4225
4226         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4227                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4228
4229         ldiskfs_dec_count(oh->ot_handle, inode);
4230         spin_unlock(&obj->oo_guard);
4231
4232         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4233         LINVRNT(osd_invariant(obj));
4234
4235         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4236
4237         return 0;
4238 }
4239
4240 /*
4241  * Concurrency: @dt is read locked.
4242  */
4243 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4244                          struct lu_buf *buf, const char *name)
4245 {
4246         struct osd_object      *obj    = osd_dt_obj(dt);
4247         struct inode           *inode  = obj->oo_inode;
4248         struct osd_thread_info *info   = osd_oti_get(env);
4249         struct dentry          *dentry = &info->oti_obj_dentry;
4250         bool cache_xattr = false;
4251         int rc;
4252
4253         LASSERT(buf);
4254
4255         /* version get is not real XATTR but uses xattr API */
4256         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4257                 dt_obj_version_t *ver = buf->lb_buf;
4258
4259                 /*
4260                  * for version we are just using xattr API but change inode
4261                  * field instead
4262                  */
4263                 if (buf->lb_len == 0)
4264                         return sizeof(dt_obj_version_t);
4265
4266                 if (buf->lb_len < sizeof(dt_obj_version_t))
4267                         return -ERANGE;
4268
4269                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4270                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4271
4272                 *ver = LDISKFS_I(inode)->i_fs_version;
4273
4274                 return sizeof(dt_obj_version_t);
4275         }
4276
4277         if (!dt_object_exists(dt))
4278                 return -ENOENT;
4279
4280         LASSERT(!dt_object_remote(dt));
4281         LASSERT(inode->i_op != NULL);
4282 #ifdef HAVE_IOP_XATTR
4283         LASSERT(inode->i_op->getxattr != NULL);
4284 #endif
4285
4286         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4287             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4288                 cache_xattr = true;
4289
4290         if (cache_xattr) {
4291                 rc = osd_oxc_get(obj, name, buf);
4292                 if (rc != -ENOENT)
4293                         return rc;
4294         }
4295
4296         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4297                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4298                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4299                 struct filter_fid *ff;
4300                 struct ost_layout *ol;
4301
4302                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4303                 if (rc)
4304                         return rc;
4305
4306                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4307
4308                 rc = sizeof(*ff);
4309                 if (buf->lb_len == 0 || !buf->lb_buf)
4310                         return rc;
4311
4312                 if (buf->lb_len < rc)
4313                         return -ERANGE;
4314
4315                 ff = buf->lb_buf;
4316                 ol = &ff->ff_layout;
4317                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4318                                                   PFID_STRIPE_IDX_BITS);
4319                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4320                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4321                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4322                 if (lma->lma_compat & LMAC_COMP_INFO) {
4323                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4324                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4325                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4326                 } else {
4327                         ol->ol_comp_start = 0;
4328                         ol->ol_comp_end = 0;
4329                         ol->ol_comp_id = 0;
4330                 }
4331         } else {
4332                 rc = __osd_xattr_get(inode, dentry, name,
4333                                      buf->lb_buf, buf->lb_len);
4334         }
4335
4336         if (cache_xattr) {
4337                 if (rc == -ENOENT || rc == -ENODATA)
4338                         osd_oxc_add(obj, name, NULL, 0);
4339                 else if (rc > 0 && buf->lb_buf != NULL)
4340                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4341         }
4342
4343         return rc;
4344 }
4345
4346 static int osd_declare_xattr_set(const struct lu_env *env,
4347                                  struct dt_object *dt,
4348                                  const struct lu_buf *buf, const char *name,
4349                                  int fl, struct thandle *handle)
4350 {
4351         struct osd_thandle *oh;
4352         int credits = 0;
4353         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4354
4355         LASSERT(handle != NULL);
4356
4357         oh = container_of(handle, struct osd_thandle, ot_super);
4358         LASSERT(oh->ot_handle == NULL);
4359
4360         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4361                 /*
4362                  * For non-upgrading case, the LMA is set first and
4363                  * usually fit inode. But for upgrade case, the LMA
4364                  * may be in another separated EA block.
4365                  */
4366                 if (dt_object_exists(dt)) {
4367                         if (fl == LU_XATTR_REPLACE)
4368                                 credits = 1;
4369                         else
4370                                 goto upgrade;
4371                 }
4372         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4373                 credits = 1;
4374         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4375                 /* We may need to delete the old PFID EA. */
4376                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4377                 if (fl == LU_XATTR_REPLACE)
4378                         credits += 1;
4379                 else
4380                         goto upgrade;
4381         } else {
4382                 /*
4383                  * If some name entry resides on remote MDT, then will create
4384                  * agent entry under remote parent. On the other hand, if the
4385                  * remote entry will be removed, then related agent entry may
4386                  * need to be removed from the remote parent. So there may be
4387                  * kinds of cases, let's declare enough credits. The credits
4388                  * for create agent entry is enough for remove case.
4389                  */
4390                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4391                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4392                         if (dt_object_exists(dt))
4393                                 credits += 1; /* For updating LMA */
4394                 }
4395
4396 upgrade:
4397                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4398
4399                 if (buf != NULL) {
4400                         ssize_t buflen;
4401
4402                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4403                                 /*
4404                                  * learn xattr size from osd_xattr_get if
4405                                  * attribute has not been read yet
4406                                  */
4407                                 buflen = __osd_xattr_get(
4408                                     osd_dt_obj(dt)->oo_inode,
4409                                     &osd_oti_get(env)->oti_obj_dentry,
4410                                     name, NULL, 0);
4411                                 if (buflen < 0)
4412                                         buflen = 0;
4413                         } else {
4414                                 buflen = buf->lb_len;
4415                         }
4416
4417                         if (buflen > sb->s_blocksize) {
4418                                 credits += osd_calc_bkmap_credits(
4419                                     sb, NULL, 0, -1,
4420                                     (buflen + sb->s_blocksize - 1) >>
4421                                     sb->s_blocksize_bits);
4422                         }
4423                 }
4424                 /*
4425                  * xattr set may involve inode quota change, reserve credits for
4426                  * dquot_initialize()
4427                  */
4428                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4429         }
4430
4431         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4432
4433         return 0;
4434 }
4435
4436 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4437                               const struct lu_buf *buf, int fl,
4438                               struct thandle *handle)
4439 {
4440         struct osd_thread_info *info = osd_oti_get(env);
4441         struct dentry *dentry = &info->oti_obj_dentry;
4442         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4443         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4444         struct inode *inode = obj->oo_inode;
4445         struct filter_fid *ff = buf->lb_buf;
4446         struct ost_layout *ol = &ff->ff_layout;
4447         int flags = XATTR_REPLACE;
4448         int rc;
4449
4450         ENTRY;
4451
4452         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4453                 RETURN(-EINVAL);
4454
4455         rc = osd_get_lma(info, inode, dentry, loa);
4456         if (rc == -ENODATA) {
4457                 /* Usually for upgarding from old device */
4458                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4459                                 LMAC_FID_ON_OST, 0);
4460                 flags = XATTR_CREATE;
4461         } else if (rc) {
4462                 RETURN(rc);
4463         }
4464
4465         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4466                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4467                         RETURN(-EEXIST);
4468
4469                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4470                         /* Separate PFID EA from LMA */
4471                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4472                         lustre_lma_swab(lma);
4473                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4474                                              sizeof(*lma), XATTR_REPLACE);
4475                         if (!rc) {
4476                                 obj->oo_pfid_in_lma = 0;
4477                                 rc = LU_XATTR_CREATE;
4478                         }
4479
4480                         RETURN(rc);
4481                 }
4482         } else {
4483                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4484                         RETURN(fl);
4485
4486                 /*
4487                  * Old client does not send stripe information,
4488                  * then store the PFID EA on disk separatedly.
4489                  */
4490                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4491                              ol->ol_stripe_size == 0))
4492                         RETURN(fl);
4493
4494                 /* Remove old PFID EA entry firstly. */
4495                 dquot_initialize(inode);
4496                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4497                 if (rc == -ENODATA) {
4498                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4499                                 RETURN(rc);
4500                 } else if (rc) {
4501                         RETURN(rc);
4502                 }
4503         }
4504
4505         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4506         if (likely(ol->ol_stripe_size != 0)) {
4507                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4508                                              PFID_STRIPE_IDX_BITS;
4509                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4510                 lma->lma_compat |= LMAC_STRIPE_INFO;
4511                 if (ol->ol_comp_id != 0) {
4512                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4513                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4514                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4515                         lma->lma_compat |= LMAC_COMP_INFO;
4516                 }
4517         }
4518
4519         lustre_loa_swab(loa, false);
4520
4521         /* Store the PFID EA inside LMA. */
4522         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4523                              flags);
4524         if (!rc)
4525                 obj->oo_pfid_in_lma = 1;
4526
4527         RETURN(rc);
4528 }
4529
4530 /*
4531  * In DNE environment, the object (in spite of regular file or directory)
4532  * and its name entry may reside on different MDTs. Under such case, we will
4533  * create an agent entry on the MDT where the object resides. The agent entry
4534  * references the object locally, that makes the object to be visible to the
4535  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4536  * such as 'tar' can handle the object properly.
4537  *
4538  * We handle the agent entry during set linkEA that is the common interface
4539  * for both regular file and directroy, can handle kinds of cases, such as
4540  * create/link/unlink/rename, and so on.
4541  *
4542  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4543  *
4544  * XXX: There are two known issues:
4545  * 1. For one object, we will create at most one agent entry even if there
4546  *    may be more than one cross-MDTs hard links on the object. So the local
4547  *    e2fsck may claim that the object's nlink is larger than the name entries
4548  *    that reference such inode. And in further, the e2fsck will fix the nlink
4549  *    attribute to match the local references. Then it will cause the object's
4550  *    nlink attribute to be inconsistent with the global references. it is bad
4551  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4552  *    other hand, the global namespace LFSCK can repair the object's attribute
4553  *    according to the linkEA.
4554  * 2. There may be too many hard links on the object as to its linkEA overflow,
4555  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4556  *    case happened, then at this point, we do not know whether there are some
4557  *    cross-MDTs reference. But there are local references, it guarantees that
4558  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4559  */
4560 static int osd_xattr_handle_linkea(const struct lu_env *env,
4561                                    struct osd_device *osd,
4562                                    struct osd_object *obj,
4563                                    const struct lu_buf *buf,
4564                                    struct thandle *handle)
4565 {
4566         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4567         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4568         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4569         struct lu_name tmpname;
4570         struct osd_thandle *oh;
4571         int rc;
4572         bool remote = false;
4573
4574         ENTRY;
4575
4576         oh = container_of(handle, struct osd_thandle, ot_super);
4577         LASSERT(oh->ot_handle != NULL);
4578
4579         rc = linkea_init_with_rec(&ldata);
4580         if (!rc) {
4581                 linkea_first_entry(&ldata);
4582                 while (ldata.ld_lee != NULL && !remote) {
4583                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4584                                             &tmpname, tfid);
4585                         if (osd_remote_fid(env, osd, tfid) > 0)
4586                                 remote = true;
4587                         else
4588                                 linkea_next_entry(&ldata);
4589                 }
4590         } else if (rc == -ENODATA) {
4591                 rc = 0;
4592         } else {
4593                 RETURN(rc);
4594         }
4595
4596         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4597                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4598                 if (rc)
4599                         CERROR("%s: failed to remove agent entry for "DFID
4600                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4601         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4602                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4603                 if (rc)
4604                         CERROR("%s: failed to create agent entry for "DFID
4605                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4606         }
4607
4608         RETURN(rc);
4609 }
4610
4611 /*
4612  * Concurrency: @dt is write locked.
4613  */
4614 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4615                          const struct lu_buf *buf, const char *name, int fl,
4616                          struct thandle *handle)
4617 {
4618         struct osd_object *obj = osd_dt_obj(dt);
4619         struct osd_device *osd = osd_obj2dev(obj);
4620         struct inode *inode = obj->oo_inode;
4621         struct osd_thread_info *info = osd_oti_get(env);
4622         int fs_flags = 0;
4623         int len;
4624         int rc;
4625
4626         ENTRY;
4627
4628         LASSERT(handle);
4629         LASSERT(buf);
4630
4631         /* version set is not real XATTR */
4632         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4633                 dt_obj_version_t *version = buf->lb_buf;
4634
4635                 /*
4636                  * for version we are just using xattr API but change inode
4637                  * field instead
4638                  */
4639                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4640
4641                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4642                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4643
4644                 LDISKFS_I(inode)->i_fs_version = *version;
4645                 /*
4646                  * Version is set after all inode operations are finished,
4647                  * so we should mark it dirty here
4648                  */
4649                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4650
4651                 RETURN(0);
4652         }
4653
4654         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4655                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4656
4657         len = buf->lb_len;
4658         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4659
4660         /*
4661          * For the OST device with 256 bytes inode size by default,
4662          * the PFID EA will be stored together with LMA EA to avoid
4663          * performance trouble. Otherwise the PFID EA can be stored
4664          * independently. LU-8998
4665          */
4666         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4667             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4668                 LASSERT(buf->lb_buf);
4669
4670                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4671                 if (fl <= 0)
4672                         RETURN(fl);
4673         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4674                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4675                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4676
4677                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4678                 if (rc)
4679                         RETURN(rc);
4680
4681                 lma->lma_incompat |= LMAI_STRIPED;
4682                 lustre_lma_swab(lma);
4683                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4684                                      sizeof(*lma), XATTR_REPLACE);
4685                 if (rc != 0)
4686                         RETURN(rc);
4687         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4688                 LASSERT(!osd->od_is_ost);
4689
4690                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4691                 if (rc)
4692                         RETURN(rc);
4693         }
4694
4695         if (fl & LU_XATTR_REPLACE)
4696                 fs_flags |= XATTR_REPLACE;
4697
4698         if (fl & LU_XATTR_CREATE)
4699                 fs_flags |= XATTR_CREATE;
4700
4701         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4702         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4703
4704         if (rc == 0 &&
4705             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4706              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4707                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4708
4709         return rc;
4710 }
4711
4712 /*
4713  * Concurrency: @dt is read locked.
4714  */
4715 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4716                           const struct lu_buf *buf)
4717 {
4718         struct osd_object *obj = osd_dt_obj(dt);
4719         struct inode *inode = obj->oo_inode;
4720         struct osd_thread_info *info = osd_oti_get(env);
4721         struct dentry *dentry = &info->oti_obj_dentry;
4722
4723         if (!dt_object_exists(dt))
4724                 return -ENOENT;
4725
4726         LASSERT(!dt_object_remote(dt));
4727         LASSERT(inode->i_op != NULL);
4728         LASSERT(inode->i_op->listxattr != NULL);
4729
4730         dentry->d_inode = inode;
4731         dentry->d_sb = inode->i_sb;
4732         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4733 }
4734
4735 static int osd_declare_xattr_del(const struct lu_env *env,
4736                                  struct dt_object *dt, const char *name,
4737                                  struct thandle *handle)
4738 {
4739         struct osd_thandle *oh;
4740         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4741
4742         LASSERT(!dt_object_remote(dt));
4743         LASSERT(handle != NULL);
4744
4745         oh = container_of(handle, struct osd_thandle, ot_super);
4746         LASSERT(oh->ot_handle == NULL);
4747
4748         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4749                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4750         /*
4751          * xattr del may involve inode quota change, reserve credits for
4752          * dquot_initialize()
4753          */
4754         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4755
4756         return 0;
4757 }
4758
4759 /*
4760  * Concurrency: @dt is write locked.
4761  */
4762 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4763                          const char *name, struct thandle *handle)
4764 {
4765         struct osd_object *obj = osd_dt_obj(dt);
4766         struct inode *inode = obj->oo_inode;
4767         struct osd_thread_info *info = osd_oti_get(env);
4768         struct dentry *dentry = &info->oti_obj_dentry;
4769         int rc;
4770
4771         if (!dt_object_exists(dt))
4772                 return -ENOENT;
4773
4774         LASSERT(!dt_object_remote(dt));
4775         LASSERT(inode->i_op != NULL);
4776         LASSERT(handle != NULL);
4777 #ifdef HAVE_IOP_XATTR
4778         LASSERT(inode->i_op->removexattr != NULL);
4779 #endif
4780
4781         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4782
4783         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4784                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4785
4786                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4787                                  &info->oti_ost_attrs);
4788                 if (!rc) {
4789                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4790
4791                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4792                         lustre_lma_swab(lma);
4793                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4794                                              sizeof(*lma), XATTR_REPLACE);
4795                         if (!rc)
4796                                 obj->oo_pfid_in_lma = 0;
4797                 }
4798         } else {
4799                 dquot_initialize(inode);
4800                 dentry->d_inode = inode;
4801                 dentry->d_sb = inode->i_sb;
4802                 rc = ll_vfs_removexattr(dentry, inode, name);
4803         }
4804
4805         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4806
4807         if (rc == 0 &&
4808             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4809              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4810                 osd_oxc_del(obj, name);
4811
4812         return rc;
4813 }
4814
4815 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4816                            __u64 start, __u64 end)
4817 {
4818         struct osd_object *obj = osd_dt_obj(dt);
4819         struct inode *inode = obj->oo_inode;
4820         struct file *file = osd_quasi_file(env, inode);
4821         int rc;
4822
4823         ENTRY;
4824
4825         rc = vfs_fsync_range(file, start, end, 0);
4826
4827         RETURN(rc);
4828 }
4829
4830 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4831 {
4832         return 0;
4833 }
4834
4835 /*
4836  * Index operations.
4837  */
4838
4839 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4840                                const struct dt_index_features *feat)
4841 {
4842         struct iam_descr *descr;
4843
4844         if (osd_object_is_root(o))
4845                 return feat == &dt_directory_features;
4846
4847         LASSERT(o->oo_dir != NULL);
4848
4849         descr = o->oo_dir->od_container.ic_descr;
4850         if (feat == &dt_directory_features) {
4851                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4852                         return 1;
4853                 else
4854                         return 0;
4855         } else {
4856                 return feat->dif_keysize_min <= descr->id_key_size &&
4857                        descr->id_key_size <= feat->dif_keysize_max &&
4858                        feat->dif_recsize_min <= descr->id_rec_size &&
4859                        descr->id_rec_size <= feat->dif_recsize_max &&
4860                        !(feat->dif_flags & (DT_IND_VARKEY |
4861                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4862                        ergo(feat->dif_flags & DT_IND_UPDATE,
4863                             1 /* XXX check that object (and fs) is writable */);
4864         }
4865 }
4866
4867 static int osd_iam_container_init(const struct lu_env *env,
4868                                   struct osd_object *obj,
4869                                   struct osd_directory *dir)
4870 {
4871         struct iam_container *bag = &dir->od_container;
4872         int result;
4873
4874         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4875         if (result != 0)
4876                 return result;
4877
4878         result = iam_container_setup(bag);
4879         if (result == 0)
4880                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4881         else
4882                 iam_container_fini(bag);
4883
4884         return result;
4885 }
4886
4887
4888 /*
4889  * Concurrency: no external locking is necessary.
4890  */
4891 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4892                          const struct dt_index_features *feat)
4893 {
4894         int result;
4895         int skip_iam = 0;
4896         struct osd_object *obj = osd_dt_obj(dt);
4897
4898         LINVRNT(osd_invariant(obj));
4899
4900         if (osd_object_is_root(obj)) {
4901                 dt->do_index_ops = &osd_index_ea_ops;
4902                 result = 0;
4903         } else if (feat == &dt_directory_features) {
4904                 dt->do_index_ops = &osd_index_ea_ops;
4905                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4906                         result = 0;
4907                 else
4908                         result = -ENOTDIR;
4909                 skip_iam = 1;
4910         } else if (unlikely(feat == &dt_otable_features)) {
4911                 dt->do_index_ops = &osd_otable_ops;
4912                 return 0;
4913         } else if (unlikely(feat == &dt_acct_features)) {
4914                 dt->do_index_ops = &osd_acct_index_ops;
4915                 result = 0;
4916                 skip_iam = 1;
4917         } else if (!osd_has_index(obj)) {
4918                 struct osd_directory *dir;
4919                 struct osd_device *osd = osd_obj2dev(obj);
4920                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4921
4922                 OBD_ALLOC_PTR(dir);
4923                 if (dir) {
4924
4925                         spin_lock(&obj->oo_guard);
4926                         if (obj->oo_dir == NULL)
4927                                 obj->oo_dir = dir;
4928                         else
4929                                 /*
4930                                  * Concurrent thread allocated container data.
4931                                  */
4932                                 OBD_FREE_PTR(dir);
4933                         spin_unlock(&obj->oo_guard);
4934                         /*
4935                          * Now, that we have container data, serialize its
4936                          * initialization.
4937                          */
4938                         down_write(&obj->oo_ext_idx_sem);
4939                         /*
4940                          * recheck under lock.
4941                          */
4942
4943                         if (osd_has_index(obj)) {
4944                                 result = 0;
4945                                 goto unlock;
4946                         }
4947
4948                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4949                         if (result || feat == &dt_lfsck_namespace_features ||
4950                             feat == &dt_lfsck_layout_orphan_features ||
4951                             feat == &dt_lfsck_layout_dangling_features)
4952                                 goto unlock;
4953
4954                         result = osd_index_register(osd, fid,
4955                                                     feat->dif_keysize_max,
4956                                                     feat->dif_recsize_max);
4957                         if (result < 0)
4958                                 CWARN("%s: failed to register index "
4959                                       DFID": rc = %d\n",
4960                                       osd_name(osd), PFID(fid), result);
4961                         else if (result > 0)
4962                                 result = 0;
4963                         else
4964                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4965                                        " (%d/%d) registered\n",
4966                                        osd_name(osd), PFID(fid),
4967                                        (int)feat->dif_keysize_max,
4968                                        (int)feat->dif_recsize_max);
4969
4970 unlock:
4971                         up_write(&obj->oo_ext_idx_sem);
4972                 } else {
4973                         result = -ENOMEM;
4974                 }
4975         } else {
4976                 result = 0;
4977         }
4978
4979         if (result == 0 && skip_iam == 0) {
4980                 if (!osd_iam_index_probe(env, obj, feat))
4981                         result = -ENOTDIR;
4982         }
4983         LINVRNT(osd_invariant(obj));
4984
4985         return result;
4986 }
4987
4988 static int osd_otable_it_attr_get(const struct lu_env *env,
4989                                  struct dt_object *dt,
4990                                  struct lu_attr *attr)
4991 {
4992         attr->la_valid = 0;
4993         return 0;
4994 }
4995
4996 static const struct dt_object_operations osd_obj_ops = {
4997         .do_read_lock           = osd_read_lock,
4998         .do_write_lock          = osd_write_lock,
4999         .do_read_unlock         = osd_read_unlock,
5000         .do_write_unlock        = osd_write_unlock,
5001         .do_write_locked        = osd_write_locked,
5002         .do_attr_get            = osd_attr_get,
5003         .do_declare_attr_set    = osd_declare_attr_set,
5004         .do_attr_set            = osd_attr_set,
5005         .do_ah_init             = osd_ah_init,
5006         .do_declare_create      = osd_declare_create,
5007         .do_create              = osd_create,
5008         .do_declare_destroy     = osd_declare_destroy,
5009         .do_destroy             = osd_destroy,
5010         .do_index_try           = osd_index_try,
5011         .do_declare_ref_add     = osd_declare_ref_add,
5012         .do_ref_add             = osd_ref_add,
5013         .do_declare_ref_del     = osd_declare_ref_del,
5014         .do_ref_del             = osd_ref_del,
5015         .do_xattr_get           = osd_xattr_get,
5016         .do_declare_xattr_set   = osd_declare_xattr_set,
5017         .do_xattr_set           = osd_xattr_set,
5018         .do_declare_xattr_del   = osd_declare_xattr_del,
5019         .do_xattr_del           = osd_xattr_del,
5020         .do_xattr_list          = osd_xattr_list,
5021         .do_object_sync         = osd_object_sync,
5022         .do_invalidate          = osd_invalidate,
5023 };
5024
5025 static const struct dt_object_operations osd_obj_otable_it_ops = {
5026         .do_attr_get    = osd_otable_it_attr_get,
5027         .do_index_try   = osd_index_try,
5028 };
5029
5030 static int osd_index_declare_iam_delete(const struct lu_env *env,
5031                                         struct dt_object *dt,
5032                                         const struct dt_key *key,
5033                                         struct thandle *handle)
5034 {
5035         struct osd_thandle *oh;
5036
5037         oh = container_of(handle, struct osd_thandle, ot_super);
5038         LASSERT(oh->ot_handle == NULL);
5039
5040         /* Recycle  may cause additional three blocks to be changed. */
5041         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5042                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5043
5044         return 0;
5045 }
5046
5047 /**
5048  *      delete a (key, value) pair from index \a dt specified by \a key
5049  *
5050  *      \param  dt      osd index object
5051  *      \param  key     key for index
5052  *      \param  rec     record reference
5053  *      \param  handle  transaction handler
5054  *
5055  *      \retval  0  success
5056  *      \retval -ve   failure
5057  */
5058 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5059                                 const struct dt_key *key,
5060                                 struct thandle *handle)
5061 {
5062         struct osd_thread_info *oti = osd_oti_get(env);
5063         struct osd_object *obj = osd_dt_obj(dt);
5064         struct osd_thandle *oh;
5065         struct iam_path_descr *ipd;
5066         struct iam_container *bag = &obj->oo_dir->od_container;
5067         int rc;
5068
5069         ENTRY;
5070
5071         if (!dt_object_exists(dt))
5072                 RETURN(-ENOENT);
5073
5074         LINVRNT(osd_invariant(obj));
5075         LASSERT(!dt_object_remote(dt));
5076         LASSERT(bag->ic_object == obj->oo_inode);
5077         LASSERT(handle != NULL);
5078
5079         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5080
5081         ipd = osd_idx_ipd_get(env, bag);
5082         if (unlikely(ipd == NULL))
5083                 RETURN(-ENOMEM);
5084
5085         oh = container_of(handle, struct osd_thandle, ot_super);
5086         LASSERT(oh->ot_handle != NULL);
5087         LASSERT(oh->ot_handle->h_transaction != NULL);
5088
5089         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5090                 /* swab quota uid/gid provided by caller */
5091                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5092                 key = (const struct dt_key *)&oti->oti_quota_id;
5093         }
5094
5095         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5096         osd_ipd_put(env, bag, ipd);
5097         LINVRNT(osd_invariant(obj));
5098         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5099         RETURN(rc);
5100 }
5101
5102 static int osd_index_declare_ea_delete(const struct lu_env *env,
5103                                        struct dt_object *dt,
5104                                        const struct dt_key *key,
5105                                        struct thandle *handle)
5106 {
5107         struct osd_thandle *oh;
5108         struct inode *inode;
5109         int rc, credits;
5110
5111         ENTRY;
5112
5113         LASSERT(!dt_object_remote(dt));
5114         LASSERT(handle != NULL);
5115
5116         oh = container_of(handle, struct osd_thandle, ot_super);
5117         LASSERT(oh->ot_handle == NULL);
5118
5119         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5120         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5121
5122         inode = osd_dt_obj(dt)->oo_inode;
5123         if (inode == NULL)
5124                 RETURN(-ENOENT);
5125
5126         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5127                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5128                                    NULL, OSD_QID_BLK);
5129         RETURN(rc);
5130 }
5131
5132 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5133                                           struct dt_rec *fid)
5134 {
5135         struct osd_fid_pack *rec;
5136         int rc = -ENODATA;
5137
5138         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5139                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5140                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5141                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5142                         rc = -EINVAL;
5143         }
5144         return rc;
5145 }
5146
5147 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5148                           const struct lu_fid *fid)
5149 {
5150         struct seq_server_site *ss = osd_seq_site(osd);
5151
5152         ENTRY;
5153
5154         /* FID seqs not in FLDB, must be local seq */
5155         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5156                 RETURN(0);
5157
5158         /*
5159          * If FLD is not being initialized yet, it only happens during the
5160          * initialization, likely during mgs initialization, and we assume
5161          * this is local FID.
5162          */
5163         if (ss == NULL || ss->ss_server_fld == NULL)
5164                 RETURN(0);
5165
5166         /* Only check the local FLDB here */
5167         if (osd_seq_exists(env, osd, fid_seq(fid)))
5168                 RETURN(0);
5169
5170         RETURN(1);
5171 }
5172
5173 static void osd_take_care_of_agent(const struct lu_env *env,
5174                                    struct osd_device *osd,
5175                                    struct osd_thandle *oh,
5176                                    struct ldiskfs_dir_entry_2 *de)
5177 {
5178         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5179         struct osd_idmap_cache *idc;
5180         int rc, schedule = 0;
5181
5182         LASSERT(de != NULL);
5183
5184         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5185         if (likely(rc == 0)) {
5186                 idc = osd_idc_find_or_init(env, osd, fid);
5187                 if (IS_ERR(idc) || idc->oic_remote)
5188                         schedule = 1;
5189         } else if (rc == -ENODATA) {
5190                 /*
5191                  * can't get FID, postpone to the end of the
5192                  * transaction when iget() is safe
5193                  */
5194                 schedule = 1;
5195         } else {
5196                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5197         }
5198         if (schedule)
5199                 osd_schedule_agent_inode_removal(env, oh,
5200                                                  le32_to_cpu(de->inode));
5201 }
5202
5203 /**
5204  * Index delete function for interoperability mode (b11826).
5205  * It will remove the directory entry added by osd_index_ea_insert().
5206  * This entry is needed to maintain name->fid mapping.
5207  *
5208  * \param key,  key i.e. file entry to be deleted
5209  *
5210  * \retval   0, on success
5211  * \retval -ve, on error
5212  */
5213 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5214                                const struct dt_key *key, struct thandle *handle)
5215 {
5216         struct osd_object *obj = osd_dt_obj(dt);
5217         struct inode *dir = obj->oo_inode;
5218         struct dentry *dentry;
5219         struct osd_thandle *oh;
5220         struct ldiskfs_dir_entry_2 *de = NULL;
5221         struct buffer_head *bh;
5222         struct htree_lock *hlock = NULL;
5223         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5224         int rc;
5225
5226         ENTRY;
5227
5228         if (!dt_object_exists(dt))
5229                 RETURN(-ENOENT);
5230
5231         LINVRNT(osd_invariant(obj));
5232         LASSERT(!dt_object_remote(dt));
5233         LASSERT(handle != NULL);
5234
5235         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5236
5237         oh = container_of(handle, struct osd_thandle, ot_super);
5238         LASSERT(oh->ot_handle != NULL);
5239         LASSERT(oh->ot_handle->h_transaction != NULL);
5240
5241         dquot_initialize(dir);
5242         dentry = osd_child_dentry_get(env, obj,
5243                                       (char *)key, strlen((char *)key));
5244
5245         if (obj->oo_hl_head != NULL) {
5246                 hlock = osd_oti_get(env)->oti_hlock;
5247                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5248                                    dir, LDISKFS_HLOCK_DEL);
5249         } else {
5250                 down_write(&obj->oo_ext_idx_sem);
5251         }
5252
5253         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5254         if (!IS_ERR(bh)) {
5255                 /*
5256                  * If this is not the ".." entry, it might be a remote DNE
5257                  * entry and  we need to check if the FID is for a remote
5258                  * MDT.  If the FID is  not in the directory entry (e.g.
5259                  * upgraded 1.8 filesystem without dirdata enabled) then
5260                  * we need to get the FID from the LMA. For a remote directory
5261                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5262                  * case.
5263                  *
5264                  * Delete the entry before the agent inode in order to
5265                  * simplify error handling.  At worst an error after deleting
5266                  * the entry first might leak the agent inode afterward. The
5267                  * reverse would need filesystem abort in case of error deleting
5268                  * the entry after the agent had been removed, or leave a
5269                  * dangling entry pointing at a random inode.
5270                  */
5271                 if (strcmp((char *)key, dotdot) != 0)
5272                         osd_take_care_of_agent(env, osd, oh, de);
5273                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5274                 brelse(bh);
5275         } else {
5276                 rc = PTR_ERR(bh);
5277         }
5278
5279         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5280             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5281                 /* NB, dirent count may not be accurate, because it's counted
5282                  * without lock.
5283                  */
5284                 if (obj->oo_dirent_count)
5285                         obj->oo_dirent_count--;
5286                 else
5287                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5288         }
5289         if (hlock != NULL)
5290                 ldiskfs_htree_unlock(hlock);
5291         else
5292                 up_write(&obj->oo_ext_idx_sem);
5293         GOTO(out, rc);
5294 out:
5295         LASSERT(osd_invariant(obj));
5296         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5297         RETURN(rc);
5298 }
5299
5300 /**
5301  *      Lookup index for \a key and copy record to \a rec.
5302  *
5303  *      \param  dt      osd index object
5304  *      \param  key     key for index
5305  *      \param  rec     record reference
5306  *
5307  *      \retval  +ve  success : exact mach
5308  *      \retval  0    return record with key not greater than \a key
5309  *      \retval -ve   failure
5310  */
5311 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5312                                 struct dt_rec *rec, const struct dt_key *key)
5313 {
5314         struct osd_object *obj = osd_dt_obj(dt);
5315         struct iam_path_descr *ipd;
5316         struct iam_container *bag = &obj->oo_dir->od_container;
5317         struct osd_thread_info *oti = osd_oti_get(env);
5318         struct iam_iterator *it = &oti->oti_idx_it;
5319         struct iam_rec *iam_rec;
5320         int rc;
5321
5322         ENTRY;
5323
5324         if (!dt_object_exists(dt))
5325                 RETURN(-ENOENT);
5326
5327         LASSERT(osd_invariant(obj));
5328         LASSERT(!dt_object_remote(dt));
5329         LASSERT(bag->ic_object == obj->oo_inode);
5330
5331         ipd = osd_idx_ipd_get(env, bag);
5332         if (IS_ERR(ipd))
5333                 RETURN(-ENOMEM);
5334
5335         /* got ipd now we can start iterator. */
5336         iam_it_init(it, bag, 0, ipd);
5337
5338         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5339                 /* swab quota uid/gid provided by caller */
5340                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5341                 key = (const struct dt_key *)&oti->oti_quota_id;
5342         }
5343
5344         rc = iam_it_get(it, (struct iam_key *)key);
5345         if (rc >= 0) {
5346                 if (S_ISDIR(obj->oo_inode->i_mode))
5347                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5348                 else
5349                         iam_rec = (struct iam_rec *)rec;
5350
5351                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5352
5353                 if (S_ISDIR(obj->oo_inode->i_mode))
5354                         osd_fid_unpack((struct lu_fid *)rec,
5355                                        (struct osd_fid_pack *)iam_rec);
5356                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5357                         osd_quota_unpack(obj, rec);
5358         }
5359
5360         iam_it_put(it);
5361         iam_it_fini(it);
5362         osd_ipd_put(env, bag, ipd);
5363
5364         LINVRNT(osd_invariant(obj));
5365
5366         RETURN(rc);
5367 }
5368
5369 static int osd_index_declare_iam_insert(const struct lu_env *env,
5370                                         struct dt_object *dt,
5371                                         const struct dt_rec *rec,
5372                                         const struct dt_key *key,
5373                                         struct thandle *handle)
5374 {
5375         struct osd_thandle *oh;
5376
5377         LASSERT(handle != NULL);
5378
5379         oh = container_of(handle, struct osd_thandle, ot_super);
5380         LASSERT(oh->ot_handle == NULL);
5381
5382         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5383                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5384
5385         return 0;
5386 }
5387
5388 /**
5389  *      Inserts (key, value) pair in \a dt index object.
5390  *
5391  *      \param  dt      osd index object
5392  *      \param  key     key for index
5393  *      \param  rec     record reference
5394  *      \param  th      transaction handler
5395  *
5396  *      \retval  0  success
5397  *      \retval -ve failure
5398  */
5399 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5400                                 const struct dt_rec *rec,
5401                                 const struct dt_key *key, struct thandle *th)
5402 {
5403         struct osd_object *obj = osd_dt_obj(dt);
5404         struct iam_path_descr *ipd;
5405         struct osd_thandle *oh;
5406         struct iam_container *bag;
5407         struct osd_thread_info *oti = osd_oti_get(env);
5408         struct iam_rec *iam_rec;
5409         int rc;
5410
5411         ENTRY;
5412
5413         if (!dt_object_exists(dt))
5414                 RETURN(-ENOENT);
5415
5416         LINVRNT(osd_invariant(obj));
5417         LASSERT(!dt_object_remote(dt));
5418
5419         bag = &obj->oo_dir->od_container;
5420         LASSERT(bag->ic_object == obj->oo_inode);
5421         LASSERT(th != NULL);
5422
5423         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5424
5425         ipd = osd_idx_ipd_get(env, bag);
5426         if (unlikely(ipd == NULL))
5427                 RETURN(-ENOMEM);
5428
5429         oh = container_of(th, struct osd_thandle, ot_super);
5430         LASSERT(oh->ot_handle != NULL);
5431         LASSERT(oh->ot_handle->h_transaction != NULL);
5432         if (S_ISDIR(obj->oo_inode->i_mode)) {
5433                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5434                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5435                              &oti->oti_fid);
5436         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5437                 /* pack quota uid/gid */
5438                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5439                 key = (const struct dt_key *)&oti->oti_quota_id;
5440                 /* pack quota record */
5441                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5442                 iam_rec = (struct iam_rec *)rec;
5443         } else {
5444                 iam_rec = (struct iam_rec *)rec;
5445         }
5446
5447         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5448                         iam_rec, ipd);
5449         osd_ipd_put(env, bag, ipd);
5450         LINVRNT(osd_invariant(obj));
5451         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5452         RETURN(rc);
5453 }
5454
5455 /**
5456  * Calls ldiskfs_add_entry() to add directory entry
5457  * into the directory. This is required for
5458  * interoperability mode (b11826)
5459  *
5460  * \retval   0, on success
5461  * \retval -ve, on error
5462  */
5463 static int __osd_ea_add_rec(struct osd_thread_info *info,
5464                             struct osd_object *pobj, struct inode  *cinode,
5465                             const char *name, const struct lu_fid *fid,
5466                             struct htree_lock *hlock, struct thandle *th)
5467 {
5468         struct ldiskfs_dentry_param *ldp;
5469         struct dentry *child;
5470         struct osd_thandle *oth;
5471         int rc;
5472
5473         oth = container_of(th, struct osd_thandle, ot_super);
5474         LASSERT(oth->ot_handle != NULL);
5475         LASSERT(oth->ot_handle->h_transaction != NULL);
5476         LASSERT(pobj->oo_inode);
5477
5478         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5479         if (unlikely(pobj->oo_inode ==
5480                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5481                 ldp->edp_magic = 0;
5482         else
5483                 osd_get_ldiskfs_dirent_param(ldp, fid);
5484         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5485         child->d_fsdata = (void *)ldp;
5486         dquot_initialize(pobj->oo_inode);
5487         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5488                                    child, cinode, hlock);
5489         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5490                 struct ldiskfs_dir_entry_2      *de;
5491                 struct buffer_head              *bh;
5492                 int                              rc1;
5493
5494                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5495                                             NULL, hlock);
5496                 if (!IS_ERR(bh)) {
5497                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5498                                                                bh);
5499                         if (rc1 == 0) {
5500                                 if (S_ISDIR(cinode->i_mode))
5501                                         de->file_type = LDISKFS_DIRENT_LUFID |
5502                                                         LDISKFS_FT_REG_FILE;
5503                                 else
5504                                         de->file_type = LDISKFS_DIRENT_LUFID |
5505                                                         LDISKFS_FT_DIR;
5506                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5507                                                               NULL, bh);
5508                         }
5509                         brelse(bh);
5510                 }
5511         }
5512
5513         RETURN(rc);
5514 }
5515
5516 /**
5517  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5518  * into the directory.Also sets flags into osd object to
5519  * indicate dot and dotdot are created. This is required for
5520  * interoperability mode (b11826)
5521  *
5522  * \param dir   directory for dot and dotdot fixup.
5523  * \param obj   child object for linking
5524  *
5525  * \retval   0, on success
5526  * \retval -ve, on error
5527  */
5528 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5529                               struct osd_object *dir,
5530                               struct inode *parent_dir, const char *name,
5531                               const struct lu_fid *dot_fid,
5532                               const struct lu_fid *dot_dot_fid,
5533                               struct thandle *th)
5534 {
5535         struct inode *inode = dir->oo_inode;
5536         struct osd_thandle *oth;
5537         int result = 0;
5538
5539         oth = container_of(th, struct osd_thandle, ot_super);
5540         LASSERT(oth->ot_handle->h_transaction != NULL);
5541         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5542
5543         if (strcmp(name, dot) == 0) {
5544                 if (dir->oo_compat_dot_created) {
5545                         result = -EEXIST;
5546                 } else {
5547                         LASSERT(inode->i_ino == parent_dir->i_ino);
5548                         dir->oo_compat_dot_created = 1;
5549                         result = 0;
5550                 }
5551         } else if (strcmp(name, dotdot) == 0) {
5552                 if (!dir->oo_compat_dot_created)
5553                         return -EINVAL;
5554                 /* in case of rename, dotdot is already created */
5555                 if (dir->oo_compat_dotdot_created) {
5556                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5557                                                 dot_dot_fid, NULL, th);
5558                 }
5559
5560                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5561                         struct lu_fid tfid = *dot_dot_fid;
5562
5563                         tfid.f_oid--;
5564                         result = osd_add_dot_dotdot_internal(info,
5565                                         dir->oo_inode, parent_dir, dot_fid,
5566                                         &tfid, oth);
5567                 } else {
5568                         result = osd_add_dot_dotdot_internal(info,
5569                                         dir->oo_inode, parent_dir, dot_fid,
5570                                         dot_dot_fid, oth);
5571                 }
5572
5573                 if (result == 0)
5574                         dir->oo_compat_dotdot_created = 1;
5575         }
5576
5577         return result;
5578 }
5579
5580
5581 /**
5582  * It will call the appropriate osd_add* function and return the
5583  * value, return by respective functions.
5584  */
5585 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5586                           struct inode *cinode, const char *name,
5587                           const struct lu_fid *fid, struct thandle *th)
5588 {
5589         struct osd_thread_info *info = osd_oti_get(env);
5590         struct htree_lock *hlock;
5591         int rc;
5592
5593         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5594
5595         if (name[0] == '.' && (name[1] == '\0' ||
5596                                (name[1] == '.' && name[2] == '\0'))) {
5597                 if (hlock != NULL) {
5598                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5599                                            pobj->oo_inode, 0);
5600                 } else {
5601                         down_write(&pobj->oo_ext_idx_sem);
5602                 }
5603
5604                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5605                                         lu_object_fid(&pobj->oo_dt.do_lu),
5606                                         fid, th);
5607         } else {
5608                 if (hlock != NULL) {
5609                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5610                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5611                 } else {
5612                         down_write(&pobj->oo_ext_idx_sem);
5613                 }
5614
5615                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5616                         struct lu_fid *tfid = &info->oti_fid;
5617
5618                         *tfid = *fid;
5619                         tfid->f_ver = ~0;
5620                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5621                                               tfid, hlock, th);
5622                 } else {
5623                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5624                                               hlock, th);
5625                 }
5626         }
5627         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5628             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5629                 pobj->oo_dirent_count++;
5630
5631         if (hlock != NULL)
5632                 ldiskfs_htree_unlock(hlock);
5633         else
5634                 up_write(&pobj->oo_ext_idx_sem);
5635
5636         return rc;
5637 }
5638
5639 static int
5640 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5641                       struct osd_idmap_cache *oic)
5642 {
5643         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5644         struct lu_fid *fid = &oic->oic_fid;
5645         struct osd_inode_id *id = &oic->oic_lid;
5646         struct inode *inode = NULL;
5647         int once = 0;
5648         bool insert;
5649         int rc;
5650
5651         ENTRY;
5652
5653         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5654                 RETURN(0);
5655
5656         if (scrub->os_running && scrub->os_pos_current > id->oii_ino)
5657                 RETURN(0);
5658
5659         if (dev->od_auto_scrub_interval == AS_NEVER ||
5660             ktime_get_real_seconds() <
5661             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5662                 RETURN(0);
5663
5664 again:
5665         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5666         if (rc == -ENOENT) {
5667                 __u32 gen = id->oii_gen;
5668
5669                 insert = true;
5670                 if (inode != NULL)
5671                         goto trigger;
5672
5673                 inode = osd_iget(oti, dev, id);
5674                 /* The inode has been removed (by race maybe). */
5675                 if (IS_ERR(inode)) {
5676                         rc = PTR_ERR(inode);
5677
5678                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5679                 }
5680
5681                 /* The OI mapping is lost. */
5682                 if (gen != OSD_OII_NOGEN)
5683                         goto trigger;
5684
5685                 /*
5686                  * The inode may has been reused by others, we do not know,
5687                  * leave it to be handled by subsequent osd_fid_lookup().
5688                  */
5689                 GOTO(out, rc = 0);
5690         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5691                 GOTO(out, rc);
5692         }
5693
5694         insert = false;
5695
5696 trigger:
5697         if (scrub->os_running) {
5698                 if (inode == NULL) {
5699                         inode = osd_iget(oti, dev, id);
5700                         /* The inode has been removed (by race maybe). */
5701                         if (IS_ERR(inode)) {
5702                                 rc = PTR_ERR(inode);
5703
5704                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5705                         }
5706                 }
5707
5708                 rc = osd_oii_insert(dev, oic, insert);
5709                 /*
5710                  * There is race condition between osd_oi_lookup and OI scrub.
5711                  * The OI scrub finished just after osd_oi_lookup() failure.
5712                  * Under such case, it is unnecessary to trigger OI scrub again,
5713                  * but try to call osd_oi_lookup() again.
5714                  */
5715                 if (unlikely(rc == -EAGAIN))
5716                         goto again;
5717
5718                 if (!S_ISDIR(inode->i_mode))
5719                         rc = 0;
5720                 else
5721                         rc = osd_check_lmv(oti, dev, inode, oic);
5722
5723                 GOTO(out, rc);
5724         }
5725
5726         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5727                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5728                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5729                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5730                        "%s: trigger partial OI scrub for RPC inconsistency "
5731                        "checking FID "DFID": rc = %d\n",
5732                        osd_dev2name(dev), PFID(fid), rc);
5733                 if (rc == 0 || rc == -EALREADY)
5734                         goto again;
5735         }
5736
5737         GOTO(out, rc);
5738
5739 out:
5740         if (inode)
5741                 iput(inode);
5742
5743         RETURN(rc);
5744 }
5745
5746 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5747                                struct osd_device *dev,
5748                                struct osd_idmap_cache *oic,
5749                                struct lu_fid *fid, __u32 ino)
5750 {
5751         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5752         struct inode *inode;
5753         int rc;
5754
5755         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5756         inode = osd_iget(oti, dev, &oic->oic_lid);
5757         if (IS_ERR(inode)) {
5758                 fid_zero(&oic->oic_fid);
5759                 return PTR_ERR(inode);
5760         }
5761
5762         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5763         iput(inode);
5764         if (rc != 0)
5765                 fid_zero(&oic->oic_fid);
5766         else
5767                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5768         return rc;
5769 }
5770
5771 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5772                       struct osd_inode_id *id, const struct lu_fid *fid)
5773 {
5774         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5775                id->oii_ino, id->oii_gen, info);
5776         info->oti_cache.oic_lid = *id;
5777         info->oti_cache.oic_fid = *fid;
5778         info->oti_cache.oic_dev = osd;
5779 }
5780
5781 /**
5782  * Get parent FID from the linkEA.
5783  *
5784  * For a directory which parent resides on remote MDT, to satisfy the
5785  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5786  * the other hand, to make the lookup(..) on the directory can return
5787  * the real parent FID, we append the real parent FID after its ".."
5788  * name entry in the /REMOTE_PARENT_DIR.
5789  *
5790  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5791  * backup. So after the restore, we cannot get the right parent FID from
5792  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5793  * we have stored the real parent FID in the directory object's linkEA,
5794  * we can parse the linkEA for the real parent FID.
5795  *
5796  * \param[in] env       pointer to the thread context
5797  * \param[in] obj       pointer to the object to be handled
5798  * \param[out]fid       pointer to the buffer to hold the parent FID
5799  *
5800  * \retval              0 for getting the real parent FID successfully
5801  * \retval              negative error number on failure
5802  */
5803 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5804                                     struct osd_object *obj,
5805                                     struct lu_fid *fid)
5806 {
5807         struct osd_thread_info *oti = osd_oti_get(env);
5808         struct lu_buf *buf = &oti->oti_big_buf;
5809         struct dentry *dentry = &oti->oti_obj_dentry;
5810         struct inode *inode = obj->oo_inode;
5811         struct linkea_data ldata = { NULL };
5812         int rc;
5813
5814         ENTRY;
5815
5816         fid_zero(fid);
5817         if (!S_ISDIR(inode->i_mode))
5818                 RETURN(-EIO);
5819
5820 again:
5821         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5822                              buf->lb_buf, buf->lb_len);
5823         if (rc == -ERANGE) {
5824                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5825                                      NULL, 0);
5826                 if (rc > 0) {
5827                         lu_buf_realloc(buf, rc);
5828                         if (buf->lb_buf == NULL)
5829                                 RETURN(-ENOMEM);
5830
5831                         goto again;
5832                 }
5833         }
5834
5835         if (unlikely(rc == 0))
5836                 RETURN(-ENODATA);
5837
5838         if (rc < 0)
5839                 RETURN(rc);
5840
5841         if (unlikely(buf->lb_buf == NULL)) {
5842                 lu_buf_realloc(buf, rc);
5843                 if (buf->lb_buf == NULL)
5844                         RETURN(-ENOMEM);
5845
5846                 goto again;
5847         }
5848
5849         ldata.ld_buf = buf;
5850         rc = linkea_init_with_rec(&ldata);
5851         if (!rc) {
5852                 linkea_first_entry(&ldata);
5853                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5854         }
5855
5856         RETURN(rc);
5857 }
5858
5859 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5860                                     struct inode *inode,
5861                                     const struct lu_fid *pfid,
5862                                     const char *name, const int namelen)
5863 {
5864         struct osd_thread_info *oti = osd_oti_get(env);
5865         struct lu_buf *buf = &oti->oti_big_buf;
5866         struct dentry *dentry = &oti->oti_obj_dentry;
5867         struct linkea_data ldata = { NULL };
5868         struct lu_name cname = { .ln_name = name,
5869                                  .ln_namelen = namelen };
5870         int rc;
5871
5872         ENTRY;
5873
5874 again:
5875         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5876                              buf->lb_buf, buf->lb_len);
5877         if (rc == -ERANGE)
5878                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5879
5880         if (rc < 0)
5881                 RETURN(rc);
5882
5883         if (unlikely(rc == 0))
5884                 RETURN(-ENODATA);
5885
5886         if (buf->lb_len < rc) {
5887                 lu_buf_realloc(buf, rc);
5888                 if (buf->lb_buf == NULL)
5889                         RETURN(-ENOMEM);
5890
5891                 goto again;
5892         }
5893
5894         ldata.ld_buf = buf;
5895         rc = linkea_init_with_rec(&ldata);
5896         if (!rc)
5897                 rc = linkea_links_find(&ldata, &cname, pfid);
5898
5899         RETURN(rc);
5900 }
5901
5902 /**
5903  * Calls ->lookup() to find dentry. From dentry get inode and
5904  * read inode's ea to get fid. This is required for  interoperability
5905  * mode (b11826)
5906  *
5907  * \retval   0, on success
5908  * \retval -ve, on error
5909  */
5910 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5911                              struct dt_rec *rec, const struct dt_key *key)
5912 {
5913         struct inode *dir = obj->oo_inode;
5914         struct dentry *dentry;
5915         struct ldiskfs_dir_entry_2 *de;
5916         struct buffer_head *bh;
5917         struct lu_fid *fid = (struct lu_fid *)rec;
5918         struct htree_lock *hlock = NULL;
5919         int ino;
5920         int rc;
5921
5922         ENTRY;
5923
5924         LASSERT(dir->i_op != NULL);
5925         LASSERT(dir->i_op->lookup != NULL);
5926
5927         dentry = osd_child_dentry_get(env, obj,
5928                                       (char *)key, strlen((char *)key));
5929
5930         if (obj->oo_hl_head != NULL) {
5931                 hlock = osd_oti_get(env)->oti_hlock;
5932                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5933                                    dir, LDISKFS_HLOCK_LOOKUP);
5934         } else {
5935                 down_read(&obj->oo_ext_idx_sem);
5936         }
5937
5938         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5939         if (!IS_ERR(bh)) {
5940                 struct osd_thread_info *oti = osd_oti_get(env);
5941                 struct osd_inode_id *id = &oti->oti_id;
5942                 struct osd_idmap_cache *oic = &oti->oti_cache;
5943                 struct osd_device *dev = osd_obj2dev(obj);
5944
5945                 ino = le32_to_cpu(de->inode);
5946                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5947                         brelse(bh);
5948                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5949                         GOTO(out, rc);
5950                 }
5951
5952                 rc = osd_get_fid_from_dentry(de, rec);
5953
5954                 /* done with de, release bh */
5955                 brelse(bh);
5956                 if (rc != 0) {
5957                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5958                                 const char *name = (const char *)key;
5959
5960                                 /*
5961                                  * If the parent is on remote MDT, and there
5962                                  * is no FID-in-dirent, then we have to get
5963                                  * the parent FID from the linkEA.
5964                                  */
5965                                 if (likely(strlen(name) == 2 &&
5966                                            name[0] == '.' && name[1] == '.'))
5967                                         rc = osd_get_pfid_from_linkea(env, obj,
5968                                                                       fid);
5969                         } else {
5970                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5971                         }
5972                 } else {
5973                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5974                 }
5975
5976                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5977                         fid_zero(&oic->oic_fid);
5978
5979                         GOTO(out, rc);
5980                 }
5981
5982                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5983                 rc = osd_consistency_check(oti, dev, oic);
5984                 if (rc == -ENOENT)
5985                         fid_zero(&oic->oic_fid);
5986                 else
5987                         /* Other error should not affect lookup result. */
5988                         rc = 0;
5989         } else {
5990                 rc = PTR_ERR(bh);
5991         }
5992
5993         GOTO(out, rc);
5994
5995 out:
5996         if (hlock != NULL)
5997                 ldiskfs_htree_unlock(hlock);
5998         else
5999                 up_read(&obj->oo_ext_idx_sem);
6000         return rc;
6001 }
6002
6003 static int osd_index_declare_ea_insert(const struct lu_env *env,
6004                                        struct dt_object *dt,
6005                                        const struct dt_rec *rec,
6006                                        const struct dt_key *key,
6007                                        struct thandle *handle)
6008 {
6009         struct osd_thandle *oh;
6010         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6011         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6012         const struct lu_fid *fid = rec1->rec_fid;
6013         int credits, rc = 0;
6014         struct osd_idmap_cache *idc;
6015
6016         ENTRY;
6017
6018         LASSERT(!dt_object_remote(dt));
6019         LASSERT(handle != NULL);
6020         LASSERT(fid != NULL);
6021         LASSERT(rec1->rec_type != 0);
6022
6023         oh = container_of(handle, struct osd_thandle, ot_super);
6024         LASSERT(oh->ot_handle == NULL);
6025
6026         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6027
6028         /*
6029          * we can't call iget() while a transactions is running
6030          * (this can lead to a deadlock), but we need to know
6031          * inum and object type. so we find this information at
6032          * declaration and cache in per-thread info
6033          */
6034         idc = osd_idc_find_or_init(env, osd, fid);
6035         if (IS_ERR(idc))
6036                 RETURN(PTR_ERR(idc));
6037         if (idc->oic_remote) {
6038                 /*
6039                  * a reference to remote inode is represented by an
6040                  * agent inode which we have to create
6041                  */
6042                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6043                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6044         }
6045
6046         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6047
6048         if (osd_dt_obj(dt)->oo_inode != NULL) {
6049                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6050
6051                 /*
6052                  * We ignore block quota on meta pool (MDTs), so needn't
6053                  * calculate how many blocks will be consumed by this index
6054                  * insert
6055                  */
6056                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6057                                            i_gid_read(inode),
6058                                            i_projid_read(inode), 0,
6059                                            oh, osd_dt_obj(dt), NULL,
6060                                            OSD_QID_BLK);
6061                 if (rc)
6062                         RETURN(rc);
6063
6064 #ifdef HAVE_PROJECT_QUOTA
6065                 /*
6066                  * Reserve credits for local agent inode to transfer
6067                  * to 0, quota enforcement is ignored in this case.
6068                  */
6069                 if (idc->oic_remote &&
6070                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6071                     i_projid_read(inode) != 0)
6072                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6073                                                   0, i_projid_read(inode),
6074                                                   0, false, PRJQUOTA, true);
6075 #endif
6076         }
6077
6078         RETURN(rc);
6079 }
6080
6081 /**
6082  * Index add function for interoperability mode (b11826).
6083  * It will add the directory entry.This entry is needed to
6084  * maintain name->fid mapping.
6085  *
6086  * \param key it is key i.e. file entry to be inserted
6087  * \param rec it is value of given key i.e. fid
6088  *
6089  * \retval   0, on success
6090  * \retval -ve, on error
6091  */
6092 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6093                                const struct dt_rec *rec,
6094                                const struct dt_key *key, struct thandle *th)
6095 {
6096         struct osd_object *obj = osd_dt_obj(dt);
6097         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6098         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6099         const struct lu_fid *fid = rec1->rec_fid;
6100         const char *name = (const char *)key;
6101         struct osd_thread_info *oti = osd_oti_get(env);
6102         struct inode *child_inode = NULL;
6103         struct osd_idmap_cache *idc;
6104         int rc;
6105
6106         ENTRY;
6107
6108         if (!dt_object_exists(dt))
6109                 RETURN(-ENOENT);
6110
6111         LASSERT(osd_invariant(obj));
6112         LASSERT(!dt_object_remote(dt));
6113         LASSERT(th != NULL);
6114
6115         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6116
6117         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6118
6119         idc = osd_idc_find(env, osd, fid);
6120         if (unlikely(idc == NULL)) {
6121                 idc = osd_idc_find_or_init(env, osd, fid);
6122                 if (IS_ERR(idc)) {
6123                         /*
6124                          * this dt_insert() wasn't declared properly, so
6125                          * FID is missing in OI cache. we better do not
6126                          * lookup FID in FLDB/OI and don't risk to deadlock,
6127                          * but in some special cases (lfsck testing, etc)
6128                          * it's much simpler than fixing a caller.
6129                          *
6130                          * normally this error should be placed after the first
6131                          * find, but migrate may attach source stripes to
6132                          * target, which doesn't create stripes.
6133                          */
6134                         CERROR("%s: "DFID" wasn't declared for insert\n",
6135                                osd_name(osd), PFID(fid));
6136                         dump_stack();
6137                         RETURN(PTR_ERR(idc));
6138                 }
6139         }
6140
6141         if (idc->oic_remote) {
6142                 /* Insert remote entry */
6143                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6144                         child_inode =
6145                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6146                 } else {
6147                         child_inode = osd_create_local_agent_inode(env, osd,
6148                                         obj, fid, rec1->rec_type & S_IFMT, th);
6149                         if (IS_ERR(child_inode))
6150                                 RETURN(PTR_ERR(child_inode));
6151                 }
6152         } else {
6153                 /* Insert local entry */
6154                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6155                         /* for a reason OI cache wasn't filled properly */
6156                         CERROR("%s: OIC for "DFID" isn't filled\n",
6157                                osd_name(osd), PFID(fid));
6158                         RETURN(-EINVAL);
6159                 }
6160                 child_inode = oti->oti_inode;
6161                 if (unlikely(child_inode == NULL)) {
6162                         struct ldiskfs_inode_info *lii;
6163
6164                         OBD_ALLOC_PTR(lii);
6165                         if (lii == NULL)
6166                                 RETURN(-ENOMEM);
6167                         child_inode = oti->oti_inode = &lii->vfs_inode;
6168                 }
6169                 child_inode->i_sb = osd_sb(osd);
6170                 child_inode->i_ino = idc->oic_lid.oii_ino;
6171                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6172         }
6173
6174         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6175
6176         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6177                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6178
6179         if (child_inode && child_inode != oti->oti_inode)
6180                 iput(child_inode);
6181         LASSERT(osd_invariant(obj));
6182         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6183
6184         RETURN(rc);
6185 }
6186
6187 /**
6188  *  Initialize osd Iterator for given osd index object.
6189  *
6190  *  \param  dt      osd index object
6191  */
6192
6193 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6194                                      struct dt_object *dt,
6195                                      __u32 unused)
6196 {
6197         struct osd_it_iam *it;
6198         struct osd_object *obj = osd_dt_obj(dt);
6199         struct lu_object *lo = &dt->do_lu;
6200         struct iam_path_descr *ipd;
6201         struct iam_container *bag = &obj->oo_dir->od_container;
6202
6203         if (!dt_object_exists(dt))
6204                 return ERR_PTR(-ENOENT);
6205
6206         OBD_ALLOC_PTR(it);
6207         if (it == NULL)
6208                 return ERR_PTR(-ENOMEM);
6209
6210         ipd = osd_it_ipd_get(env, bag);
6211         if (likely(ipd != NULL)) {
6212                 it->oi_obj = obj;
6213                 it->oi_ipd = ipd;
6214                 lu_object_get(lo);
6215                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6216                 return (struct dt_it *)it;
6217         } else {
6218                 OBD_FREE_PTR(it);
6219                 return ERR_PTR(-ENOMEM);
6220         }
6221 }
6222
6223 /**
6224  * free given Iterator.
6225  */
6226 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6227 {
6228         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6229         struct osd_object *obj = it->oi_obj;
6230
6231         iam_it_fini(&it->oi_it);
6232         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6233         osd_object_put(env, obj);
6234         OBD_FREE_PTR(it);
6235 }
6236
6237 /**
6238  *  Move Iterator to record specified by \a key
6239  *
6240  *  \param  di      osd iterator
6241  *  \param  key     key for index
6242  *
6243  *  \retval +ve  di points to record with least key not larger than key
6244  *  \retval  0   di points to exact matched key
6245  *  \retval -ve  failure
6246  */
6247
6248 static int osd_it_iam_get(const struct lu_env *env,
6249                           struct dt_it *di, const struct dt_key *key)
6250 {
6251         struct osd_thread_info *oti = osd_oti_get(env);
6252         struct osd_it_iam *it = (struct osd_it_iam *)di;
6253
6254         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6255                 /* swab quota uid/gid */
6256                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6257                 key = (struct dt_key *)&oti->oti_quota_id;
6258         }
6259
6260         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6261 }
6262
6263 /**
6264  *  Release Iterator
6265  *
6266  *  \param  di      osd iterator
6267  */
6268 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6269 {
6270         struct osd_it_iam *it = (struct osd_it_iam *)di;
6271
6272         iam_it_put(&it->oi_it);
6273 }
6274
6275 /**
6276  *  Move iterator by one record
6277  *
6278  *  \param  di      osd iterator
6279  *
6280  *  \retval +1   end of container reached
6281  *  \retval  0   success
6282  *  \retval -ve  failure
6283  */
6284
6285 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6286 {
6287         struct osd_it_iam *it = (struct osd_it_iam *)di;
6288
6289         return iam_it_next(&it->oi_it);
6290 }
6291
6292 /**
6293  * Return pointer to the key under iterator.
6294  */
6295
6296 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6297                                      const struct dt_it *di)
6298 {
6299         struct osd_thread_info *oti = osd_oti_get(env);
6300         struct osd_it_iam *it = (struct osd_it_iam *)di;
6301         struct osd_object *obj = it->oi_obj;
6302         struct dt_key *key;
6303
6304         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6305
6306         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6307                 /* swab quota uid/gid */
6308                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6309                 key = (struct dt_key *)&oti->oti_quota_id;
6310         }
6311
6312         return key;
6313 }
6314
6315 /**
6316  * Return size of key under iterator (in bytes)
6317  */
6318
6319 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6320 {
6321         struct osd_it_iam *it = (struct osd_it_iam *)di;
6322
6323         return iam_it_key_size(&it->oi_it);
6324 }
6325
6326 static inline void
6327 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6328 {
6329         /* check if file type is required */
6330         if (ent->lde_attrs & LUDA_TYPE) {
6331                 struct luda_type *lt;
6332                 int align = sizeof(*lt) - 1;
6333
6334                 len = (len + align) & ~align;
6335                 lt = (struct luda_type *)(ent->lde_name + len);
6336                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6337         }
6338
6339         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6340 }
6341
6342 /**
6343  * build lu direct from backend fs dirent.
6344  */
6345
6346 static inline void
6347 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6348                    char *name, __u16 namelen, __u16 type, __u32 attr)
6349 {
6350         ent->lde_attrs = attr | LUDA_FID;
6351         fid_cpu_to_le(&ent->lde_fid, fid);
6352
6353         ent->lde_hash = cpu_to_le64(offset);
6354         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6355
6356         strncpy(ent->lde_name, name, namelen);
6357         ent->lde_name[namelen] = '\0';
6358         ent->lde_namelen = cpu_to_le16(namelen);
6359
6360         /* append lustre attributes */
6361         osd_it_append_attrs(ent, namelen, type);
6362 }
6363
6364 /**
6365  * Return pointer to the record under iterator.
6366  */
6367 static int osd_it_iam_rec(const struct lu_env *env,
6368                           const struct dt_it *di,
6369                           struct dt_rec *dtrec, __u32 attr)
6370 {
6371         struct osd_it_iam *it = (struct osd_it_iam *)di;
6372         struct osd_thread_info *info = osd_oti_get(env);
6373
6374         ENTRY;
6375
6376         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6377                 const struct osd_fid_pack *rec;
6378                 struct lu_fid *fid = &info->oti_fid;
6379                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6380                 char *name;
6381                 int namelen;
6382                 __u64 hash;
6383                 int rc;
6384
6385                 name = (char *)iam_it_key_get(&it->oi_it);
6386                 if (IS_ERR(name))
6387                         RETURN(PTR_ERR(name));
6388
6389                 namelen = iam_it_key_size(&it->oi_it);
6390
6391                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6392                 if (IS_ERR(rec))
6393                         RETURN(PTR_ERR(rec));
6394
6395                 rc = osd_fid_unpack(fid, rec);
6396                 if (rc)
6397                         RETURN(rc);
6398
6399                 hash = iam_it_store(&it->oi_it);
6400
6401                 /* IAM does not store object type in IAM index (dir) */
6402                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6403                                    0, LUDA_FID);
6404         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6405                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6406                            (struct iam_rec *)dtrec);
6407                 osd_quota_unpack(it->oi_obj, dtrec);
6408         } else {
6409                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6410                            (struct iam_rec *)dtrec);
6411         }
6412
6413         RETURN(0);
6414 }
6415
6416 /**
6417  * Returns cookie for current Iterator position.
6418  */
6419 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6420 {
6421         struct osd_it_iam *it = (struct osd_it_iam *)di;
6422
6423         return iam_it_store(&it->oi_it);
6424 }
6425
6426 /**
6427  * Restore iterator from cookie.
6428  *
6429  * \param  di      osd iterator
6430  * \param  hash    Iterator location cookie
6431  *
6432  * \retval +ve  di points to record with least key not larger than key.
6433  * \retval  0   di points to exact matched key
6434  * \retval -ve  failure
6435  */
6436
6437 static int osd_it_iam_load(const struct lu_env *env,
6438                            const struct dt_it *di, __u64 hash)
6439 {
6440         struct osd_it_iam *it = (struct osd_it_iam *)di;
6441
6442         return iam_it_load(&it->oi_it, hash);
6443 }
6444
6445 static const struct dt_index_operations osd_index_iam_ops = {
6446         .dio_lookup         = osd_index_iam_lookup,
6447         .dio_declare_insert = osd_index_declare_iam_insert,
6448         .dio_insert         = osd_index_iam_insert,
6449         .dio_declare_delete = osd_index_declare_iam_delete,
6450         .dio_delete         = osd_index_iam_delete,
6451         .dio_it     = {
6452                 .init     = osd_it_iam_init,
6453                 .fini     = osd_it_iam_fini,
6454                 .get      = osd_it_iam_get,
6455                 .put      = osd_it_iam_put,
6456                 .next     = osd_it_iam_next,
6457                 .key      = osd_it_iam_key,
6458                 .key_size = osd_it_iam_key_size,
6459                 .rec      = osd_it_iam_rec,
6460                 .store    = osd_it_iam_store,
6461                 .load     = osd_it_iam_load
6462         }
6463 };
6464
6465
6466 /**
6467  * Creates or initializes iterator context.
6468  *
6469  * \retval struct osd_it_ea, iterator structure on success
6470  *
6471  */
6472 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6473                                     struct dt_object *dt,
6474                                     __u32 attr)
6475 {
6476         struct osd_object *obj = osd_dt_obj(dt);
6477         struct osd_thread_info *info = osd_oti_get(env);
6478         struct osd_it_ea *oie;
6479         struct file *file;
6480         struct lu_object *lo = &dt->do_lu;
6481         struct dentry *obj_dentry;
6482
6483         ENTRY;
6484
6485         if (!dt_object_exists(dt) || obj->oo_destroyed)
6486                 RETURN(ERR_PTR(-ENOENT));
6487
6488         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6489         if (oie == NULL)
6490                 RETURN(ERR_PTR(-ENOMEM));
6491         obj_dentry = &oie->oie_dentry;
6492
6493         obj_dentry->d_inode = obj->oo_inode;
6494         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6495         obj_dentry->d_name.hash = 0;
6496
6497         oie->oie_rd_dirent       = 0;
6498         oie->oie_it_dirent       = 0;
6499         oie->oie_dirent          = NULL;
6500         if (unlikely(!info->oti_it_ea_buf_used)) {
6501                 oie->oie_buf = info->oti_it_ea_buf;
6502                 info->oti_it_ea_buf_used = 1;
6503         } else {
6504                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6505                 if (oie->oie_buf == NULL)
6506                         RETURN(ERR_PTR(-ENOMEM));
6507         }
6508         oie->oie_obj = obj;
6509
6510         file = &oie->oie_file;
6511
6512         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6513         if (attr & LUDA_64BITHASH)
6514                 file->f_mode    = FMODE_64BITHASH;
6515         else
6516                 file->f_mode    = FMODE_32BITHASH;
6517         file->f_path.dentry     = obj_dentry;
6518         file->f_mapping         = obj->oo_inode->i_mapping;
6519         file->f_op              = obj->oo_inode->i_fop;
6520         file->f_inode = obj->oo_inode;
6521
6522         lu_object_get(lo);
6523         RETURN((struct dt_it *)oie);
6524 }
6525
6526 /**
6527  * Destroy or finishes iterator context.
6528  *
6529  * \param di iterator structure to be destroyed
6530  */
6531 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6532 {
6533         struct osd_thread_info *info = osd_oti_get(env);
6534         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6535         struct osd_object *obj = oie->oie_obj;
6536         struct inode *inode = obj->oo_inode;
6537
6538         ENTRY;
6539         oie->oie_file.f_op->release(inode, &oie->oie_file);
6540         osd_object_put(env, obj);
6541         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6542                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6543         else
6544                 info->oti_it_ea_buf_used = 0;
6545         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6546         EXIT;
6547 }
6548
6549 /**
6550  * It position the iterator at given key, so that next lookup continues from
6551  * that key Or it is similar to dio_it->load() but based on a key,
6552  * rather than file position.
6553  *
6554  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6555  * to the beginning.
6556  *
6557  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6558  */
6559 static int osd_it_ea_get(const struct lu_env *env,
6560                          struct dt_it *di, const struct dt_key *key)
6561 {
6562         struct osd_it_ea *it = (struct osd_it_ea *)di;
6563
6564         ENTRY;
6565         LASSERT(((const char *)key)[0] == '\0');
6566         it->oie_file.f_pos = 0;
6567         it->oie_rd_dirent = 0;
6568         it->oie_it_dirent = 0;
6569         it->oie_dirent = NULL;
6570
6571         RETURN(+1);
6572 }
6573
6574 /**
6575  * Does nothing
6576  */
6577 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6578 {
6579 }
6580
6581 struct osd_filldir_cbs {
6582         struct dir_context ctx;
6583         struct osd_it_ea  *it;
6584 };
6585 /**
6586  * It is called internally by ->iterate*(). It fills the
6587  * iterator's in-memory data structure with required
6588  * information i.e. name, namelen, rec_size etc.
6589  *
6590  * \param buf in which information to be filled in.
6591  * \param name name of the file in given dir
6592  *
6593  * \retval 0 on success
6594  * \retval 1 on buffer full
6595  */
6596 #ifdef HAVE_FILLDIR_USE_CTX
6597 static int osd_ldiskfs_filldir(struct dir_context *buf,
6598 #else
6599 static int osd_ldiskfs_filldir(void *buf,
6600 #endif
6601                                const char *name, int namelen,
6602                                loff_t offset, __u64 ino, unsigned int d_type)
6603 {
6604         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6605         struct osd_object *obj = it->oie_obj;
6606         struct osd_it_ea_dirent *ent = it->oie_dirent;
6607         struct lu_fid *fid = &ent->oied_fid;
6608         struct osd_fid_pack *rec;
6609
6610         ENTRY;
6611
6612 /* this should never happen */
6613         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6614                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6615                 RETURN(-EIO);
6616         }
6617
6618         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6619             OSD_IT_EA_BUFSIZE)
6620                 RETURN(1);
6621
6622         /* "." is just the object itself. */
6623         if (namelen == 1 && name[0] == '.') {
6624                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6625         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6626                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6627                 if (osd_fid_unpack(fid, rec) != 0)
6628                         fid_zero(fid);
6629         } else {
6630                 fid_zero(fid);
6631         }
6632         d_type &= ~LDISKFS_DIRENT_LUFID;
6633
6634         /* NOT export local root. */
6635         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6636                 ino = obj->oo_inode->i_ino;
6637                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6638         }
6639
6640         ent->oied_ino     = ino;
6641         ent->oied_off     = offset;
6642         ent->oied_namelen = namelen;
6643         ent->oied_type    = d_type;
6644
6645         memcpy(ent->oied_name, name, namelen);
6646
6647         it->oie_rd_dirent++;
6648         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6649         RETURN(0);
6650 }
6651
6652 /**
6653  * Calls ->iterate*() to load a directory entry at a time
6654  * and stored it in iterator's in-memory data structure.
6655  *
6656  * \param di iterator's in memory structure
6657  *
6658  * \retval   0 on success
6659  * \retval -ve on error
6660  * \retval +1 reach the end of entry
6661  */
6662 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6663                                const struct dt_it *di)
6664 {
6665         struct osd_it_ea *it = (struct osd_it_ea *)di;
6666         struct osd_object *obj = it->oie_obj;
6667         struct inode *inode = obj->oo_inode;
6668         struct htree_lock *hlock = NULL;
6669         struct file *filp = &it->oie_file;
6670         int rc = 0;
6671         struct osd_filldir_cbs buf = {
6672                 .ctx.actor = osd_ldiskfs_filldir,
6673                 .it = it
6674         };
6675
6676         ENTRY;
6677         it->oie_dirent = it->oie_buf;
6678         it->oie_rd_dirent = 0;
6679
6680         if (obj->oo_hl_head != NULL) {
6681                 hlock = osd_oti_get(env)->oti_hlock;
6682                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6683                                    inode, LDISKFS_HLOCK_READDIR);
6684         } else {
6685                 down_read(&obj->oo_ext_idx_sem);
6686         }
6687
6688         filp->f_cred = current_cred();
6689         rc = osd_security_file_alloc(filp);
6690         if (rc)
6691                 RETURN(rc);
6692
6693         filp->f_flags |= O_NOATIME;
6694         filp->f_mode |= FMODE_NONOTIFY;
6695         rc = iterate_dir(filp, &buf.ctx);
6696         if (rc)
6697                 RETURN(rc);
6698
6699         if (hlock != NULL)
6700                 ldiskfs_htree_unlock(hlock);
6701         else
6702                 up_read(&obj->oo_ext_idx_sem);
6703
6704         if (it->oie_rd_dirent == 0) {
6705                 /*
6706                  * If it does not get any dirent, it means it has been reached
6707                  * to the end of the dir
6708                  */
6709                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6710                 if (rc == 0)
6711                         rc = 1;
6712         } else {
6713                 it->oie_dirent = it->oie_buf;
6714                 it->oie_it_dirent = 1;
6715         }
6716
6717         RETURN(rc);
6718 }
6719
6720 /**
6721  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6722  * to load a directory entry at a time and stored it in
6723  * iterator's in-memory data structure.
6724  *
6725  * \param di iterator's in memory structure
6726  *
6727  * \retval +ve iterator reached to end
6728  * \retval   0 iterator not reached to end
6729  * \retval -ve on error
6730  */
6731 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6732 {
6733         struct osd_it_ea *it = (struct osd_it_ea *)di;
6734         int rc;
6735
6736         ENTRY;
6737
6738         if (it->oie_it_dirent < it->oie_rd_dirent) {
6739                 it->oie_dirent =
6740                         (void *)it->oie_dirent +
6741                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6742                                        it->oie_dirent->oied_namelen);
6743                 it->oie_it_dirent++;
6744                 rc = 0;
6745         } else {
6746                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6747                         rc = 1;
6748                 else
6749                         rc = osd_ldiskfs_it_fill(env, di);
6750         }
6751
6752         RETURN(rc);
6753 }
6754
6755 /**
6756  * Returns the key at current position from iterator's in memory structure.
6757  *
6758  * \param di iterator's in memory structure
6759  *
6760  * \retval key i.e. struct dt_key on success
6761  */
6762 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6763                                     const struct dt_it *di)
6764 {
6765         struct osd_it_ea *it = (struct osd_it_ea *)di;
6766
6767         return (struct dt_key *)it->oie_dirent->oied_name;
6768 }
6769
6770 /**
6771  * Returns key's size at current position from iterator's in memory structure.
6772  *
6773  * \param di iterator's in memory structure
6774  *
6775  * \retval key_size i.e. struct dt_key on success
6776  */
6777 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6778 {
6779         struct osd_it_ea *it = (struct osd_it_ea *)di;
6780
6781         return it->oie_dirent->oied_namelen;
6782 }
6783
6784 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6785 #undef LDISKFS_DIR_REC_LEN
6786 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6787 #endif
6788
6789 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6790 {
6791         if (LDISKFS_DIR_REC_LEN(de) >=
6792             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6793                 return true;
6794
6795         return false;
6796 }
6797
6798 static inline bool
6799 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6800                      unsigned int blocksize, bool dotdot)
6801 {
6802         if (dotdot)
6803                 return osd_dotdot_has_space(de);
6804
6805         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6806             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6807                 return true;
6808
6809         return false;
6810 }
6811
6812 static int
6813 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6814                     handle_t *jh, struct dentry *dentry,
6815                     const struct lu_fid *fid, struct buffer_head *bh,
6816                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6817                     bool dotdot)
6818 {
6819         struct inode *dir = dentry->d_parent->d_inode;
6820         struct inode *inode = dentry->d_inode;
6821         struct osd_fid_pack *rec;
6822         struct ldiskfs_dentry_param *ldp;
6823         int namelen = dentry->d_name.len;
6824         int rc;
6825         struct osd_thread_info *info = osd_oti_get(env);
6826
6827         ENTRY;
6828
6829         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6830                 RETURN(0);
6831
6832         /* There is enough space to hold the FID-in-dirent. */
6833         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6834                 rc = ldiskfs_journal_get_write_access(jh, bh);
6835                 if (rc != 0)
6836                         RETURN(rc);
6837
6838                 de->name[namelen] = 0;
6839                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6840                 rec->fp_len = sizeof(struct lu_fid) + 1;
6841                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6842                 de->file_type |= LDISKFS_DIRENT_LUFID;
6843                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6844
6845                 RETURN(rc);
6846         }
6847
6848         LASSERT(!dotdot);
6849
6850         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6851         if (rc != 0)
6852                 RETURN(rc);
6853
6854         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6855         osd_get_ldiskfs_dirent_param(ldp, fid);
6856         dentry->d_fsdata = (void *)ldp;
6857         dquot_initialize(dir);
6858         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6859         /*
6860          * It is too bad, we cannot reinsert the name entry back.
6861          * That means we lose it!
6862          */
6863         if (rc != 0)
6864                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6865                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6866                        osd_ino2name(inode),
6867                        dir->i_ino, dir->i_generation, namelen,
6868                        dentry->d_name.name, PFID(fid), rc);
6869
6870         RETURN(rc);
6871 }
6872
6873 static int
6874 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6875                         struct osd_it_ea *it, struct lu_fid *fid,
6876                         struct osd_inode_id *id, __u32 *attr)
6877 {
6878         struct osd_thread_info *info = osd_oti_get(env);
6879         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6880         struct osd_device *dev = osd_obj2dev(obj);
6881         struct super_block *sb = osd_sb(dev);
6882         const char *devname = osd_name(dev);
6883         struct osd_it_ea_dirent *ent = it->oie_dirent;
6884         struct inode *dir = obj->oo_inode;
6885         struct htree_lock *hlock = NULL;
6886         struct buffer_head *bh = NULL;
6887         handle_t *jh = NULL;
6888         struct ldiskfs_dir_entry_2 *de;
6889         struct dentry *dentry;
6890         struct inode *inode;
6891         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6892         int credits;
6893         int rc;
6894         bool dotdot = false;
6895         bool dirty = false;
6896
6897         ENTRY;
6898
6899         if (ent->oied_name[0] == '.') {
6900                 if (ent->oied_namelen == 1)
6901                         RETURN(0);
6902
6903                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6904                         dotdot = true;
6905         }
6906
6907         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6908         inode = osd_iget(info, dev, id);
6909         if (IS_ERR(inode)) {
6910                 rc = PTR_ERR(inode);
6911                 if (rc == -ENOENT || rc == -ESTALE) {
6912                         /*
6913                          * Maybe dangling name entry, or
6914                          * corrupted directory entry.
6915                          */
6916                         *attr |= LUDA_UNKNOWN;
6917                         rc = 0;
6918                 } else {
6919                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6920                                "check_repair, dir = %lu/%u, name = %.*s, "
6921                                "ino = %llu, rc = %d\n",
6922                                devname, dir->i_ino, dir->i_generation,
6923                                ent->oied_namelen, ent->oied_name,
6924                                ent->oied_ino, rc);
6925                 }
6926
6927                 RETURN(rc);
6928         }
6929
6930         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6931                                            ent->oied_namelen);
6932         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6933         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6934                 lma = NULL;
6935         else if (rc != 0)
6936                 GOTO(out, rc);
6937
6938         /*
6939          * We need to ensure that the name entry is still valid.
6940          * Because it may be removed or renamed by other already.
6941          *
6942          * The unlink or rename operation will start journal before PDO lock,
6943          * so to avoid deadlock, here we need to start journal handle before
6944          * related PDO lock also. But because we do not know whether there
6945          * will be something to be repaired before PDO lock, we just start
6946          * journal without conditions.
6947          *
6948          * We may need to remove the name entry firstly, then insert back.
6949          * One credit is for user quota file update.
6950          * One credit is for group quota file update.
6951          * Two credits are for dirty inode.
6952          */
6953         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6954                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6955
6956         if (dev->od_dirent_journal != 0) {
6957
6958 again:
6959                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6960                 if (IS_ERR(jh)) {
6961                         rc = PTR_ERR(jh);
6962                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6963                                "check_repair, dir = %lu/%u, credits = %d, "
6964                                "name = %.*s, ino = %llu: rc = %d\n",
6965                                devname, dir->i_ino, dir->i_generation, credits,
6966                                ent->oied_namelen, ent->oied_name,
6967                                ent->oied_ino, rc);
6968
6969                         GOTO(out_inode, rc);
6970                 }
6971
6972                 if (obj->oo_hl_head != NULL) {
6973                         hlock = osd_oti_get(env)->oti_hlock;
6974                         /*
6975                          * "0" means exclusive lock for the whole directory.
6976                          * We need to prevent others access such name entry
6977                          * during the delete + insert. Neither HLOCK_ADD nor
6978                          * HLOCK_DEL cannot guarantee the atomicity.
6979                          */
6980                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6981                 } else {
6982                         down_write(&obj->oo_ext_idx_sem);
6983                 }
6984         } else {
6985                 if (obj->oo_hl_head != NULL) {
6986                         hlock = osd_oti_get(env)->oti_hlock;
6987                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6988                                            LDISKFS_HLOCK_LOOKUP);
6989                 } else {
6990                         down_read(&obj->oo_ext_idx_sem);
6991                 }
6992         }
6993
6994         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6995         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6996                 *attr |= LUDA_IGNORE;
6997
6998                 GOTO(out, rc = 0);
6999         }
7000
7001         /*
7002          * For dotdot entry, if there is not enough space to hold the
7003          * FID-in-dirent, just keep them there. It only happens when the
7004          * device upgraded from 1.8 or restored from MDT file-level backup.
7005          * For the whole directory, only dotdot entry have no FID-in-dirent
7006          * and needs to get FID from LMA when readdir, it will not affect the
7007          * performance much.
7008          */
7009         if (dotdot && !osd_dotdot_has_space(de)) {
7010                 *attr |= LUDA_UNKNOWN;
7011
7012                 GOTO(out, rc = 0);
7013         }
7014
7015         if (lma != NULL) {
7016                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7017                         GOTO(out, rc = 0);
7018
7019                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7020                         struct lu_fid *tfid = &lma->lma_self_fid;
7021
7022                         if (likely(dotdot &&
7023                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7024                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7025                                 /*
7026                                  * It must be REMOTE_PARENT_DIR and as the
7027                                  * 'dotdot' entry of remote directory
7028                                  */
7029                                 *attr |= LUDA_IGNORE;
7030                         } else {
7031                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7032                                        "parent directory, but got %.*s under "
7033                                        "dir = %lu/%u with the FID "DFID"\n",
7034                                        devname, ent->oied_namelen,
7035                                        ent->oied_name, dir->i_ino,
7036                                        dir->i_generation, PFID(tfid));
7037
7038                                 *attr |= LUDA_UNKNOWN;
7039                         }
7040
7041                         GOTO(out, rc = 0);
7042                 }
7043         }
7044
7045         if (!fid_is_zero(fid)) {
7046                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7047                                               ent->oied_namelen);
7048                 if (rc == -ENOENT ||
7049                     (rc == -ENODATA &&
7050                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7051                         /*
7052                          * linkEA does not recognize the dirent entry,
7053                          * it may because the dirent entry corruption
7054                          * and points to other's inode.
7055                          */
7056                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7057                                "recognize the dirent, dir = %lu/%u, "
7058                                " name = %.*s, ino = %llu, "
7059                                DFID": rc = %d\n", devname, dir->i_ino,
7060                                dir->i_generation, ent->oied_namelen,
7061                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7062                         *attr |= LUDA_UNKNOWN;
7063
7064                         GOTO(out, rc = 0);
7065                 }
7066
7067                 if (rc && rc != -ENODATA) {
7068                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7069                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7070                                DFID": rc = %d\n", devname, dir->i_ino,
7071                                dir->i_generation, ent->oied_namelen,
7072                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7073                         *attr |= LUDA_UNKNOWN;
7074
7075                         GOTO(out, rc = 0);
7076                 }
7077         }
7078
7079         if (lma != NULL) {
7080                 /*
7081                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7082                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7083                  */
7084                 if (*attr & LUDA_VERIFY_DRYRUN) {
7085                         *fid = lma->lma_self_fid;
7086                         *attr |= LUDA_REPAIR;
7087
7088                         GOTO(out, rc = 0);
7089                 }
7090
7091                 if (jh == NULL) {
7092                         brelse(bh);
7093                         dev->od_dirent_journal = 1;
7094                         if (hlock != NULL) {
7095                                 ldiskfs_htree_unlock(hlock);
7096                                 hlock = NULL;
7097                         } else {
7098                                 up_read(&obj->oo_ext_idx_sem);
7099                         }
7100
7101                         goto again;
7102                 }
7103
7104                 *fid = lma->lma_self_fid;
7105                 dirty = true;
7106                 /* Update or append the FID-in-dirent. */
7107                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7108                                          bh, de, hlock, dotdot);
7109                 if (rc == 0)
7110                         *attr |= LUDA_REPAIR;
7111                 else
7112                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7113                                "the dirent, dir = %lu/%u, name = %.*s, "
7114                                "ino = %llu, "DFID": rc = %d\n",
7115                                devname, dir->i_ino, dir->i_generation,
7116                                ent->oied_namelen, ent->oied_name,
7117                                ent->oied_ino, PFID(fid), rc);
7118         } else {
7119                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7120                 if (*attr & LUDA_VERIFY_DRYRUN) {
7121                         if (fid_is_sane(fid)) {
7122                                 *attr |= LUDA_REPAIR;
7123                         } else if (dev->od_index == 0) {
7124                                 lu_igif_build(fid, inode->i_ino,
7125                                               inode->i_generation);
7126                                 *attr |= LUDA_UPGRADE;
7127                         }
7128
7129                         GOTO(out, rc = 0);
7130                 }
7131
7132                 if (jh == NULL) {
7133                         brelse(bh);
7134                         dev->od_dirent_journal = 1;
7135                         if (hlock != NULL) {
7136                                 ldiskfs_htree_unlock(hlock);
7137                                 hlock = NULL;
7138                         } else {
7139                                 up_read(&obj->oo_ext_idx_sem);
7140                         }
7141
7142                         goto again;
7143                 }
7144
7145                 dirty = true;
7146                 if (unlikely(fid_is_sane(fid))) {
7147                         /*
7148                          * FID-in-dirent exists, but FID-in-LMA is lost.
7149                          * Trust the FID-in-dirent, and add FID-in-LMA.
7150                          */
7151                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7152                         if (rc == 0)
7153                                 *attr |= LUDA_REPAIR;
7154                         else
7155                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7156                                        "update dirent, dir = %lu/%u, "
7157                                        "name = %.*s, ino = %llu, "
7158                                        DFID": rc = %d\n",
7159                                        devname, dir->i_ino, dir->i_generation,
7160                                        ent->oied_namelen, ent->oied_name,
7161                                        ent->oied_ino, PFID(fid), rc);
7162                 } else if (dev->od_index == 0) {
7163                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7164                         /*
7165                          * It is probably IGIF object. Only aappend the
7166                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7167                          */
7168                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7169                                                  bh, de, hlock, dotdot);
7170                         if (rc == 0)
7171                                 *attr |= LUDA_UPGRADE;
7172                         else
7173                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7174                                        "after the dirent, dir = %lu/%u, "
7175                                        "name = %.*s, ino = %llu, "
7176                                        DFID": rc = %d\n",
7177                                        devname, dir->i_ino, dir->i_generation,
7178                                        ent->oied_namelen, ent->oied_name,
7179                                        ent->oied_ino, PFID(fid), rc);
7180                 }
7181         }
7182
7183         GOTO(out, rc);
7184
7185 out:
7186         if (!IS_ERR(bh))
7187                 brelse(bh);
7188         if (hlock != NULL) {
7189                 ldiskfs_htree_unlock(hlock);
7190         } else {
7191                 if (dev->od_dirent_journal != 0)
7192                         up_write(&obj->oo_ext_idx_sem);
7193                 else
7194                         up_read(&obj->oo_ext_idx_sem);
7195         }
7196
7197         if (jh != NULL)
7198                 ldiskfs_journal_stop(jh);
7199
7200 out_inode:
7201         iput(inode);
7202         if (rc >= 0 && !dirty)
7203                 dev->od_dirent_journal = 0;
7204
7205         return rc;
7206 }
7207
7208 /**
7209  * Returns the value at current position from iterator's in memory structure.
7210  *
7211  * \param di struct osd_it_ea, iterator's in memory structure
7212  * \param attr attr requested for dirent.
7213  * \param lde lustre dirent
7214  *
7215  * \retval   0 no error and \param lde has correct lustre dirent.
7216  * \retval -ve on error
7217  */
7218 static inline int osd_it_ea_rec(const struct lu_env *env,
7219                                 const struct dt_it *di,
7220                                 struct dt_rec *dtrec, __u32 attr)
7221 {
7222         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7223         struct osd_object      *obj   = it->oie_obj;
7224         struct osd_device      *dev   = osd_obj2dev(obj);
7225         struct osd_thread_info *oti   = osd_oti_get(env);
7226         struct osd_inode_id    *id    = &oti->oti_id;
7227         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7228         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7229         __u32 ino = it->oie_dirent->oied_ino;
7230         int rc = 0;
7231
7232         ENTRY;
7233
7234         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7235
7236         if (attr & LUDA_VERIFY) {
7237                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7238                         attr |= LUDA_IGNORE;
7239                         /*
7240                          * If the parent is on remote MDT, and there
7241                          * is no FID-in-dirent, then we have to get
7242                          * the parent FID from the linkEA.
7243                          */
7244                         if (!fid_is_sane(fid) &&
7245                             it->oie_dirent->oied_namelen == 2 &&
7246                             it->oie_dirent->oied_name[0] == '.' &&
7247                             it->oie_dirent->oied_name[1] == '.')
7248                                 osd_get_pfid_from_linkea(env, obj, fid);
7249                 } else {
7250                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7251                                                      &attr);
7252                 }
7253
7254                 if (!fid_is_sane(fid))
7255                         attr |= LUDA_UNKNOWN;
7256         } else {
7257                 attr &= ~LU_DIRENT_ATTRS_MASK;
7258                 if (!fid_is_sane(fid)) {
7259                         bool is_dotdot = false;
7260
7261                         if (it->oie_dirent->oied_namelen == 2 &&
7262                             it->oie_dirent->oied_name[0] == '.' &&
7263                             it->oie_dirent->oied_name[1] == '.')
7264                                 is_dotdot = true;
7265                         /*
7266                          * If the parent is on remote MDT, and there
7267                          * is no FID-in-dirent, then we have to get
7268                          * the parent FID from the linkEA.
7269                          */
7270                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7271                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7272                         } else {
7273                                 if (is_dotdot == false &&
7274                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7275                                         RETURN(-ENOENT);
7276
7277                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7278                         }
7279                 }
7280         }
7281
7282         /* Pack the entry anyway, at least the offset is right. */
7283         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7284                            it->oie_dirent->oied_name,
7285                            it->oie_dirent->oied_namelen,
7286                            it->oie_dirent->oied_type, attr);
7287
7288         RETURN(rc > 0 ? 0 : rc);
7289 }
7290
7291 /**
7292  * Returns the record size size at current position.
7293  *
7294  * This function will return record(lu_dirent) size in bytes.
7295  *
7296  * \param[in] env       execution environment
7297  * \param[in] di        iterator's in memory structure
7298  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7299  *                      calculate the lu_dirent size.
7300  *
7301  * \retval      record size(in bytes & in memory) of the current lu_dirent
7302  *              entry.
7303  */
7304 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7305                               __u32 attr)
7306 {
7307         struct osd_it_ea *it = (struct osd_it_ea *)di;
7308
7309         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7310 }
7311
7312 /**
7313  * Returns a cookie for current position of the iterator head, so that
7314  * user can use this cookie to load/start the iterator next time.
7315  *
7316  * \param di iterator's in memory structure
7317  *
7318  * \retval cookie for current position, on success
7319  */
7320 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7321 {
7322         struct osd_it_ea *it = (struct osd_it_ea *)di;
7323
7324         return it->oie_dirent->oied_off;
7325 }
7326
7327 /**
7328  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7329  * to load a directory entry at a time and stored it i inn,
7330  * in iterator's in-memory data structure.
7331  *
7332  * \param di struct osd_it_ea, iterator's in memory structure
7333  *
7334  * \retval +ve on success
7335  * \retval -ve on error
7336  */
7337 static int osd_it_ea_load(const struct lu_env *env,
7338                           const struct dt_it *di, __u64 hash)
7339 {
7340         struct osd_it_ea *it = (struct osd_it_ea *)di;
7341         int rc;
7342
7343         ENTRY;
7344         it->oie_file.f_pos = hash;
7345
7346         rc =  osd_ldiskfs_it_fill(env, di);
7347         if (rc > 0)
7348                 rc = -ENODATA;
7349
7350         if (rc == 0)
7351                 rc = 1;
7352
7353         RETURN(rc);
7354 }
7355
7356 /**
7357  * Index lookup function for interoperability mode (b11826).
7358  *
7359  * \param key,  key i.e. file name to be searched
7360  *
7361  * \retval +ve, on success
7362  * \retval -ve, on error
7363  */
7364 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7365                                struct dt_rec *rec, const struct dt_key *key)
7366 {
7367         struct osd_object *obj = osd_dt_obj(dt);
7368         int rc = 0;
7369
7370         ENTRY;
7371
7372         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7373         LINVRNT(osd_invariant(obj));
7374
7375         rc = osd_ea_lookup_rec(env, obj, rec, key);
7376         if (rc == 0)
7377                 rc = 1;
7378         RETURN(rc);
7379 }
7380
7381 /**
7382  * Index and Iterator operations for interoperability
7383  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7384  */
7385 static const struct dt_index_operations osd_index_ea_ops = {
7386         .dio_lookup         = osd_index_ea_lookup,
7387         .dio_declare_insert = osd_index_declare_ea_insert,
7388         .dio_insert         = osd_index_ea_insert,
7389         .dio_declare_delete = osd_index_declare_ea_delete,
7390         .dio_delete         = osd_index_ea_delete,
7391         .dio_it     = {
7392                 .init     = osd_it_ea_init,
7393                 .fini     = osd_it_ea_fini,
7394                 .get      = osd_it_ea_get,
7395                 .put      = osd_it_ea_put,
7396                 .next     = osd_it_ea_next,
7397                 .key      = osd_it_ea_key,
7398                 .key_size = osd_it_ea_key_size,
7399                 .rec      = osd_it_ea_rec,
7400                 .rec_size = osd_it_ea_rec_size,
7401                 .store    = osd_it_ea_store,
7402                 .load     = osd_it_ea_load
7403         }
7404 };
7405
7406 static void *osd_key_init(const struct lu_context *ctx,
7407                           struct lu_context_key *key)
7408 {
7409         struct osd_thread_info *info;
7410
7411         OBD_ALLOC_PTR(info);
7412         if (info == NULL)
7413                 return ERR_PTR(-ENOMEM);
7414
7415         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7416         if (info->oti_it_ea_buf == NULL)
7417                 goto out_free_info;
7418
7419         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7420
7421         info->oti_hlock = ldiskfs_htree_lock_alloc();
7422         if (info->oti_hlock == NULL)
7423                 goto out_free_ea;
7424
7425         return info;
7426
7427 out_free_ea:
7428         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7429 out_free_info:
7430         OBD_FREE_PTR(info);
7431         return ERR_PTR(-ENOMEM);
7432 }
7433
7434 static void osd_key_fini(const struct lu_context *ctx,
7435                          struct lu_context_key *key, void *data)
7436 {
7437         struct osd_thread_info *info = data;
7438         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7439         struct osd_idmap_cache *idc = info->oti_ins_cache;
7440
7441         if (info->oti_dio_pages) {
7442                 int i;
7443                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7444                         struct page *page = info->oti_dio_pages[i];
7445                         if (page) {
7446                                 LASSERT(PagePrivate2(page));
7447                                 LASSERT(PageLocked(page));
7448                                 ClearPagePrivate2(page);
7449                                 unlock_page(page);
7450                                 __free_page(page);
7451                         }
7452                 }
7453                 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7454                                          PTLRPC_MAX_BRW_PAGES);
7455         }
7456
7457         if (info->oti_inode != NULL)
7458                 OBD_FREE_PTR(lli);
7459         if (info->oti_hlock != NULL)
7460                 ldiskfs_htree_lock_free(info->oti_hlock);
7461         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7462         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7463         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7464         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7465         lu_buf_free(&info->oti_big_buf);
7466         if (idc != NULL) {
7467                 LASSERT(info->oti_ins_cache_size > 0);
7468                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7469                 info->oti_ins_cache = NULL;
7470                 info->oti_ins_cache_size = 0;
7471         }
7472         OBD_FREE_PTR(info);
7473 }
7474
7475 static void osd_key_exit(const struct lu_context *ctx,
7476                          struct lu_context_key *key, void *data)
7477 {
7478         struct osd_thread_info *info = data;
7479
7480         LASSERT(info->oti_r_locks == 0);
7481         LASSERT(info->oti_w_locks == 0);
7482         LASSERT(info->oti_txns    == 0);
7483 }
7484
7485 /* type constructor/destructor: osd_type_init, osd_type_fini */
7486 LU_TYPE_INIT_FINI(osd, &osd_key);
7487
7488 struct lu_context_key osd_key = {
7489         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7490         .lct_init = osd_key_init,
7491         .lct_fini = osd_key_fini,
7492         .lct_exit = osd_key_exit
7493 };
7494
7495
7496 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7497                            const char *name, struct lu_device *next)
7498 {
7499         struct osd_device *osd = osd_dev(d);
7500
7501         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7502             sizeof(osd->od_svname))
7503                 return -E2BIG;
7504         return osd_procfs_init(osd, name);
7505 }
7506
7507 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7508 {
7509         struct seq_server_site *ss = osd_seq_site(osd);
7510         int rc = 0;
7511
7512         ENTRY;
7513
7514         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7515                 RETURN(0);
7516
7517         if (unlikely(ss == NULL))
7518                 RETURN(-ENODEV);
7519
7520         OBD_ALLOC_PTR(osd->od_cl_seq);
7521         if (osd->od_cl_seq == NULL)
7522                 RETURN(-ENOMEM);
7523
7524         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7525                         osd->od_svname, ss->ss_server_seq);
7526
7527         if (ss->ss_node_id == 0) {
7528                 /*
7529                  * If the OSD on the sequence controller(MDT0), then allocate
7530                  * sequence here, otherwise allocate sequence after connected
7531                  * to MDT0 (see mdt_register_lwp_callback()).
7532                  */
7533                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7534                                    &osd->od_cl_seq->lcs_space, env);
7535         }
7536
7537         RETURN(rc);
7538 }
7539
7540 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7541 {
7542         if (osd->od_cl_seq == NULL)
7543                 return;
7544
7545         seq_client_fini(osd->od_cl_seq);
7546         OBD_FREE_PTR(osd->od_cl_seq);
7547         osd->od_cl_seq = NULL;
7548 }
7549
7550 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7551 {
7552         ENTRY;
7553
7554         /* shutdown quota slave instance associated with the device */
7555         if (o->od_quota_slave_md != NULL) {
7556                 struct qsd_instance *qsd = o->od_quota_slave_md;
7557
7558                 o->od_quota_slave_md = NULL;
7559                 qsd_fini(env, qsd);
7560         }
7561
7562         if (o->od_quota_slave_dt != NULL) {
7563                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7564
7565                 o->od_quota_slave_dt = NULL;
7566                 qsd_fini(env, qsd);
7567         }
7568
7569         osd_fid_fini(env, o);
7570         osd_scrub_cleanup(env, o);
7571
7572         RETURN(0);
7573 }
7574
7575 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7576 {
7577         ENTRY;
7578
7579         if (o->od_mnt != NULL) {
7580                 shrink_dcache_sb(osd_sb(o));
7581                 osd_sync(env, &o->od_dt_dev);
7582
7583                 mntput(o->od_mnt);
7584                 o->od_mnt = NULL;
7585         }
7586
7587         EXIT;
7588 }
7589
7590 static int osd_mount(const struct lu_env *env,
7591                      struct osd_device *o, struct lustre_cfg *cfg)
7592 {
7593         const char *name = lustre_cfg_string(cfg, 0);
7594         const char *dev = lustre_cfg_string(cfg, 1);
7595         const char *opts;
7596         unsigned long page, s_flags, lmd_flags = 0;
7597         struct page *__page;
7598         struct file_system_type *type;
7599         char *options = NULL;
7600         char *str;
7601         struct osd_thread_info *info = osd_oti_get(env);
7602         struct lu_fid *fid = &info->oti_fid;
7603         struct inode *inode;
7604         int rc = 0, force_over_1024tb = 0;
7605
7606         ENTRY;
7607
7608         if (o->od_mnt != NULL)
7609                 RETURN(0);
7610
7611         if (strlen(dev) >= sizeof(o->od_mntdev))
7612                 RETURN(-E2BIG);
7613         strcpy(o->od_mntdev, dev);
7614
7615         str = lustre_cfg_string(cfg, 2);
7616         s_flags = simple_strtoul(str, NULL, 0);
7617         str = strstr(str, ":");
7618         if (str)
7619                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7620         opts = lustre_cfg_string(cfg, 3);
7621 #ifdef __BIG_ENDIAN
7622         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7623                 CERROR("%s: device %s extents feature is not guaranteed to "
7624                        "work on big-endian systems. Use \"bigendian_extents\" "
7625                        "mount option to override.\n", name, dev);
7626                 RETURN(-EINVAL);
7627         }
7628 #endif
7629 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7630         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7631                 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");
7632         }
7633 #endif
7634 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7635         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7636                 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");
7637         }
7638 #endif
7639 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7640         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7641                 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");
7642         }
7643 #endif
7644
7645         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7646                 force_over_1024tb = 1;
7647
7648         __page = alloc_page(GFP_KERNEL);
7649         if (__page == NULL)
7650                 GOTO(out, rc = -ENOMEM);
7651         page = (unsigned long)page_address(__page);
7652         options = (char *)page;
7653         *options = '\0';
7654         if (opts != NULL) {
7655                 /* strip out the options for back compatiblity */
7656                 static const char * const sout[] = {
7657                         "mballoc",
7658                         "iopen",
7659                         "noiopen",
7660                         "iopen_nopriv",
7661                         "extents",
7662                         "noextents",
7663                         /* strip out option we processed in osd */
7664                         "bigendian_extents",
7665                         "force_over_128tb",
7666                         "force_over_256tb",
7667                         "force_over_512tb",
7668                         "force_over_1024tb",
7669                         NULL
7670                 };
7671                 strncat(options, opts, PAGE_SIZE);
7672                 for (rc = 0, str = options; sout[rc]; ) {
7673                         char *op = strstr(str, sout[rc]);
7674
7675                         if (op == NULL) {
7676                                 rc++;
7677                                 str = options;
7678                                 continue;
7679                         }
7680                         if (op == options || *(op - 1) == ',') {
7681                                 str = op + strlen(sout[rc]);
7682                                 if (*str == ',' || *str == '\0') {
7683                                         *str == ',' ? str++ : str;
7684                                         memmove(op, str, strlen(str) + 1);
7685                                 }
7686                         }
7687                         for (str = op; *str != ',' && *str != '\0'; str++)
7688                                 ;
7689                 }
7690         } else {
7691                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7692         }
7693
7694         /* Glom up mount options */
7695         if (*options != '\0')
7696                 strncat(options, ",", PAGE_SIZE);
7697         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7698
7699         type = get_fs_type("ldiskfs");
7700         if (!type) {
7701                 CERROR("%s: cannot find ldiskfs module\n", name);
7702                 GOTO(out, rc = -ENODEV);
7703         }
7704
7705         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7706         module_put(type->owner);
7707
7708         if (IS_ERR(o->od_mnt)) {
7709                 rc = PTR_ERR(o->od_mnt);
7710                 o->od_mnt = NULL;
7711                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7712                 GOTO(out, rc);
7713         }
7714
7715         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7716                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7717                                  force_over_1024tb == 0) {
7718                 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",
7719                        name, dev);
7720                 GOTO(out_mnt, rc = -EINVAL);
7721         }
7722
7723         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7724                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7725                               name);
7726
7727                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7728         }
7729
7730         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7731                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7732                 GOTO(out_mnt, rc = -EINVAL);
7733         }
7734
7735 #ifdef LDISKFS_MOUNT_DIRDATA
7736         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7737                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7738         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7739                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7740                       "enabling the dirdata feature. If you do not want to "
7741                       "downgrade to Lustre-1.x again, you can enable it via "
7742                       "'tune2fs -O dirdata device'\n", name, dev);
7743 #endif
7744         inode = osd_sb(o)->s_root->d_inode;
7745         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7746         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7747         if (rc != 0) {
7748                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7749                 GOTO(out_mnt, rc);
7750         }
7751
7752         if (lmd_flags & LMD_FLG_NOSCRUB)
7753                 o->od_auto_scrub_interval = AS_NEVER;
7754
7755         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7756                 /* do not use pagecache with flash-backed storage */
7757                 o->od_writethrough_cache = 0;
7758                 o->od_read_cache = 0;
7759         }
7760
7761         GOTO(out, rc = 0);
7762
7763 out_mnt:
7764         mntput(o->od_mnt);
7765         o->od_mnt = NULL;
7766
7767 out:
7768         if (__page)
7769                 __free_page(__page);
7770
7771         return rc;
7772 }
7773
7774 static struct lu_device *osd_device_fini(const struct lu_env *env,
7775                                          struct lu_device *d)
7776 {
7777         struct osd_device *o = osd_dev(d);
7778
7779         ENTRY;
7780
7781         osd_index_backup(env, o, false);
7782         osd_shutdown(env, o);
7783         osd_procfs_fini(o);
7784         if (o->od_oi_table != NULL)
7785                 osd_oi_fini(osd_oti_get(env), o);
7786         osd_obj_map_fini(o);
7787         osd_umount(env, o);
7788
7789         RETURN(NULL);
7790 }
7791
7792 static int osd_device_init0(const struct lu_env *env,
7793                             struct osd_device *o,
7794                             struct lustre_cfg *cfg)
7795 {
7796         struct lu_device *l = osd2lu_dev(o);
7797         struct osd_thread_info *info;
7798         int rc;
7799         int cplen = 0;
7800
7801         /* if the module was re-loaded, env can loose its keys */
7802         rc = lu_env_refill((struct lu_env *)env);
7803         if (rc)
7804                 GOTO(out, rc);
7805         info = osd_oti_get(env);
7806         LASSERT(info);
7807
7808         l->ld_ops = &osd_lu_ops;
7809         o->od_dt_dev.dd_ops = &osd_dt_ops;
7810
7811         spin_lock_init(&o->od_osfs_lock);
7812         mutex_init(&o->od_otable_mutex);
7813         INIT_LIST_HEAD(&o->od_orphan_list);
7814         INIT_LIST_HEAD(&o->od_index_backup_list);
7815         INIT_LIST_HEAD(&o->od_index_restore_list);
7816         spin_lock_init(&o->od_lock);
7817         o->od_index_backup_policy = LIBP_NONE;
7818         o->od_t10_type = 0;
7819
7820         o->od_read_cache = 1;
7821         o->od_writethrough_cache = 1;
7822         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7823         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7824         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7825         o->od_auto_scrub_interval = AS_DEFAULT;
7826
7827         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7828                         sizeof(o->od_svname));
7829         if (cplen >= sizeof(o->od_svname)) {
7830                 rc = -E2BIG;
7831                 GOTO(out, rc);
7832         }
7833
7834         o->od_index_backup_stop = 0;
7835         o->od_index = -1; /* -1 means index is invalid */
7836         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7837         if (rc == LDD_F_SV_TYPE_OST)
7838                 o->od_is_ost = 1;
7839
7840         o->od_full_scrub_ratio = OFSR_DEFAULT;
7841         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7842         rc = osd_mount(env, o, cfg);
7843         if (rc != 0)
7844                 GOTO(out, rc);
7845
7846         /* Can only check block device after mount */
7847         o->od_nonrotational =
7848                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7849
7850         rc = osd_obj_map_init(env, o);
7851         if (rc != 0)
7852                 GOTO(out_mnt, rc);
7853
7854         rc = lu_site_init(&o->od_site, l);
7855         if (rc != 0)
7856                 GOTO(out_compat, rc);
7857         o->od_site.ls_bottom_dev = l;
7858
7859         rc = lu_site_init_finish(&o->od_site);
7860         if (rc != 0)
7861                 GOTO(out_site, rc);
7862
7863         INIT_LIST_HEAD(&o->od_ios_list);
7864         /* setup scrub, including OI files initialization */
7865         o->od_in_init = 1;
7866         rc = osd_scrub_setup(env, o);
7867         o->od_in_init = 0;
7868         if (rc < 0)
7869                 GOTO(out_site, rc);
7870
7871         rc = osd_procfs_init(o, o->od_svname);
7872         if (rc != 0) {
7873                 CERROR("%s: can't initialize procfs: rc = %d\n",
7874                        o->od_svname, rc);
7875                 GOTO(out_scrub, rc);
7876         }
7877
7878         LASSERT(l->ld_site->ls_linkage.next != NULL);
7879         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7880
7881         /* initialize quota slave instance */
7882         /* currently it's no need to prepare qsd_instance_md for OST */
7883         if (!o->od_is_ost) {
7884                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7885                                                 &o->od_dt_dev,
7886                                                 o->od_proc_entry, true);
7887                 if (IS_ERR(o->od_quota_slave_md)) {
7888                         rc = PTR_ERR(o->od_quota_slave_md);
7889                         o->od_quota_slave_md = NULL;
7890                         GOTO(out_procfs, rc);
7891                 }
7892         }
7893
7894         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7895                                         o->od_proc_entry, false);
7896
7897         if (IS_ERR(o->od_quota_slave_dt)) {
7898                 if (o->od_quota_slave_md != NULL) {
7899                         qsd_fini(env, o->od_quota_slave_md);
7900                         o->od_quota_slave_md = NULL;
7901                 }
7902
7903                 rc = PTR_ERR(o->od_quota_slave_dt);
7904                 o->od_quota_slave_dt = NULL;
7905                 GOTO(out_procfs, rc);
7906         }
7907
7908         RETURN(0);
7909
7910 out_procfs:
7911         osd_procfs_fini(o);
7912 out_scrub:
7913         osd_scrub_cleanup(env, o);
7914 out_site:
7915         lu_site_fini(&o->od_site);
7916 out_compat:
7917         osd_obj_map_fini(o);
7918 out_mnt:
7919         osd_umount(env, o);
7920 out:
7921         return rc;
7922 }
7923
7924 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7925                                           struct lu_device_type *t,
7926                                           struct lustre_cfg *cfg)
7927 {
7928         struct osd_device *o;
7929         int rc;
7930
7931         OBD_ALLOC_PTR(o);
7932         if (o == NULL)
7933                 return ERR_PTR(-ENOMEM);
7934
7935         rc = dt_device_init(&o->od_dt_dev, t);
7936         if (rc == 0) {
7937                 /*
7938                  * Because the ctx might be revived in dt_device_init,
7939                  * refill the env here
7940                  */
7941                 lu_env_refill((struct lu_env *)env);
7942                 rc = osd_device_init0(env, o, cfg);
7943                 if (rc)
7944                         dt_device_fini(&o->od_dt_dev);
7945         }
7946
7947         if (unlikely(rc != 0))
7948                 OBD_FREE_PTR(o);
7949
7950         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7951 }
7952
7953 static struct lu_device *osd_device_free(const struct lu_env *env,
7954                                          struct lu_device *d)
7955 {
7956         struct osd_device *o = osd_dev(d);
7957
7958         ENTRY;
7959
7960         /* XXX: make osd top device in order to release reference */
7961         d->ld_site->ls_top_dev = d;
7962         lu_site_purge(env, d->ld_site, -1);
7963         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
7964                       D_ERROR, lu_cdebug_printer);
7965         lu_site_fini(&o->od_site);
7966         dt_device_fini(&o->od_dt_dev);
7967         OBD_FREE_PTR(o);
7968         RETURN(NULL);
7969 }
7970
7971 static int osd_process_config(const struct lu_env *env,
7972                               struct lu_device *d, struct lustre_cfg *cfg)
7973 {
7974         struct osd_device *o = osd_dev(d);
7975         ssize_t count;
7976         int rc;
7977
7978         ENTRY;
7979
7980         switch (cfg->lcfg_command) {
7981         case LCFG_SETUP:
7982                 rc = osd_mount(env, o, cfg);
7983                 break;
7984         case LCFG_CLEANUP:
7985                 /*
7986                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7987                  * that may happend if hit failure during mount process.
7988                  */
7989                 osd_index_backup(env, o, false);
7990                 lu_dev_del_linkage(d->ld_site, d);
7991                 rc = osd_shutdown(env, o);
7992                 break;
7993         case LCFG_PARAM:
7994                 LASSERT(&o->od_dt_dev);
7995                 count  = class_modify_config(cfg, PARAM_OSD,
7996                                              &o->od_dt_dev.dd_kobj);
7997                 if (count < 0)
7998                         count = class_modify_config(cfg, PARAM_OST,
7999                                                     &o->od_dt_dev.dd_kobj);
8000                 rc = count > 0 ? 0 : count;
8001                 break;
8002         case LCFG_PRE_CLEANUP:
8003                 osd_scrub_stop(o);
8004                 osd_index_backup(env, o,
8005                                  o->od_index_backup_policy != LIBP_NONE);
8006                 rc = 0;
8007                 break;
8008         default:
8009                 rc = -ENOSYS;
8010         }
8011
8012         RETURN(rc);
8013 }
8014
8015 static int osd_recovery_complete(const struct lu_env *env,
8016                                  struct lu_device *d)
8017 {
8018         struct osd_device *osd = osd_dev(d);
8019         int rc = 0;
8020
8021         ENTRY;
8022
8023         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8024                 RETURN(0);
8025
8026         /*
8027          * start qsd instance on recovery completion, this notifies the quota
8028          * slave code that we are about to process new requests now
8029          */
8030         rc = qsd_start(env, osd->od_quota_slave_dt);
8031         if (rc == 0 && osd->od_quota_slave_md != NULL)
8032                 rc = qsd_start(env, osd->od_quota_slave_md);
8033
8034         RETURN(rc);
8035 }
8036
8037 /*
8038  * we use exports to track all osd users
8039  */
8040 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8041                            struct obd_device *obd, struct obd_uuid *cluuid,
8042                            struct obd_connect_data *data, void *localdata)
8043 {
8044         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8045         struct lustre_handle conn;
8046         int rc;
8047
8048         ENTRY;
8049
8050         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8051
8052         rc = class_connect(&conn, obd, cluuid);
8053         if (rc)
8054                 RETURN(rc);
8055
8056         *exp = class_conn2export(&conn);
8057
8058         spin_lock(&osd->od_osfs_lock);
8059         osd->od_connects++;
8060         spin_unlock(&osd->od_osfs_lock);
8061
8062         RETURN(0);
8063 }
8064
8065 /*
8066  * once last export (we don't count self-export) disappeared
8067  * osd can be released
8068  */
8069 static int osd_obd_disconnect(struct obd_export *exp)
8070 {
8071         struct obd_device *obd = exp->exp_obd;
8072         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8073         int rc, release = 0;
8074
8075         ENTRY;
8076
8077         /* Only disconnect the underlying layers on the final disconnect. */
8078         spin_lock(&osd->od_osfs_lock);
8079         osd->od_connects--;
8080         if (osd->od_connects == 0)
8081                 release = 1;
8082         spin_unlock(&osd->od_osfs_lock);
8083
8084         rc = class_disconnect(exp); /* bz 9811 */
8085
8086         if (rc == 0 && release)
8087                 class_manual_cleanup(obd);
8088         RETURN(rc);
8089 }
8090
8091 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8092                        struct lu_device *dev)
8093 {
8094         struct osd_device *osd = osd_dev(dev);
8095         struct lr_server_data *lsd =
8096                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8097         int result = 0;
8098
8099         ENTRY;
8100
8101         if (osd->od_quota_slave_md != NULL) {
8102                 /* set up quota slave objects for inode */
8103                 result = qsd_prepare(env, osd->od_quota_slave_md);
8104                 if (result != 0)
8105                         RETURN(result);
8106         }
8107
8108         if (osd->od_quota_slave_dt != NULL) {
8109                 /* set up quota slave objects for block */
8110                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8111                 if (result != 0)
8112                         RETURN(result);
8113         }
8114
8115
8116         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8117 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8118                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8119                         osd->od_index_in_idif = 1;
8120                 } else {
8121                         osd->od_index_in_idif = 0;
8122                         result = osd_register_proc_index_in_idif(osd);
8123                         if (result != 0)
8124                                 RETURN(result);
8125                 }
8126 #else
8127                 osd->od_index_in_idif = 1;
8128 #endif
8129         }
8130
8131         result = osd_fid_init(env, osd);
8132
8133         RETURN(result);
8134 }
8135
8136 /**
8137  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8138  *
8139  * Allocate FID.
8140  *
8141  * see include/lu_object.h for the details.
8142  */
8143 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8144                          struct lu_fid *fid, struct lu_object *parent,
8145                          const struct lu_name *name)
8146 {
8147         struct osd_device *osd = osd_dev(d);
8148
8149         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8150 }
8151
8152 static const struct lu_object_operations osd_lu_obj_ops = {
8153         .loo_object_init      = osd_object_init,
8154         .loo_object_delete    = osd_object_delete,
8155         .loo_object_release   = osd_object_release,
8156         .loo_object_free      = osd_object_free,
8157         .loo_object_print     = osd_object_print,
8158         .loo_object_invariant = osd_object_invariant
8159 };
8160
8161 const struct lu_device_operations osd_lu_ops = {
8162         .ldo_object_alloc      = osd_object_alloc,
8163         .ldo_process_config    = osd_process_config,
8164         .ldo_recovery_complete = osd_recovery_complete,
8165         .ldo_prepare           = osd_prepare,
8166         .ldo_fid_alloc         = osd_fid_alloc,
8167 };
8168
8169 static const struct lu_device_type_operations osd_device_type_ops = {
8170         .ldto_init = osd_type_init,
8171         .ldto_fini = osd_type_fini,
8172
8173         .ldto_start = osd_type_start,
8174         .ldto_stop  = osd_type_stop,
8175
8176         .ldto_device_alloc = osd_device_alloc,
8177         .ldto_device_free  = osd_device_free,
8178
8179         .ldto_device_init = osd_device_init,
8180         .ldto_device_fini = osd_device_fini
8181 };
8182
8183 static struct lu_device_type osd_device_type = {
8184         .ldt_tags     = LU_DEVICE_DT,
8185         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8186         .ldt_ops      = &osd_device_type_ops,
8187         .ldt_ctx_tags = LCT_LOCAL,
8188 };
8189
8190 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8191 {
8192         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8193         struct super_block *sb = osd_sb(osd);
8194
8195         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8196 }
8197
8198 /*
8199  * lprocfs legacy support.
8200  */
8201 static const struct obd_ops osd_obd_device_ops = {
8202         .o_owner = THIS_MODULE,
8203         .o_connect      = osd_obd_connect,
8204         .o_disconnect   = osd_obd_disconnect,
8205         .o_health_check = osd_health_check,
8206 };
8207
8208 static ssize_t track_declares_assert_show(struct kobject *kobj,
8209                                    struct attribute *attr,
8210                                    char *buf)
8211 {
8212         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8213 }
8214
8215 static ssize_t track_declares_assert_store(struct kobject *kobj,
8216                                            struct attribute *attr,
8217                                            const char *buffer, size_t count)
8218 {
8219         bool track_declares_assert;
8220         int rc;
8221
8222         rc = kstrtobool(buffer, &track_declares_assert);
8223         if (rc)
8224                 return rc;
8225
8226         ldiskfs_track_declares_assert = track_declares_assert;
8227
8228         return count;
8229 }
8230 LUSTRE_RW_ATTR(track_declares_assert);
8231
8232 static int __init osd_init(void)
8233 {
8234         struct kobject *kobj;
8235         int rc;
8236
8237         BUILD_BUG_ON(BH_DXLock >=
8238                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8239 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8240         /* please, try to keep osd_thread_info smaller than a page */
8241         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8242 #endif
8243
8244         osd_oi_mod_init();
8245
8246         rc = lu_kmem_init(ldiskfs_caches);
8247         if (rc)
8248                 return rc;
8249
8250 #ifdef CONFIG_KALLSYMS
8251         priv_security_file_alloc =
8252                 (void *)kallsyms_lookup_name("security_file_alloc");
8253 #endif
8254
8255         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8256                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8257         if (rc) {
8258                 lu_kmem_fini(ldiskfs_caches);
8259                 return rc;
8260         }
8261
8262         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8263         if (kobj) {
8264                 rc = sysfs_create_file(kobj,
8265                                        &lustre_attr_track_declares_assert.attr);
8266                 kobject_put(kobj);
8267                 if (rc) {
8268                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8269                         rc = 0;
8270                 }
8271         }
8272         return rc;
8273 }
8274
8275 static void __exit osd_exit(void)
8276 {
8277         struct kobject *kobj;
8278
8279         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8280         if (kobj) {
8281                 sysfs_remove_file(kobj,
8282                                   &lustre_attr_track_declares_assert.attr);
8283                 kobject_put(kobj);
8284         }
8285         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8286         lu_kmem_fini(ldiskfs_caches);
8287 }
8288
8289 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8290 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8291 MODULE_VERSION(LUSTRE_VERSION_STRING);
8292 MODULE_LICENSE("GPL");
8293
8294 module_init(osd_init);
8295 module_exit(osd_exit);