lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/fs_struct.h>
  43 #include <linux/kallsyms.h>
  44 #include <linux/module.h>
  45 #include <linux/user_namespace.h>
  46 #include <linux/uidgid.h>
  47
  48 /* prerequisite for linux/xattr.h */
  49 #include <linux/types.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/fs.h>
  52 /* XATTR_{REPLACE,CREATE} */
  53 #include <linux/xattr.h>
  54
  55 #include <ldiskfs/ldiskfs.h>
  56 #include <ldiskfs/xattr.h>
  57 #include <ldiskfs/ldiskfs_extents.h>
  58 #undef ENTRY
  59 /*
  60  * struct OBD_{ALLOC,FREE}*()
  61  * OBD_FAIL_CHECK
  62  */
  63 #include <obd_support.h>
  64 /* struct ptlrpc_thread */
  65 #include <lustre_net.h>
  66 #include <lustre_fid.h>
  67 /* process_config */
  68 #include <uapi/linux/lustre/lustre_param.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_dynlocks.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75 #include <lustre_quota.h>
  76
  77 #include <lustre_linkea.h>
  78
  79 /* Maximum EA size is limited by LNET_MTU for remote objects */
  80 #define OSD_MAX_EA_SIZE 1048364
  81
  82 int ldiskfs_pdo = 1;
  83 module_param(ldiskfs_pdo, int, 0644);
  84 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  85
  86 int ldiskfs_track_declares_assert;
  87 module_param(ldiskfs_track_declares_assert, int, 0644);
  88 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  89
  90 /* Slab to allocate dynlocks */
  91 struct kmem_cache *dynlock_cachep;
  92
  93 /* Slab to allocate osd_it_ea */
  94 struct kmem_cache *osd_itea_cachep;
  95
  96 static struct lu_kmem_descr ldiskfs_caches[] = {
  97         {
  98                 .ckd_cache = &dynlock_cachep,
  99                 .ckd_name  = "dynlock_cache",
 100                 .ckd_size  = sizeof(struct dynlock_handle)
 101         },
 102         {
 103                 .ckd_cache = &osd_itea_cachep,
 104                 .ckd_name  = "osd_itea_cache",
 105                 .ckd_size  = sizeof(struct osd_it_ea)
 106         },
 107         {
 108                 .ckd_cache = NULL
 109         }
 110 };
 111
 112 static const char dot[] = ".";
 113 static const char dotdot[] = "..";
 114
 115 static const struct lu_object_operations      osd_lu_obj_ops;
 116 static const struct dt_object_operations      osd_obj_ops;
 117 static const struct dt_object_operations      osd_obj_otable_it_ops;
 118 static const struct dt_index_operations       osd_index_iam_ops;
 119 static const struct dt_index_operations       osd_index_ea_ops;
 120
 121 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 122                           const struct lu_fid *fid);
 123 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 124                                                 struct osd_device *osd);
 125
 126 int osd_trans_declare_op2rb[] = {
 127         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 128         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 129         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 130         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 131         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 132         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 133         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 134         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 135         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 136         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 137         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 138 };
 139
 140 static int osd_has_index(const struct osd_object *obj)
 141 {
 142         return obj->oo_dt.do_index_ops != NULL;
 143 }
 144
 145 static int osd_object_invariant(const struct lu_object *l)
 146 {
 147         return osd_invariant(osd_obj(l));
 148 }
 149
 150 /*
 151  * Concurrency: doesn't matter
 152  */
 153 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 154 {
 155         struct osd_thread_info *oti = osd_oti_get(env);
 156
 157         return oti->oti_w_locks > 0 && o->oo_owner == env;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't access mutable data
 162  */
 163 static int osd_root_get(const struct lu_env *env,
 164                         struct dt_device *dev, struct lu_fid *f)
 165 {
 166         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 167         return 0;
 168 }
 169
 170 /*
 171  * the following set of functions are used to maintain per-thread
 172  * cache of FID->ino mapping. this mechanism is needed to resolve
 173  * FID to inode at dt_insert() which in turn stores ino in the
 174  * directory entries to keep ldiskfs compatible with ext[34].
 175  * due to locking-originated restrictions we can't lookup ino
 176  * using LU cache (deadlock is possible). lookup using OI is quite
 177  * expensive. so instead we maintain this cache and methods like
 178  * dt_create() fill it. so in the majority of cases dt_insert() is
 179  * able to find needed mapping in lockless manner.
 180  */
 181 static struct osd_idmap_cache *
 182 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 183              const struct lu_fid *fid)
 184 {
 185         struct osd_thread_info *oti = osd_oti_get(env);
 186         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 187         int i;
 188
 189         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 190                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 191                         continue;
 192                 if (idc[i].oic_dev != osd)
 193                         continue;
 194
 195                 return idc + i;
 196         }
 197
 198         return NULL;
 199 }
 200
 201 static struct osd_idmap_cache *
 202 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 203             const struct lu_fid *fid)
 204 {
 205         struct osd_thread_info *oti   = osd_oti_get(env);
 206         struct osd_idmap_cache *idc;
 207         int i;
 208
 209         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 210                 i = oti->oti_ins_cache_size * 2;
 211                 if (i == 0)
 212                         i = OSD_INS_CACHE_SIZE;
 213                 OBD_ALLOC_PTR_ARRAY(idc, i);
 214                 if (idc == NULL)
 215                         return ERR_PTR(-ENOMEM);
 216                 if (oti->oti_ins_cache != NULL) {
 217                         memcpy(idc, oti->oti_ins_cache,
 218                                oti->oti_ins_cache_used * sizeof(*idc));
 219                         OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
 220                                            oti->oti_ins_cache_used);
 221                 }
 222                 oti->oti_ins_cache = idc;
 223                 oti->oti_ins_cache_size = i;
 224         }
 225
 226         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 227         idc->oic_fid = *fid;
 228         idc->oic_dev = osd;
 229         idc->oic_lid.oii_ino = 0;
 230         idc->oic_lid.oii_gen = 0;
 231         idc->oic_remote = 0;
 232
 233         return idc;
 234 }
 235
 236 /*
 237  * lookup mapping for the given fid in the cache, initialize a
 238  * new one if not found. the initialization checks whether the
 239  * object is local or remote. for local objects, OI is used to
 240  * learn ino/generation. the function is used when the caller
 241  * has no information about the object, e.g. at dt_insert().
 242  */
 243 static struct osd_idmap_cache *
 244 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 245                      const struct lu_fid *fid)
 246 {
 247         struct osd_idmap_cache *idc;
 248         int rc;
 249
 250         idc = osd_idc_find(env, osd, fid);
 251         LASSERT(!IS_ERR(idc));
 252         if (idc != NULL)
 253                 return idc;
 254
 255         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 256                osd->od_svname, PFID(fid));
 257
 258         /* new mapping is needed */
 259         idc = osd_idc_add(env, osd, fid);
 260         if (IS_ERR(idc)) {
 261                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 262                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 263                 return idc;
 264         }
 265
 266         /* initialize it */
 267         rc = osd_remote_fid(env, osd, fid);
 268         if (unlikely(rc < 0))
 269                 return ERR_PTR(rc);
 270
 271         if (rc == 0) {
 272                 /* the object is local, lookup in OI */
 273                 /* XXX: probably cheaper to lookup in LU first? */
 274                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 275                                    &idc->oic_lid, 0);
 276                 if (unlikely(rc < 0)) {
 277                         CERROR("can't lookup: rc = %d\n", rc);
 278                         return ERR_PTR(rc);
 279                 }
 280         } else {
 281                 /* the object is remote */
 282                 idc->oic_remote = 1;
 283         }
 284
 285         return idc;
 286 }
 287
 288 /*
 289  * lookup mapping for given FID and fill it from the given object.
 290  * the object is lolcal by definition.
 291  */
 292 static int osd_idc_find_and_init(const struct lu_env *env,
 293                                  struct osd_device *osd,
 294                                  struct osd_object *obj)
 295 {
 296         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 297         struct osd_idmap_cache *idc;
 298
 299         idc = osd_idc_find(env, osd, fid);
 300         LASSERT(!IS_ERR(idc));
 301         if (idc != NULL) {
 302                 if (obj->oo_inode == NULL)
 303                         return 0;
 304                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 305                         LASSERT(idc->oic_lid.oii_ino == 0);
 306                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 307                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 308                 }
 309                 return 0;
 310         }
 311
 312         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 313                osd->od_svname, PFID(fid));
 314
 315         /* new mapping is needed */
 316         idc = osd_idc_add(env, osd, fid);
 317         if (IS_ERR(idc)) {
 318                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 319                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 320                 return PTR_ERR(idc);
 321         }
 322
 323         if (obj->oo_inode != NULL) {
 324                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 325                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 326         }
 327         return 0;
 328 }
 329
 330 /*
 331  * OSD object methods.
 332  */
 333
 334 /*
 335  * Concurrency: no concurrent access is possible that early in object
 336  * life-cycle.
 337  */
 338 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 339                                           const struct lu_object_header *hdr,
 340                                           struct lu_device *d)
 341 {
 342         struct osd_object *mo;
 343
 344         OBD_ALLOC_PTR(mo);
 345         if (mo != NULL) {
 346                 struct lu_object *l;
 347                 struct lu_object_header *h;
 348                 struct osd_device *o = osd_dev(d);
 349
 350                 l = &mo->oo_dt.do_lu;
 351                 if (unlikely(o->od_in_init)) {
 352                         OBD_ALLOC_PTR(h);
 353                         if (!h) {
 354                                 OBD_FREE_PTR(mo);
 355                                 return NULL;
 356                         }
 357
 358                         lu_object_header_init(h);
 359                         lu_object_init(l, h, d);
 360                         lu_object_add_top(h, l);
 361                         mo->oo_header = h;
 362                 } else {
 363                         dt_object_init(&mo->oo_dt, NULL, d);
 364                         mo->oo_header = NULL;
 365                 }
 366
 367                 mo->oo_dt.do_ops = &osd_obj_ops;
 368                 l->lo_ops = &osd_lu_obj_ops;
 369                 init_rwsem(&mo->oo_sem);
 370                 init_rwsem(&mo->oo_ext_idx_sem);
 371                 spin_lock_init(&mo->oo_guard);
 372                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 373                 return l;
 374         }
 375         return NULL;
 376 }
 377
 378 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 379                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 380 {
 381         int rc;
 382
 383         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 384                              (void *)loa, sizeof(*loa));
 385         if (rc > 0) {
 386                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 387
 388                 if (rc < sizeof(*lma))
 389                         return -EINVAL;
 390
 391                 rc = 0;
 392                 lustre_loa_swab(loa, true);
 393                 /* Check LMA compatibility */
 394                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 395                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 396                               "for fid = "DFID", ino = %lu\n",
 397                               osd_ino2name(inode),
 398                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 399                               PFID(&lma->lma_self_fid), inode->i_ino);
 400                         rc = -EOPNOTSUPP;
 401                 }
 402         } else if (rc == 0) {
 403                 rc = -ENODATA;
 404         }
 405
 406         return rc;
 407 }
 408
 409 /*
 410  * retrieve object from backend ext fs.
 411  **/
 412 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 413                        struct osd_inode_id *id)
 414 {
 415         int rc;
 416         struct inode *inode = NULL;
 417
 418         /*
 419          * if we look for an inode withing a running
 420          * transaction, then we risk to deadlock
 421          * osd_dirent_check_repair() breaks this
 422          */
 423          /* LASSERT(current->journal_info == NULL); */
 424
 425         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 426         if (IS_ERR(inode)) {
 427                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 428                        id->oii_ino, PTR_ERR(inode));
 429         } else if (id->oii_gen != OSD_OII_NOGEN &&
 430                    inode->i_generation != id->oii_gen) {
 431                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 432                        "i_generation = %u\n",
 433                        id->oii_ino, id->oii_gen, inode->i_generation);
 434                 iput(inode);
 435                 inode = ERR_PTR(-ESTALE);
 436         } else if (inode->i_nlink == 0) {
 437                 /*
 438                  * due to parallel readdir and unlink,
 439                  * we can have dead inode here.
 440                  */
 441                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 442                 iput(inode);
 443                 inode = ERR_PTR(-ESTALE);
 444         } else if (is_bad_inode(inode)) {
 445                 CWARN("%s: bad inode: ino = %u\n",
 446                       osd_dev2name(dev), id->oii_ino);
 447                 iput(inode);
 448                 inode = ERR_PTR(-ENOENT);
 449         } else if ((rc = osd_attach_jinode(inode))) {
 450                 iput(inode);
 451                 inode = ERR_PTR(rc);
 452         } else {
 453                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 454                 if (id->oii_gen == OSD_OII_NOGEN)
 455                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 456
 457                 /*
 458                  * Do not update file c/mtime in ldiskfs.
 459                  * NB: we don't have any lock to protect this because we don't
 460                  * have reference on osd_object now, but contention with
 461                  * another lookup + attr_set can't happen in the tiny window
 462                  * between if (...) and set S_NOCMTIME.
 463                  */
 464                 if (!(inode->i_flags & S_NOCMTIME))
 465                         inode->i_flags |= S_NOCMTIME;
 466         }
 467         return inode;
 468 }
 469
 470 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 471                           handle_t *handle, struct dentry *child,
 472                           struct inode *inode, struct htree_lock *hlock)
 473 {
 474         int rc, rc2;
 475
 476         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 477         if (rc == -ENOBUFS || rc == -ENOSPC) {
 478                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 479                 struct inode *parent = child->d_parent->d_inode;
 480                 struct lu_fid *fid = NULL;
 481                 char fidstr[FID_LEN + 1] = "unknown";
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
 499
 500                 /* below message is checked in sanity.sh test_129 */
 501                 if (rc == -ENOSPC) {
 502                         CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
 503                               osd_name(osd), parent->i_ino, fidstr);
 504                 } else {
 505                         rc = 0; /* ignore such error now */
 506                         CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
 507                               osd_name(osd), parent->i_ino, fidstr);
 508                 }
 509
 510         }
 511
 512         return rc;
 513 }
 514
 515
 516 struct inode *
 517 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 518              struct osd_inode_id *id, struct lu_fid *fid)
 519 {
 520         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 521         struct inode *inode;
 522         int rc;
 523
 524         inode = osd_iget(info, dev, id);
 525         if (IS_ERR(inode))
 526                 return inode;
 527
 528         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 529         if (!rc) {
 530                 *fid = loa->loa_lma.lma_self_fid;
 531         } else if (rc == -ENODATA) {
 532                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 533                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 534                 else
 535                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 536         } else {
 537                 iput(inode);
 538                 inode = ERR_PTR(rc);
 539         }
 540         return inode;
 541 }
 542
 543 static struct inode *osd_iget_check(struct osd_thread_info *info,
 544                                     struct osd_device *dev,
 545                                     const struct lu_fid *fid,
 546                                     struct osd_inode_id *id,
 547                                     bool trusted)
 548 {
 549         struct inode *inode;
 550         int rc = 0;
 551
 552         ENTRY;
 553
 554         /*
 555          * The cached OI mapping is trustable. If we cannot locate the inode
 556          * via the cached OI mapping, then return the failure to the caller
 557          * directly without further OI checking.
 558          */
 559
 560 again:
 561         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 562         if (IS_ERR(inode)) {
 563                 rc = PTR_ERR(inode);
 564                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 565                         goto check_oi;
 566
 567                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 568                        PFID(fid), id->oii_ino, rc);
 569                 GOTO(put, rc);
 570         }
 571
 572         if (is_bad_inode(inode)) {
 573                 rc = -ENOENT;
 574                 if (!trusted)
 575                         goto check_oi;
 576
 577                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 578                        PFID(fid), id->oii_ino);
 579                 GOTO(put, rc);
 580         }
 581
 582         if (id->oii_gen != OSD_OII_NOGEN &&
 583             inode->i_generation != id->oii_gen) {
 584                 rc = -ESTALE;
 585                 if (!trusted)
 586                         goto check_oi;
 587
 588                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 589                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 590                        id->oii_ino, id->oii_gen, inode->i_generation);
 591                 GOTO(put, rc);
 592         }
 593
 594         if (inode->i_nlink == 0) {
 595                 rc = -ENOENT;
 596                 if (!trusted)
 597                         goto check_oi;
 598
 599                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 600                        PFID(fid), id->oii_ino);
 601                 GOTO(put, rc);
 602         }
 603
 604         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 605
 606 check_oi:
 607         if (rc != 0) {
 608                 __u32 saved_ino = id->oii_ino;
 609                 __u32 saved_gen = id->oii_gen;
 610
 611                 LASSERT(!trusted);
 612                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 613
 614                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 615                 /*
 616                  * XXX: There are four possible cases:
 617                  *      1. rc = 0.
 618                  *         Backup/restore caused the OI invalid.
 619                  *      2. rc = 0.
 620                  *         Someone unlinked the object but NOT removed
 621                  *         the OI mapping, such as mount target device
 622                  *         as ldiskfs, and modify something directly.
 623                  *      3. rc = -ENOENT.
 624                  *         Someone just removed the object between the
 625                  *         former oi_lookup and the iget. It is normal.
 626                  *      4. Other failure cases.
 627                  *
 628                  *      Generally, when the device is mounted, it will
 629                  *      auto check whether the system is restored from
 630                  *      file-level backup or not. We trust such detect
 631                  *      to distinguish the 1st case from the 2nd case:
 632                  *      if the OI files are consistent but may contain
 633                  *      stale OI mappings because of case 2, if iget()
 634                  *      returns -ENOENT or -ESTALE, then it should be
 635                  *      the case 2.
 636                  */
 637                 if (rc != 0)
 638                         /*
 639                          * If the OI mapping was in OI file before the
 640                          * osd_iget_check(), but now, it is disappear,
 641                          * then it must be removed by race. That is a
 642                          * normal race case.
 643                          */
 644                         GOTO(put, rc);
 645
 646                 /*
 647                  * It is the OI scrub updated the OI mapping by race.
 648                  * The new OI mapping must be valid.
 649                  */
 650                 if (saved_ino != id->oii_ino ||
 651                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 652                         if (!IS_ERR(inode))
 653                                 iput(inode);
 654
 655                         trusted = true;
 656                         goto again;
 657                 }
 658
 659                 if (IS_ERR(inode)) {
 660                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 661                             SF_INCONSISTENT)
 662                                 /*
 663                                  * It still can be the case 2, but we cannot
 664                                  * distinguish it from the case 1. So return
 665                                  * -EREMCHG to block current operation until
 666                                  *  OI scrub rebuilt the OI mappings.
 667                                  */
 668                                 rc = -EREMCHG;
 669                         else
 670                                 rc = -ENOENT;
 671
 672                         GOTO(put, rc);
 673                 }
 674
 675                 if (inode->i_generation == id->oii_gen)
 676                         rc = -ENOENT;
 677                 else
 678                         rc = -EREMCHG;
 679         } else {
 680                 if (id->oii_gen == OSD_OII_NOGEN)
 681                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 682
 683                 /*
 684                  * Do not update file c/mtime in ldiskfs.
 685                  * NB: we don't have any lock to protect this because we don't
 686                  * have reference on osd_object now, but contention with
 687                  * another lookup + attr_set can't happen in the tiny window
 688                  * between if (...) and set S_NOCMTIME.
 689                  */
 690                 if (!(inode->i_flags & S_NOCMTIME))
 691                         inode->i_flags |= S_NOCMTIME;
 692         }
 693
 694         GOTO(put, rc);
 695
 696 put:
 697         if (rc != 0) {
 698                 if (!IS_ERR(inode))
 699                         iput(inode);
 700
 701                 inode = ERR_PTR(rc);
 702         }
 703
 704         return inode;
 705 }
 706
 707 /**
 708  * \retval +v: new filter_fid does not contain self-fid
 709  * \retval 0:  filter_fid_18_23, contains self-fid
 710  * \retval -v: other failure cases
 711  */
 712 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 713                  struct dentry *dentry, struct lu_fid *fid)
 714 {
 715         struct filter_fid *ff = &info->oti_ff;
 716         struct ost_id *ostid = &info->oti_ostid;
 717         int rc;
 718
 719         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 720         if (rc == sizeof(struct filter_fid_18_23)) {
 721                 struct filter_fid_18_23 *ff_old = (void *)ff;
 722
 723                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 724                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 725                 /*
 726                  * XXX: use 0 as the index for compatibility, the caller will
 727                  * handle index related issues when necessary.
 728                  */
 729                 if (!rc)
 730                         ostid_to_fid(fid, ostid, 0);
 731         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 732                 rc = 1;
 733         } else if (rc >= 0) {
 734                 rc = -EINVAL;
 735         }
 736
 737         return rc;
 738 }
 739
 740 static int osd_lma_self_repair(struct osd_thread_info *info,
 741                                struct osd_device *osd, struct inode *inode,
 742                                const struct lu_fid *fid, __u32 compat)
 743 {
 744         handle_t *jh;
 745         int rc;
 746
 747         LASSERT(current->journal_info == NULL);
 748
 749         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 750                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 751         if (IS_ERR(jh)) {
 752                 rc = PTR_ERR(jh);
 753                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 754                       osd_name(osd), rc);
 755                 return rc;
 756         }
 757
 758         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 759         if (rc != 0)
 760                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 761                       osd_name(osd), rc);
 762         ldiskfs_journal_stop(jh);
 763         return rc;
 764 }
 765
 766 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 767 {
 768         struct osd_thread_info *info = osd_oti_get(env);
 769         struct osd_device *osd = osd_obj2dev(obj);
 770         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 771         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 772         struct inode *inode = obj->oo_inode;
 773         struct dentry *dentry = &info->oti_obj_dentry;
 774         struct lu_fid *fid = NULL;
 775         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 776         int rc;
 777
 778         ENTRY;
 779
 780         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 781                              (void *)loa, sizeof(*loa));
 782         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 783                 fid = &lma->lma_self_fid;
 784                 rc = osd_get_idif(info, inode, dentry, fid);
 785                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 786                         /*
 787                          * For the given OST-object, if it has neither LMA nor
 788                          * FID in XATTR_NAME_FID, then the given FID (which is
 789                          * contained in the @obj, from client RPC for locating
 790                          * the OST-object) is trusted. We use it to generate
 791                          * the LMA.
 792                          */
 793                         osd_lma_self_repair(info, osd, inode, rfid,
 794                                             LMAC_FID_ON_OST);
 795                         RETURN(0);
 796                 }
 797         }
 798
 799         if (rc < 0)
 800                 RETURN(rc);
 801
 802         if (rc > 0) {
 803                 rc = 0;
 804                 lustre_lma_swab(lma);
 805                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 806                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 807                               S_ISREG(inode->i_mode)))) {
 808                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 809                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 810                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 811                               PFID(rfid), inode->i_ino);
 812                         rc = -EOPNOTSUPP;
 813                 } else {
 814                         fid = &lma->lma_self_fid;
 815                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 816                             osd->od_is_ost)
 817                                 obj->oo_pfid_in_lma = 1;
 818                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 819                             !osd->od_is_ost)
 820                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 821                 }
 822         }
 823
 824         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 825                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 826                         struct ost_id   *oi   = &info->oti_ostid;
 827                         struct lu_fid   *fid1 = &info->oti_fid3;
 828                         __u32            idx  = fid_idif_ost_idx(rfid);
 829
 830                         /*
 831                          * For old IDIF, the OST index is not part of the IDIF,
 832                          * Means that different OSTs may have the same IDIFs.
 833                          * Under such case, we need to make some compatible
 834                          * check to make sure to trigger OI scrub properly.
 835                          */
 836                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 837                                 /* Given @rfid is new, LMA is old. */
 838                                 fid_to_ostid(fid, oi);
 839                                 ostid_to_fid(fid1, oi, idx);
 840                                 if (lu_fid_eq(fid1, rfid)) {
 841                                         if (osd->od_index_in_idif)
 842                                                 osd_lma_self_repair(info, osd,
 843                                                         inode, rfid,
 844                                                         LMAC_FID_ON_OST);
 845                                         RETURN(0);
 846                                 }
 847                         }
 848                 }
 849
 850                 rc = -EREMCHG;
 851         }
 852
 853         RETURN(rc);
 854 }
 855
 856 struct osd_check_lmv_buf {
 857         /* please keep it as first member */
 858         struct dir_context ctx;
 859         struct osd_thread_info *oclb_info;
 860         struct osd_device *oclb_dev;
 861         struct osd_idmap_cache *oclb_oic;
 862         int oclb_items;
 863         bool oclb_found;
 864 };
 865
 866 /**
 867  * It is called internally by ->iterate*() to filter out the
 868  * local slave object's FID of the striped directory.
 869  *
 870  * \retval      1 found the local slave's FID
 871  * \retval      0 continue to check next item
 872  * \retval      -ve for failure
 873  */
 874 #ifdef HAVE_FILLDIR_USE_CTX
 875 static int osd_stripe_dir_filldir(struct dir_context *buf,
 876 #else
 877 static int osd_stripe_dir_filldir(void *buf,
 878 #endif
 879                                   const char *name, int namelen,
 880                                   loff_t offset, __u64 ino, unsigned int d_type)
 881 {
 882         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 883         struct osd_thread_info *oti = oclb->oclb_info;
 884         struct lu_fid *fid = &oti->oti_fid3;
 885         struct osd_inode_id *id = &oti->oti_id3;
 886         struct osd_device *dev = oclb->oclb_dev;
 887         struct osd_idmap_cache *oic = oclb->oclb_oic;
 888         struct inode *inode;
 889
 890         oclb->oclb_items++;
 891
 892         if (name[0] == '.')
 893                 return 0;
 894
 895         fid_zero(fid);
 896         sscanf(name + 1, SFID, RFID(fid));
 897         if (!fid_is_sane(fid))
 898                 return 0;
 899
 900         if (osd_remote_fid(oti->oti_env, dev, fid))
 901                 return 0;
 902
 903         osd_id_gen(id, ino, OSD_OII_NOGEN);
 904         inode = osd_iget(oti, dev, id);
 905         if (IS_ERR(inode))
 906                 return PTR_ERR(inode);
 907
 908         iput(inode);
 909         osd_add_oi_cache(oti, dev, id, fid);
 910         oic->oic_fid = *fid;
 911         oic->oic_lid = *id;
 912         oic->oic_dev = dev;
 913         osd_oii_insert(dev, oic, true);
 914         oclb->oclb_found = true;
 915
 916         return 1;
 917 }
 918
 919 /*
 920  * When lookup item under striped directory, we need to locate the master
 921  * MDT-object of the striped directory firstly, then the client will send
 922  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 923  * and the item's name. If the system is restored from MDT file level backup,
 924  * then before the OI scrub completely built the OI files, the OI mappings of
 925  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 926  * not a problem for the master MDT-object. Because when locate the master
 927  * MDT-object, we will do name based lookup (for the striped directory itself)
 928  * firstly, during such process we can setup the correct OI mapping for the
 929  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 930  * the client will not trigger name based lookup on the MDT to locate the slave
 931  * MDT-object before locating item under the striped directory, then when
 932  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 933  * is invalid and does not know what the right OI mapping is, then the MDT has
 934  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 935  * the OI file, related OI mapping is unknown yet, please try again later. And
 936  * then client will re-try the RPC again and again until related OI mapping has
 937  * been updated. That is quite inefficient.
 938  *
 939  * To resolve above trouble, we will handle it as the following two cases:
 940  *
 941  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 942  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 943  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 944  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 945  *    directly. Please check osd_fid_lookup().
 946  *
 947  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 948  *    Under such case, during lookup the master MDT-object, we will lookup the
 949  *    slave MDT-object via readdir against the master MDT-object, because the
 950  *    slave MDT-objects information are stored as sub-directories with the name
 951  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 952  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 953  *    the correct OI mapping for the slave MDT-object.
 954  */
 955 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 956                          struct inode *inode, struct osd_idmap_cache *oic)
 957 {
 958         struct lu_buf *buf = &oti->oti_big_buf;
 959         struct dentry *dentry = &oti->oti_obj_dentry;
 960         struct file *filp = &oti->oti_file;
 961         const struct file_operations *fops;
 962         struct lmv_mds_md_v1 *lmv1;
 963         struct osd_check_lmv_buf oclb = {
 964                 .ctx.actor = osd_stripe_dir_filldir,
 965                 .oclb_info = oti,
 966                 .oclb_dev = dev,
 967                 .oclb_oic = oic,
 968                 .oclb_found = false,
 969         };
 970         int rc = 0;
 971
 972         ENTRY;
 973
 974 again:
 975         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 976                              buf->lb_len);
 977         if (rc == -ERANGE) {
 978                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 979                 if (rc > 0) {
 980                         lu_buf_realloc(buf, rc);
 981                         if (buf->lb_buf == NULL)
 982                                 GOTO(out, rc = -ENOMEM);
 983
 984                         goto again;
 985                 }
 986         }
 987
 988         if (unlikely(rc == 0 || rc == -ENODATA))
 989                 GOTO(out, rc = 0);
 990
 991         if (rc < 0)
 992                 GOTO(out, rc);
 993
 994         if (unlikely(buf->lb_buf == NULL)) {
 995                 lu_buf_realloc(buf, rc);
 996                 if (buf->lb_buf == NULL)
 997                         GOTO(out, rc = -ENOMEM);
 998
 999                 goto again;
1000         }
1001
1002         lmv1 = buf->lb_buf;
1003         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1004                 GOTO(out, rc = 0);
1005
1006         fops = inode->i_fop;
1007         dentry->d_inode = inode;
1008         dentry->d_sb = inode->i_sb;
1009         filp->f_pos = 0;
1010         filp->f_path.dentry = dentry;
1011         filp->f_flags |= O_NOATIME;
1012         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1013         filp->f_mapping = inode->i_mapping;
1014         filp->f_op = fops;
1015         filp->private_data = NULL;
1016         filp->f_cred = current_cred();
1017         filp->f_inode = inode;
1018         rc = osd_security_file_alloc(filp);
1019         if (rc)
1020                 goto out;
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024                 rc = iterate_dir(filp, &oclb.ctx);
1025         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1026                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1027         fops->release(inode, filp);
1028
1029 out:
1030         if (rc < 0)
1031                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1032                        DFID": rc = %d\n", osd_ino2name(inode),
1033                        inode->i_ino, inode->i_generation,
1034                        PFID(&oic->oic_fid), rc);
1035         else
1036                 rc = 0;
1037
1038         RETURN(rc);
1039 }
1040
1041 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1042                           const struct lu_fid *fid,
1043                           const struct lu_object_conf *conf)
1044 {
1045         struct osd_thread_info *info;
1046         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1047         struct osd_device *dev;
1048         struct osd_idmap_cache *oic;
1049         struct osd_inode_id *id;
1050         struct inode *inode = NULL;
1051         struct lustre_scrub *scrub;
1052         struct scrub_file *sf;
1053         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1054         __u32 saved_ino;
1055         __u32 saved_gen;
1056         int result = 0;
1057         int rc1 = 0;
1058         bool remote = false;
1059         bool trusted = true;
1060         bool updated = false;
1061         bool checked = false;
1062
1063         ENTRY;
1064
1065         LINVRNT(osd_invariant(obj));
1066         LASSERT(obj->oo_inode == NULL);
1067         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1068
1069         dev = osd_dev(ldev);
1070         scrub = &dev->od_scrub.os_scrub;
1071         sf = &scrub->os_file;
1072         info = osd_oti_get(env);
1073         LASSERT(info);
1074         oic = &info->oti_cache;
1075
1076         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1077                 RETURN(-ENOENT);
1078
1079         /*
1080          * For the object is created as locking anchor, or for the object to
1081          * be created on disk. No need to osd_oi_lookup() at here because FID
1082          * shouldn't never be re-used, if it's really a duplicate FID from
1083          * unexpected reason, we should be able to detect it later by calling
1084          * do_create->osd_oi_insert().
1085          */
1086         if (conf && conf->loc_flags & LOC_F_NEW)
1087                 GOTO(out, result = 0);
1088
1089         /* Search order: 1. per-thread cache. */
1090         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1091                 id = &oic->oic_lid;
1092                 goto iget;
1093         }
1094
1095         id = &info->oti_id;
1096         if (!list_empty(&scrub->os_inconsistent_items)) {
1097                 /* Search order: 2. OI scrub pending list. */
1098                 result = osd_oii_lookup(dev, fid, id);
1099                 if (!result)
1100                         goto iget;
1101         }
1102
1103         /*
1104          * The OI mapping in the OI file can be updated by the OI scrub
1105          * when we locate the inode via FID. So it may be not trustable.
1106          */
1107         trusted = false;
1108
1109         /* Search order: 3. OI files. */
1110         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1111         if (result == -ENOENT) {
1112                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1113                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1114                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1115                                       sf->sf_oi_bitmap))
1116                         GOTO(out, result = 0);
1117
1118                 goto trigger;
1119         }
1120
1121         /* -ESTALE is returned if inode of OST object doesn't exist */
1122         if (result == -ESTALE &&
1123             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1124                 GOTO(out, result = 0);
1125         }
1126
1127         if (result)
1128                 GOTO(out, result);
1129
1130 iget:
1131         obj->oo_inode = NULL;
1132         /* for later passes through checks, not true on first pass */
1133         if (!IS_ERR_OR_NULL(inode))
1134                 iput(inode);
1135
1136         inode = osd_iget_check(info, dev, fid, id, trusted);
1137         if (!IS_ERR(inode)) {
1138                 obj->oo_inode = inode;
1139                 result = 0;
1140                 if (remote)
1141                         goto trigger;
1142
1143                 goto check_lma;
1144         }
1145
1146         result = PTR_ERR(inode);
1147         if (result == -ENOENT || result == -ESTALE)
1148                 GOTO(out, result = 0);
1149
1150         if (result != -EREMCHG)
1151                 GOTO(out, result);
1152
1153 trigger:
1154         /*
1155          * We still have chance to get the valid inode: for the
1156          * object which is referenced by remote name entry, the
1157          * object on the local MDT will be linked under the dir
1158          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1159          *
1160          * We do not know whether the object for the given FID
1161          * is referenced by some remote name entry or not, and
1162          * especially for DNE II, a multiple-linked object may
1163          * have many name entries reside on many MDTs.
1164          *
1165          * To simplify the operation, OSD will not distinguish
1166          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1167          * only happened for the RPC from other MDT during the
1168          * OI scrub, or for the client side RPC with FID only,
1169          * such as FID to path, or from old connected client.
1170          */
1171         if (!remote) {
1172                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1173                 if (!rc1) {
1174                         remote = true;
1175                         trusted = true;
1176                         flags |= SS_AUTO_PARTIAL;
1177                         flags &= ~SS_AUTO_FULL;
1178                         goto iget;
1179                 }
1180         }
1181
1182         if (scrub->os_running) {
1183                 if (scrub->os_partial_scan && !scrub->os_in_join)
1184                         goto join;
1185
1186                 osd_add_oi_cache(info, dev, id, fid);
1187                 if (IS_ERR_OR_NULL(inode) || result) {
1188                         osd_oii_insert(dev, oic, result == -ENOENT);
1189                         GOTO(out, result = -EINPROGRESS);
1190                 }
1191
1192                 LASSERT(remote);
1193                 LASSERT(obj->oo_inode == inode);
1194
1195                 osd_oii_insert(dev, oic, true);
1196                 goto found;
1197         }
1198
1199         if (dev->od_auto_scrub_interval == AS_NEVER) {
1200                 if (!remote)
1201                         GOTO(out, result = -EREMCHG);
1202
1203                 LASSERT(!result);
1204                 LASSERT(obj->oo_inode == inode);
1205
1206                 osd_add_oi_cache(info, dev, id, fid);
1207                 goto found;
1208         }
1209
1210 join:
1211         rc1 = osd_scrub_start(env, dev, flags);
1212         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1213                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1214         if (rc1 && rc1 != -EALREADY)
1215                 GOTO(out, result = -EREMCHG);
1216
1217         osd_add_oi_cache(info, dev, id, fid);
1218         if (IS_ERR_OR_NULL(inode) || result) {
1219                 osd_oii_insert(dev, oic, result == -ENOENT);
1220                 GOTO(out, result = -EINPROGRESS);
1221         }
1222
1223         LASSERT(remote);
1224         LASSERT(obj->oo_inode == inode);
1225
1226         osd_oii_insert(dev, oic, true);
1227         goto found;
1228
1229 check_lma:
1230         checked = true;
1231         if (unlikely(obj->oo_header))
1232                 goto found;
1233
1234         result = osd_check_lma(env, obj);
1235         if (!result)
1236                 goto found;
1237
1238         LASSERTF(id->oii_ino == inode->i_ino &&
1239                  id->oii_gen == inode->i_generation,
1240                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1241                  PFID(fid), id->oii_ino, id->oii_gen,
1242                  inode->i_ino, inode->i_generation);
1243
1244         saved_ino = inode->i_ino;
1245         saved_gen = inode->i_generation;
1246
1247         if (unlikely(result == -ENODATA)) {
1248                 /*
1249                  * If the OI scrub updated the OI mapping by race, it
1250                  * must be valid. Trust the inode that has no LMA EA.
1251                  */
1252                 if (updated)
1253                         goto found;
1254
1255                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1256                 if (!result) {
1257                         /*
1258                          * The OI mapping is still there, the inode is still
1259                          * valid. It is just becaues the inode has no LMA EA.
1260                          */
1261                         if (saved_ino == id->oii_ino &&
1262                             saved_gen == id->oii_gen)
1263                                 goto found;
1264
1265                         /*
1266                          * It is the OI scrub updated the OI mapping by race.
1267                          * The new OI mapping must be valid.
1268                          */
1269                         trusted = true;
1270                         updated = true;
1271                         goto iget;
1272                 }
1273
1274                 /*
1275                  * "result == -ENOENT" means that the OI mappinghas been
1276                  * removed by race, so the inode belongs to other object.
1277                  *
1278                  * Others error can be returned  directly.
1279                  */
1280                 if (result == -ENOENT) {
1281                         LASSERT(trusted);
1282
1283                         obj->oo_inode = NULL;
1284                         result = 0;
1285                 }
1286         }
1287
1288         if (result != -EREMCHG)
1289                 GOTO(out, result);
1290
1291         LASSERT(!updated);
1292
1293         /*
1294          * if two OST objects map to the same inode, and inode mode is
1295          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1296          * reserved by precreate, and not written yet, in this case, don't
1297          * set inode for the object whose FID mismatch, so that it can create
1298          * inode and not block precreate.
1299          */
1300         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1301             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1302                 obj->oo_inode = NULL;
1303                 GOTO(out, result = 0);
1304         }
1305
1306         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1307         /*
1308          * "result == -ENOENT" means the cached OI mapping has been removed
1309          * from the OI file by race, above inode belongs to other object.
1310          */
1311         if (result == -ENOENT) {
1312                 LASSERT(trusted);
1313
1314                 obj->oo_inode = NULL;
1315                 GOTO(out, result = 0);
1316         }
1317
1318         if (result)
1319                 GOTO(out, result);
1320
1321         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1322                 result = -EREMCHG;
1323                 goto trigger;
1324         }
1325
1326         /*
1327          * It is the OI scrub updated the OI mapping by race.
1328          * The new OI mapping must be valid.
1329          */
1330         trusted = true;
1331         updated = true;
1332         goto iget;
1333
1334 found:
1335         if (!checked) {
1336                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1337                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1338
1339                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1340                 if (!result) {
1341                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1342                             dev->od_is_ost)
1343                                 obj->oo_pfid_in_lma = 1;
1344                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1345                             !dev->od_is_ost)
1346                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1347                 } else if (result != -ENODATA) {
1348                         GOTO(out, result);
1349                 }
1350         }
1351
1352         obj->oo_compat_dot_created = 1;
1353         obj->oo_compat_dotdot_created = 1;
1354
1355         if (S_ISDIR(inode->i_mode) &&
1356             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1357                 osd_check_lmv(info, dev, inode, oic);
1358
1359         result = osd_attach_jinode(inode);
1360         if (result)
1361                 GOTO(out, result);
1362
1363         if (!ldiskfs_pdo)
1364                 GOTO(out, result = 0);
1365
1366         LASSERT(!obj->oo_hl_head);
1367         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1368
1369         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1370
1371 out:
1372         if (result || !obj->oo_inode) {
1373                 if (!IS_ERR_OR_NULL(inode))
1374                         iput(inode);
1375
1376                 obj->oo_inode = NULL;
1377                 if (trusted)
1378                         fid_zero(&oic->oic_fid);
1379         }
1380
1381         LINVRNT(osd_invariant(obj));
1382         return result;
1383 }
1384
1385 /*
1386  * Concurrency: shouldn't matter.
1387  */
1388 static void osd_object_init0(struct osd_object *obj)
1389 {
1390         LASSERT(obj->oo_inode != NULL);
1391         obj->oo_dt.do_body_ops = &osd_body_ops;
1392         obj->oo_dt.do_lu.lo_header->loh_attr |=
1393                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1394 }
1395
1396 /*
1397  * Concurrency: no concurrent access is possible that early in object
1398  * life-cycle.
1399  */
1400 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1401                            const struct lu_object_conf *conf)
1402 {
1403         struct osd_object *obj = osd_obj(l);
1404         int result;
1405
1406         LINVRNT(osd_invariant(obj));
1407
1408         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1409             cfs_fail_val == 2) {
1410                 struct osd_thread_info *info = osd_oti_get(env);
1411                 struct osd_idmap_cache *oic = &info->oti_cache;
1412                 /* invalidate thread cache */
1413                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1414         }
1415         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1416                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1417                 l->lo_header->loh_attr |= LOHA_EXISTS;
1418                 return 0;
1419         }
1420
1421         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1422         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1423         if (result == 0 && obj->oo_inode != NULL) {
1424                 struct osd_thread_info *oti = osd_oti_get(env);
1425                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1426
1427                 osd_object_init0(obj);
1428                 if (unlikely(obj->oo_header))
1429                         return 0;
1430
1431                 result = osd_get_lma(oti, obj->oo_inode,
1432                                      &oti->oti_obj_dentry, loa);
1433                 if (!result) {
1434                         /*
1435                          * Convert LMAI flags to lustre LMA flags
1436                          * and cache it to oo_lma_flags
1437                          */
1438                         obj->oo_lma_flags =
1439                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1440                 } else if (result == -ENODATA) {
1441                         result = 0;
1442                 }
1443         }
1444         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1445
1446         LINVRNT(osd_invariant(obj));
1447         return result;
1448 }
1449
1450 /*
1451  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1452  * The left part is for value, binary mode.
1453  */
1454 struct osd_xattr_entry {
1455         struct list_head        oxe_list;
1456         size_t                  oxe_len;
1457         size_t                  oxe_namelen;
1458         bool                    oxe_exist;
1459         struct rcu_head         oxe_rcu;
1460         char                    oxe_buf[0];
1461 };
1462
1463 static int osd_oxc_get(struct osd_object *obj, const char *name,
1464                        struct lu_buf *buf)
1465 {
1466         struct osd_xattr_entry *tmp;
1467         struct osd_xattr_entry *oxe = NULL;
1468         size_t namelen = strlen(name);
1469         int rc;
1470
1471         rcu_read_lock();
1472         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1473                 if (namelen == tmp->oxe_namelen &&
1474                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1475                         oxe = tmp;
1476                         break;
1477                 }
1478         }
1479
1480         if (oxe == NULL)
1481                 GOTO(out, rc = -ENOENT);
1482
1483         if (!oxe->oxe_exist)
1484                 GOTO(out, rc = -ENODATA);
1485
1486         /* vallen */
1487         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1488         LASSERT(rc > 0);
1489
1490         if (buf->lb_buf == NULL)
1491                 GOTO(out, rc);
1492
1493         if (buf->lb_len < rc)
1494                 GOTO(out, rc = -ERANGE);
1495
1496         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1497 out:
1498         rcu_read_unlock();
1499
1500         return rc;
1501 }
1502
1503 static void osd_oxc_free(struct rcu_head *head)
1504 {
1505         struct osd_xattr_entry *oxe;
1506
1507         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1508         OBD_FREE(oxe, oxe->oxe_len);
1509 }
1510
1511 static void osd_oxc_add(struct osd_object *obj, const char *name,
1512                         const char *buf, int buflen)
1513 {
1514         struct osd_xattr_entry *oxe;
1515         struct osd_xattr_entry *old = NULL;
1516         struct osd_xattr_entry *tmp;
1517         size_t namelen = strlen(name);
1518         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1519
1520         OBD_ALLOC(oxe, len);
1521         if (oxe == NULL)
1522                 return;
1523
1524         INIT_LIST_HEAD(&oxe->oxe_list);
1525         oxe->oxe_len = len;
1526         oxe->oxe_namelen = namelen;
1527         memcpy(oxe->oxe_buf, name, namelen);
1528         if (buflen > 0) {
1529                 LASSERT(buf != NULL);
1530                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1531                 oxe->oxe_exist = true;
1532         } else {
1533                 oxe->oxe_exist = false;
1534         }
1535
1536         /* this should be rarely called, just remove old and add new */
1537         spin_lock(&obj->oo_guard);
1538         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1539                 if (namelen == tmp->oxe_namelen &&
1540                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1541                         old = tmp;
1542                         break;
1543                 }
1544         }
1545         if (old != NULL) {
1546                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1547                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1548         } else {
1549                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1550         }
1551         spin_unlock(&obj->oo_guard);
1552 }
1553
1554 static void osd_oxc_del(struct osd_object *obj, const char *name)
1555 {
1556         struct osd_xattr_entry *oxe;
1557         size_t namelen = strlen(name);
1558
1559         spin_lock(&obj->oo_guard);
1560         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1561                 if (namelen == oxe->oxe_namelen &&
1562                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1563                         list_del_rcu(&oxe->oxe_list);
1564                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1565                         break;
1566                 }
1567         }
1568         spin_unlock(&obj->oo_guard);
1569 }
1570
1571 static void osd_oxc_fini(struct osd_object *obj)
1572 {
1573         struct osd_xattr_entry *oxe, *next;
1574
1575         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1576                 list_del(&oxe->oxe_list);
1577                 OBD_FREE(oxe, oxe->oxe_len);
1578         }
1579 }
1580
1581 /*
1582  * Concurrency: no concurrent access is possible that late in object
1583  * life-cycle.
1584  */
1585 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1586 {
1587         struct osd_object *obj = osd_obj(l);
1588         struct lu_object_header *h = obj->oo_header;
1589
1590         LINVRNT(osd_invariant(obj));
1591
1592         osd_oxc_fini(obj);
1593         dt_object_fini(&obj->oo_dt);
1594         if (obj->oo_hl_head != NULL)
1595                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1596         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1597         OBD_FREE_PTR(obj);
1598         if (unlikely(h)) {
1599                 lu_object_header_fini(h);
1600                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1601                 kfree_rcu(h, loh_rcu);
1602         }
1603 }
1604
1605 /*
1606  * Concurrency: no concurrent access is possible that late in object
1607  * life-cycle.
1608  */
1609 static void osd_index_fini(struct osd_object *o)
1610 {
1611         struct iam_container *bag;
1612
1613         if (o->oo_dir != NULL) {
1614                 bag = &o->oo_dir->od_container;
1615                 if (o->oo_inode != NULL) {
1616                         if (bag->ic_object == o->oo_inode)
1617                                 iam_container_fini(bag);
1618                 }
1619                 OBD_FREE_PTR(o->oo_dir);
1620                 o->oo_dir = NULL;
1621         }
1622 }
1623
1624 /*
1625  * Concurrency: no concurrent access is possible that late in object
1626  * life-cycle (for all existing callers, that is. New callers have to provide
1627  * their own locking.)
1628  */
1629 static int osd_inode_unlinked(const struct inode *inode)
1630 {
1631         return inode->i_nlink == 0;
1632 }
1633
1634 enum {
1635         OSD_TXN_OI_DELETE_CREDITS    = 20,
1636         OSD_TXN_INODE_DELETE_CREDITS = 20
1637 };
1638
1639 /*
1640  * Journal
1641  */
1642
1643 #if OSD_THANDLE_STATS
1644 /**
1645  * Set time when the handle is allocated
1646  */
1647 static void osd_th_alloced(struct osd_thandle *oth)
1648 {
1649         oth->oth_alloced = ktime_get();
1650 }
1651
1652 /**
1653  * Set time when the handle started
1654  */
1655 static void osd_th_started(struct osd_thandle *oth)
1656 {
1657         oth->oth_started = ktime_get();
1658 }
1659
1660 /**
1661  * Check whether the we deal with this handle for too long.
1662  */
1663 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1664                                 ktime_t alloced, ktime_t started,
1665                                 ktime_t closed)
1666 {
1667         ktime_t now = ktime_get();
1668
1669         LASSERT(dev != NULL);
1670
1671         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1672                             ktime_us_delta(started, alloced));
1673         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1674                             ktime_us_delta(closed, started));
1675         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1676                             ktime_us_delta(now, closed));
1677
1678         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1679                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1680                                 oth, now, alloced, started, closed);
1681                 libcfs_debug_dumpstack(NULL);
1682         }
1683 }
1684
1685 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1686 {                                                                       \
1687         ktime_t __closed = ktime_get();                                 \
1688         ktime_t __alloced = oth->oth_alloced;                           \
1689         ktime_t __started = oth->oth_started;                           \
1690                                                                         \
1691         expr;                                                           \
1692         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1693 }
1694
1695 #else /* OSD_THANDLE_STATS */
1696
1697 #define osd_th_alloced(h)                  do {} while(0)
1698 #define osd_th_started(h)                  do {} while(0)
1699 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1700
1701 #endif /* OSD_THANDLE_STATS */
1702
1703 /*
1704  * Concurrency: doesn't access mutable data.
1705  */
1706 static int osd_param_is_not_sane(const struct osd_device *dev,
1707                                  const struct thandle *th)
1708 {
1709         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1710
1711         return oh->ot_credits > osd_transaction_size(dev);
1712 }
1713
1714 /*
1715  * Concurrency: shouldn't matter.
1716  */
1717 static void osd_trans_commit_cb(struct super_block *sb,
1718                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1719 {
1720         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1721         struct thandle *th = &oh->ot_super;
1722         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1723         struct dt_txn_commit_cb *dcb, *tmp;
1724
1725         LASSERT(oh->ot_handle == NULL);
1726
1727         if (error)
1728                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1729
1730         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1731         /* call per-transaction callbacks if any */
1732         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1733                                  dcb_linkage) {
1734                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1735                          "commit callback entry: magic=%x name='%s'\n",
1736                          dcb->dcb_magic, dcb->dcb_name);
1737                 list_del_init(&dcb->dcb_linkage);
1738                 dcb->dcb_func(NULL, th, dcb, error);
1739         }
1740
1741         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1742         lu_device_put(lud);
1743         th->th_dev = NULL;
1744
1745         OBD_FREE_PTR(oh);
1746 }
1747
1748 static struct thandle *osd_trans_create(const struct lu_env *env,
1749                                         struct dt_device *d)
1750 {
1751         struct osd_thread_info *oti = osd_oti_get(env);
1752         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1753         struct osd_thandle *oh;
1754         struct thandle *th;
1755
1756         ENTRY;
1757
1758         if (d->dd_rdonly) {
1759                 CERROR("%s: someone try to start transaction under "
1760                        "readonly mode, should be disabled.\n",
1761                        osd_name(osd_dt_dev(d)));
1762                 dump_stack();
1763                 RETURN(ERR_PTR(-EROFS));
1764         }
1765
1766         /* on pending IO in this thread should left from prev. request */
1767         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1768
1769         sb_start_write(osd_sb(osd_dt_dev(d)));
1770
1771         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1772         if (!oh) {
1773                 sb_end_write(osd_sb(osd_dt_dev(d)));
1774                 RETURN(ERR_PTR(-ENOMEM));
1775         }
1776
1777         oh->ot_quota_trans = &oti->oti_quota_trans;
1778         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1779         th = &oh->ot_super;
1780         th->th_dev = d;
1781         th->th_result = 0;
1782         oh->ot_credits = 0;
1783         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1784         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1785         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1786         osd_th_alloced(oh);
1787
1788         memset(oti->oti_declare_ops, 0,
1789                sizeof(oti->oti_declare_ops));
1790         memset(oti->oti_declare_ops_cred, 0,
1791                sizeof(oti->oti_declare_ops_cred));
1792         memset(oti->oti_declare_ops_used, 0,
1793                sizeof(oti->oti_declare_ops_used));
1794
1795         oti->oti_ins_cache_depth++;
1796
1797         RETURN(th);
1798 }
1799
1800 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1801 {
1802         struct osd_thread_info *oti = osd_oti_get(env);
1803         struct osd_thandle *oh;
1804
1805         oh = container_of(th, struct osd_thandle, ot_super);
1806         LASSERT(oh != NULL);
1807
1808         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1809               oti->oti_declare_ops[OSD_OT_CREATE],
1810               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1811               oti->oti_declare_ops_used[OSD_OT_CREATE],
1812               oti->oti_declare_ops[OSD_OT_DESTROY],
1813               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1814               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1815         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1816               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1817               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1818               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1819               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1820               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1821               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1822         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1823               oti->oti_declare_ops[OSD_OT_WRITE],
1824               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1825               oti->oti_declare_ops_used[OSD_OT_WRITE],
1826               oti->oti_declare_ops[OSD_OT_PUNCH],
1827               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1828               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1829               oti->oti_declare_ops[OSD_OT_QUOTA],
1830               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1831               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1832         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1833               oti->oti_declare_ops[OSD_OT_INSERT],
1834               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1835               oti->oti_declare_ops_used[OSD_OT_INSERT],
1836               oti->oti_declare_ops[OSD_OT_DELETE],
1837               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1838               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1839         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1840               oti->oti_declare_ops[OSD_OT_REF_ADD],
1841               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1842               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1843               oti->oti_declare_ops[OSD_OT_REF_DEL],
1844               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1845               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1846 }
1847
1848 /*
1849  * Concurrency: shouldn't matter.
1850  */
1851 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1852                            struct thandle *th)
1853 {
1854         struct osd_thread_info *oti = osd_oti_get(env);
1855         struct osd_device *dev = osd_dt_dev(d);
1856         handle_t *jh;
1857         struct osd_thandle *oh;
1858         int rc;
1859
1860         ENTRY;
1861
1862         LASSERT(current->journal_info == NULL);
1863
1864         oh = container_of(th, struct osd_thandle, ot_super);
1865         LASSERT(oh != NULL);
1866         LASSERT(oh->ot_handle == NULL);
1867
1868         rc = dt_txn_hook_start(env, d, th);
1869         if (rc != 0)
1870                 GOTO(out, rc);
1871
1872         if (unlikely(osd_param_is_not_sane(dev, th))) {
1873                 static unsigned long last_printed;
1874                 static int last_credits;
1875
1876                 /*
1877                  * don't make noise on a tiny testing systems
1878                  * actual credits misuse will be caught anyway
1879                  */
1880                 if (last_credits != oh->ot_credits &&
1881                     time_after(jiffies, last_printed +
1882                                cfs_time_seconds(60)) &&
1883                     osd_transaction_size(dev) > 512) {
1884                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1885                               oh->ot_credits, osd_transaction_size(dev));
1886                         osd_trans_dump_creds(env, th);
1887                         libcfs_debug_dumpstack(NULL);
1888                         last_credits = oh->ot_credits;
1889                         last_printed = jiffies;
1890                 }
1891                 /*
1892                  * XXX Limit the credits to 'max_transaction_buffers', and
1893                  *     let the underlying filesystem to catch the error if
1894                  *     we really need so many credits.
1895                  *
1896                  *     This should be removed when we can calculate the
1897                  *     credits precisely.
1898                  */
1899                 oh->ot_credits = osd_transaction_size(dev);
1900         } else if (ldiskfs_track_declares_assert != 0) {
1901                 /*
1902                  * reserve few credits to prevent an assertion in JBD
1903                  * our debugging mechanism will be able to detected
1904                  * overuse. this can help to debug single-update
1905                  * transactions
1906                  */
1907                 oh->ot_credits += 10;
1908                 if (unlikely(osd_param_is_not_sane(dev, th)))
1909                         oh->ot_credits = osd_transaction_size(dev);
1910         }
1911
1912         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1913                 GOTO(out, rc = -EIO);
1914
1915         /*
1916          * XXX temporary stuff. Some abstraction layer should
1917          * be used.
1918          */
1919         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1920         osd_th_started(oh);
1921         if (!IS_ERR(jh)) {
1922                 oh->ot_handle = jh;
1923                 LASSERT(oti->oti_txns == 0);
1924
1925                 lu_device_get(&d->dd_lu_dev);
1926                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1927                               "osd-tx", th);
1928                 oti->oti_txns++;
1929                 rc = 0;
1930         } else {
1931                 rc = PTR_ERR(jh);
1932         }
1933 out:
1934         RETURN(rc);
1935 }
1936
1937 static int osd_seq_exists(const struct lu_env *env,
1938                           struct osd_device *osd, u64 seq)
1939 {
1940         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1941         struct seq_server_site *ss = osd_seq_site(osd);
1942         int rc;
1943
1944         ENTRY;
1945
1946         LASSERT(ss != NULL);
1947         LASSERT(ss->ss_server_fld != NULL);
1948
1949         rc = osd_fld_lookup(env, osd, seq, range);
1950         if (rc != 0) {
1951                 if (rc != -ENOENT)
1952                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1953                                osd_name(osd), seq, rc);
1954                 RETURN(0);
1955         }
1956
1957         RETURN(ss->ss_node_id == range->lsr_index);
1958 }
1959
1960 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1961 {
1962         struct dt_txn_commit_cb *dcb;
1963         struct dt_txn_commit_cb *tmp;
1964
1965         /* call per-transaction stop callbacks if any */
1966         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1967                                  dcb_linkage) {
1968                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1969                          "commit callback entry: magic=%x name='%s'\n",
1970                          dcb->dcb_magic, dcb->dcb_name);
1971                 list_del_init(&dcb->dcb_linkage);
1972                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1973         }
1974 }
1975
1976 /*
1977  * Concurrency: shouldn't matter.
1978  */
1979 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1980                           struct thandle *th)
1981 {
1982         struct osd_thread_info *oti = osd_oti_get(env);
1983         struct osd_thandle *oh;
1984         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1985         struct osd_device *osd = osd_dt_dev(th->th_dev);
1986         struct qsd_instance *qsd = osd_def_qsd(osd);
1987         struct lquota_trans *qtrans;
1988         LIST_HEAD(truncates);
1989         int rc = 0, remove_agents = 0;
1990
1991         ENTRY;
1992
1993         oh = container_of(th, struct osd_thandle, ot_super);
1994
1995         remove_agents = oh->ot_remove_agents;
1996
1997         qtrans = oh->ot_quota_trans;
1998         oh->ot_quota_trans = NULL;
1999
2000         /* move locks to local list, stop tx, execute truncates */
2001         list_splice(&oh->ot_trunc_locks, &truncates);
2002
2003         if (oh->ot_handle != NULL) {
2004                 int rc2;
2005
2006                 handle_t *hdl = oh->ot_handle;
2007
2008                 /*
2009                  * add commit callback
2010                  * notice we don't do this in osd_trans_start()
2011                  * as underlying transaction can change during truncate
2012                  */
2013                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2014                                              &oh->ot_jcb);
2015
2016                 LASSERT(oti->oti_txns == 1);
2017                 oti->oti_txns--;
2018
2019                 rc = dt_txn_hook_stop(env, th);
2020                 if (rc != 0)
2021                         CERROR("%s: failed in transaction hook: rc = %d\n",
2022                                osd_name(osd), rc);
2023
2024                 osd_trans_stop_cb(oh, rc);
2025                 /* hook functions might modify th_sync */
2026                 hdl->h_sync = th->th_sync;
2027
2028                 oh->ot_handle = NULL;
2029                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2030                 if (rc2 != 0)
2031                         CERROR("%s: failed to stop transaction: rc = %d\n",
2032                                osd_name(osd), rc2);
2033                 if (!rc)
2034                         rc = rc2;
2035
2036                 osd_process_truncates(&truncates);
2037         } else {
2038                 osd_trans_stop_cb(oh, th->th_result);
2039                 OBD_FREE_PTR(oh);
2040         }
2041
2042         osd_trunc_unlock_all(env, &truncates);
2043
2044         /* inform the quota slave device that the transaction is stopping */
2045         qsd_op_end(env, qsd, qtrans);
2046
2047         /*
2048          * as we want IO to journal and data IO be concurrent, we don't block
2049          * awaiting data IO completion in osd_do_bio(), instead we wait here
2050          * once transaction is submitted to the journal. all reqular requests
2051          * don't do direct IO (except read/write), thus this wait_event becomes
2052          * no-op for them.
2053          *
2054          * IMPORTANT: we have to wait till any IO submited by the thread is
2055          * completed otherwise iobuf may be corrupted by different request
2056          */
2057         wait_event(iobuf->dr_wait,
2058                        atomic_read(&iobuf->dr_numreqs) == 0);
2059         osd_fini_iobuf(osd, iobuf);
2060         if (!rc)
2061                 rc = iobuf->dr_error;
2062
2063         if (unlikely(remove_agents != 0))
2064                 osd_process_scheduled_agent_removals(env, osd);
2065
2066         oti->oti_ins_cache_depth--;
2067         /* reset OI cache for safety */
2068         if (oti->oti_ins_cache_depth == 0)
2069                 oti->oti_ins_cache_used = 0;
2070
2071         sb_end_write(osd_sb(osd));
2072
2073         RETURN(rc);
2074 }
2075
2076 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2077 {
2078         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2079                                               ot_super);
2080
2081         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2082         LASSERT(&dcb->dcb_func != NULL);
2083         if (dcb->dcb_flags & DCB_TRANS_STOP)
2084                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2085         else
2086                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2087
2088         return 0;
2089 }
2090
2091 /*
2092  * Called just before object is freed. Releases all resources except for
2093  * object itself (that is released by osd_object_free()).
2094  *
2095  * Concurrency: no concurrent access is possible that late in object
2096  * life-cycle.
2097  */
2098 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2099 {
2100         struct osd_object *obj = osd_obj(l);
2101         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2102         struct inode *inode = obj->oo_inode;
2103         __u64 projid;
2104         qid_t uid;
2105         qid_t gid;
2106
2107         LINVRNT(osd_invariant(obj));
2108
2109         /*
2110          * If object is unlinked remove fid->ino mapping from object index.
2111          */
2112
2113         osd_index_fini(obj);
2114
2115         if (!inode)
2116                 return;
2117
2118         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2119                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2120
2121         uid = i_uid_read(inode);
2122         gid = i_gid_read(inode);
2123         projid = i_projid_read(inode);
2124
2125         obj->oo_inode = NULL;
2126         iput(inode);
2127
2128         /* do not rebalance quota if the caller needs to release memory
2129          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2130          * transaction and risk to deadlock - LU-12178 */
2131         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2132                 return;
2133
2134         if (!obj->oo_header && qsd) {
2135                 struct osd_thread_info *info = osd_oti_get(env);
2136                 struct lquota_id_info *qi = &info->oti_qi;
2137
2138                 /* Release granted quota to master if necessary */
2139                 qi->lqi_id.qid_uid = uid;
2140                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2141
2142                 qi->lqi_id.qid_uid = gid;
2143                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2144
2145                 qi->lqi_id.qid_uid = projid;
2146                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2147         }
2148 }
2149
2150 /*
2151  * Concurrency: ->loo_object_release() is called under site spin-lock.
2152  */
2153 static void osd_object_release(const struct lu_env *env,
2154                                struct lu_object *l)
2155 {
2156         struct osd_object *o = osd_obj(l);
2157
2158         /*
2159          * nobody should be releasing a non-destroyed object with nlink=0
2160          * the API allows this, but ldiskfs doesn't like and then report
2161          * this inode as deleted
2162          */
2163         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2164                   o->oo_inode->i_nlink == 0));
2165 }
2166
2167 /*
2168  * Concurrency: shouldn't matter.
2169  */
2170 static int osd_object_print(const struct lu_env *env, void *cookie,
2171                             lu_printer_t p, const struct lu_object *l)
2172 {
2173         struct osd_object *o = osd_obj(l);
2174         struct iam_descr *d;
2175
2176         if (o->oo_dir != NULL)
2177                 d = o->oo_dir->od_container.ic_descr;
2178         else
2179                 d = NULL;
2180         return (*p)(env, cookie,
2181                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2182                     o, o->oo_inode,
2183                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2184                     o->oo_inode ? o->oo_inode->i_generation : 0,
2185                     d ? d->id_ops->id_name : "plain");
2186 }
2187
2188 /*
2189  * Concurrency: shouldn't matter.
2190  */
2191 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2192                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2193 {
2194         struct osd_device *osd = osd_dt_dev(d);
2195         struct super_block *sb = osd_sb(osd);
2196         struct kstatfs *ksfs;
2197         __u64 reserved;
2198         int result = 0;
2199
2200         if (unlikely(osd->od_mnt == NULL))
2201                 return -EINPROGRESS;
2202
2203         /* osd_lproc.c call this without env, allocate ksfs for that case */
2204         if (unlikely(env == NULL)) {
2205                 OBD_ALLOC_PTR(ksfs);
2206                 if (ksfs == NULL)
2207                         return -ENOMEM;
2208         } else {
2209                 ksfs = &osd_oti_get(env)->oti_ksfs;
2210         }
2211
2212         result = sb->s_op->statfs(sb->s_root, ksfs);
2213         if (result)
2214                 goto out;
2215
2216         statfs_pack(sfs, ksfs);
2217         if (unlikely(sb->s_flags & SB_RDONLY))
2218                 sfs->os_state |= OS_STATFS_READONLY;
2219
2220         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2221
2222         if (ldiskfs_has_feature_extents(sb))
2223                 sfs->os_maxbytes = sb->s_maxbytes;
2224         else
2225                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2226
2227         /*
2228          * Reserve some space so to avoid fragmenting the filesystem too much.
2229          * Fragmentation not only impacts performance, but can also increase
2230          * metadata overhead significantly, causing grant calculation to be
2231          * wrong.
2232          *
2233          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2234          */
2235         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2236         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2237         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2238                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2239
2240         sfs->os_blocks -= reserved;
2241         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2242         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2243
2244 out:
2245         if (unlikely(env == NULL))
2246                 OBD_FREE_PTR(ksfs);
2247         return result;
2248 }
2249
2250 /**
2251  * Estimate space needed for file creations. We assume the largest filename
2252  * which is 2^64 - 1, hence a filename of 20 chars.
2253  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2254  */
2255 #ifdef __LDISKFS_DIR_REC_LEN
2256 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2257 #else
2258 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2259 #endif
2260
2261 /*
2262  * Concurrency: doesn't access mutable data.
2263  */
2264 static void osd_conf_get(const struct lu_env *env,
2265                          const struct dt_device *dev,
2266                          struct dt_device_param *param)
2267 {
2268         struct osd_device *d = osd_dt_dev(dev);
2269         struct super_block *sb = osd_sb(d);
2270         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2271         const char *name;
2272         int ea_overhead;
2273
2274         /*
2275          * XXX should be taken from not-yet-existing fs abstraction layer.
2276          */
2277         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2278         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2279         param->ddp_symlink_max  = sb->s_blocksize;
2280         param->ddp_mount_type   = LDD_MT_LDISKFS;
2281         if (ldiskfs_has_feature_extents(sb))
2282                 param->ddp_maxbytes = sb->s_maxbytes;
2283         else
2284                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2285         /*
2286          * inode are statically allocated, so per-inode space consumption
2287          * is the space consumed by the directory entry
2288          */
2289         param->ddp_inodespace     = PER_OBJ_USAGE;
2290         /*
2291          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2292          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2293          * maximum length to not under-count the actual number of extents
2294          * needed for writing a file if there are sub-optimal block allocations.
2295          */
2296         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2297         /* worst-case extent insertion metadata overhead */
2298         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2299         param->ddp_mntopts = 0;
2300         if (test_opt(sb, XATTR_USER))
2301                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2302         if (test_opt(sb, POSIX_ACL))
2303                 param->ddp_mntopts |= MNTOPT_ACL;
2304
2305         /*
2306          * LOD might calculate the max stripe count based on max_ea_size,
2307          * so we need take account in the overhead as well,
2308          * xattr_header + magic + xattr_entry_head
2309          */
2310         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2311                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2312
2313 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2314         if (ldiskfs_has_feature_ea_inode(sb))
2315                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2316                                                                 ea_overhead;
2317         else
2318 #endif
2319                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2320
2321         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2322                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2323
2324         /*
2325          * Preferred RPC size for efficient disk IO.  4MB shows good
2326          * all-around performance for ldiskfs, but use bigalloc chunk size
2327          * by default if larger.
2328          */
2329 #if defined(LDISKFS_CLUSTER_SIZE)
2330         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2331                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2332         else
2333 #endif
2334                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2335
2336         param->ddp_t10_cksum_type = 0;
2337         if (bi) {
2338                 unsigned short interval = blk_integrity_interval(bi);
2339                 name = blk_integrity_name(bi);
2340                 /*
2341                  * Expected values:
2342                  * T10-DIF-TYPE1-CRC
2343                  * T10-DIF-TYPE3-CRC
2344                  * T10-DIF-TYPE1-IP
2345                  * T10-DIF-TYPE3-IP
2346                  */
2347                 if (strncmp(name, "T10-DIF-TYPE",
2348                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2349                         /* also skip "1/3-" at end */
2350                         const int type_off = sizeof("T10-DIF-TYPE.");
2351                         char type_number = name[type_off - 2];
2352
2353                         if (interval != 512 && interval != 4096) {
2354                                 CERROR("%s: unsupported T10PI sector size %u\n",
2355                                        d->od_svname, interval);
2356                         } else if (type_number != '1' && type_number != '3') {
2357                                 CERROR("%s: unsupported T10PI type %s\n",
2358                                        d->od_svname, name);
2359                         } else if (strcmp(name + type_off, "CRC") == 0) {
2360                                 d->od_t10_type = type_number == '1' ?
2361                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2362                                 param->ddp_t10_cksum_type = interval == 512 ?
2363                                         OBD_CKSUM_T10CRC512 :
2364                                         OBD_CKSUM_T10CRC4K;
2365                         } else if (strcmp(name + type_off, "IP") == 0) {
2366                                 d->od_t10_type = type_number == '1' ?
2367                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2368                                 param->ddp_t10_cksum_type = interval == 512 ?
2369                                         OBD_CKSUM_T10IP512 :
2370                                         OBD_CKSUM_T10IP4K;
2371                         } else {
2372                                 CERROR("%s: unsupported checksum type of "
2373                                        "T10PI type '%s'",
2374                                        d->od_svname, name);
2375                         }
2376
2377                 } else {
2378                         CERROR("%s: unsupported T10PI type '%s'",
2379                                d->od_svname, name);
2380                 }
2381         }
2382 }
2383
2384 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2385 {
2386         return osd_sb(osd_dt_dev(d));
2387 }
2388
2389 /*
2390  * Concurrency: shouldn't matter.
2391  */
2392 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2393 {
2394         int rc;
2395         struct super_block *s = osd_sb(osd_dt_dev(d));
2396         ENTRY;
2397
2398         down_read(&s->s_umount);
2399         rc = s->s_op->sync_fs(s, 1);
2400         up_read(&s->s_umount);
2401
2402         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2403                rc);
2404
2405         return rc;
2406 }
2407
2408 /**
2409  * Start commit for OSD device.
2410  *
2411  * An implementation of dt_commit_async method for OSD device.
2412  * Asychronously starts underlayng fs sync and thereby a transaction
2413  * commit.
2414  *
2415  * \param env environment
2416  * \param d dt device
2417  *
2418  * \see dt_device_operations
2419  */
2420 static int osd_commit_async(const struct lu_env *env,
2421                             struct dt_device *d)
2422 {
2423         struct super_block *s = osd_sb(osd_dt_dev(d));
2424         int rc;
2425
2426         ENTRY;
2427
2428         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2429         down_read(&s->s_umount);
2430         rc = s->s_op->sync_fs(s, 0);
2431         up_read(&s->s_umount);
2432
2433         RETURN(rc);
2434 }
2435
2436 static int (*priv_security_file_alloc)(struct file *file);
2437
2438 int osd_security_file_alloc(struct file *file)
2439 {
2440         if (priv_security_file_alloc)
2441                 return priv_security_file_alloc(file);
2442         return 0;
2443 }
2444
2445 /*
2446  * Concurrency: shouldn't matter.
2447  */
2448 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2449 {
2450         struct super_block *sb = osd_sb(osd_dt_dev(d));
2451         struct block_device *dev = sb->s_bdev;
2452         int rc = -EOPNOTSUPP;
2453
2454         ENTRY;
2455
2456         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2457                osd_dt_dev(d)->od_svname, (long)dev, rc);
2458
2459         RETURN(rc);
2460 }
2461
2462 /**
2463  * Note: we do not count into QUOTA here.
2464  * If we mount with --data_journal we may need more.
2465  */
2466 const int osd_dto_credits_noquota[DTO_NR] = {
2467         /**
2468          * Insert.
2469          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2470          * SINGLEDATA_TRANS_BLOCKS(8)
2471          * XXX Note: maybe iam need more, since iam have more level than
2472          *           EXT3 htree.
2473          */
2474         [DTO_INDEX_INSERT]  = 16,
2475         /**
2476          * Delete
2477          * just modify a single entry, probably merge few within a block
2478          */
2479         [DTO_INDEX_DELETE]  = 1,
2480         /**
2481          * Used for OI scrub
2482          */
2483         [DTO_INDEX_UPDATE]  = 16,
2484         /**
2485          * 4(inode, inode bits, groups, GDT)
2486          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2487          */
2488         [DTO_OBJECT_CREATE] = 4,
2489         /**
2490          * 4(inode, inode bits, groups, GDT)
2491          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2492          */
2493         [DTO_OBJECT_DELETE] = 4,
2494         /**
2495          * Attr set credits (inode)
2496          */
2497         [DTO_ATTR_SET_BASE] = 1,
2498         /**
2499          * Xattr set. The same as xattr of EXT3.
2500          * DATA_TRANS_BLOCKS(14)
2501          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2502          * are also counted in. Do not know why?
2503          */
2504         [DTO_XATTR_SET]     = 14,
2505         /**
2506          * credits for inode change during write.
2507          */
2508         [DTO_WRITE_BASE]    = 3,
2509         /**
2510          * credits for single block write.
2511          */
2512         [DTO_WRITE_BLOCK]   = 14,
2513         /**
2514          * Attr set credits for chown.
2515          * This is extra credits for setattr, and it is null without quota
2516          */
2517         [DTO_ATTR_SET_CHOWN] = 0
2518 };
2519
2520 static const struct dt_device_operations osd_dt_ops = {
2521         .dt_root_get       = osd_root_get,
2522         .dt_statfs         = osd_statfs,
2523         .dt_trans_create   = osd_trans_create,
2524         .dt_trans_start    = osd_trans_start,
2525         .dt_trans_stop     = osd_trans_stop,
2526         .dt_trans_cb_add   = osd_trans_cb_add,
2527         .dt_conf_get       = osd_conf_get,
2528         .dt_mnt_sb_get     = osd_mnt_sb_get,
2529         .dt_sync           = osd_sync,
2530         .dt_ro             = osd_ro,
2531         .dt_commit_async   = osd_commit_async,
2532 };
2533
2534 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2535                           unsigned int role)
2536 {
2537         struct osd_object *obj = osd_dt_obj(dt);
2538         struct osd_thread_info *oti = osd_oti_get(env);
2539
2540         LINVRNT(osd_invariant(obj));
2541
2542         LASSERT(obj->oo_owner != env);
2543         down_read_nested(&obj->oo_sem, role);
2544
2545         LASSERT(obj->oo_owner == NULL);
2546         oti->oti_r_locks++;
2547 }
2548
2549 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2550                            unsigned int role)
2551 {
2552         struct osd_object *obj = osd_dt_obj(dt);
2553         struct osd_thread_info *oti = osd_oti_get(env);
2554
2555         LINVRNT(osd_invariant(obj));
2556
2557         LASSERT(obj->oo_owner != env);
2558         down_write_nested(&obj->oo_sem, role);
2559
2560         LASSERT(obj->oo_owner == NULL);
2561         obj->oo_owner = env;
2562         oti->oti_w_locks++;
2563 }
2564
2565 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2566 {
2567         struct osd_object *obj = osd_dt_obj(dt);
2568         struct osd_thread_info *oti = osd_oti_get(env);
2569
2570         LINVRNT(osd_invariant(obj));
2571
2572         LASSERT(oti->oti_r_locks > 0);
2573         oti->oti_r_locks--;
2574         up_read(&obj->oo_sem);
2575 }
2576
2577 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2578 {
2579         struct osd_object *obj = osd_dt_obj(dt);
2580         struct osd_thread_info *oti = osd_oti_get(env);
2581
2582         LINVRNT(osd_invariant(obj));
2583
2584         LASSERT(obj->oo_owner == env);
2585         LASSERT(oti->oti_w_locks > 0);
2586         oti->oti_w_locks--;
2587         obj->oo_owner = NULL;
2588         up_write(&obj->oo_sem);
2589 }
2590
2591 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2592 {
2593         struct osd_object *obj = osd_dt_obj(dt);
2594
2595         LINVRNT(osd_invariant(obj));
2596
2597         return obj->oo_owner == env;
2598 }
2599
2600 static void osd_inode_getattr(const struct lu_env *env,
2601                               struct inode *inode, struct lu_attr *attr)
2602 {
2603         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2604                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2605                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2606                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2607
2608         attr->la_atime = inode->i_atime.tv_sec;
2609         attr->la_mtime = inode->i_mtime.tv_sec;
2610         attr->la_ctime = inode->i_ctime.tv_sec;
2611         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2612         attr->la_mode    = inode->i_mode;
2613         attr->la_size    = i_size_read(inode);
2614         attr->la_blocks  = inode->i_blocks;
2615         attr->la_uid     = i_uid_read(inode);
2616         attr->la_gid     = i_gid_read(inode);
2617         attr->la_projid  = i_projid_read(inode);
2618         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2619         attr->la_nlink   = inode->i_nlink;
2620         attr->la_rdev    = inode->i_rdev;
2621         attr->la_blksize = 1 << inode->i_blkbits;
2622         attr->la_blkbits = inode->i_blkbits;
2623         /*
2624          * Ext4 did not transfer inherit flags from raw inode
2625          * to inode flags, and ext4 internally test raw inode
2626          * @i_flags directly. Instead of patching ext4, we do it here.
2627          */
2628         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2629                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2630 }
2631
2632 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2633                             u64 *count)
2634 {
2635         struct osd_object *obj = osd_dt_obj(dt);
2636         const struct dt_it_ops *iops;
2637         struct dt_it *it;
2638         int rc;
2639
2640         ENTRY;
2641
2642         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2643         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2644
2645         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2646                 *count = obj->oo_dirent_count;
2647                 RETURN(0);
2648         }
2649
2650         /* directory not initialized yet */
2651         if (!dt->do_index_ops) {
2652                 *count = 0;
2653                 RETURN(0);
2654         }
2655
2656         iops = &dt->do_index_ops->dio_it;
2657         it = iops->init(env, dt, LUDA_64BITHASH);
2658         if (IS_ERR(it))
2659                 RETURN(PTR_ERR(it));
2660
2661         rc = iops->load(env, it, 0);
2662         if (rc < 0) {
2663                 if (rc == -ENODATA) {
2664                         rc = 0;
2665                         *count = 0;
2666                 }
2667                 GOTO(out, rc);
2668         }
2669         if (rc > 0)
2670                 rc = iops->next(env, it);
2671
2672         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2673                 if (rc == -ESTALE)
2674                         continue;
2675
2676                 if (iops->key_size(env, it) == 0)
2677                         continue;
2678
2679                 (*count)++;
2680         }
2681         if (rc == 1) {
2682                 obj->oo_dirent_count = *count;
2683                 rc = 0;
2684         }
2685 out:
2686         iops->put(env, it);
2687         iops->fini(env, it);
2688
2689         RETURN(rc);
2690 }
2691
2692 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2693                         struct lu_attr *attr)
2694 {
2695         struct osd_object *obj = osd_dt_obj(dt);
2696         int rc = 0;
2697
2698         if (unlikely(!dt_object_exists(dt)))
2699                 return -ENOENT;
2700         if (unlikely(obj->oo_destroyed))
2701                 return -ENOENT;
2702
2703         LASSERT(!dt_object_remote(dt));
2704         LINVRNT(osd_invariant(obj));
2705
2706         spin_lock(&obj->oo_guard);
2707         osd_inode_getattr(env, obj->oo_inode, attr);
2708         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2709                 attr->la_valid |= LA_FLAGS;
2710                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2711         }
2712         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2713                 attr->la_valid |= LA_FLAGS;
2714                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2715         }
2716         spin_unlock(&obj->oo_guard);
2717
2718         if (S_ISDIR(obj->oo_inode->i_mode) &&
2719             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2720                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2721
2722         return rc;
2723 }
2724
2725 static int osd_declare_attr_qid(const struct lu_env *env,
2726                                 struct osd_object *obj,
2727                                 struct osd_thandle *oh, long long bspace,
2728                                 qid_t old_id, qid_t new_id, bool enforce,
2729                                 unsigned int type, bool ignore_edquot)
2730 {
2731         int rc;
2732         struct osd_thread_info *info = osd_oti_get(env);
2733         struct lquota_id_info  *qi = &info->oti_qi;
2734
2735         qi->lqi_type = type;
2736         /* inode accounting */
2737         qi->lqi_is_blk = false;
2738
2739         /* one more inode for the new id ... */
2740         qi->lqi_id.qid_uid = new_id;
2741         qi->lqi_space      = 1;
2742         /* Reserve credits for the new id */
2743         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2744         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2745                 rc = 0;
2746         if (rc)
2747                 RETURN(rc);
2748
2749         /* and one less inode for the current id */
2750         qi->lqi_id.qid_uid = old_id;
2751         qi->lqi_space = -1;
2752         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2753         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2754                 rc = 0;
2755         if (rc)
2756                 RETURN(rc);
2757
2758         /* block accounting */
2759         qi->lqi_is_blk = true;
2760
2761         /* more blocks for the new id ... */
2762         qi->lqi_id.qid_uid = new_id;
2763         qi->lqi_space      = bspace;
2764         /*
2765          * Credits for the new uid has been reserved, re-use "obj"
2766          * to save credit reservation.
2767          */
2768         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2769         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2770                 rc = 0;
2771         if (rc)
2772                 RETURN(rc);
2773
2774         /* and finally less blocks for the current uid */
2775         qi->lqi_id.qid_uid = old_id;
2776         qi->lqi_space      = -bspace;
2777         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2778         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2779                 rc = 0;
2780
2781         RETURN(rc);
2782 }
2783
2784 static int osd_declare_attr_set(const struct lu_env *env,
2785                                 struct dt_object *dt,
2786                                 const struct lu_attr *attr,
2787                                 struct thandle *handle)
2788 {
2789         struct osd_thandle *oh;
2790         struct osd_object *obj;
2791         qid_t uid;
2792         qid_t gid;
2793         long long bspace;
2794         int rc = 0;
2795         bool enforce;
2796
2797         ENTRY;
2798
2799         LASSERT(dt != NULL);
2800         LASSERT(handle != NULL);
2801
2802         obj = osd_dt_obj(dt);
2803         LASSERT(osd_invariant(obj));
2804
2805         oh = container_of(handle, struct osd_thandle, ot_super);
2806         LASSERT(oh->ot_handle == NULL);
2807
2808         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2809                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2810
2811         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2812                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2813
2814         if (attr == NULL || obj->oo_inode == NULL)
2815                 RETURN(rc);
2816
2817         bspace   = obj->oo_inode->i_blocks << 9;
2818         bspace   = toqb(bspace);
2819
2820         /*
2821          * Changing ownership is always preformed by super user, it should not
2822          * fail with EDQUOT unless required explicitly.
2823          *
2824          * We still need to call the osd_declare_qid() to calculate the journal
2825          * credits for updating quota accounting files and to trigger quota
2826          * space adjustment once the operation is completed.
2827          */
2828         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2829                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2830
2831                 if (!ignore_edquot)
2832                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2833                                "(the quota space is %lld)\n",
2834                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2835                                attr->la_gid, bspace);
2836
2837                 /* USERQUOTA */
2838                 uid = i_uid_read(obj->oo_inode);
2839                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2840                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2841                                           attr->la_uid, enforce, USRQUOTA,
2842                                           true);
2843                 if (rc)
2844                         RETURN(rc);
2845
2846                 gid = i_gid_read(obj->oo_inode);
2847                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2848                        attr->la_uid, gid, attr->la_gid);
2849                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2850                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2851                                           attr->la_gid, enforce, GRPQUOTA,
2852                                           ignore_edquot);
2853                 if (rc)
2854                         RETURN(rc);
2855
2856         }
2857 #ifdef HAVE_PROJECT_QUOTA
2858         if (attr->la_valid & LA_PROJID) {
2859                 __u32 projid = i_projid_read(obj->oo_inode);
2860
2861                 enforce = (attr->la_valid & LA_PROJID) &&
2862                                         (attr->la_projid != projid);
2863                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2864                                           (qid_t)projid, (qid_t)attr->la_projid,
2865                                           enforce, PRJQUOTA, true);
2866                 if (rc)
2867                         RETURN(rc);
2868         }
2869 #endif
2870         RETURN(rc);
2871 }
2872
2873 static int osd_inode_setattr(const struct lu_env *env,
2874                              struct inode *inode, const struct lu_attr *attr)
2875 {
2876         __u64 bits = attr->la_valid;
2877
2878         /* Only allow set size for regular file */
2879         if (!S_ISREG(inode->i_mode))
2880                 bits &= ~(LA_SIZE | LA_BLOCKS);
2881
2882         if (bits == 0)
2883                 return 0;
2884
2885         if (bits & LA_ATIME)
2886                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2887         if (bits & LA_CTIME)
2888                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2889         if (bits & LA_MTIME)
2890                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2891         if (bits & LA_SIZE) {
2892                 spin_lock(&inode->i_lock);
2893                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2894                 i_size_write(inode, attr->la_size);
2895                 spin_unlock(&inode->i_lock);
2896         }
2897
2898         /*
2899          * OSD should not change "i_blocks" which is used by quota.
2900          * "i_blocks" should be changed by ldiskfs only.
2901          */
2902         if (bits & LA_MODE)
2903                 inode->i_mode = (inode->i_mode & S_IFMT) |
2904                                 (attr->la_mode & ~S_IFMT);
2905         if (bits & LA_UID)
2906                 i_uid_write(inode, attr->la_uid);
2907         if (bits & LA_GID)
2908                 i_gid_write(inode, attr->la_gid);
2909         if (bits & LA_PROJID)
2910                 i_projid_write(inode, attr->la_projid);
2911         if (bits & LA_NLINK)
2912                 set_nlink(inode, attr->la_nlink);
2913         if (bits & LA_RDEV)
2914                 inode->i_rdev = attr->la_rdev;
2915
2916         if (bits & LA_FLAGS) {
2917                 /* always keep S_NOCMTIME */
2918                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2919                                  S_NOCMTIME;
2920 #if defined(S_ENCRYPTED)
2921                 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2922                  * aware of encryption status. It is just stored into LMA
2923                  * so that it can be forwared to client side.
2924                  */
2925                 inode->i_flags &= ~S_ENCRYPTED;
2926 #endif
2927                 /*
2928                  * Ext4 did not transfer inherit flags from
2929                  * @inode->i_flags to raw inode i_flags when writing
2930                  * flags, we do it explictly here.
2931                  */
2932                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2933                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2934                 else
2935                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2936         }
2937         return 0;
2938 }
2939
2940 #ifdef HAVE_PROJECT_QUOTA
2941 static int osd_transfer_project(struct inode *inode, __u32 projid,
2942                                 struct thandle *handle)
2943 {
2944         struct super_block *sb = inode->i_sb;
2945         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2946         int err;
2947         kprojid_t kprojid;
2948         struct ldiskfs_iloc iloc;
2949         struct ldiskfs_inode *raw_inode;
2950         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2951
2952         if (!ldiskfs_has_feature_project(sb)) {
2953                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2954                         == LDISKFS_DEF_PROJID);
2955                 if (projid != LDISKFS_DEF_PROJID)
2956                         return -EOPNOTSUPP;
2957                 else
2958                         return 0;
2959         }
2960
2961         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2962                 return -EOPNOTSUPP;
2963
2964         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2965         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2966                 return 0;
2967
2968         err = ldiskfs_get_inode_loc(inode, &iloc);
2969         if (err)
2970                 return err;
2971
2972         raw_inode = ldiskfs_raw_inode(&iloc);
2973         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2974                 struct osd_thandle *oh =
2975                                 container_of0(handle, struct osd_thandle,
2976                                               ot_super);
2977                 /**
2978                  * try to expand inode size automatically.
2979                  */
2980                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
2981                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2982                         err = -EOVERFLOW;
2983                         brelse(iloc.bh);
2984                         return err;
2985                 }
2986         }
2987         brelse(iloc.bh);
2988
2989         dquot_initialize(inode);
2990         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2991         if (transfer_to[PRJQUOTA]) {
2992                 err = __dquot_transfer(inode, transfer_to);
2993                 dqput(transfer_to[PRJQUOTA]);
2994                 if (err)
2995                         return err;
2996         }
2997
2998         return err;
2999 }
3000 #endif
3001
3002 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3003                               struct thandle *handle)
3004 {
3005         int rc;
3006
3007         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3008             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3009                 struct iattr iattr;
3010
3011                 CDEBUG(D_QUOTA,
3012                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3013                        inode->i_ino, i_uid_read(inode), attr->la_uid,
3014                        i_gid_read(inode), attr->la_gid);
3015
3016                 dquot_initialize(inode);
3017                 iattr.ia_valid = 0;
3018                 if (attr->la_valid & LA_UID)
3019                         iattr.ia_valid |= ATTR_UID;
3020                 if (attr->la_valid & LA_GID)
3021                         iattr.ia_valid |= ATTR_GID;
3022                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3023                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3024
3025                 rc = dquot_transfer(inode, &iattr);
3026                 if (rc) {
3027                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
3028                                "enforcement enabled on the ldiskfs "
3029                                "filesystem?\n", inode->i_sb->s_id, rc);
3030                         return rc;
3031                 }
3032         }
3033
3034         /* Handle project id transfer here properly */
3035         if (attr->la_valid & LA_PROJID &&
3036             attr->la_projid != i_projid_read(inode)) {
3037 #ifdef HAVE_PROJECT_QUOTA
3038                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3039 #else
3040                 rc = -ENOTSUPP;
3041 #endif
3042                 if (rc) {
3043                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3044                                "enforcement enabled on the ldiskfs "
3045                                "filesystem?\n", inode->i_sb->s_id, rc);
3046                         return rc;
3047                 }
3048         }
3049         return 0;
3050 }
3051
3052 static int osd_attr_set(const struct lu_env *env,
3053                         struct dt_object *dt,
3054                         const struct lu_attr *attr,
3055                         struct thandle *handle)
3056 {
3057         struct osd_object *obj = osd_dt_obj(dt);
3058         struct inode *inode;
3059         int rc;
3060
3061         if (!dt_object_exists(dt))
3062                 return -ENOENT;
3063
3064         LASSERT(handle != NULL);
3065         LASSERT(!dt_object_remote(dt));
3066         LASSERT(osd_invariant(obj));
3067
3068         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3069
3070         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3071             !osd_obj2dev(obj)->od_is_ost) {
3072                 struct osd_thread_info *oti = osd_oti_get(env);
3073                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3074                 struct lu_fid *fid1 = &oti->oti_fid;
3075                 struct osd_inode_id *id = &oti->oti_id;
3076                 struct iam_path_descr *ipd;
3077                 struct iam_container *bag;
3078                 struct osd_thandle *oh;
3079                 int rc;
3080
3081                 fid_cpu_to_be(fid1, fid0);
3082                 memset(id, 1, sizeof(*id));
3083                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3084                                   fid0)->oi_dir.od_container;
3085                 ipd = osd_idx_ipd_get(env, bag);
3086                 if (unlikely(ipd == NULL))
3087                         RETURN(-ENOMEM);
3088
3089                 oh = container_of(handle, struct osd_thandle, ot_super);
3090                 rc = iam_update(oh->ot_handle, bag,
3091                                 (const struct iam_key *)fid1,
3092                                 (const struct iam_rec *)id, ipd);
3093                 osd_ipd_put(env, bag, ipd);
3094                 return(rc > 0 ? 0 : rc);
3095         }
3096
3097         inode = obj->oo_inode;
3098
3099         rc = osd_quota_transfer(inode, attr, handle);
3100         if (rc)
3101                 return rc;
3102
3103         spin_lock(&obj->oo_guard);
3104         rc = osd_inode_setattr(env, inode, attr);
3105         spin_unlock(&obj->oo_guard);
3106         if (rc != 0)
3107                 GOTO(out, rc);
3108
3109         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3110
3111         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3112
3113         if (!(attr->la_valid & LA_FLAGS))
3114                 GOTO(out, rc);
3115
3116         /* Let's check if there are extra flags need to be set into LMA */
3117         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3118                 struct osd_thread_info *info = osd_oti_get(env);
3119                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3120
3121                 LASSERT(!obj->oo_pfid_in_lma);
3122
3123                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3124                                  &info->oti_ost_attrs);
3125                 if (rc)
3126                         GOTO(out, rc);
3127
3128                 lma->lma_incompat |=
3129                         lustre_to_lma_flags(attr->la_flags);
3130                 lustre_lma_swab(lma);
3131
3132                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3133
3134                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3135                                      lma, sizeof(*lma), XATTR_REPLACE);
3136                 if (rc != 0) {
3137                         struct osd_device *osd = osd_obj2dev(obj);
3138
3139                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3140                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3141                               lma->lma_incompat, rc);
3142                 } else {
3143                         obj->oo_lma_flags =
3144                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3145                 }
3146                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3147         }
3148 out:
3149
3150         return rc;
3151 }
3152
3153 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3154                                            struct osd_object *obj,
3155                                            const char *name, const int namelen)
3156 {
3157         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3158 }
3159
3160 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3161                       umode_t mode, struct dt_allocation_hint *hint,
3162                       struct thandle *th, struct lu_attr *attr)
3163 {
3164         int result;
3165         struct osd_device *osd = osd_obj2dev(obj);
3166         struct osd_thandle *oth;
3167         struct dt_object *parent = NULL;
3168         struct inode *inode;
3169         uid_t owner[2] = {0, 0};
3170
3171         if (attr->la_valid & LA_UID)
3172                 owner[0] = attr->la_uid;
3173         if (attr->la_valid & LA_GID)
3174                 owner[1] = attr->la_gid;
3175
3176         LINVRNT(osd_invariant(obj));
3177         LASSERT(obj->oo_inode == NULL);
3178         LASSERT(obj->oo_hl_head == NULL);
3179
3180         if (S_ISDIR(mode) && ldiskfs_pdo) {
3181                 obj->oo_hl_head =
3182                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3183                 if (obj->oo_hl_head == NULL)
3184                         return -ENOMEM;
3185         }
3186
3187         oth = container_of(th, struct osd_thandle, ot_super);
3188         LASSERT(oth->ot_handle->h_transaction != NULL);
3189
3190         if (hint != NULL && hint->dah_parent != NULL &&
3191             !dt_object_remote(hint->dah_parent))
3192                 parent = hint->dah_parent;
3193
3194         inode = ldiskfs_create_inode(oth->ot_handle,
3195                                      parent ? osd_dt_obj(parent)->oo_inode :
3196                                               osd_sb(osd)->s_root->d_inode,
3197                                      mode, owner);
3198         if (!IS_ERR(inode)) {
3199                 /* Do not update file c/mtime in ldiskfs. */
3200                 inode->i_flags |= S_NOCMTIME;
3201
3202                 /*
3203                  * For new created object, it must be consistent,
3204                  * and it is unnecessary to scrub against it.
3205                  */
3206                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3207
3208                 obj->oo_inode = inode;
3209                 result = 0;
3210         } else {
3211                 if (obj->oo_hl_head != NULL) {
3212                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3213                         obj->oo_hl_head = NULL;
3214                 }
3215                 result = PTR_ERR(inode);
3216         }
3217         LINVRNT(osd_invariant(obj));
3218         return result;
3219 }
3220
3221 enum {
3222         OSD_NAME_LEN = 255
3223 };
3224
3225 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3226                      struct lu_attr *attr,
3227                      struct dt_allocation_hint *hint,
3228                      struct dt_object_format *dof,
3229                      struct thandle *th)
3230 {
3231         int result;
3232         struct osd_thandle *oth;
3233         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3234
3235         LASSERT(S_ISDIR(attr->la_mode));
3236
3237         oth = container_of(th, struct osd_thandle, ot_super);
3238         LASSERT(oth->ot_handle->h_transaction != NULL);
3239         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3240                 obj->oo_dirent_count = 0;
3241         result = osd_mkfile(info, obj, mode, hint, th, attr);
3242
3243         return result;
3244 }
3245
3246 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3247                         struct lu_attr *attr,
3248                         struct dt_allocation_hint *hint,
3249                         struct dt_object_format *dof,
3250                         struct thandle *th)
3251 {
3252         int result;
3253         struct osd_thandle *oth;
3254         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3255
3256         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3257
3258         LASSERT(S_ISREG(attr->la_mode));
3259
3260         oth = container_of(th, struct osd_thandle, ot_super);
3261         LASSERT(oth->ot_handle->h_transaction != NULL);
3262
3263         result = osd_mkfile(info, obj, mode, hint, th, attr);
3264         if (result == 0) {
3265                 LASSERT(obj->oo_inode != NULL);
3266                 if (feat->dif_flags & DT_IND_VARKEY)
3267                         result = iam_lvar_create(obj->oo_inode,
3268                                                  feat->dif_keysize_max,
3269                                                  feat->dif_ptrsize,
3270                                                  feat->dif_recsize_max,
3271                                                  oth->ot_handle);
3272                 else
3273                         result = iam_lfix_create(obj->oo_inode,
3274                                                  feat->dif_keysize_max,
3275                                                  feat->dif_ptrsize,
3276                                                  feat->dif_recsize_max,
3277                                                  oth->ot_handle);
3278         }
3279         return result;
3280 }
3281
3282 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3283                      struct lu_attr *attr,
3284                      struct dt_allocation_hint *hint,
3285                      struct dt_object_format *dof,
3286                      struct thandle *th)
3287 {
3288         LASSERT(S_ISREG(attr->la_mode));
3289         return osd_mkfile(info, obj, (attr->la_mode &
3290                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3291                           attr);
3292 }
3293
3294 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3295                      struct lu_attr *attr,
3296                      struct dt_allocation_hint *hint,
3297                      struct dt_object_format *dof,
3298                      struct thandle *th)
3299 {
3300         LASSERT(S_ISLNK(attr->la_mode));
3301         return osd_mkfile(info, obj, (attr->la_mode &
3302                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3303                           attr);
3304 }
3305
3306 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3307                      struct lu_attr *attr,
3308                      struct dt_allocation_hint *hint,
3309                      struct dt_object_format *dof,
3310                      struct thandle *th)
3311 {
3312         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3313         int result;
3314
3315         LINVRNT(osd_invariant(obj));
3316         LASSERT(obj->oo_inode == NULL);
3317         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3318                 S_ISFIFO(mode) || S_ISSOCK(mode));
3319
3320         result = osd_mkfile(info, obj, mode, hint, th, attr);
3321         if (result == 0) {
3322                 LASSERT(obj->oo_inode != NULL);
3323                 /*
3324                  * This inode should be marked dirty for i_rdev.  Currently
3325                  * that is done in the osd_attr_init().
3326                  */
3327                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3328                                    attr->la_rdev);
3329         }
3330         LINVRNT(osd_invariant(obj));
3331         return result;
3332 }
3333
3334 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3335                               struct lu_attr *,
3336                               struct dt_allocation_hint *hint,
3337                               struct dt_object_format *dof,
3338                               struct thandle *);
3339
3340 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3341 {
3342         osd_obj_type_f result;
3343
3344         switch (type) {
3345         case DFT_DIR:
3346                 result = osd_mkdir;
3347                 break;
3348         case DFT_REGULAR:
3349                 result = osd_mkreg;
3350                 break;
3351         case DFT_SYM:
3352                 result = osd_mksym;
3353                 break;
3354         case DFT_NODE:
3355                 result = osd_mknod;
3356                 break;
3357         case DFT_INDEX:
3358                 result = osd_mk_index;
3359                 break;
3360
3361         default:
3362                 LBUG();
3363                 break;
3364         }
3365         return result;
3366 }
3367
3368
3369 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3370                         struct dt_object *parent, struct dt_object *child,
3371                         umode_t child_mode)
3372 {
3373         LASSERT(ah);
3374
3375         ah->dah_parent = parent;
3376         ah->dah_mode = child_mode;
3377
3378         if (parent != NULL && !dt_object_remote(parent)) {
3379                 /* will help to find FID->ino at dt_insert("..") */
3380                 struct osd_object *pobj = osd_dt_obj(parent);
3381
3382                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3383         }
3384 }
3385
3386 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3387                           struct lu_attr *attr, struct dt_object_format *dof,
3388                           struct thandle *handle)
3389 {
3390         struct inode *inode = obj->oo_inode;
3391         __u64 valid = attr->la_valid;
3392         int result;
3393
3394         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3395
3396         if (dof->dof_type != DFT_NODE)
3397                 attr->la_valid &= ~LA_RDEV;
3398         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3399                 attr->la_valid &= ~LA_ATIME;
3400         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3401                 attr->la_valid &= ~LA_CTIME;
3402         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3403                 attr->la_valid &= ~LA_MTIME;
3404
3405         result = osd_quota_transfer(inode, attr, handle);
3406         if (result)
3407                 return;
3408
3409         if (attr->la_valid != 0) {
3410                 result = osd_inode_setattr(info->oti_env, inode, attr);
3411                 /*
3412                  * The osd_inode_setattr() should always succeed here.  The
3413                  * only error that could be returned is EDQUOT when we are
3414                  * trying to change the UID or GID of the inode. However, this
3415                  * should not happen since quota enforcement is no longer
3416                  * enabled on ldiskfs (lquota takes care of it).
3417                  */
3418                 LASSERTF(result == 0, "%d\n", result);
3419                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3420         }
3421
3422         attr->la_valid = valid;
3423 }
3424
3425 /**
3426  * Helper function for osd_create()
3427  *
3428  * \retval 0, on success
3429  */
3430 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3431                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3432                         struct dt_object_format *dof, struct thandle *th)
3433 {
3434         int result;
3435         __u32 umask;
3436
3437         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3438
3439         /* we drop umask so that permissions we pass are not affected */
3440         umask = current->fs->umask;
3441         current->fs->umask = 0;
3442
3443         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3444                                                   th);
3445         if (likely(obj->oo_inode != NULL)) {
3446                 LASSERT(obj->oo_inode->i_state & I_NEW);
3447
3448                 /*
3449                  * Unlock the inode before attr initialization to avoid
3450                  * unnecessary dqget operations. LU-6378
3451                  */
3452                 unlock_new_inode(obj->oo_inode);
3453         }
3454
3455         if (likely(result == 0)) {
3456                 osd_attr_init(info, obj, attr, dof, th);
3457                 osd_object_init0(obj);
3458         }
3459
3460         /* restore previous umask value */
3461         current->fs->umask = umask;
3462
3463         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3464
3465         return result;
3466 }
3467
3468 /**
3469  * Helper function for osd_create()
3470  *
3471  * \retval 0, on success
3472  */
3473 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3474                            const struct lu_fid *fid, struct thandle *th)
3475 {
3476         struct osd_thread_info *info = osd_oti_get(env);
3477         struct osd_inode_id    *id   = &info->oti_id;
3478         struct osd_device      *osd  = osd_obj2dev(obj);
3479         struct osd_thandle     *oh;
3480         int rc;
3481
3482         LASSERT(obj->oo_inode != NULL);
3483
3484         oh = container_of(th, struct osd_thandle, ot_super);
3485         LASSERT(oh->ot_handle);
3486         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3487
3488         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3489         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3490                            OI_CHECK_FLD, NULL);
3491         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3492                 struct lu_fid next_fid = *fid;
3493
3494                 /* insert next object in advance, and map to the same inode */
3495                 next_fid.f_oid++;
3496                 if (next_fid.f_oid != 0) {
3497                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3498                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3499                                       OI_CHECK_FLD, NULL);
3500                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3501                 }
3502         }
3503
3504         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3505
3506         return rc;
3507 }
3508
3509 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3510                    u64 seq, struct lu_seq_range *range)
3511 {
3512         struct seq_server_site *ss = osd_seq_site(osd);
3513
3514         if (fid_seq_is_idif(seq)) {
3515                 fld_range_set_ost(range);
3516                 range->lsr_index = idif_ost_idx(seq);
3517                 return 0;
3518         }
3519
3520         if (!fid_seq_in_fldb(seq)) {
3521                 fld_range_set_mdt(range);
3522                 if (ss != NULL)
3523                         /*
3524                          * FIXME: If ss is NULL, it suppose not get lsr_index
3525                          * at all
3526                          */
3527                         range->lsr_index = ss->ss_node_id;
3528                 return 0;
3529         }
3530
3531         LASSERT(ss != NULL);
3532         fld_range_set_any(range);
3533         /* OSD will only do local fld lookup */
3534         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3535 }
3536
3537 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3538                               struct lu_attr *attr,
3539                               struct dt_allocation_hint *hint,
3540                               struct dt_object_format *dof,
3541                               struct thandle *handle)
3542 {
3543         struct osd_thandle *oh;
3544         int rc;
3545
3546         ENTRY;
3547
3548         LASSERT(handle != NULL);
3549
3550         oh = container_of(handle, struct osd_thandle, ot_super);
3551         LASSERT(oh->ot_handle == NULL);
3552
3553         /*
3554          * EA object consumes more credits than regular object: osd_mk_index
3555          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3556          * leaf_node, could involves the block, block bitmap, groups, GDT
3557          * change for each block, so add 4 * 2 credits in that case.
3558          */
3559         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3560                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3561                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3562         /*
3563          * Reuse idle OI block may cause additional one OI block
3564          * to be changed.
3565          */
3566         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3567                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3568         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3569                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3570                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3571
3572         /* will help to find FID->ino mapping at dt_insert() */
3573         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3574                                    osd_dt_obj(dt));
3575         if (rc != 0)
3576                 RETURN(rc);
3577
3578         if (!attr)
3579                 RETURN(0);
3580
3581         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3582                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3583                                    NULL, OSD_QID_INODE);
3584         if (rc != 0)
3585                 RETURN(rc);
3586
3587         RETURN(rc);
3588 }
3589
3590 /**
3591  * Called to destroy on-disk representation of the object
3592  *
3593  * Concurrency: must be locked
3594  */
3595 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3596                                struct thandle *th)
3597 {
3598         struct osd_object *obj = osd_dt_obj(dt);
3599         struct inode *inode = obj->oo_inode;
3600         struct osd_thandle *oh;
3601         int rc;
3602
3603         ENTRY;
3604
3605         if (inode == NULL)
3606                 RETURN(-ENOENT);
3607
3608         oh = container_of(th, struct osd_thandle, ot_super);
3609         LASSERT(oh->ot_handle == NULL);
3610
3611         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3612                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3613
3614         /* For removing agent entry */
3615         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3616                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3617
3618         /*
3619          * Recycle idle OI leaf may cause additional three OI blocks
3620          * to be changed.
3621          */
3622         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3623                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3624                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3625         /* one less inode */
3626         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3627                                    i_projid_read(inode), -1, oh, obj, NULL,
3628                                    OSD_QID_INODE);
3629         if (rc)
3630                 RETURN(rc);
3631         /* data to be truncated */
3632         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3633                                    i_projid_read(inode), 0, oh, obj, NULL,
3634                                    OSD_QID_BLK);
3635         if (rc)
3636                 RETURN(rc);
3637
3638         /*
3639          * will help to find FID->ino when this object is being
3640          * added to PENDING
3641          */
3642         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3643
3644         RETURN(rc);
3645 }
3646
3647 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3648                        struct thandle *th)
3649 {
3650         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3651         struct osd_object *obj = osd_dt_obj(dt);
3652         struct inode *inode = obj->oo_inode;
3653         struct osd_device *osd = osd_obj2dev(obj);
3654         struct osd_thandle *oh;
3655         int result;
3656
3657         ENTRY;
3658
3659         oh = container_of(th, struct osd_thandle, ot_super);
3660         LASSERT(oh->ot_handle);
3661         LASSERT(inode);
3662         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3663
3664         if (unlikely(fid_is_acct(fid)))
3665                 RETURN(-EPERM);
3666
3667         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3668                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3669                 if (result != 0)
3670                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3671                                osd_name(osd), PFID(fid), result);
3672         }
3673
3674         if (S_ISDIR(inode->i_mode)) {
3675                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3676                         inode->i_nlink == 2);
3677
3678                 spin_lock(&obj->oo_guard);
3679                 clear_nlink(inode);
3680                 spin_unlock(&obj->oo_guard);
3681                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3682         }
3683
3684         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3685
3686         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3687
3688         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3689                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3690                                        oh->ot_handle, OI_CHECK_FLD);
3691
3692         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3693         /* XXX: add to ext3 orphan list */
3694         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3695
3696         /* not needed in the cache anymore */
3697         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3698         obj->oo_destroyed = 1;
3699
3700         RETURN(0);
3701 }
3702
3703 /**
3704  * Put the fid into lustre_mdt_attrs, and then place the structure
3705  * inode's ea. This fid should not be altered during the life time
3706  * of the inode.
3707  *
3708  * \retval +ve, on success
3709  * \retval -ve, on error
3710  *
3711  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3712  */
3713 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3714                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3715 {
3716         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3717         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3718         int rc;
3719
3720         ENTRY;
3721
3722         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3723                 RETURN(0);
3724
3725         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3726                 rc = -ENOMEM;
3727
3728         lustre_loa_init(loa, fid, compat, incompat);
3729         lustre_loa_swab(loa, false);
3730
3731         /*
3732          * For the OST device with 256 bytes inode size by default,
3733          * the PFID EA will be stored together with LMA EA to avoid
3734          * performance trouble. Otherwise the PFID EA can be stored
3735          * independently. LU-8998
3736          */
3737         if ((compat & LMAC_FID_ON_OST) &&
3738             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3739                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3740                                      sizeof(*loa), XATTR_CREATE);
3741         else
3742                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3743                                      sizeof(*lma), XATTR_CREATE);
3744         /*
3745          * LMA may already exist, but we need to check that all the
3746          * desired compat/incompat flags have been added.
3747          */
3748         if (unlikely(rc == -EEXIST)) {
3749                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3750                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3751                 if (rc < 0)
3752                         RETURN(rc);
3753
3754                 if (rc < sizeof(*lma))
3755                         RETURN(-EINVAL);
3756
3757                 lustre_loa_swab(loa, true);
3758                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3759                     ((compat == 0 && incompat == 0) ||
3760                      (!(~lma->lma_compat & compat) &&
3761                       !(~lma->lma_incompat & incompat))))
3762                         RETURN(0);
3763
3764                 lma->lma_self_fid = *fid;
3765                 lma->lma_compat |= compat;
3766                 lma->lma_incompat |= incompat;
3767                 if (rc == sizeof(*lma)) {
3768                         lustre_lma_swab(lma);
3769                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3770                                              sizeof(*lma), XATTR_REPLACE);
3771                 } else {
3772                         lustre_loa_swab(loa, false);
3773                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3774                                              sizeof(*loa), XATTR_REPLACE);
3775                 }
3776         }
3777
3778         RETURN(rc);
3779 }
3780
3781 /**
3782  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3783  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3784  * To have compatilibility with 1.8 ldiskfs driver we need to have
3785  * magic number at start of fid data.
3786  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3787  * its inmemory API.
3788  */
3789 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3790                                          const struct lu_fid *fid)
3791 {
3792         if (!fid_is_namespace_visible(fid) ||
3793             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3794                 param->edp_magic = 0;
3795                 return;
3796         }
3797
3798         param->edp_magic = LDISKFS_LUFID_MAGIC;
3799         param->edp_len =  sizeof(struct lu_fid) + 1;
3800         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3801 }
3802
3803 /**
3804  * Try to read the fid from inode ea into dt_rec.
3805  *
3806  * \param fid object fid.
3807  *
3808  * \retval 0 on success
3809  */
3810 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3811                           __u32 ino, struct lu_fid *fid,
3812                           struct osd_inode_id *id)
3813 {
3814         struct osd_thread_info *info  = osd_oti_get(env);
3815         struct inode *inode;
3816
3817         ENTRY;
3818
3819         osd_id_gen(id, ino, OSD_OII_NOGEN);
3820         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3821         if (IS_ERR(inode))
3822                 RETURN(PTR_ERR(inode));
3823
3824         iput(inode);
3825         RETURN(0);
3826 }
3827
3828 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3829                                         struct inode *dir,
3830                                         struct inode *parent_dir,
3831                                         const struct lu_fid *dot_fid,
3832                                         const struct lu_fid *dot_dot_fid,
3833                                         struct osd_thandle *oth)
3834 {
3835         struct ldiskfs_dentry_param *dot_ldp;
3836         struct ldiskfs_dentry_param *dot_dot_ldp;
3837         __u32 saved_nlink = dir->i_nlink;
3838         int rc;
3839
3840         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3841         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3842
3843         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3844         dot_ldp->edp_magic = 0;
3845
3846         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3847                                     dir, dot_ldp, dot_dot_ldp);
3848         /*
3849          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3850          * the subseqent ref_add() will increase the dir->i_nlink
3851          * as 3. That is incorrect for new created directory.
3852          *
3853          * It looks like hack, because we want to make the OSD API
3854          * to be order-independent for new created directory object
3855          * between dt_insert(..) and ref_add() operations.
3856          *
3857          * Here, we only restore the in-RAM dir-inode's nlink attr,
3858          * becuase if the nlink attr is not 2, then there will be
3859          * ref_add() called following the dt_insert(..), such call
3860          * will make both the in-RAM and on-disk dir-inode's nlink
3861          * attr to be set as 2. LU-7447
3862          */
3863         set_nlink(dir, saved_nlink);
3864         return rc;
3865 }
3866
3867 /**
3868  * Create an local agent inode for remote entry
3869  */
3870 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3871                                                   struct osd_device *osd,
3872                                                   struct osd_object *pobj,
3873                                                   const struct lu_fid *fid,
3874                                                   __u32 type,
3875                                                   struct thandle *th)
3876 {
3877         struct osd_thread_info *info = osd_oti_get(env);
3878         struct inode *local;
3879         struct osd_thandle *oh;
3880         int rc;
3881
3882         ENTRY;
3883
3884         LASSERT(th);
3885         oh = container_of(th, struct osd_thandle, ot_super);
3886         LASSERT(oh->ot_handle->h_transaction != NULL);
3887
3888         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3889                                      NULL);
3890         if (IS_ERR(local)) {
3891                 CERROR("%s: create local error %d\n", osd_name(osd),
3892                        (int)PTR_ERR(local));
3893                 RETURN(local);
3894         }
3895
3896         /*
3897          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3898          * correct gid on remote file, not agent here
3899          */
3900         local->i_gid = current_fsgid();
3901         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3902
3903         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3904          * That gives e2fsck something to look at and be happy, and allows
3905          * debugging if we need to determine where this symlink came from.
3906          */
3907         if (S_ISLNK(type)) {
3908                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3909                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3910                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3911
3912                 i_size_write(local, rc);
3913                 LDISKFS_I(local)->i_disksize = rc;
3914         }
3915         unlock_new_inode(local);
3916
3917         /* Agent inode should not have project ID */
3918 #ifdef  HAVE_PROJECT_QUOTA
3919         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3920             i_projid_read(pobj->oo_inode) != 0) {
3921                 rc = osd_transfer_project(local, 0, th);
3922                 if (rc) {
3923                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3924                                "quota enforcement enabled on the ldiskfs "
3925                                "filesystem?\n", local->i_sb->s_id, rc);
3926                         RETURN(ERR_PTR(rc));
3927                 }
3928         }
3929 #endif
3930         /* Set special LMA flag for local agent inode */
3931         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3932         if (rc != 0) {
3933                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3934                        osd_name(osd), PFID(fid), rc);
3935                 RETURN(ERR_PTR(rc));
3936         }
3937
3938         if (!S_ISDIR(type))
3939                 RETURN(local);
3940
3941         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3942                                          lu_object_fid(&pobj->oo_dt.do_lu),
3943                                          fid, oh);
3944         if (rc != 0) {
3945                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3946                         osd_name(osd), PFID(fid), rc);
3947                 RETURN(ERR_PTR(rc));
3948         }
3949
3950         RETURN(local);
3951 }
3952
3953 /**
3954  * when direntry is deleted, we have to take care of possible agent inode
3955  * referenced by that. unfortunately we can't do this at that point:
3956  * iget() within a running transaction leads to deadlock and we better do
3957  * not call that every delete declaration to save performance. so we put
3958  * a potention agent inode on a list and process that once the transaction
3959  * is over. Notice it's not any worse in terms of real orphans as regular
3960  * object destroy doesn't put inodes on the on-disk orphan list. this should
3961  * be addressed separately
3962  */
3963 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3964                                             struct osd_thandle *oh,
3965                                             __u32 ino)
3966 {
3967         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3968         struct osd_obj_orphan *oor;
3969
3970         OBD_ALLOC_PTR(oor);
3971         if (oor == NULL)
3972                 return -ENOMEM;
3973
3974         oor->oor_ino = ino;
3975         oor->oor_env = (struct lu_env *)env;
3976         spin_lock(&osd->od_osfs_lock);
3977         list_add(&oor->oor_list, &osd->od_orphan_list);
3978         spin_unlock(&osd->od_osfs_lock);
3979
3980         oh->ot_remove_agents = 1;
3981
3982         return 0;
3983
3984 }
3985
3986 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3987                                                 struct osd_device *osd)
3988 {
3989         struct osd_thread_info *info = osd_oti_get(env);
3990         struct osd_obj_orphan *oor, *tmp;
3991         struct osd_inode_id id;
3992         LIST_HEAD(list);
3993         struct inode *inode;
3994         struct lu_fid fid;
3995         handle_t *jh;
3996         __u32 ino;
3997
3998         spin_lock(&osd->od_osfs_lock);
3999         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4000                 if (oor->oor_env == env)
4001                         list_move(&oor->oor_list, &list);
4002         }
4003         spin_unlock(&osd->od_osfs_lock);
4004
4005         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4006
4007                 ino = oor->oor_ino;
4008
4009                 list_del(&oor->oor_list);
4010                 OBD_FREE_PTR(oor);
4011
4012                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4013                 inode = osd_iget_fid(info, osd, &id, &fid);
4014                 if (IS_ERR(inode))
4015                         continue;
4016
4017                 if (!osd_remote_fid(env, osd, &fid)) {
4018                         iput(inode);
4019                         continue;
4020                 }
4021
4022                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4023                 clear_nlink(inode);
4024                 mark_inode_dirty(inode);
4025                 ldiskfs_journal_stop(jh);
4026                 iput(inode);
4027         }
4028
4029         return 0;
4030 }
4031
4032 /**
4033  * OSD layer object create function for OST objects (b=11826).
4034  *
4035  * The FID is inserted into inode xattr here.
4036  *
4037  * \retval   0, on success
4038  * \retval -ve, on error
4039  */
4040 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4041                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4042                       struct dt_object_format *dof, struct thandle *th)
4043 {
4044         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4045         struct osd_object *obj = osd_dt_obj(dt);
4046         struct osd_thread_info *info = osd_oti_get(env);
4047         int result, on_ost = 0;
4048
4049         ENTRY;
4050
4051         if (dt_object_exists(dt))
4052                 RETURN(-EEXIST);
4053
4054         LINVRNT(osd_invariant(obj));
4055         LASSERT(!dt_object_remote(dt));
4056         LASSERT(osd_is_write_locked(env, obj));
4057         LASSERT(th != NULL);
4058
4059         if (unlikely(fid_is_acct(fid)))
4060                 /*
4061                  * Quota files can't be created from the kernel any more,
4062                  * 'tune2fs -O quota' will take care of creating them
4063                  */
4064                 RETURN(-EPERM);
4065
4066         result = __osd_create(info, obj, attr, hint, dof, th);
4067         if (result == 0) {
4068                 if (fid_is_idif(fid) &&
4069                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4070                         struct lu_fid *tfid = &info->oti_fid;
4071                         struct ost_id *oi   = &info->oti_ostid;
4072
4073                         fid_to_ostid(fid, oi);
4074                         ostid_to_fid(tfid, oi, 0);
4075                         on_ost = 1;
4076                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4077                                                 LMAC_FID_ON_OST, 0);
4078                 } else {
4079                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4080                                                fid, OI_CHECK_FLD);
4081                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4082                                                 on_ost ? LMAC_FID_ON_OST : 0,
4083                                                 0);
4084                 }
4085                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4086                         obj->oo_dt.do_body_ops = &osd_body_ops;
4087         }
4088
4089         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4090                 result = __osd_oi_insert(env, obj, fid, th);
4091
4092         /*
4093          * a small optimization - dt_insert() isn't usually applied
4094          * to OST objects, so we don't need to cache OI mapping for
4095          * OST objects
4096          */
4097         if (result == 0 && on_ost == 0) {
4098                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4099
4100                 result = osd_idc_find_and_init(env, osd, obj);
4101                 LASSERT(result == 0);
4102         }
4103
4104         LASSERT(ergo(result == 0,
4105                      dt_object_exists(dt) && !dt_object_remote(dt)));
4106         LINVRNT(osd_invariant(obj));
4107         RETURN(result);
4108 }
4109
4110 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4111                                struct thandle *handle)
4112 {
4113         struct osd_thandle *oh;
4114
4115         /* it's possible that object doesn't exist yet */
4116         LASSERT(handle != NULL);
4117
4118         oh = container_of(handle, struct osd_thandle, ot_super);
4119         LASSERT(oh->ot_handle == NULL);
4120
4121         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4122                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4123
4124         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4125
4126         return 0;
4127 }
4128
4129 /*
4130  * Concurrency: @dt is write locked.
4131  */
4132 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4133                        struct thandle *th)
4134 {
4135         struct osd_object *obj = osd_dt_obj(dt);
4136         struct inode *inode = obj->oo_inode;
4137         struct osd_thandle *oh;
4138         int rc = 0;
4139
4140         if (!dt_object_exists(dt) || obj->oo_destroyed)
4141                 return -ENOENT;
4142
4143         LINVRNT(osd_invariant(obj));
4144         LASSERT(!dt_object_remote(dt));
4145         LASSERT(osd_is_write_locked(env, obj));
4146         LASSERT(th != NULL);
4147
4148         oh = container_of(th, struct osd_thandle, ot_super);
4149         LASSERT(oh->ot_handle != NULL);
4150
4151         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4152
4153         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4154                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4155         /*
4156          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4157          * (65000) subdirectories by storing "1" in i_nlink if the link count
4158          * would otherwise overflow. Directory tranversal tools understand
4159          * that (st_nlink == 1) indicates that the filesystem dose not track
4160          * hard links count on the directory, and will not abort subdirectory
4161          * scanning early once (st_nlink - 2) subdirs have been found.
4162          *
4163          * This also has to properly handle the case of inodes with nlink == 0
4164          * in case they are being linked into the PENDING directory
4165          */
4166         spin_lock(&obj->oo_guard);
4167         if (unlikely(inode->i_nlink == 0))
4168                 /* inc_nlink from 0 may cause WARN_ON */
4169                 set_nlink(inode, 1);
4170         else {
4171                 ldiskfs_inc_count(oh->ot_handle, inode);
4172                 if (!S_ISDIR(inode->i_mode))
4173                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4174         }
4175         spin_unlock(&obj->oo_guard);
4176
4177         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4178         LINVRNT(osd_invariant(obj));
4179
4180         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4181
4182         return rc;
4183 }
4184
4185 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4186                                struct thandle *handle)
4187 {
4188         struct osd_thandle *oh;
4189
4190         if (!dt_object_exists(dt))
4191                 return -ENOENT;
4192
4193         LASSERT(!dt_object_remote(dt));
4194         LASSERT(handle != NULL);
4195
4196         oh = container_of(handle, struct osd_thandle, ot_super);
4197         LASSERT(oh->ot_handle == NULL);
4198
4199         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4200                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4201
4202         return 0;
4203 }
4204
4205 /*
4206  * Concurrency: @dt is write locked.
4207  */
4208 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4209                        struct thandle *th)
4210 {
4211         struct osd_object *obj = osd_dt_obj(dt);
4212         struct inode *inode = obj->oo_inode;
4213         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4214         struct osd_thandle *oh;
4215
4216         if (!dt_object_exists(dt))
4217                 return -ENOENT;
4218
4219         LINVRNT(osd_invariant(obj));
4220         LASSERT(!dt_object_remote(dt));
4221         LASSERT(osd_is_write_locked(env, obj));
4222         LASSERT(th != NULL);
4223
4224         oh = container_of(th, struct osd_thandle, ot_super);
4225         LASSERT(oh->ot_handle != NULL);
4226
4227         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4228
4229         spin_lock(&obj->oo_guard);
4230         /*
4231          * That can be result of upgrade from old Lustre version and
4232          * applied only to local files.  Just skip this ref_del call.
4233          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4234          */
4235         if (inode->i_nlink == 0) {
4236                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4237                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4238                              ", maybe an upgraded file? (LU-3915)\n",
4239                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4240                 spin_unlock(&obj->oo_guard);
4241                 return 0;
4242         }
4243
4244         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4245                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4246
4247         ldiskfs_dec_count(oh->ot_handle, inode);
4248         spin_unlock(&obj->oo_guard);
4249
4250         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4251         LINVRNT(osd_invariant(obj));
4252
4253         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4254
4255         return 0;
4256 }
4257
4258 /*
4259  * Concurrency: @dt is read locked.
4260  */
4261 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4262                          struct lu_buf *buf, const char *name)
4263 {
4264         struct osd_object      *obj    = osd_dt_obj(dt);
4265         struct inode           *inode  = obj->oo_inode;
4266         struct osd_thread_info *info   = osd_oti_get(env);
4267         struct dentry          *dentry = &info->oti_obj_dentry;
4268         bool cache_xattr = false;
4269         int rc;
4270
4271         LASSERT(buf);
4272
4273         /* version get is not real XATTR but uses xattr API */
4274         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4275                 dt_obj_version_t *ver = buf->lb_buf;
4276
4277                 /*
4278                  * for version we are just using xattr API but change inode
4279                  * field instead
4280                  */
4281                 if (buf->lb_len == 0)
4282                         return sizeof(dt_obj_version_t);
4283
4284                 if (buf->lb_len < sizeof(dt_obj_version_t))
4285                         return -ERANGE;
4286
4287                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4288                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4289
4290                 *ver = LDISKFS_I(inode)->i_fs_version;
4291
4292                 return sizeof(dt_obj_version_t);
4293         }
4294
4295         if (!dt_object_exists(dt))
4296                 return -ENOENT;
4297
4298         LASSERT(!dt_object_remote(dt));
4299         LASSERT(inode->i_op != NULL);
4300 #ifdef HAVE_IOP_XATTR
4301         LASSERT(inode->i_op->getxattr != NULL);
4302 #endif
4303
4304         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4305             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4306                 cache_xattr = true;
4307
4308         if (cache_xattr) {
4309                 rc = osd_oxc_get(obj, name, buf);
4310                 if (rc != -ENOENT)
4311                         return rc;
4312         }
4313
4314         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4315                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4316                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4317                 struct filter_fid *ff;
4318                 struct ost_layout *ol;
4319
4320                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4321                 if (rc)
4322                         return rc;
4323
4324                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4325
4326                 rc = sizeof(*ff);
4327                 if (buf->lb_len == 0 || !buf->lb_buf)
4328                         return rc;
4329
4330                 if (buf->lb_len < rc)
4331                         return -ERANGE;
4332
4333                 ff = buf->lb_buf;
4334                 ol = &ff->ff_layout;
4335                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4336                                                   PFID_STRIPE_IDX_BITS);
4337                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4338                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4339                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4340                 if (lma->lma_compat & LMAC_COMP_INFO) {
4341                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4342                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4343                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4344                 } else {
4345                         ol->ol_comp_start = 0;
4346                         ol->ol_comp_end = 0;
4347                         ol->ol_comp_id = 0;
4348                 }
4349         } else {
4350                 rc = __osd_xattr_get(inode, dentry, name,
4351                                      buf->lb_buf, buf->lb_len);
4352         }
4353
4354         if (cache_xattr) {
4355                 if (rc == -ENOENT || rc == -ENODATA)
4356                         osd_oxc_add(obj, name, NULL, 0);
4357                 else if (rc > 0 && buf->lb_buf != NULL)
4358                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4359         }
4360
4361         return rc;
4362 }
4363
4364 static int osd_declare_xattr_set(const struct lu_env *env,
4365                                  struct dt_object *dt,
4366                                  const struct lu_buf *buf, const char *name,
4367                                  int fl, struct thandle *handle)
4368 {
4369         struct osd_thandle *oh;
4370         int credits = 0;
4371         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4372
4373         LASSERT(handle != NULL);
4374
4375         oh = container_of(handle, struct osd_thandle, ot_super);
4376         LASSERT(oh->ot_handle == NULL);
4377
4378         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4379                 /*
4380                  * For non-upgrading case, the LMA is set first and
4381                  * usually fit inode. But for upgrade case, the LMA
4382                  * may be in another separated EA block.
4383                  */
4384                 if (dt_object_exists(dt)) {
4385                         if (fl == LU_XATTR_REPLACE)
4386                                 credits = 1;
4387                         else
4388                                 goto upgrade;
4389                 }
4390         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4391                 credits = 1;
4392         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4393                 /* We may need to delete the old PFID EA. */
4394                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4395                 if (fl == LU_XATTR_REPLACE)
4396                         credits += 1;
4397                 else
4398                         goto upgrade;
4399         } else {
4400                 /*
4401                  * If some name entry resides on remote MDT, then will create
4402                  * agent entry under remote parent. On the other hand, if the
4403                  * remote entry will be removed, then related agent entry may
4404                  * need to be removed from the remote parent. So there may be
4405                  * kinds of cases, let's declare enough credits. The credits
4406                  * for create agent entry is enough for remove case.
4407                  */
4408                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4409                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4410                         if (dt_object_exists(dt))
4411                                 credits += 1; /* For updating LMA */
4412                 }
4413
4414 upgrade:
4415                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4416
4417                 if (buf != NULL) {
4418                         ssize_t buflen;
4419
4420                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4421                                 /*
4422                                  * learn xattr size from osd_xattr_get if
4423                                  * attribute has not been read yet
4424                                  */
4425                                 buflen = __osd_xattr_get(
4426                                     osd_dt_obj(dt)->oo_inode,
4427                                     &osd_oti_get(env)->oti_obj_dentry,
4428                                     name, NULL, 0);
4429                                 if (buflen < 0)
4430                                         buflen = 0;
4431                         } else {
4432                                 buflen = buf->lb_len;
4433                         }
4434
4435                         if (buflen > sb->s_blocksize) {
4436                                 credits += osd_calc_bkmap_credits(
4437                                     sb, NULL, 0, -1,
4438                                     (buflen + sb->s_blocksize - 1) >>
4439                                     sb->s_blocksize_bits);
4440                         }
4441                 }
4442                 /*
4443                  * xattr set may involve inode quota change, reserve credits for
4444                  * dquot_initialize()
4445                  */
4446                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4447         }
4448
4449         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4450
4451         return 0;
4452 }
4453
4454 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4455                               const struct lu_buf *buf, int fl,
4456                               struct thandle *handle)
4457 {
4458         struct osd_thread_info *info = osd_oti_get(env);
4459         struct dentry *dentry = &info->oti_obj_dentry;
4460         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4461         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4462         struct inode *inode = obj->oo_inode;
4463         struct filter_fid *ff = buf->lb_buf;
4464         struct ost_layout *ol = &ff->ff_layout;
4465         int flags = XATTR_REPLACE;
4466         int rc;
4467
4468         ENTRY;
4469
4470         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4471                 RETURN(-EINVAL);
4472
4473         rc = osd_get_lma(info, inode, dentry, loa);
4474         if (rc == -ENODATA) {
4475                 /* Usually for upgarding from old device */
4476                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4477                                 LMAC_FID_ON_OST, 0);
4478                 flags = XATTR_CREATE;
4479         } else if (rc) {
4480                 RETURN(rc);
4481         }
4482
4483         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4484                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4485                         RETURN(-EEXIST);
4486
4487                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4488                         /* Separate PFID EA from LMA */
4489                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4490                         lustre_lma_swab(lma);
4491                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4492                                              sizeof(*lma), XATTR_REPLACE);
4493                         if (!rc) {
4494                                 obj->oo_pfid_in_lma = 0;
4495                                 rc = LU_XATTR_CREATE;
4496                         }
4497
4498                         RETURN(rc);
4499                 }
4500         } else {
4501                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4502                         RETURN(fl);
4503
4504                 /*
4505                  * Old client does not send stripe information,
4506                  * then store the PFID EA on disk separatedly.
4507                  */
4508                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4509                              ol->ol_stripe_size == 0))
4510                         RETURN(fl);
4511
4512                 /* Remove old PFID EA entry firstly. */
4513                 dquot_initialize(inode);
4514                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4515                 if (rc == -ENODATA) {
4516                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4517                                 RETURN(rc);
4518                 } else if (rc) {
4519                         RETURN(rc);
4520                 }
4521         }
4522
4523         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4524         if (likely(ol->ol_stripe_size != 0)) {
4525                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4526                                              PFID_STRIPE_IDX_BITS;
4527                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4528                 lma->lma_compat |= LMAC_STRIPE_INFO;
4529                 if (ol->ol_comp_id != 0) {
4530                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4531                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4532                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4533                         lma->lma_compat |= LMAC_COMP_INFO;
4534                 }
4535         }
4536
4537         lustre_loa_swab(loa, false);
4538
4539         /* Store the PFID EA inside LMA. */
4540         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4541                              flags);
4542         if (!rc)
4543                 obj->oo_pfid_in_lma = 1;
4544
4545         RETURN(rc);
4546 }
4547
4548 /*
4549  * In DNE environment, the object (in spite of regular file or directory)
4550  * and its name entry may reside on different MDTs. Under such case, we will
4551  * create an agent entry on the MDT where the object resides. The agent entry
4552  * references the object locally, that makes the object to be visible to the
4553  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4554  * such as 'tar' can handle the object properly.
4555  *
4556  * We handle the agent entry during set linkEA that is the common interface
4557  * for both regular file and directroy, can handle kinds of cases, such as
4558  * create/link/unlink/rename, and so on.
4559  *
4560  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4561  *
4562  * XXX: There are two known issues:
4563  * 1. For one object, we will create at most one agent entry even if there
4564  *    may be more than one cross-MDTs hard links on the object. So the local
4565  *    e2fsck may claim that the object's nlink is larger than the name entries
4566  *    that reference such inode. And in further, the e2fsck will fix the nlink
4567  *    attribute to match the local references. Then it will cause the object's
4568  *    nlink attribute to be inconsistent with the global references. it is bad
4569  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4570  *    other hand, the global namespace LFSCK can repair the object's attribute
4571  *    according to the linkEA.
4572  * 2. There may be too many hard links on the object as to its linkEA overflow,
4573  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4574  *    case happened, then at this point, we do not know whether there are some
4575  *    cross-MDTs reference. But there are local references, it guarantees that
4576  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4577  */
4578 static int osd_xattr_handle_linkea(const struct lu_env *env,
4579                                    struct osd_device *osd,
4580                                    struct osd_object *obj,
4581                                    const struct lu_buf *buf,
4582                                    struct thandle *handle)
4583 {
4584         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4585         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4586         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4587         struct lu_name tmpname;
4588         struct osd_thandle *oh;
4589         int rc;
4590         bool remote = false;
4591
4592         ENTRY;
4593
4594         oh = container_of(handle, struct osd_thandle, ot_super);
4595         LASSERT(oh->ot_handle != NULL);
4596
4597         rc = linkea_init_with_rec(&ldata);
4598         if (!rc) {
4599                 linkea_first_entry(&ldata);
4600                 while (ldata.ld_lee != NULL && !remote) {
4601                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4602                                             &tmpname, tfid);
4603                         if (osd_remote_fid(env, osd, tfid) > 0)
4604                                 remote = true;
4605                         else
4606                                 linkea_next_entry(&ldata);
4607                 }
4608         } else if (rc == -ENODATA) {
4609                 rc = 0;
4610         } else {
4611                 RETURN(rc);
4612         }
4613
4614         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4615                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4616                 if (rc)
4617                         CERROR("%s: failed to remove agent entry for "DFID
4618                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4619         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4620                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4621                 if (rc)
4622                         CERROR("%s: failed to create agent entry for "DFID
4623                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4624         }
4625
4626         RETURN(rc);
4627 }
4628
4629 /*
4630  * Concurrency: @dt is write locked.
4631  */
4632 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4633                          const struct lu_buf *buf, const char *name, int fl,
4634                          struct thandle *handle)
4635 {
4636         struct osd_object *obj = osd_dt_obj(dt);
4637         struct osd_device *osd = osd_obj2dev(obj);
4638         struct inode *inode = obj->oo_inode;
4639         struct osd_thread_info *info = osd_oti_get(env);
4640         int fs_flags = 0;
4641         int len;
4642         int rc;
4643
4644         ENTRY;
4645
4646         LASSERT(handle);
4647         LASSERT(buf);
4648
4649         /* version set is not real XATTR */
4650         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4651                 dt_obj_version_t *version = buf->lb_buf;
4652
4653                 /*
4654                  * for version we are just using xattr API but change inode
4655                  * field instead
4656                  */
4657                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4658
4659                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4660                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4661
4662                 LDISKFS_I(inode)->i_fs_version = *version;
4663                 /*
4664                  * Version is set after all inode operations are finished,
4665                  * so we should mark it dirty here
4666                  */
4667                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4668
4669                 RETURN(0);
4670         }
4671
4672         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4673                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4674
4675         len = buf->lb_len;
4676         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4677
4678         /*
4679          * For the OST device with 256 bytes inode size by default,
4680          * the PFID EA will be stored together with LMA EA to avoid
4681          * performance trouble. Otherwise the PFID EA can be stored
4682          * independently. LU-8998
4683          */
4684         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4685             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4686                 LASSERT(buf->lb_buf);
4687
4688                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4689                 if (fl <= 0)
4690                         RETURN(fl);
4691         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4692                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4693                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4694
4695                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4696                 if (rc)
4697                         RETURN(rc);
4698
4699                 lma->lma_incompat |= LMAI_STRIPED;
4700                 lustre_lma_swab(lma);
4701                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4702                                      sizeof(*lma), XATTR_REPLACE);
4703                 if (rc != 0)
4704                         RETURN(rc);
4705         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4706                 LASSERT(!osd->od_is_ost);
4707
4708                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4709                 if (rc)
4710                         RETURN(rc);
4711         }
4712
4713         if (fl & LU_XATTR_REPLACE)
4714                 fs_flags |= XATTR_REPLACE;
4715
4716         if (fl & LU_XATTR_CREATE)
4717                 fs_flags |= XATTR_CREATE;
4718
4719         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4720         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4721
4722         if (rc == 0 &&
4723             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4724              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4725                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4726
4727         return rc;
4728 }
4729
4730 /*
4731  * Concurrency: @dt is read locked.
4732  */
4733 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4734                           const struct lu_buf *buf)
4735 {
4736         struct osd_object *obj = osd_dt_obj(dt);
4737         struct inode *inode = obj->oo_inode;
4738         struct osd_thread_info *info = osd_oti_get(env);
4739         struct dentry *dentry = &info->oti_obj_dentry;
4740
4741         if (!dt_object_exists(dt))
4742                 return -ENOENT;
4743
4744         LASSERT(!dt_object_remote(dt));
4745         LASSERT(inode->i_op != NULL);
4746         LASSERT(inode->i_op->listxattr != NULL);
4747
4748         dentry->d_inode = inode;
4749         dentry->d_sb = inode->i_sb;
4750         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4751 }
4752
4753 static int osd_declare_xattr_del(const struct lu_env *env,
4754                                  struct dt_object *dt, const char *name,
4755                                  struct thandle *handle)
4756 {
4757         struct osd_thandle *oh;
4758         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4759
4760         LASSERT(!dt_object_remote(dt));
4761         LASSERT(handle != NULL);
4762
4763         oh = container_of(handle, struct osd_thandle, ot_super);
4764         LASSERT(oh->ot_handle == NULL);
4765
4766         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4767                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4768         /*
4769          * xattr del may involve inode quota change, reserve credits for
4770          * dquot_initialize()
4771          */
4772         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4773
4774         return 0;
4775 }
4776
4777 /*
4778  * Concurrency: @dt is write locked.
4779  */
4780 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4781                          const char *name, struct thandle *handle)
4782 {
4783         struct osd_object *obj = osd_dt_obj(dt);
4784         struct inode *inode = obj->oo_inode;
4785         struct osd_thread_info *info = osd_oti_get(env);
4786         struct dentry *dentry = &info->oti_obj_dentry;
4787         int rc;
4788
4789         if (!dt_object_exists(dt))
4790                 return -ENOENT;
4791
4792         LASSERT(!dt_object_remote(dt));
4793         LASSERT(inode->i_op != NULL);
4794         LASSERT(handle != NULL);
4795 #ifdef HAVE_IOP_XATTR
4796         LASSERT(inode->i_op->removexattr != NULL);
4797 #endif
4798
4799         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4800
4801         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4802                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4803
4804                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4805                                  &info->oti_ost_attrs);
4806                 if (!rc) {
4807                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4808
4809                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4810                         lustre_lma_swab(lma);
4811                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4812                                              sizeof(*lma), XATTR_REPLACE);
4813                         if (!rc)
4814                                 obj->oo_pfid_in_lma = 0;
4815                 }
4816         } else {
4817                 dquot_initialize(inode);
4818                 dentry->d_inode = inode;
4819                 dentry->d_sb = inode->i_sb;
4820                 rc = ll_vfs_removexattr(dentry, inode, name);
4821         }
4822
4823         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4824
4825         if (rc == 0 &&
4826             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4827              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4828                 osd_oxc_del(obj, name);
4829
4830         return rc;
4831 }
4832
4833 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4834                            __u64 start, __u64 end)
4835 {
4836         struct osd_object *obj = osd_dt_obj(dt);
4837         struct inode *inode = obj->oo_inode;
4838         struct osd_thread_info *info = osd_oti_get(env);
4839         struct dentry *dentry = &info->oti_obj_dentry;
4840         struct file *file = &info->oti_file;
4841         int rc;
4842
4843         ENTRY;
4844
4845         dentry->d_inode = inode;
4846         dentry->d_sb = inode->i_sb;
4847         file->f_path.dentry = dentry;
4848         file->f_mapping = inode->i_mapping;
4849         file->f_op = inode->i_fop;
4850         file->f_inode = inode;
4851
4852         rc = vfs_fsync_range(file, start, end, 0);
4853
4854         RETURN(rc);
4855 }
4856
4857 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4858 {
4859         return 0;
4860 }
4861
4862 /*
4863  * Index operations.
4864  */
4865
4866 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4867                                const struct dt_index_features *feat)
4868 {
4869         struct iam_descr *descr;
4870
4871         if (osd_object_is_root(o))
4872                 return feat == &dt_directory_features;
4873
4874         LASSERT(o->oo_dir != NULL);
4875
4876         descr = o->oo_dir->od_container.ic_descr;
4877         if (feat == &dt_directory_features) {
4878                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4879                         return 1;
4880                 else
4881                         return 0;
4882         } else {
4883                 return feat->dif_keysize_min <= descr->id_key_size &&
4884                        descr->id_key_size <= feat->dif_keysize_max &&
4885                        feat->dif_recsize_min <= descr->id_rec_size &&
4886                        descr->id_rec_size <= feat->dif_recsize_max &&
4887                        !(feat->dif_flags & (DT_IND_VARKEY |
4888                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4889                        ergo(feat->dif_flags & DT_IND_UPDATE,
4890                             1 /* XXX check that object (and fs) is writable */);
4891         }
4892 }
4893
4894 static int osd_iam_container_init(const struct lu_env *env,
4895                                   struct osd_object *obj,
4896                                   struct osd_directory *dir)
4897 {
4898         struct iam_container *bag = &dir->od_container;
4899         int result;
4900
4901         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4902         if (result != 0)
4903                 return result;
4904
4905         result = iam_container_setup(bag);
4906         if (result == 0)
4907                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4908         else
4909                 iam_container_fini(bag);
4910
4911         return result;
4912 }
4913
4914
4915 /*
4916  * Concurrency: no external locking is necessary.
4917  */
4918 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4919                          const struct dt_index_features *feat)
4920 {
4921         int result;
4922         int skip_iam = 0;
4923         struct osd_object *obj = osd_dt_obj(dt);
4924
4925         LINVRNT(osd_invariant(obj));
4926
4927         if (osd_object_is_root(obj)) {
4928                 dt->do_index_ops = &osd_index_ea_ops;
4929                 result = 0;
4930         } else if (feat == &dt_directory_features) {
4931                 dt->do_index_ops = &osd_index_ea_ops;
4932                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4933                         result = 0;
4934                 else
4935                         result = -ENOTDIR;
4936                 skip_iam = 1;
4937         } else if (unlikely(feat == &dt_otable_features)) {
4938                 dt->do_index_ops = &osd_otable_ops;
4939                 return 0;
4940         } else if (unlikely(feat == &dt_acct_features)) {
4941                 dt->do_index_ops = &osd_acct_index_ops;
4942                 result = 0;
4943                 skip_iam = 1;
4944         } else if (!osd_has_index(obj)) {
4945                 struct osd_directory *dir;
4946                 struct osd_device *osd = osd_obj2dev(obj);
4947                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4948
4949                 OBD_ALLOC_PTR(dir);
4950                 if (dir) {
4951
4952                         spin_lock(&obj->oo_guard);
4953                         if (obj->oo_dir == NULL)
4954                                 obj->oo_dir = dir;
4955                         else
4956                                 /*
4957                                  * Concurrent thread allocated container data.
4958                                  */
4959                                 OBD_FREE_PTR(dir);
4960                         spin_unlock(&obj->oo_guard);
4961                         /*
4962                          * Now, that we have container data, serialize its
4963                          * initialization.
4964                          */
4965                         down_write(&obj->oo_ext_idx_sem);
4966                         /*
4967                          * recheck under lock.
4968                          */
4969
4970                         if (osd_has_index(obj)) {
4971                                 result = 0;
4972                                 goto unlock;
4973                         }
4974
4975                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4976                         if (result || feat == &dt_lfsck_namespace_features ||
4977                             feat == &dt_lfsck_layout_orphan_features ||
4978                             feat == &dt_lfsck_layout_dangling_features)
4979                                 goto unlock;
4980
4981                         result = osd_index_register(osd, fid,
4982                                                     feat->dif_keysize_max,
4983                                                     feat->dif_recsize_max);
4984                         if (result < 0)
4985                                 CWARN("%s: failed to register index "
4986                                       DFID": rc = %d\n",
4987                                       osd_name(osd), PFID(fid), result);
4988                         else if (result > 0)
4989                                 result = 0;
4990                         else
4991                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4992                                        " (%d/%d) registered\n",
4993                                        osd_name(osd), PFID(fid),
4994                                        (int)feat->dif_keysize_max,
4995                                        (int)feat->dif_recsize_max);
4996
4997 unlock:
4998                         up_write(&obj->oo_ext_idx_sem);
4999                 } else {
5000                         result = -ENOMEM;
5001                 }
5002         } else {
5003                 result = 0;
5004         }
5005
5006         if (result == 0 && skip_iam == 0) {
5007                 if (!osd_iam_index_probe(env, obj, feat))
5008                         result = -ENOTDIR;
5009         }
5010         LINVRNT(osd_invariant(obj));
5011
5012         return result;
5013 }
5014
5015 static int osd_otable_it_attr_get(const struct lu_env *env,
5016                                  struct dt_object *dt,
5017                                  struct lu_attr *attr)
5018 {
5019         attr->la_valid = 0;
5020         return 0;
5021 }
5022
5023 static const struct dt_object_operations osd_obj_ops = {
5024         .do_read_lock           = osd_read_lock,
5025         .do_write_lock          = osd_write_lock,
5026         .do_read_unlock         = osd_read_unlock,
5027         .do_write_unlock        = osd_write_unlock,
5028         .do_write_locked        = osd_write_locked,
5029         .do_attr_get            = osd_attr_get,
5030         .do_declare_attr_set    = osd_declare_attr_set,
5031         .do_attr_set            = osd_attr_set,
5032         .do_ah_init             = osd_ah_init,
5033         .do_declare_create      = osd_declare_create,
5034         .do_create              = osd_create,
5035         .do_declare_destroy     = osd_declare_destroy,
5036         .do_destroy             = osd_destroy,
5037         .do_index_try           = osd_index_try,
5038         .do_declare_ref_add     = osd_declare_ref_add,
5039         .do_ref_add             = osd_ref_add,
5040         .do_declare_ref_del     = osd_declare_ref_del,
5041         .do_ref_del             = osd_ref_del,
5042         .do_xattr_get           = osd_xattr_get,
5043         .do_declare_xattr_set   = osd_declare_xattr_set,
5044         .do_xattr_set           = osd_xattr_set,
5045         .do_declare_xattr_del   = osd_declare_xattr_del,
5046         .do_xattr_del           = osd_xattr_del,
5047         .do_xattr_list          = osd_xattr_list,
5048         .do_object_sync         = osd_object_sync,
5049         .do_invalidate          = osd_invalidate,
5050 };
5051
5052 static const struct dt_object_operations osd_obj_otable_it_ops = {
5053         .do_attr_get    = osd_otable_it_attr_get,
5054         .do_index_try   = osd_index_try,
5055 };
5056
5057 static int osd_index_declare_iam_delete(const struct lu_env *env,
5058                                         struct dt_object *dt,
5059                                         const struct dt_key *key,
5060                                         struct thandle *handle)
5061 {
5062         struct osd_thandle *oh;
5063
5064         oh = container_of(handle, struct osd_thandle, ot_super);
5065         LASSERT(oh->ot_handle == NULL);
5066
5067         /* Recycle  may cause additional three blocks to be changed. */
5068         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5069                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5070
5071         return 0;
5072 }
5073
5074 /**
5075  *      delete a (key, value) pair from index \a dt specified by \a key
5076  *
5077  *      \param  dt      osd index object
5078  *      \param  key     key for index
5079  *      \param  rec     record reference
5080  *      \param  handle  transaction handler
5081  *
5082  *      \retval  0  success
5083  *      \retval -ve   failure
5084  */
5085 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5086                                 const struct dt_key *key,
5087                                 struct thandle *handle)
5088 {
5089         struct osd_thread_info *oti = osd_oti_get(env);
5090         struct osd_object *obj = osd_dt_obj(dt);
5091         struct osd_thandle *oh;
5092         struct iam_path_descr *ipd;
5093         struct iam_container *bag = &obj->oo_dir->od_container;
5094         int rc;
5095
5096         ENTRY;
5097
5098         if (!dt_object_exists(dt))
5099                 RETURN(-ENOENT);
5100
5101         LINVRNT(osd_invariant(obj));
5102         LASSERT(!dt_object_remote(dt));
5103         LASSERT(bag->ic_object == obj->oo_inode);
5104         LASSERT(handle != NULL);
5105
5106         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5107
5108         ipd = osd_idx_ipd_get(env, bag);
5109         if (unlikely(ipd == NULL))
5110                 RETURN(-ENOMEM);
5111
5112         oh = container_of(handle, struct osd_thandle, ot_super);
5113         LASSERT(oh->ot_handle != NULL);
5114         LASSERT(oh->ot_handle->h_transaction != NULL);
5115
5116         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5117                 /* swab quota uid/gid provided by caller */
5118                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5119                 key = (const struct dt_key *)&oti->oti_quota_id;
5120         }
5121
5122         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5123         osd_ipd_put(env, bag, ipd);
5124         LINVRNT(osd_invariant(obj));
5125         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5126         RETURN(rc);
5127 }
5128
5129 static int osd_index_declare_ea_delete(const struct lu_env *env,
5130                                        struct dt_object *dt,
5131                                        const struct dt_key *key,
5132                                        struct thandle *handle)
5133 {
5134         struct osd_thandle *oh;
5135         struct inode *inode;
5136         int rc, credits;
5137
5138         ENTRY;
5139
5140         LASSERT(!dt_object_remote(dt));
5141         LASSERT(handle != NULL);
5142
5143         oh = container_of(handle, struct osd_thandle, ot_super);
5144         LASSERT(oh->ot_handle == NULL);
5145
5146         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5147         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5148
5149         inode = osd_dt_obj(dt)->oo_inode;
5150         if (inode == NULL)
5151                 RETURN(-ENOENT);
5152
5153         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5154                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5155                                    NULL, OSD_QID_BLK);
5156         RETURN(rc);
5157 }
5158
5159 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5160                                           struct dt_rec *fid)
5161 {
5162         struct osd_fid_pack *rec;
5163         int rc = -ENODATA;
5164
5165         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5166                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5167                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5168                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5169                         rc = -EINVAL;
5170         }
5171         return rc;
5172 }
5173
5174 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5175                           const struct lu_fid *fid)
5176 {
5177         struct seq_server_site *ss = osd_seq_site(osd);
5178
5179         ENTRY;
5180
5181         /* FID seqs not in FLDB, must be local seq */
5182         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5183                 RETURN(0);
5184
5185         /*
5186          * If FLD is not being initialized yet, it only happens during the
5187          * initialization, likely during mgs initialization, and we assume
5188          * this is local FID.
5189          */
5190         if (ss == NULL || ss->ss_server_fld == NULL)
5191                 RETURN(0);
5192
5193         /* Only check the local FLDB here */
5194         if (osd_seq_exists(env, osd, fid_seq(fid)))
5195                 RETURN(0);
5196
5197         RETURN(1);
5198 }
5199
5200 static void osd_take_care_of_agent(const struct lu_env *env,
5201                                    struct osd_device *osd,
5202                                    struct osd_thandle *oh,
5203                                    struct ldiskfs_dir_entry_2 *de)
5204 {
5205         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5206         struct osd_idmap_cache *idc;
5207         int rc, schedule = 0;
5208
5209         LASSERT(de != NULL);
5210
5211         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5212         if (likely(rc == 0)) {
5213                 idc = osd_idc_find_or_init(env, osd, fid);
5214                 if (IS_ERR(idc) || idc->oic_remote)
5215                         schedule = 1;
5216         } else if (rc == -ENODATA) {
5217                 /*
5218                  * can't get FID, postpone to the end of the
5219                  * transaction when iget() is safe
5220                  */
5221                 schedule = 1;
5222         } else {
5223                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5224         }
5225         if (schedule)
5226                 osd_schedule_agent_inode_removal(env, oh,
5227                                                  le32_to_cpu(de->inode));
5228 }
5229
5230 /**
5231  * Index delete function for interoperability mode (b11826).
5232  * It will remove the directory entry added by osd_index_ea_insert().
5233  * This entry is needed to maintain name->fid mapping.
5234  *
5235  * \param key,  key i.e. file entry to be deleted
5236  *
5237  * \retval   0, on success
5238  * \retval -ve, on error
5239  */
5240 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5241                                const struct dt_key *key, struct thandle *handle)
5242 {
5243         struct osd_object *obj = osd_dt_obj(dt);
5244         struct inode *dir = obj->oo_inode;
5245         struct dentry *dentry;
5246         struct osd_thandle *oh;
5247         struct ldiskfs_dir_entry_2 *de = NULL;
5248         struct buffer_head *bh;
5249         struct htree_lock *hlock = NULL;
5250         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5251         int rc;
5252
5253         ENTRY;
5254
5255         if (!dt_object_exists(dt))
5256                 RETURN(-ENOENT);
5257
5258         LINVRNT(osd_invariant(obj));
5259         LASSERT(!dt_object_remote(dt));
5260         LASSERT(handle != NULL);
5261
5262         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5263
5264         oh = container_of(handle, struct osd_thandle, ot_super);
5265         LASSERT(oh->ot_handle != NULL);
5266         LASSERT(oh->ot_handle->h_transaction != NULL);
5267
5268         dquot_initialize(dir);
5269         dentry = osd_child_dentry_get(env, obj,
5270                                       (char *)key, strlen((char *)key));
5271
5272         if (obj->oo_hl_head != NULL) {
5273                 hlock = osd_oti_get(env)->oti_hlock;
5274                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5275                                    dir, LDISKFS_HLOCK_DEL);
5276         } else {
5277                 down_write(&obj->oo_ext_idx_sem);
5278         }
5279
5280         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5281         if (!IS_ERR(bh)) {
5282                 /*
5283                  * If this is not the ".." entry, it might be a remote DNE
5284                  * entry and  we need to check if the FID is for a remote
5285                  * MDT.  If the FID is  not in the directory entry (e.g.
5286                  * upgraded 1.8 filesystem without dirdata enabled) then
5287                  * we need to get the FID from the LMA. For a remote directory
5288                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5289                  * case.
5290                  *
5291                  * Delete the entry before the agent inode in order to
5292                  * simplify error handling.  At worst an error after deleting
5293                  * the entry first might leak the agent inode afterward. The
5294                  * reverse would need filesystem abort in case of error deleting
5295                  * the entry after the agent had been removed, or leave a
5296                  * dangling entry pointing at a random inode.
5297                  */
5298                 if (strcmp((char *)key, dotdot) != 0)
5299                         osd_take_care_of_agent(env, osd, oh, de);
5300                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5301                 brelse(bh);
5302         } else {
5303                 rc = PTR_ERR(bh);
5304         }
5305
5306         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5307             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5308                 /* NB, dirent count may not be accurate, because it's counted
5309                  * without lock.
5310                  */
5311                 if (obj->oo_dirent_count)
5312                         obj->oo_dirent_count--;
5313                 else
5314                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5315         }
5316         if (hlock != NULL)
5317                 ldiskfs_htree_unlock(hlock);
5318         else
5319                 up_write(&obj->oo_ext_idx_sem);
5320         GOTO(out, rc);
5321 out:
5322         LASSERT(osd_invariant(obj));
5323         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5324         RETURN(rc);
5325 }
5326
5327 /**
5328  *      Lookup index for \a key and copy record to \a rec.
5329  *
5330  *      \param  dt      osd index object
5331  *      \param  key     key for index
5332  *      \param  rec     record reference
5333  *
5334  *      \retval  +ve  success : exact mach
5335  *      \retval  0    return record with key not greater than \a key
5336  *      \retval -ve   failure
5337  */
5338 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5339                                 struct dt_rec *rec, const struct dt_key *key)
5340 {
5341         struct osd_object *obj = osd_dt_obj(dt);
5342         struct iam_path_descr *ipd;
5343         struct iam_container *bag = &obj->oo_dir->od_container;
5344         struct osd_thread_info *oti = osd_oti_get(env);
5345         struct iam_iterator *it = &oti->oti_idx_it;
5346         struct iam_rec *iam_rec;
5347         int rc;
5348
5349         ENTRY;
5350
5351         if (!dt_object_exists(dt))
5352                 RETURN(-ENOENT);
5353
5354         LASSERT(osd_invariant(obj));
5355         LASSERT(!dt_object_remote(dt));
5356         LASSERT(bag->ic_object == obj->oo_inode);
5357
5358         ipd = osd_idx_ipd_get(env, bag);
5359         if (IS_ERR(ipd))
5360                 RETURN(-ENOMEM);
5361
5362         /* got ipd now we can start iterator. */
5363         iam_it_init(it, bag, 0, ipd);
5364
5365         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5366                 /* swab quota uid/gid provided by caller */
5367                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5368                 key = (const struct dt_key *)&oti->oti_quota_id;
5369         }
5370
5371         rc = iam_it_get(it, (struct iam_key *)key);
5372         if (rc >= 0) {
5373                 if (S_ISDIR(obj->oo_inode->i_mode))
5374                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5375                 else
5376                         iam_rec = (struct iam_rec *)rec;
5377
5378                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5379
5380                 if (S_ISDIR(obj->oo_inode->i_mode))
5381                         osd_fid_unpack((struct lu_fid *)rec,
5382                                        (struct osd_fid_pack *)iam_rec);
5383                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5384                         osd_quota_unpack(obj, rec);
5385         }
5386
5387         iam_it_put(it);
5388         iam_it_fini(it);
5389         osd_ipd_put(env, bag, ipd);
5390
5391         LINVRNT(osd_invariant(obj));
5392
5393         RETURN(rc);
5394 }
5395
5396 static int osd_index_declare_iam_insert(const struct lu_env *env,
5397                                         struct dt_object *dt,
5398                                         const struct dt_rec *rec,
5399                                         const struct dt_key *key,
5400                                         struct thandle *handle)
5401 {
5402         struct osd_thandle *oh;
5403
5404         LASSERT(handle != NULL);
5405
5406         oh = container_of(handle, struct osd_thandle, ot_super);
5407         LASSERT(oh->ot_handle == NULL);
5408
5409         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5410                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5411
5412         return 0;
5413 }
5414
5415 /**
5416  *      Inserts (key, value) pair in \a dt index object.
5417  *
5418  *      \param  dt      osd index object
5419  *      \param  key     key for index
5420  *      \param  rec     record reference
5421  *      \param  th      transaction handler
5422  *
5423  *      \retval  0  success
5424  *      \retval -ve failure
5425  */
5426 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5427                                 const struct dt_rec *rec,
5428                                 const struct dt_key *key, struct thandle *th)
5429 {
5430         struct osd_object *obj = osd_dt_obj(dt);
5431         struct iam_path_descr *ipd;
5432         struct osd_thandle *oh;
5433         struct iam_container *bag;
5434         struct osd_thread_info *oti = osd_oti_get(env);
5435         struct iam_rec *iam_rec;
5436         int rc;
5437
5438         ENTRY;
5439
5440         if (!dt_object_exists(dt))
5441                 RETURN(-ENOENT);
5442
5443         LINVRNT(osd_invariant(obj));
5444         LASSERT(!dt_object_remote(dt));
5445
5446         bag = &obj->oo_dir->od_container;
5447         LASSERT(bag->ic_object == obj->oo_inode);
5448         LASSERT(th != NULL);
5449
5450         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5451
5452         ipd = osd_idx_ipd_get(env, bag);
5453         if (unlikely(ipd == NULL))
5454                 RETURN(-ENOMEM);
5455
5456         oh = container_of(th, struct osd_thandle, ot_super);
5457         LASSERT(oh->ot_handle != NULL);
5458         LASSERT(oh->ot_handle->h_transaction != NULL);
5459         if (S_ISDIR(obj->oo_inode->i_mode)) {
5460                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5461                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5462                              &oti->oti_fid);
5463         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5464                 /* pack quota uid/gid */
5465                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5466                 key = (const struct dt_key *)&oti->oti_quota_id;
5467                 /* pack quota record */
5468                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5469                 iam_rec = (struct iam_rec *)rec;
5470         } else {
5471                 iam_rec = (struct iam_rec *)rec;
5472         }
5473
5474         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5475                         iam_rec, ipd);
5476         osd_ipd_put(env, bag, ipd);
5477         LINVRNT(osd_invariant(obj));
5478         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5479         RETURN(rc);
5480 }
5481
5482 /**
5483  * Calls ldiskfs_add_entry() to add directory entry
5484  * into the directory. This is required for
5485  * interoperability mode (b11826)
5486  *
5487  * \retval   0, on success
5488  * \retval -ve, on error
5489  */
5490 static int __osd_ea_add_rec(struct osd_thread_info *info,
5491                             struct osd_object *pobj, struct inode  *cinode,
5492                             const char *name, const struct lu_fid *fid,
5493                             struct htree_lock *hlock, struct thandle *th)
5494 {
5495         struct ldiskfs_dentry_param *ldp;
5496         struct dentry *child;
5497         struct osd_thandle *oth;
5498         int rc;
5499
5500         oth = container_of(th, struct osd_thandle, ot_super);
5501         LASSERT(oth->ot_handle != NULL);
5502         LASSERT(oth->ot_handle->h_transaction != NULL);
5503         LASSERT(pobj->oo_inode);
5504
5505         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5506         if (unlikely(pobj->oo_inode ==
5507                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5508                 ldp->edp_magic = 0;
5509         else
5510                 osd_get_ldiskfs_dirent_param(ldp, fid);
5511         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5512         child->d_fsdata = (void *)ldp;
5513         dquot_initialize(pobj->oo_inode);
5514         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5515                                    child, cinode, hlock);
5516         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5517                 struct ldiskfs_dir_entry_2      *de;
5518                 struct buffer_head              *bh;
5519                 int                              rc1;
5520
5521                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5522                                             NULL, hlock);
5523                 if (!IS_ERR(bh)) {
5524                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5525                                                                bh);
5526                         if (rc1 == 0) {
5527                                 if (S_ISDIR(cinode->i_mode))
5528                                         de->file_type = LDISKFS_DIRENT_LUFID |
5529                                                         LDISKFS_FT_REG_FILE;
5530                                 else
5531                                         de->file_type = LDISKFS_DIRENT_LUFID |
5532                                                         LDISKFS_FT_DIR;
5533                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5534                                                               NULL, bh);
5535                         }
5536                         brelse(bh);
5537                 }
5538         }
5539
5540         RETURN(rc);
5541 }
5542
5543 /**
5544  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5545  * into the directory.Also sets flags into osd object to
5546  * indicate dot and dotdot are created. This is required for
5547  * interoperability mode (b11826)
5548  *
5549  * \param dir   directory for dot and dotdot fixup.
5550  * \param obj   child object for linking
5551  *
5552  * \retval   0, on success
5553  * \retval -ve, on error
5554  */
5555 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5556                               struct osd_object *dir,
5557                               struct inode *parent_dir, const char *name,
5558                               const struct lu_fid *dot_fid,
5559                               const struct lu_fid *dot_dot_fid,
5560                               struct thandle *th)
5561 {
5562         struct inode *inode = dir->oo_inode;
5563         struct osd_thandle *oth;
5564         int result = 0;
5565
5566         oth = container_of(th, struct osd_thandle, ot_super);
5567         LASSERT(oth->ot_handle->h_transaction != NULL);
5568         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5569
5570         if (strcmp(name, dot) == 0) {
5571                 if (dir->oo_compat_dot_created) {
5572                         result = -EEXIST;
5573                 } else {
5574                         LASSERT(inode->i_ino == parent_dir->i_ino);
5575                         dir->oo_compat_dot_created = 1;
5576                         result = 0;
5577                 }
5578         } else if (strcmp(name, dotdot) == 0) {
5579                 if (!dir->oo_compat_dot_created)
5580                         return -EINVAL;
5581                 /* in case of rename, dotdot is already created */
5582                 if (dir->oo_compat_dotdot_created) {
5583                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5584                                                 dot_dot_fid, NULL, th);
5585                 }
5586
5587                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5588                         struct lu_fid tfid = *dot_dot_fid;
5589
5590                         tfid.f_oid--;
5591                         result = osd_add_dot_dotdot_internal(info,
5592                                         dir->oo_inode, parent_dir, dot_fid,
5593                                         &tfid, oth);
5594                 } else {
5595                         result = osd_add_dot_dotdot_internal(info,
5596                                         dir->oo_inode, parent_dir, dot_fid,
5597                                         dot_dot_fid, oth);
5598                 }
5599
5600                 if (result == 0)
5601                         dir->oo_compat_dotdot_created = 1;
5602         }
5603
5604         return result;
5605 }
5606
5607
5608 /**
5609  * It will call the appropriate osd_add* function and return the
5610  * value, return by respective functions.
5611  */
5612 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5613                           struct inode *cinode, const char *name,
5614                           const struct lu_fid *fid, struct thandle *th)
5615 {
5616         struct osd_thread_info *info = osd_oti_get(env);
5617         struct htree_lock *hlock;
5618         int rc;
5619
5620         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5621
5622         if (name[0] == '.' && (name[1] == '\0' ||
5623                                (name[1] == '.' && name[2] == '\0'))) {
5624                 if (hlock != NULL) {
5625                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5626                                            pobj->oo_inode, 0);
5627                 } else {
5628                         down_write(&pobj->oo_ext_idx_sem);
5629                 }
5630
5631                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5632                                         lu_object_fid(&pobj->oo_dt.do_lu),
5633                                         fid, th);
5634         } else {
5635                 if (hlock != NULL) {
5636                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5637                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5638                 } else {
5639                         down_write(&pobj->oo_ext_idx_sem);
5640                 }
5641
5642                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5643                         struct lu_fid *tfid = &info->oti_fid;
5644
5645                         *tfid = *fid;
5646                         tfid->f_ver = ~0;
5647                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5648                                               tfid, hlock, th);
5649                 } else {
5650                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5651                                               hlock, th);
5652                 }
5653         }
5654         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5655             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5656                 pobj->oo_dirent_count++;
5657
5658         if (hlock != NULL)
5659                 ldiskfs_htree_unlock(hlock);
5660         else
5661                 up_write(&pobj->oo_ext_idx_sem);
5662
5663         return rc;
5664 }
5665
5666 static int
5667 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5668                       struct osd_idmap_cache *oic)
5669 {
5670         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5671         struct lu_fid *fid = &oic->oic_fid;
5672         struct osd_inode_id *id = &oic->oic_lid;
5673         struct inode *inode = NULL;
5674         int once = 0;
5675         bool insert;
5676         int rc;
5677
5678         ENTRY;
5679
5680         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5681                 RETURN(0);
5682
5683         if (scrub->os_running && scrub->os_pos_current > id->oii_ino)
5684                 RETURN(0);
5685
5686         if (dev->od_auto_scrub_interval == AS_NEVER ||
5687             ktime_get_real_seconds() <
5688             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5689                 RETURN(0);
5690
5691 again:
5692         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5693         if (rc == -ENOENT) {
5694                 __u32 gen = id->oii_gen;
5695
5696                 insert = true;
5697                 if (inode != NULL)
5698                         goto trigger;
5699
5700                 inode = osd_iget(oti, dev, id);
5701                 /* The inode has been removed (by race maybe). */
5702                 if (IS_ERR(inode)) {
5703                         rc = PTR_ERR(inode);
5704
5705                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5706                 }
5707
5708                 /* The OI mapping is lost. */
5709                 if (gen != OSD_OII_NOGEN)
5710                         goto trigger;
5711
5712                 /*
5713                  * The inode may has been reused by others, we do not know,
5714                  * leave it to be handled by subsequent osd_fid_lookup().
5715                  */
5716                 GOTO(out, rc = 0);
5717         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5718                 GOTO(out, rc);
5719         }
5720
5721         insert = false;
5722
5723 trigger:
5724         if (scrub->os_running) {
5725                 if (inode == NULL) {
5726                         inode = osd_iget(oti, dev, id);
5727                         /* The inode has been removed (by race maybe). */
5728                         if (IS_ERR(inode)) {
5729                                 rc = PTR_ERR(inode);
5730
5731                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5732                         }
5733                 }
5734
5735                 rc = osd_oii_insert(dev, oic, insert);
5736                 /*
5737                  * There is race condition between osd_oi_lookup and OI scrub.
5738                  * The OI scrub finished just after osd_oi_lookup() failure.
5739                  * Under such case, it is unnecessary to trigger OI scrub again,
5740                  * but try to call osd_oi_lookup() again.
5741                  */
5742                 if (unlikely(rc == -EAGAIN))
5743                         goto again;
5744
5745                 if (!S_ISDIR(inode->i_mode))
5746                         rc = 0;
5747                 else
5748                         rc = osd_check_lmv(oti, dev, inode, oic);
5749
5750                 GOTO(out, rc);
5751         }
5752
5753         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5754                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5755                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5756                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5757                        "%s: trigger partial OI scrub for RPC inconsistency "
5758                        "checking FID "DFID": rc = %d\n",
5759                        osd_dev2name(dev), PFID(fid), rc);
5760                 if (rc == 0 || rc == -EALREADY)
5761                         goto again;
5762         }
5763
5764         GOTO(out, rc);
5765
5766 out:
5767         if (inode)
5768                 iput(inode);
5769
5770         RETURN(rc);
5771 }
5772
5773 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5774                                struct osd_device *dev,
5775                                struct osd_idmap_cache *oic,
5776                                struct lu_fid *fid, __u32 ino)
5777 {
5778         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5779         struct inode *inode;
5780         int rc;
5781
5782         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5783         inode = osd_iget(oti, dev, &oic->oic_lid);
5784         if (IS_ERR(inode)) {
5785                 fid_zero(&oic->oic_fid);
5786                 return PTR_ERR(inode);
5787         }
5788
5789         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5790         iput(inode);
5791         if (rc != 0)
5792                 fid_zero(&oic->oic_fid);
5793         else
5794                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5795         return rc;
5796 }
5797
5798 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5799                       struct osd_inode_id *id, const struct lu_fid *fid)
5800 {
5801         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5802                id->oii_ino, id->oii_gen, info);
5803         info->oti_cache.oic_lid = *id;
5804         info->oti_cache.oic_fid = *fid;
5805         info->oti_cache.oic_dev = osd;
5806 }
5807
5808 /**
5809  * Get parent FID from the linkEA.
5810  *
5811  * For a directory which parent resides on remote MDT, to satisfy the
5812  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5813  * the other hand, to make the lookup(..) on the directory can return
5814  * the real parent FID, we append the real parent FID after its ".."
5815  * name entry in the /REMOTE_PARENT_DIR.
5816  *
5817  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5818  * backup. So after the restore, we cannot get the right parent FID from
5819  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5820  * we have stored the real parent FID in the directory object's linkEA,
5821  * we can parse the linkEA for the real parent FID.
5822  *
5823  * \param[in] env       pointer to the thread context
5824  * \param[in] obj       pointer to the object to be handled
5825  * \param[out]fid       pointer to the buffer to hold the parent FID
5826  *
5827  * \retval              0 for getting the real parent FID successfully
5828  * \retval              negative error number on failure
5829  */
5830 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5831                                     struct osd_object *obj,
5832                                     struct lu_fid *fid)
5833 {
5834         struct osd_thread_info *oti = osd_oti_get(env);
5835         struct lu_buf *buf = &oti->oti_big_buf;
5836         struct dentry *dentry = &oti->oti_obj_dentry;
5837         struct inode *inode = obj->oo_inode;
5838         struct linkea_data ldata = { NULL };
5839         int rc;
5840
5841         ENTRY;
5842
5843         fid_zero(fid);
5844         if (!S_ISDIR(inode->i_mode))
5845                 RETURN(-EIO);
5846
5847 again:
5848         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5849                              buf->lb_buf, buf->lb_len);
5850         if (rc == -ERANGE) {
5851                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5852                                      NULL, 0);
5853                 if (rc > 0) {
5854                         lu_buf_realloc(buf, rc);
5855                         if (buf->lb_buf == NULL)
5856                                 RETURN(-ENOMEM);
5857
5858                         goto again;
5859                 }
5860         }
5861
5862         if (unlikely(rc == 0))
5863                 RETURN(-ENODATA);
5864
5865         if (rc < 0)
5866                 RETURN(rc);
5867
5868         if (unlikely(buf->lb_buf == NULL)) {
5869                 lu_buf_realloc(buf, rc);
5870                 if (buf->lb_buf == NULL)
5871                         RETURN(-ENOMEM);
5872
5873                 goto again;
5874         }
5875
5876         ldata.ld_buf = buf;
5877         rc = linkea_init_with_rec(&ldata);
5878         if (!rc) {
5879                 linkea_first_entry(&ldata);
5880                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5881         }
5882
5883         RETURN(rc);
5884 }
5885
5886 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5887                                     struct inode *inode,
5888                                     const struct lu_fid *pfid,
5889                                     const char *name, const int namelen)
5890 {
5891         struct osd_thread_info *oti = osd_oti_get(env);
5892         struct lu_buf *buf = &oti->oti_big_buf;
5893         struct dentry *dentry = &oti->oti_obj_dentry;
5894         struct linkea_data ldata = { NULL };
5895         struct lu_name cname = { .ln_name = name,
5896                                  .ln_namelen = namelen };
5897         int rc;
5898
5899         ENTRY;
5900
5901 again:
5902         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5903                              buf->lb_buf, buf->lb_len);
5904         if (rc == -ERANGE)
5905                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5906
5907         if (rc < 0)
5908                 RETURN(rc);
5909
5910         if (unlikely(rc == 0))
5911                 RETURN(-ENODATA);
5912
5913         if (buf->lb_len < rc) {
5914                 lu_buf_realloc(buf, rc);
5915                 if (buf->lb_buf == NULL)
5916                         RETURN(-ENOMEM);
5917
5918                 goto again;
5919         }
5920
5921         ldata.ld_buf = buf;
5922         rc = linkea_init_with_rec(&ldata);
5923         if (!rc)
5924                 rc = linkea_links_find(&ldata, &cname, pfid);
5925
5926         RETURN(rc);
5927 }
5928
5929 /**
5930  * Calls ->lookup() to find dentry. From dentry get inode and
5931  * read inode's ea to get fid. This is required for  interoperability
5932  * mode (b11826)
5933  *
5934  * \retval   0, on success
5935  * \retval -ve, on error
5936  */
5937 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5938                              struct dt_rec *rec, const struct dt_key *key)
5939 {
5940         struct inode *dir = obj->oo_inode;
5941         struct dentry *dentry;
5942         struct ldiskfs_dir_entry_2 *de;
5943         struct buffer_head *bh;
5944         struct lu_fid *fid = (struct lu_fid *)rec;
5945         struct htree_lock *hlock = NULL;
5946         int ino;
5947         int rc;
5948
5949         ENTRY;
5950
5951         LASSERT(dir->i_op != NULL);
5952         LASSERT(dir->i_op->lookup != NULL);
5953
5954         dentry = osd_child_dentry_get(env, obj,
5955                                       (char *)key, strlen((char *)key));
5956
5957         if (obj->oo_hl_head != NULL) {
5958                 hlock = osd_oti_get(env)->oti_hlock;
5959                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5960                                    dir, LDISKFS_HLOCK_LOOKUP);
5961         } else {
5962                 down_read(&obj->oo_ext_idx_sem);
5963         }
5964
5965         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5966         if (!IS_ERR(bh)) {
5967                 struct osd_thread_info *oti = osd_oti_get(env);
5968                 struct osd_inode_id *id = &oti->oti_id;
5969                 struct osd_idmap_cache *oic = &oti->oti_cache;
5970                 struct osd_device *dev = osd_obj2dev(obj);
5971
5972                 ino = le32_to_cpu(de->inode);
5973                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5974                         brelse(bh);
5975                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5976                         GOTO(out, rc);
5977                 }
5978
5979                 rc = osd_get_fid_from_dentry(de, rec);
5980
5981                 /* done with de, release bh */
5982                 brelse(bh);
5983                 if (rc != 0) {
5984                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5985                                 const char *name = (const char *)key;
5986
5987                                 /*
5988                                  * If the parent is on remote MDT, and there
5989                                  * is no FID-in-dirent, then we have to get
5990                                  * the parent FID from the linkEA.
5991                                  */
5992                                 if (likely(strlen(name) == 2 &&
5993                                            name[0] == '.' && name[1] == '.'))
5994                                         rc = osd_get_pfid_from_linkea(env, obj,
5995                                                                       fid);
5996                         } else {
5997                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5998                         }
5999                 } else {
6000                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6001                 }
6002
6003                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
6004                         fid_zero(&oic->oic_fid);
6005
6006                         GOTO(out, rc);
6007                 }
6008
6009                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
6010                 rc = osd_consistency_check(oti, dev, oic);
6011                 if (rc == -ENOENT)
6012                         fid_zero(&oic->oic_fid);
6013                 else
6014                         /* Other error should not affect lookup result. */
6015                         rc = 0;
6016         } else {
6017                 rc = PTR_ERR(bh);
6018         }
6019
6020         GOTO(out, rc);
6021
6022 out:
6023         if (hlock != NULL)
6024                 ldiskfs_htree_unlock(hlock);
6025         else
6026                 up_read(&obj->oo_ext_idx_sem);
6027         return rc;
6028 }
6029
6030 static int osd_index_declare_ea_insert(const struct lu_env *env,
6031                                        struct dt_object *dt,
6032                                        const struct dt_rec *rec,
6033                                        const struct dt_key *key,
6034                                        struct thandle *handle)
6035 {
6036         struct osd_thandle *oh;
6037         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6038         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6039         const struct lu_fid *fid = rec1->rec_fid;
6040         int credits, rc = 0;
6041         struct osd_idmap_cache *idc;
6042
6043         ENTRY;
6044
6045         LASSERT(!dt_object_remote(dt));
6046         LASSERT(handle != NULL);
6047         LASSERT(fid != NULL);
6048         LASSERT(rec1->rec_type != 0);
6049
6050         oh = container_of(handle, struct osd_thandle, ot_super);
6051         LASSERT(oh->ot_handle == NULL);
6052
6053         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6054
6055         /*
6056          * we can't call iget() while a transactions is running
6057          * (this can lead to a deadlock), but we need to know
6058          * inum and object type. so we find this information at
6059          * declaration and cache in per-thread info
6060          */
6061         idc = osd_idc_find_or_init(env, osd, fid);
6062         if (IS_ERR(idc))
6063                 RETURN(PTR_ERR(idc));
6064         if (idc->oic_remote) {
6065                 /*
6066                  * a reference to remote inode is represented by an
6067                  * agent inode which we have to create
6068                  */
6069                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6070                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6071         }
6072
6073         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6074
6075         if (osd_dt_obj(dt)->oo_inode != NULL) {
6076                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6077
6078                 /*
6079                  * We ignore block quota on meta pool (MDTs), so needn't
6080                  * calculate how many blocks will be consumed by this index
6081                  * insert
6082                  */
6083                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6084                                            i_gid_read(inode),
6085                                            i_projid_read(inode), 0,
6086                                            oh, osd_dt_obj(dt), NULL,
6087                                            OSD_QID_BLK);
6088                 if (rc)
6089                         RETURN(rc);
6090
6091 #ifdef HAVE_PROJECT_QUOTA
6092                 /*
6093                  * Reserve credits for local agent inode to transfer
6094                  * to 0, quota enforcement is ignored in this case.
6095                  */
6096                 if (idc->oic_remote &&
6097                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6098                     i_projid_read(inode) != 0)
6099                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6100                                                   0, i_projid_read(inode),
6101                                                   0, false, PRJQUOTA, true);
6102 #endif
6103         }
6104
6105         RETURN(rc);
6106 }
6107
6108 /**
6109  * Index add function for interoperability mode (b11826).
6110  * It will add the directory entry.This entry is needed to
6111  * maintain name->fid mapping.
6112  *
6113  * \param key it is key i.e. file entry to be inserted
6114  * \param rec it is value of given key i.e. fid
6115  *
6116  * \retval   0, on success
6117  * \retval -ve, on error
6118  */
6119 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6120                                const struct dt_rec *rec,
6121                                const struct dt_key *key, struct thandle *th)
6122 {
6123         struct osd_object *obj = osd_dt_obj(dt);
6124         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6125         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6126         const struct lu_fid *fid = rec1->rec_fid;
6127         const char *name = (const char *)key;
6128         struct osd_thread_info *oti = osd_oti_get(env);
6129         struct inode *child_inode = NULL;
6130         struct osd_idmap_cache *idc;
6131         int rc;
6132
6133         ENTRY;
6134
6135         if (!dt_object_exists(dt))
6136                 RETURN(-ENOENT);
6137
6138         LASSERT(osd_invariant(obj));
6139         LASSERT(!dt_object_remote(dt));
6140         LASSERT(th != NULL);
6141
6142         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6143
6144         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6145
6146         idc = osd_idc_find(env, osd, fid);
6147         if (unlikely(idc == NULL)) {
6148                 idc = osd_idc_find_or_init(env, osd, fid);
6149                 if (IS_ERR(idc)) {
6150                         /*
6151                          * this dt_insert() wasn't declared properly, so
6152                          * FID is missing in OI cache. we better do not
6153                          * lookup FID in FLDB/OI and don't risk to deadlock,
6154                          * but in some special cases (lfsck testing, etc)
6155                          * it's much simpler than fixing a caller.
6156                          *
6157                          * normally this error should be placed after the first
6158                          * find, but migrate may attach source stripes to
6159                          * target, which doesn't create stripes.
6160                          */
6161                         CERROR("%s: "DFID" wasn't declared for insert\n",
6162                                osd_name(osd), PFID(fid));
6163                         dump_stack();
6164                         RETURN(PTR_ERR(idc));
6165                 }
6166         }
6167
6168         if (idc->oic_remote) {
6169                 /* Insert remote entry */
6170                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6171                         child_inode =
6172                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6173                 } else {
6174                         child_inode = osd_create_local_agent_inode(env, osd,
6175                                         obj, fid, rec1->rec_type & S_IFMT, th);
6176                         if (IS_ERR(child_inode))
6177                                 RETURN(PTR_ERR(child_inode));
6178                 }
6179         } else {
6180                 /* Insert local entry */
6181                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6182                         /* for a reason OI cache wasn't filled properly */
6183                         CERROR("%s: OIC for "DFID" isn't filled\n",
6184                                osd_name(osd), PFID(fid));
6185                         RETURN(-EINVAL);
6186                 }
6187                 child_inode = oti->oti_inode;
6188                 if (unlikely(child_inode == NULL)) {
6189                         struct ldiskfs_inode_info *lii;
6190
6191                         OBD_ALLOC_PTR(lii);
6192                         if (lii == NULL)
6193                                 RETURN(-ENOMEM);
6194                         child_inode = oti->oti_inode = &lii->vfs_inode;
6195                 }
6196                 child_inode->i_sb = osd_sb(osd);
6197                 child_inode->i_ino = idc->oic_lid.oii_ino;
6198                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6199         }
6200
6201         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6202
6203         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6204                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6205
6206         if (child_inode && child_inode != oti->oti_inode)
6207                 iput(child_inode);
6208         LASSERT(osd_invariant(obj));
6209         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6210
6211         RETURN(rc);
6212 }
6213
6214 /**
6215  *  Initialize osd Iterator for given osd index object.
6216  *
6217  *  \param  dt      osd index object
6218  */
6219
6220 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6221                                      struct dt_object *dt,
6222                                      __u32 unused)
6223 {
6224         struct osd_it_iam *it;
6225         struct osd_object *obj = osd_dt_obj(dt);
6226         struct lu_object *lo = &dt->do_lu;
6227         struct iam_path_descr *ipd;
6228         struct iam_container *bag = &obj->oo_dir->od_container;
6229
6230         if (!dt_object_exists(dt))
6231                 return ERR_PTR(-ENOENT);
6232
6233         OBD_ALLOC_PTR(it);
6234         if (it == NULL)
6235                 return ERR_PTR(-ENOMEM);
6236
6237         ipd = osd_it_ipd_get(env, bag);
6238         if (likely(ipd != NULL)) {
6239                 it->oi_obj = obj;
6240                 it->oi_ipd = ipd;
6241                 lu_object_get(lo);
6242                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6243                 return (struct dt_it *)it;
6244         } else {
6245                 OBD_FREE_PTR(it);
6246                 return ERR_PTR(-ENOMEM);
6247         }
6248 }
6249
6250 /**
6251  * free given Iterator.
6252  */
6253 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6254 {
6255         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6256         struct osd_object *obj = it->oi_obj;
6257
6258         iam_it_fini(&it->oi_it);
6259         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6260         osd_object_put(env, obj);
6261         OBD_FREE_PTR(it);
6262 }
6263
6264 /**
6265  *  Move Iterator to record specified by \a key
6266  *
6267  *  \param  di      osd iterator
6268  *  \param  key     key for index
6269  *
6270  *  \retval +ve  di points to record with least key not larger than key
6271  *  \retval  0   di points to exact matched key
6272  *  \retval -ve  failure
6273  */
6274
6275 static int osd_it_iam_get(const struct lu_env *env,
6276                           struct dt_it *di, const struct dt_key *key)
6277 {
6278         struct osd_thread_info *oti = osd_oti_get(env);
6279         struct osd_it_iam *it = (struct osd_it_iam *)di;
6280
6281         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6282                 /* swab quota uid/gid */
6283                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6284                 key = (struct dt_key *)&oti->oti_quota_id;
6285         }
6286
6287         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6288 }
6289
6290 /**
6291  *  Release Iterator
6292  *
6293  *  \param  di      osd iterator
6294  */
6295 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6296 {
6297         struct osd_it_iam *it = (struct osd_it_iam *)di;
6298
6299         iam_it_put(&it->oi_it);
6300 }
6301
6302 /**
6303  *  Move iterator by one record
6304  *
6305  *  \param  di      osd iterator
6306  *
6307  *  \retval +1   end of container reached
6308  *  \retval  0   success
6309  *  \retval -ve  failure
6310  */
6311
6312 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6313 {
6314         struct osd_it_iam *it = (struct osd_it_iam *)di;
6315
6316         return iam_it_next(&it->oi_it);
6317 }
6318
6319 /**
6320  * Return pointer to the key under iterator.
6321  */
6322
6323 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6324                                      const struct dt_it *di)
6325 {
6326         struct osd_thread_info *oti = osd_oti_get(env);
6327         struct osd_it_iam *it = (struct osd_it_iam *)di;
6328         struct osd_object *obj = it->oi_obj;
6329         struct dt_key *key;
6330
6331         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6332
6333         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6334                 /* swab quota uid/gid */
6335                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6336                 key = (struct dt_key *)&oti->oti_quota_id;
6337         }
6338
6339         return key;
6340 }
6341
6342 /**
6343  * Return size of key under iterator (in bytes)
6344  */
6345
6346 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6347 {
6348         struct osd_it_iam *it = (struct osd_it_iam *)di;
6349
6350         return iam_it_key_size(&it->oi_it);
6351 }
6352
6353 static inline void
6354 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6355 {
6356         /* check if file type is required */
6357         if (ent->lde_attrs & LUDA_TYPE) {
6358                 struct luda_type *lt;
6359                 int align = sizeof(*lt) - 1;
6360
6361                 len = (len + align) & ~align;
6362                 lt = (struct luda_type *)(ent->lde_name + len);
6363                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6364         }
6365
6366         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6367 }
6368
6369 /**
6370  * build lu direct from backend fs dirent.
6371  */
6372
6373 static inline void
6374 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6375                    char *name, __u16 namelen, __u16 type, __u32 attr)
6376 {
6377         ent->lde_attrs = attr | LUDA_FID;
6378         fid_cpu_to_le(&ent->lde_fid, fid);
6379
6380         ent->lde_hash = cpu_to_le64(offset);
6381         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6382
6383         strncpy(ent->lde_name, name, namelen);
6384         ent->lde_name[namelen] = '\0';
6385         ent->lde_namelen = cpu_to_le16(namelen);
6386
6387         /* append lustre attributes */
6388         osd_it_append_attrs(ent, namelen, type);
6389 }
6390
6391 /**
6392  * Return pointer to the record under iterator.
6393  */
6394 static int osd_it_iam_rec(const struct lu_env *env,
6395                           const struct dt_it *di,
6396                           struct dt_rec *dtrec, __u32 attr)
6397 {
6398         struct osd_it_iam *it = (struct osd_it_iam *)di;
6399         struct osd_thread_info *info = osd_oti_get(env);
6400
6401         ENTRY;
6402
6403         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6404                 const struct osd_fid_pack *rec;
6405                 struct lu_fid *fid = &info->oti_fid;
6406                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6407                 char *name;
6408                 int namelen;
6409                 __u64 hash;
6410                 int rc;
6411
6412                 name = (char *)iam_it_key_get(&it->oi_it);
6413                 if (IS_ERR(name))
6414                         RETURN(PTR_ERR(name));
6415
6416                 namelen = iam_it_key_size(&it->oi_it);
6417
6418                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6419                 if (IS_ERR(rec))
6420                         RETURN(PTR_ERR(rec));
6421
6422                 rc = osd_fid_unpack(fid, rec);
6423                 if (rc)
6424                         RETURN(rc);
6425
6426                 hash = iam_it_store(&it->oi_it);
6427
6428                 /* IAM does not store object type in IAM index (dir) */
6429                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6430                                    0, LUDA_FID);
6431         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6432                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6433                            (struct iam_rec *)dtrec);
6434                 osd_quota_unpack(it->oi_obj, dtrec);
6435         } else {
6436                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6437                            (struct iam_rec *)dtrec);
6438         }
6439
6440         RETURN(0);
6441 }
6442
6443 /**
6444  * Returns cookie for current Iterator position.
6445  */
6446 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6447 {
6448         struct osd_it_iam *it = (struct osd_it_iam *)di;
6449
6450         return iam_it_store(&it->oi_it);
6451 }
6452
6453 /**
6454  * Restore iterator from cookie.
6455  *
6456  * \param  di      osd iterator
6457  * \param  hash    Iterator location cookie
6458  *
6459  * \retval +ve  di points to record with least key not larger than key.
6460  * \retval  0   di points to exact matched key
6461  * \retval -ve  failure
6462  */
6463
6464 static int osd_it_iam_load(const struct lu_env *env,
6465                            const struct dt_it *di, __u64 hash)
6466 {
6467         struct osd_it_iam *it = (struct osd_it_iam *)di;
6468
6469         return iam_it_load(&it->oi_it, hash);
6470 }
6471
6472 static const struct dt_index_operations osd_index_iam_ops = {
6473         .dio_lookup         = osd_index_iam_lookup,
6474         .dio_declare_insert = osd_index_declare_iam_insert,
6475         .dio_insert         = osd_index_iam_insert,
6476         .dio_declare_delete = osd_index_declare_iam_delete,
6477         .dio_delete         = osd_index_iam_delete,
6478         .dio_it     = {
6479                 .init     = osd_it_iam_init,
6480                 .fini     = osd_it_iam_fini,
6481                 .get      = osd_it_iam_get,
6482                 .put      = osd_it_iam_put,
6483                 .next     = osd_it_iam_next,
6484                 .key      = osd_it_iam_key,
6485                 .key_size = osd_it_iam_key_size,
6486                 .rec      = osd_it_iam_rec,
6487                 .store    = osd_it_iam_store,
6488                 .load     = osd_it_iam_load
6489         }
6490 };
6491
6492
6493 /**
6494  * Creates or initializes iterator context.
6495  *
6496  * \retval struct osd_it_ea, iterator structure on success
6497  *
6498  */
6499 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6500                                     struct dt_object *dt,
6501                                     __u32 attr)
6502 {
6503         struct osd_object *obj = osd_dt_obj(dt);
6504         struct osd_thread_info *info = osd_oti_get(env);
6505         struct osd_it_ea *oie;
6506         struct file *file;
6507         struct lu_object *lo = &dt->do_lu;
6508         struct dentry *obj_dentry;
6509
6510         ENTRY;
6511
6512         if (!dt_object_exists(dt) || obj->oo_destroyed)
6513                 RETURN(ERR_PTR(-ENOENT));
6514
6515         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6516         if (oie == NULL)
6517                 RETURN(ERR_PTR(-ENOMEM));
6518         obj_dentry = &oie->oie_dentry;
6519
6520         obj_dentry->d_inode = obj->oo_inode;
6521         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6522         obj_dentry->d_name.hash = 0;
6523
6524         oie->oie_rd_dirent       = 0;
6525         oie->oie_it_dirent       = 0;
6526         oie->oie_dirent          = NULL;
6527         if (unlikely(!info->oti_it_ea_buf_used)) {
6528                 oie->oie_buf = info->oti_it_ea_buf;
6529                 info->oti_it_ea_buf_used = 1;
6530         } else {
6531                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6532                 if (oie->oie_buf == NULL)
6533                         RETURN(ERR_PTR(-ENOMEM));
6534         }
6535         oie->oie_obj = obj;
6536
6537         file = &oie->oie_file;
6538
6539         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6540         if (attr & LUDA_64BITHASH)
6541                 file->f_mode    = FMODE_64BITHASH;
6542         else
6543                 file->f_mode    = FMODE_32BITHASH;
6544         file->f_path.dentry     = obj_dentry;
6545         file->f_mapping         = obj->oo_inode->i_mapping;
6546         file->f_op              = obj->oo_inode->i_fop;
6547         file->f_inode = obj->oo_inode;
6548
6549         lu_object_get(lo);
6550         RETURN((struct dt_it *)oie);
6551 }
6552
6553 /**
6554  * Destroy or finishes iterator context.
6555  *
6556  * \param di iterator structure to be destroyed
6557  */
6558 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6559 {
6560         struct osd_thread_info *info = osd_oti_get(env);
6561         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6562         struct osd_object *obj = oie->oie_obj;
6563         struct inode *inode = obj->oo_inode;
6564
6565         ENTRY;
6566         oie->oie_file.f_op->release(inode, &oie->oie_file);
6567         osd_object_put(env, obj);
6568         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6569                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6570         else
6571                 info->oti_it_ea_buf_used = 0;
6572         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6573         EXIT;
6574 }
6575
6576 /**
6577  * It position the iterator at given key, so that next lookup continues from
6578  * that key Or it is similar to dio_it->load() but based on a key,
6579  * rather than file position.
6580  *
6581  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6582  * to the beginning.
6583  *
6584  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6585  */
6586 static int osd_it_ea_get(const struct lu_env *env,
6587                          struct dt_it *di, const struct dt_key *key)
6588 {
6589         struct osd_it_ea *it = (struct osd_it_ea *)di;
6590
6591         ENTRY;
6592         LASSERT(((const char *)key)[0] == '\0');
6593         it->oie_file.f_pos = 0;
6594         it->oie_rd_dirent = 0;
6595         it->oie_it_dirent = 0;
6596         it->oie_dirent = NULL;
6597
6598         RETURN(+1);
6599 }
6600
6601 /**
6602  * Does nothing
6603  */
6604 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6605 {
6606 }
6607
6608 struct osd_filldir_cbs {
6609         struct dir_context ctx;
6610         struct osd_it_ea  *it;
6611 };
6612 /**
6613  * It is called internally by ->iterate*(). It fills the
6614  * iterator's in-memory data structure with required
6615  * information i.e. name, namelen, rec_size etc.
6616  *
6617  * \param buf in which information to be filled in.
6618  * \param name name of the file in given dir
6619  *
6620  * \retval 0 on success
6621  * \retval 1 on buffer full
6622  */
6623 #ifdef HAVE_FILLDIR_USE_CTX
6624 static int osd_ldiskfs_filldir(struct dir_context *buf,
6625 #else
6626 static int osd_ldiskfs_filldir(void *buf,
6627 #endif
6628                                const char *name, int namelen,
6629                                loff_t offset, __u64 ino, unsigned int d_type)
6630 {
6631         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6632         struct osd_object *obj = it->oie_obj;
6633         struct osd_it_ea_dirent *ent = it->oie_dirent;
6634         struct lu_fid *fid = &ent->oied_fid;
6635         struct osd_fid_pack *rec;
6636
6637         ENTRY;
6638
6639 /* this should never happen */
6640         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6641                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6642                 RETURN(-EIO);
6643         }
6644
6645         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6646             OSD_IT_EA_BUFSIZE)
6647                 RETURN(1);
6648
6649         /* "." is just the object itself. */
6650         if (namelen == 1 && name[0] == '.') {
6651                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6652         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6653                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6654                 if (osd_fid_unpack(fid, rec) != 0)
6655                         fid_zero(fid);
6656         } else {
6657                 fid_zero(fid);
6658         }
6659         d_type &= ~LDISKFS_DIRENT_LUFID;
6660
6661         /* NOT export local root. */
6662         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6663                 ino = obj->oo_inode->i_ino;
6664                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6665         }
6666
6667         ent->oied_ino     = ino;
6668         ent->oied_off     = offset;
6669         ent->oied_namelen = namelen;
6670         ent->oied_type    = d_type;
6671
6672         memcpy(ent->oied_name, name, namelen);
6673
6674         it->oie_rd_dirent++;
6675         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6676         RETURN(0);
6677 }
6678
6679 /**
6680  * Calls ->iterate*() to load a directory entry at a time
6681  * and stored it in iterator's in-memory data structure.
6682  *
6683  * \param di iterator's in memory structure
6684  *
6685  * \retval   0 on success
6686  * \retval -ve on error
6687  * \retval +1 reach the end of entry
6688  */
6689 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6690                                const struct dt_it *di)
6691 {
6692         struct osd_it_ea *it = (struct osd_it_ea *)di;
6693         struct osd_object *obj = it->oie_obj;
6694         struct inode *inode = obj->oo_inode;
6695         struct htree_lock *hlock = NULL;
6696         struct file *filp = &it->oie_file;
6697         int rc = 0;
6698         struct osd_filldir_cbs buf = {
6699                 .ctx.actor = osd_ldiskfs_filldir,
6700                 .it = it
6701         };
6702
6703         ENTRY;
6704         it->oie_dirent = it->oie_buf;
6705         it->oie_rd_dirent = 0;
6706
6707         if (obj->oo_hl_head != NULL) {
6708                 hlock = osd_oti_get(env)->oti_hlock;
6709                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6710                                    inode, LDISKFS_HLOCK_READDIR);
6711         } else {
6712                 down_read(&obj->oo_ext_idx_sem);
6713         }
6714
6715         filp->f_cred = current_cred();
6716         rc = osd_security_file_alloc(filp);
6717         if (rc)
6718                 RETURN(rc);
6719
6720         filp->f_flags |= O_NOATIME;
6721         filp->f_mode |= FMODE_NONOTIFY;
6722         rc = iterate_dir(filp, &buf.ctx);
6723         if (rc)
6724                 RETURN(rc);
6725
6726         if (hlock != NULL)
6727                 ldiskfs_htree_unlock(hlock);
6728         else
6729                 up_read(&obj->oo_ext_idx_sem);
6730
6731         if (it->oie_rd_dirent == 0) {
6732                 /*
6733                  * If it does not get any dirent, it means it has been reached
6734                  * to the end of the dir
6735                  */
6736                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6737                 if (rc == 0)
6738                         rc = 1;
6739         } else {
6740                 it->oie_dirent = it->oie_buf;
6741                 it->oie_it_dirent = 1;
6742         }
6743
6744         RETURN(rc);
6745 }
6746
6747 /**
6748  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6749  * to load a directory entry at a time and stored it in
6750  * iterator's in-memory data structure.
6751  *
6752  * \param di iterator's in memory structure
6753  *
6754  * \retval +ve iterator reached to end
6755  * \retval   0 iterator not reached to end
6756  * \retval -ve on error
6757  */
6758 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6759 {
6760         struct osd_it_ea *it = (struct osd_it_ea *)di;
6761         int rc;
6762
6763         ENTRY;
6764
6765         if (it->oie_it_dirent < it->oie_rd_dirent) {
6766                 it->oie_dirent =
6767                         (void *)it->oie_dirent +
6768                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6769                                        it->oie_dirent->oied_namelen);
6770                 it->oie_it_dirent++;
6771                 rc = 0;
6772         } else {
6773                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6774                         rc = 1;
6775                 else
6776                         rc = osd_ldiskfs_it_fill(env, di);
6777         }
6778
6779         RETURN(rc);
6780 }
6781
6782 /**
6783  * Returns the key at current position from iterator's in memory structure.
6784  *
6785  * \param di iterator's in memory structure
6786  *
6787  * \retval key i.e. struct dt_key on success
6788  */
6789 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6790                                     const struct dt_it *di)
6791 {
6792         struct osd_it_ea *it = (struct osd_it_ea *)di;
6793
6794         return (struct dt_key *)it->oie_dirent->oied_name;
6795 }
6796
6797 /**
6798  * Returns key's size at current position from iterator's in memory structure.
6799  *
6800  * \param di iterator's in memory structure
6801  *
6802  * \retval key_size i.e. struct dt_key on success
6803  */
6804 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6805 {
6806         struct osd_it_ea *it = (struct osd_it_ea *)di;
6807
6808         return it->oie_dirent->oied_namelen;
6809 }
6810
6811 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6812 #undef LDISKFS_DIR_REC_LEN
6813 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6814 #endif
6815
6816 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6817 {
6818         if (LDISKFS_DIR_REC_LEN(de) >=
6819             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6820                 return true;
6821
6822         return false;
6823 }
6824
6825 static inline bool
6826 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6827                      unsigned int blocksize, bool dotdot)
6828 {
6829         if (dotdot)
6830                 return osd_dotdot_has_space(de);
6831
6832         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6833             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6834                 return true;
6835
6836         return false;
6837 }
6838
6839 static int
6840 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6841                     handle_t *jh, struct dentry *dentry,
6842                     const struct lu_fid *fid, struct buffer_head *bh,
6843                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6844                     bool dotdot)
6845 {
6846         struct inode *dir = dentry->d_parent->d_inode;
6847         struct inode *inode = dentry->d_inode;
6848         struct osd_fid_pack *rec;
6849         struct ldiskfs_dentry_param *ldp;
6850         int namelen = dentry->d_name.len;
6851         int rc;
6852         struct osd_thread_info *info = osd_oti_get(env);
6853
6854         ENTRY;
6855
6856         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6857                 RETURN(0);
6858
6859         /* There is enough space to hold the FID-in-dirent. */
6860         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6861                 rc = ldiskfs_journal_get_write_access(jh, bh);
6862                 if (rc != 0)
6863                         RETURN(rc);
6864
6865                 de->name[namelen] = 0;
6866                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6867                 rec->fp_len = sizeof(struct lu_fid) + 1;
6868                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6869                 de->file_type |= LDISKFS_DIRENT_LUFID;
6870                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6871
6872                 RETURN(rc);
6873         }
6874
6875         LASSERT(!dotdot);
6876
6877         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6878         if (rc != 0)
6879                 RETURN(rc);
6880
6881         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6882         osd_get_ldiskfs_dirent_param(ldp, fid);
6883         dentry->d_fsdata = (void *)ldp;
6884         dquot_initialize(dir);
6885         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6886         /*
6887          * It is too bad, we cannot reinsert the name entry back.
6888          * That means we lose it!
6889          */
6890         if (rc != 0)
6891                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6892                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6893                        osd_ino2name(inode),
6894                        dir->i_ino, dir->i_generation, namelen,
6895                        dentry->d_name.name, PFID(fid), rc);
6896
6897         RETURN(rc);
6898 }
6899
6900 static int
6901 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6902                         struct osd_it_ea *it, struct lu_fid *fid,
6903                         struct osd_inode_id *id, __u32 *attr)
6904 {
6905         struct osd_thread_info *info = osd_oti_get(env);
6906         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6907         struct osd_device *dev = osd_obj2dev(obj);
6908         struct super_block *sb = osd_sb(dev);
6909         const char *devname = osd_name(dev);
6910         struct osd_it_ea_dirent *ent = it->oie_dirent;
6911         struct inode *dir = obj->oo_inode;
6912         struct htree_lock *hlock = NULL;
6913         struct buffer_head *bh = NULL;
6914         handle_t *jh = NULL;
6915         struct ldiskfs_dir_entry_2 *de;
6916         struct dentry *dentry;
6917         struct inode *inode;
6918         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6919         int credits;
6920         int rc;
6921         bool dotdot = false;
6922         bool dirty = false;
6923
6924         ENTRY;
6925
6926         if (ent->oied_name[0] == '.') {
6927                 if (ent->oied_namelen == 1)
6928                         RETURN(0);
6929
6930                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6931                         dotdot = true;
6932         }
6933
6934         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6935         inode = osd_iget(info, dev, id);
6936         if (IS_ERR(inode)) {
6937                 rc = PTR_ERR(inode);
6938                 if (rc == -ENOENT || rc == -ESTALE) {
6939                         /*
6940                          * Maybe dangling name entry, or
6941                          * corrupted directory entry.
6942                          */
6943                         *attr |= LUDA_UNKNOWN;
6944                         rc = 0;
6945                 } else {
6946                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6947                                "check_repair, dir = %lu/%u, name = %.*s, "
6948                                "ino = %llu, rc = %d\n",
6949                                devname, dir->i_ino, dir->i_generation,
6950                                ent->oied_namelen, ent->oied_name,
6951                                ent->oied_ino, rc);
6952                 }
6953
6954                 RETURN(rc);
6955         }
6956
6957         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6958                                            ent->oied_namelen);
6959         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6960         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6961                 lma = NULL;
6962         else if (rc != 0)
6963                 GOTO(out, rc);
6964
6965         /*
6966          * We need to ensure that the name entry is still valid.
6967          * Because it may be removed or renamed by other already.
6968          *
6969          * The unlink or rename operation will start journal before PDO lock,
6970          * so to avoid deadlock, here we need to start journal handle before
6971          * related PDO lock also. But because we do not know whether there
6972          * will be something to be repaired before PDO lock, we just start
6973          * journal without conditions.
6974          *
6975          * We may need to remove the name entry firstly, then insert back.
6976          * One credit is for user quota file update.
6977          * One credit is for group quota file update.
6978          * Two credits are for dirty inode.
6979          */
6980         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6981                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6982
6983         if (dev->od_dirent_journal != 0) {
6984
6985 again:
6986                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6987                 if (IS_ERR(jh)) {
6988                         rc = PTR_ERR(jh);
6989                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6990                                "check_repair, dir = %lu/%u, credits = %d, "
6991                                "name = %.*s, ino = %llu: rc = %d\n",
6992                                devname, dir->i_ino, dir->i_generation, credits,
6993                                ent->oied_namelen, ent->oied_name,
6994                                ent->oied_ino, rc);
6995
6996                         GOTO(out_inode, rc);
6997                 }
6998
6999                 if (obj->oo_hl_head != NULL) {
7000                         hlock = osd_oti_get(env)->oti_hlock;
7001                         /*
7002                          * "0" means exclusive lock for the whole directory.
7003                          * We need to prevent others access such name entry
7004                          * during the delete + insert. Neither HLOCK_ADD nor
7005                          * HLOCK_DEL cannot guarantee the atomicity.
7006                          */
7007                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7008                 } else {
7009                         down_write(&obj->oo_ext_idx_sem);
7010                 }
7011         } else {
7012                 if (obj->oo_hl_head != NULL) {
7013                         hlock = osd_oti_get(env)->oti_hlock;
7014                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7015                                            LDISKFS_HLOCK_LOOKUP);
7016                 } else {
7017                         down_read(&obj->oo_ext_idx_sem);
7018                 }
7019         }
7020
7021         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7022         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7023                 *attr |= LUDA_IGNORE;
7024
7025                 GOTO(out, rc = 0);
7026         }
7027
7028         /*
7029          * For dotdot entry, if there is not enough space to hold the
7030          * FID-in-dirent, just keep them there. It only happens when the
7031          * device upgraded from 1.8 or restored from MDT file-level backup.
7032          * For the whole directory, only dotdot entry have no FID-in-dirent
7033          * and needs to get FID from LMA when readdir, it will not affect the
7034          * performance much.
7035          */
7036         if (dotdot && !osd_dotdot_has_space(de)) {
7037                 *attr |= LUDA_UNKNOWN;
7038
7039                 GOTO(out, rc = 0);
7040         }
7041
7042         if (lma != NULL) {
7043                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7044                         GOTO(out, rc = 0);
7045
7046                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7047                         struct lu_fid *tfid = &lma->lma_self_fid;
7048
7049                         if (likely(dotdot &&
7050                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7051                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7052                                 /*
7053                                  * It must be REMOTE_PARENT_DIR and as the
7054                                  * 'dotdot' entry of remote directory
7055                                  */
7056                                 *attr |= LUDA_IGNORE;
7057                         } else {
7058                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7059                                        "parent directory, but got %.*s under "
7060                                        "dir = %lu/%u with the FID "DFID"\n",
7061                                        devname, ent->oied_namelen,
7062                                        ent->oied_name, dir->i_ino,
7063                                        dir->i_generation, PFID(tfid));
7064
7065                                 *attr |= LUDA_UNKNOWN;
7066                         }
7067
7068                         GOTO(out, rc = 0);
7069                 }
7070         }
7071
7072         if (!fid_is_zero(fid)) {
7073                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7074                                               ent->oied_namelen);
7075                 if (rc == -ENOENT ||
7076                     (rc == -ENODATA &&
7077                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7078                         /*
7079                          * linkEA does not recognize the dirent entry,
7080                          * it may because the dirent entry corruption
7081                          * and points to other's inode.
7082                          */
7083                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7084                                "recognize the dirent, dir = %lu/%u, "
7085                                " name = %.*s, ino = %llu, "
7086                                DFID": rc = %d\n", devname, dir->i_ino,
7087                                dir->i_generation, ent->oied_namelen,
7088                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7089                         *attr |= LUDA_UNKNOWN;
7090
7091                         GOTO(out, rc = 0);
7092                 }
7093
7094                 if (rc && rc != -ENODATA) {
7095                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7096                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7097                                DFID": rc = %d\n", devname, dir->i_ino,
7098                                dir->i_generation, ent->oied_namelen,
7099                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7100                         *attr |= LUDA_UNKNOWN;
7101
7102                         GOTO(out, rc = 0);
7103                 }
7104         }
7105
7106         if (lma != NULL) {
7107                 /*
7108                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7109                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7110                  */
7111                 if (*attr & LUDA_VERIFY_DRYRUN) {
7112                         *fid = lma->lma_self_fid;
7113                         *attr |= LUDA_REPAIR;
7114
7115                         GOTO(out, rc = 0);
7116                 }
7117
7118                 if (jh == NULL) {
7119                         brelse(bh);
7120                         dev->od_dirent_journal = 1;
7121                         if (hlock != NULL) {
7122                                 ldiskfs_htree_unlock(hlock);
7123                                 hlock = NULL;
7124                         } else {
7125                                 up_read(&obj->oo_ext_idx_sem);
7126                         }
7127
7128                         goto again;
7129                 }
7130
7131                 *fid = lma->lma_self_fid;
7132                 dirty = true;
7133                 /* Update or append the FID-in-dirent. */
7134                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7135                                          bh, de, hlock, dotdot);
7136                 if (rc == 0)
7137                         *attr |= LUDA_REPAIR;
7138                 else
7139                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7140                                "the dirent, dir = %lu/%u, name = %.*s, "
7141                                "ino = %llu, "DFID": rc = %d\n",
7142                                devname, dir->i_ino, dir->i_generation,
7143                                ent->oied_namelen, ent->oied_name,
7144                                ent->oied_ino, PFID(fid), rc);
7145         } else {
7146                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7147                 if (*attr & LUDA_VERIFY_DRYRUN) {
7148                         if (fid_is_sane(fid)) {
7149                                 *attr |= LUDA_REPAIR;
7150                         } else if (dev->od_index == 0) {
7151                                 lu_igif_build(fid, inode->i_ino,
7152                                               inode->i_generation);
7153                                 *attr |= LUDA_UPGRADE;
7154                         }
7155
7156                         GOTO(out, rc = 0);
7157                 }
7158
7159                 if (jh == NULL) {
7160                         brelse(bh);
7161                         dev->od_dirent_journal = 1;
7162                         if (hlock != NULL) {
7163                                 ldiskfs_htree_unlock(hlock);
7164                                 hlock = NULL;
7165                         } else {
7166                                 up_read(&obj->oo_ext_idx_sem);
7167                         }
7168
7169                         goto again;
7170                 }
7171
7172                 dirty = true;
7173                 if (unlikely(fid_is_sane(fid))) {
7174                         /*
7175                          * FID-in-dirent exists, but FID-in-LMA is lost.
7176                          * Trust the FID-in-dirent, and add FID-in-LMA.
7177                          */
7178                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7179                         if (rc == 0)
7180                                 *attr |= LUDA_REPAIR;
7181                         else
7182                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7183                                        "update dirent, dir = %lu/%u, "
7184                                        "name = %.*s, ino = %llu, "
7185                                        DFID": rc = %d\n",
7186                                        devname, dir->i_ino, dir->i_generation,
7187                                        ent->oied_namelen, ent->oied_name,
7188                                        ent->oied_ino, PFID(fid), rc);
7189                 } else if (dev->od_index == 0) {
7190                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7191                         /*
7192                          * It is probably IGIF object. Only aappend the
7193                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7194                          */
7195                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7196                                                  bh, de, hlock, dotdot);
7197                         if (rc == 0)
7198                                 *attr |= LUDA_UPGRADE;
7199                         else
7200                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7201                                        "after the dirent, dir = %lu/%u, "
7202                                        "name = %.*s, ino = %llu, "
7203                                        DFID": rc = %d\n",
7204                                        devname, dir->i_ino, dir->i_generation,
7205                                        ent->oied_namelen, ent->oied_name,
7206                                        ent->oied_ino, PFID(fid), rc);
7207                 }
7208         }
7209
7210         GOTO(out, rc);
7211
7212 out:
7213         if (!IS_ERR(bh))
7214                 brelse(bh);
7215         if (hlock != NULL) {
7216                 ldiskfs_htree_unlock(hlock);
7217         } else {
7218                 if (dev->od_dirent_journal != 0)
7219                         up_write(&obj->oo_ext_idx_sem);
7220                 else
7221                         up_read(&obj->oo_ext_idx_sem);
7222         }
7223
7224         if (jh != NULL)
7225                 ldiskfs_journal_stop(jh);
7226
7227 out_inode:
7228         iput(inode);
7229         if (rc >= 0 && !dirty)
7230                 dev->od_dirent_journal = 0;
7231
7232         return rc;
7233 }
7234
7235 /**
7236  * Returns the value at current position from iterator's in memory structure.
7237  *
7238  * \param di struct osd_it_ea, iterator's in memory structure
7239  * \param attr attr requested for dirent.
7240  * \param lde lustre dirent
7241  *
7242  * \retval   0 no error and \param lde has correct lustre dirent.
7243  * \retval -ve on error
7244  */
7245 static inline int osd_it_ea_rec(const struct lu_env *env,
7246                                 const struct dt_it *di,
7247                                 struct dt_rec *dtrec, __u32 attr)
7248 {
7249         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7250         struct osd_object      *obj   = it->oie_obj;
7251         struct osd_device      *dev   = osd_obj2dev(obj);
7252         struct osd_thread_info *oti   = osd_oti_get(env);
7253         struct osd_inode_id    *id    = &oti->oti_id;
7254         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7255         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7256         __u32 ino = it->oie_dirent->oied_ino;
7257         int rc = 0;
7258
7259         ENTRY;
7260
7261         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7262
7263         if (attr & LUDA_VERIFY) {
7264                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7265                         attr |= LUDA_IGNORE;
7266                         /*
7267                          * If the parent is on remote MDT, and there
7268                          * is no FID-in-dirent, then we have to get
7269                          * the parent FID from the linkEA.
7270                          */
7271                         if (!fid_is_sane(fid) &&
7272                             it->oie_dirent->oied_namelen == 2 &&
7273                             it->oie_dirent->oied_name[0] == '.' &&
7274                             it->oie_dirent->oied_name[1] == '.')
7275                                 osd_get_pfid_from_linkea(env, obj, fid);
7276                 } else {
7277                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7278                                                      &attr);
7279                 }
7280
7281                 if (!fid_is_sane(fid))
7282                         attr |= LUDA_UNKNOWN;
7283         } else {
7284                 attr &= ~LU_DIRENT_ATTRS_MASK;
7285                 if (!fid_is_sane(fid)) {
7286                         bool is_dotdot = false;
7287
7288                         if (it->oie_dirent->oied_namelen == 2 &&
7289                             it->oie_dirent->oied_name[0] == '.' &&
7290                             it->oie_dirent->oied_name[1] == '.')
7291                                 is_dotdot = true;
7292                         /*
7293                          * If the parent is on remote MDT, and there
7294                          * is no FID-in-dirent, then we have to get
7295                          * the parent FID from the linkEA.
7296                          */
7297                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7298                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7299                         } else {
7300                                 if (is_dotdot == false &&
7301                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7302                                         RETURN(-ENOENT);
7303
7304                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7305                         }
7306                 } else {
7307                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7308                 }
7309         }
7310
7311         /* Pack the entry anyway, at least the offset is right. */
7312         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7313                            it->oie_dirent->oied_name,
7314                            it->oie_dirent->oied_namelen,
7315                            it->oie_dirent->oied_type, attr);
7316
7317         if (rc < 0)
7318                 RETURN(rc);
7319
7320         if (osd_remote_fid(env, dev, fid))
7321                 RETURN(0);
7322
7323         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7324                 osd_add_oi_cache(oti, dev, id, fid);
7325
7326         RETURN(rc > 0 ? 0 : rc);
7327 }
7328
7329 /**
7330  * Returns the record size size at current position.
7331  *
7332  * This function will return record(lu_dirent) size in bytes.
7333  *
7334  * \param[in] env       execution environment
7335  * \param[in] di        iterator's in memory structure
7336  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7337  *                      calculate the lu_dirent size.
7338  *
7339  * \retval      record size(in bytes & in memory) of the current lu_dirent
7340  *              entry.
7341  */
7342 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7343                               __u32 attr)
7344 {
7345         struct osd_it_ea *it = (struct osd_it_ea *)di;
7346
7347         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7348 }
7349
7350 /**
7351  * Returns a cookie for current position of the iterator head, so that
7352  * user can use this cookie to load/start the iterator next time.
7353  *
7354  * \param di iterator's in memory structure
7355  *
7356  * \retval cookie for current position, on success
7357  */
7358 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7359 {
7360         struct osd_it_ea *it = (struct osd_it_ea *)di;
7361
7362         return it->oie_dirent->oied_off;
7363 }
7364
7365 /**
7366  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7367  * to load a directory entry at a time and stored it i inn,
7368  * in iterator's in-memory data structure.
7369  *
7370  * \param di struct osd_it_ea, iterator's in memory structure
7371  *
7372  * \retval +ve on success
7373  * \retval -ve on error
7374  */
7375 static int osd_it_ea_load(const struct lu_env *env,
7376                           const struct dt_it *di, __u64 hash)
7377 {
7378         struct osd_it_ea *it = (struct osd_it_ea *)di;
7379         int rc;
7380
7381         ENTRY;
7382         it->oie_file.f_pos = hash;
7383
7384         rc =  osd_ldiskfs_it_fill(env, di);
7385         if (rc > 0)
7386                 rc = -ENODATA;
7387
7388         if (rc == 0)
7389                 rc = 1;
7390
7391         RETURN(rc);
7392 }
7393
7394 /**
7395  * Index lookup function for interoperability mode (b11826).
7396  *
7397  * \param key,  key i.e. file name to be searched
7398  *
7399  * \retval +ve, on success
7400  * \retval -ve, on error
7401  */
7402 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7403                                struct dt_rec *rec, const struct dt_key *key)
7404 {
7405         struct osd_object *obj = osd_dt_obj(dt);
7406         int rc = 0;
7407
7408         ENTRY;
7409
7410         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7411         LINVRNT(osd_invariant(obj));
7412
7413         rc = osd_ea_lookup_rec(env, obj, rec, key);
7414         if (rc == 0)
7415                 rc = 1;
7416         RETURN(rc);
7417 }
7418
7419 /**
7420  * Index and Iterator operations for interoperability
7421  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7422  */
7423 static const struct dt_index_operations osd_index_ea_ops = {
7424         .dio_lookup         = osd_index_ea_lookup,
7425         .dio_declare_insert = osd_index_declare_ea_insert,
7426         .dio_insert         = osd_index_ea_insert,
7427         .dio_declare_delete = osd_index_declare_ea_delete,
7428         .dio_delete         = osd_index_ea_delete,
7429         .dio_it     = {
7430                 .init     = osd_it_ea_init,
7431                 .fini     = osd_it_ea_fini,
7432                 .get      = osd_it_ea_get,
7433                 .put      = osd_it_ea_put,
7434                 .next     = osd_it_ea_next,
7435                 .key      = osd_it_ea_key,
7436                 .key_size = osd_it_ea_key_size,
7437                 .rec      = osd_it_ea_rec,
7438                 .rec_size = osd_it_ea_rec_size,
7439                 .store    = osd_it_ea_store,
7440                 .load     = osd_it_ea_load
7441         }
7442 };
7443
7444 static void *osd_key_init(const struct lu_context *ctx,
7445                           struct lu_context_key *key)
7446 {
7447         struct osd_thread_info *info;
7448
7449         OBD_ALLOC_PTR(info);
7450         if (info == NULL)
7451                 return ERR_PTR(-ENOMEM);
7452
7453         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7454         if (info->oti_it_ea_buf == NULL)
7455                 goto out_free_info;
7456
7457         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7458
7459         info->oti_hlock = ldiskfs_htree_lock_alloc();
7460         if (info->oti_hlock == NULL)
7461                 goto out_free_ea;
7462
7463         return info;
7464
7465 out_free_ea:
7466         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7467 out_free_info:
7468         OBD_FREE_PTR(info);
7469         return ERR_PTR(-ENOMEM);
7470 }
7471
7472 static void osd_key_fini(const struct lu_context *ctx,
7473                          struct lu_context_key *key, void *data)
7474 {
7475         struct osd_thread_info *info = data;
7476         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7477         struct osd_idmap_cache *idc = info->oti_ins_cache;
7478
7479         if (info->oti_dio_pages) {
7480                 int i;
7481                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7482                         struct page *page = info->oti_dio_pages[i];
7483                         if (page) {
7484                                 LASSERT(PagePrivate2(page));
7485                                 LASSERT(PageLocked(page));
7486                                 ClearPagePrivate2(page);
7487                                 unlock_page(page);
7488                                 __free_page(page);
7489                         }
7490                 }
7491                 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7492                                          PTLRPC_MAX_BRW_PAGES);
7493         }
7494
7495         if (info->oti_inode != NULL)
7496                 OBD_FREE_PTR(lli);
7497         if (info->oti_hlock != NULL)
7498                 ldiskfs_htree_lock_free(info->oti_hlock);
7499         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7500         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7501         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7502         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7503         lu_buf_free(&info->oti_big_buf);
7504         if (idc != NULL) {
7505                 LASSERT(info->oti_ins_cache_size > 0);
7506                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7507                 info->oti_ins_cache = NULL;
7508                 info->oti_ins_cache_size = 0;
7509         }
7510         OBD_FREE_PTR(info);
7511 }
7512
7513 static void osd_key_exit(const struct lu_context *ctx,
7514                          struct lu_context_key *key, void *data)
7515 {
7516         struct osd_thread_info *info = data;
7517
7518         LASSERT(info->oti_r_locks == 0);
7519         LASSERT(info->oti_w_locks == 0);
7520         LASSERT(info->oti_txns    == 0);
7521 }
7522
7523 /* type constructor/destructor: osd_type_init, osd_type_fini */
7524 LU_TYPE_INIT_FINI(osd, &osd_key);
7525
7526 struct lu_context_key osd_key = {
7527         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7528         .lct_init = osd_key_init,
7529         .lct_fini = osd_key_fini,
7530         .lct_exit = osd_key_exit
7531 };
7532
7533
7534 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7535                            const char *name, struct lu_device *next)
7536 {
7537         struct osd_device *osd = osd_dev(d);
7538
7539         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7540             sizeof(osd->od_svname))
7541                 return -E2BIG;
7542         return osd_procfs_init(osd, name);
7543 }
7544
7545 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7546 {
7547         struct seq_server_site *ss = osd_seq_site(osd);
7548         int rc = 0;
7549
7550         ENTRY;
7551
7552         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7553                 RETURN(0);
7554
7555         if (unlikely(ss == NULL))
7556                 RETURN(-ENODEV);
7557
7558         OBD_ALLOC_PTR(osd->od_cl_seq);
7559         if (osd->od_cl_seq == NULL)
7560                 RETURN(-ENOMEM);
7561
7562         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7563                         osd->od_svname, ss->ss_server_seq);
7564
7565         if (ss->ss_node_id == 0) {
7566                 /*
7567                  * If the OSD on the sequence controller(MDT0), then allocate
7568                  * sequence here, otherwise allocate sequence after connected
7569                  * to MDT0 (see mdt_register_lwp_callback()).
7570                  */
7571                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7572                                    &osd->od_cl_seq->lcs_space, env);
7573         }
7574
7575         RETURN(rc);
7576 }
7577
7578 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7579 {
7580         if (osd->od_cl_seq == NULL)
7581                 return;
7582
7583         seq_client_fini(osd->od_cl_seq);
7584         OBD_FREE_PTR(osd->od_cl_seq);
7585         osd->od_cl_seq = NULL;
7586 }
7587
7588 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7589 {
7590         ENTRY;
7591
7592         /* shutdown quota slave instance associated with the device */
7593         if (o->od_quota_slave_md != NULL) {
7594                 struct qsd_instance *qsd = o->od_quota_slave_md;
7595
7596                 o->od_quota_slave_md = NULL;
7597                 qsd_fini(env, qsd);
7598         }
7599
7600         if (o->od_quota_slave_dt != NULL) {
7601                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7602
7603                 o->od_quota_slave_dt = NULL;
7604                 qsd_fini(env, qsd);
7605         }
7606
7607         osd_fid_fini(env, o);
7608         osd_scrub_cleanup(env, o);
7609
7610         RETURN(0);
7611 }
7612
7613 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7614 {
7615         ENTRY;
7616
7617         if (o->od_mnt != NULL) {
7618                 shrink_dcache_sb(osd_sb(o));
7619                 osd_sync(env, &o->od_dt_dev);
7620
7621                 mntput(o->od_mnt);
7622                 o->od_mnt = NULL;
7623         }
7624
7625         EXIT;
7626 }
7627
7628 static int osd_mount(const struct lu_env *env,
7629                      struct osd_device *o, struct lustre_cfg *cfg)
7630 {
7631         const char *name = lustre_cfg_string(cfg, 0);
7632         const char *dev = lustre_cfg_string(cfg, 1);
7633         const char *opts;
7634         unsigned long page, s_flags, lmd_flags = 0;
7635         struct page *__page;
7636         struct file_system_type *type;
7637         char *options = NULL;
7638         char *str;
7639         struct osd_thread_info *info = osd_oti_get(env);
7640         struct lu_fid *fid = &info->oti_fid;
7641         struct inode *inode;
7642         int rc = 0, force_over_1024tb = 0;
7643
7644         ENTRY;
7645
7646         if (o->od_mnt != NULL)
7647                 RETURN(0);
7648
7649         if (strlen(dev) >= sizeof(o->od_mntdev))
7650                 RETURN(-E2BIG);
7651         strcpy(o->od_mntdev, dev);
7652
7653         str = lustre_cfg_string(cfg, 2);
7654         s_flags = simple_strtoul(str, NULL, 0);
7655         str = strstr(str, ":");
7656         if (str)
7657                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7658         opts = lustre_cfg_string(cfg, 3);
7659 #ifdef __BIG_ENDIAN
7660         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7661                 CERROR("%s: device %s extents feature is not guaranteed to "
7662                        "work on big-endian systems. Use \"bigendian_extents\" "
7663                        "mount option to override.\n", name, dev);
7664                 RETURN(-EINVAL);
7665         }
7666 #endif
7667 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7668         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7669                 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");
7670         }
7671 #endif
7672 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7673         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7674                 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");
7675         }
7676 #endif
7677 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7678         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7679                 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");
7680         }
7681 #endif
7682
7683         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7684                 force_over_1024tb = 1;
7685
7686         __page = alloc_page(GFP_KERNEL);
7687         if (__page == NULL)
7688                 GOTO(out, rc = -ENOMEM);
7689         page = (unsigned long)page_address(__page);
7690         options = (char *)page;
7691         *options = '\0';
7692         if (opts != NULL) {
7693                 /* strip out the options for back compatiblity */
7694                 static const char * const sout[] = {
7695                         "mballoc",
7696                         "iopen",
7697                         "noiopen",
7698                         "iopen_nopriv",
7699                         "extents",
7700                         "noextents",
7701                         /* strip out option we processed in osd */
7702                         "bigendian_extents",
7703                         "force_over_128tb",
7704                         "force_over_256tb",
7705                         "force_over_512tb",
7706                         "force_over_1024tb",
7707                         NULL
7708                 };
7709                 strncat(options, opts, PAGE_SIZE);
7710                 for (rc = 0, str = options; sout[rc]; ) {
7711                         char *op = strstr(str, sout[rc]);
7712
7713                         if (op == NULL) {
7714                                 rc++;
7715                                 str = options;
7716                                 continue;
7717                         }
7718                         if (op == options || *(op - 1) == ',') {
7719                                 str = op + strlen(sout[rc]);
7720                                 if (*str == ',' || *str == '\0') {
7721                                         *str == ',' ? str++ : str;
7722                                         memmove(op, str, strlen(str) + 1);
7723                                 }
7724                         }
7725                         for (str = op; *str != ',' && *str != '\0'; str++)
7726                                 ;
7727                 }
7728         } else {
7729                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7730         }
7731
7732         /* Glom up mount options */
7733         if (*options != '\0')
7734                 strncat(options, ",", PAGE_SIZE);
7735         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7736
7737         type = get_fs_type("ldiskfs");
7738         if (!type) {
7739                 CERROR("%s: cannot find ldiskfs module\n", name);
7740                 GOTO(out, rc = -ENODEV);
7741         }
7742
7743         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7744         module_put(type->owner);
7745
7746         if (IS_ERR(o->od_mnt)) {
7747                 rc = PTR_ERR(o->od_mnt);
7748                 o->od_mnt = NULL;
7749                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7750                 GOTO(out, rc);
7751         }
7752
7753         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7754                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7755                                  force_over_1024tb == 0) {
7756                 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",
7757                        name, dev);
7758                 GOTO(out_mnt, rc = -EINVAL);
7759         }
7760
7761         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7762                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7763                               name);
7764
7765                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7766         }
7767
7768         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7769                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7770                 GOTO(out_mnt, rc = -EINVAL);
7771         }
7772
7773 #ifdef LDISKFS_MOUNT_DIRDATA
7774         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7775                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7776         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7777                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7778                       "enabling the dirdata feature. If you do not want to "
7779                       "downgrade to Lustre-1.x again, you can enable it via "
7780                       "'tune2fs -O dirdata device'\n", name, dev);
7781 #endif
7782         inode = osd_sb(o)->s_root->d_inode;
7783         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7784         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7785         if (rc != 0) {
7786                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7787                 GOTO(out_mnt, rc);
7788         }
7789
7790         if (lmd_flags & LMD_FLG_NOSCRUB)
7791                 o->od_auto_scrub_interval = AS_NEVER;
7792
7793         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7794                 /* do not use pagecache with flash-backed storage */
7795                 o->od_writethrough_cache = 0;
7796                 o->od_read_cache = 0;
7797         }
7798
7799         GOTO(out, rc = 0);
7800
7801 out_mnt:
7802         mntput(o->od_mnt);
7803         o->od_mnt = NULL;
7804
7805 out:
7806         if (__page)
7807                 __free_page(__page);
7808
7809         return rc;
7810 }
7811
7812 static struct lu_device *osd_device_fini(const struct lu_env *env,
7813                                          struct lu_device *d)
7814 {
7815         struct osd_device *o = osd_dev(d);
7816
7817         ENTRY;
7818
7819         osd_index_backup(env, o, false);
7820         osd_shutdown(env, o);
7821         osd_procfs_fini(o);
7822         if (o->od_oi_table != NULL)
7823                 osd_oi_fini(osd_oti_get(env), o);
7824         osd_obj_map_fini(o);
7825         osd_umount(env, o);
7826
7827         RETURN(NULL);
7828 }
7829
7830 static int osd_device_init0(const struct lu_env *env,
7831                             struct osd_device *o,
7832                             struct lustre_cfg *cfg)
7833 {
7834         struct lu_device *l = osd2lu_dev(o);
7835         struct osd_thread_info *info;
7836         int rc;
7837         int cplen = 0;
7838
7839         /* if the module was re-loaded, env can loose its keys */
7840         rc = lu_env_refill((struct lu_env *)env);
7841         if (rc)
7842                 GOTO(out, rc);
7843         info = osd_oti_get(env);
7844         LASSERT(info);
7845
7846         l->ld_ops = &osd_lu_ops;
7847         o->od_dt_dev.dd_ops = &osd_dt_ops;
7848
7849         spin_lock_init(&o->od_osfs_lock);
7850         mutex_init(&o->od_otable_mutex);
7851         INIT_LIST_HEAD(&o->od_orphan_list);
7852         INIT_LIST_HEAD(&o->od_index_backup_list);
7853         INIT_LIST_HEAD(&o->od_index_restore_list);
7854         spin_lock_init(&o->od_lock);
7855         o->od_index_backup_policy = LIBP_NONE;
7856         o->od_t10_type = 0;
7857
7858         o->od_read_cache = 1;
7859         o->od_writethrough_cache = 1;
7860         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7861         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7862         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7863         o->od_auto_scrub_interval = AS_DEFAULT;
7864
7865         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7866                         sizeof(o->od_svname));
7867         if (cplen >= sizeof(o->od_svname)) {
7868                 rc = -E2BIG;
7869                 GOTO(out, rc);
7870         }
7871
7872         o->od_index_backup_stop = 0;
7873         o->od_index = -1; /* -1 means index is invalid */
7874         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7875         if (rc == LDD_F_SV_TYPE_OST)
7876                 o->od_is_ost = 1;
7877
7878         o->od_full_scrub_ratio = OFSR_DEFAULT;
7879         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7880         rc = osd_mount(env, o, cfg);
7881         if (rc != 0)
7882                 GOTO(out, rc);
7883
7884         /* Can only check block device after mount */
7885         o->od_nonrotational =
7886                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7887
7888         rc = osd_obj_map_init(env, o);
7889         if (rc != 0)
7890                 GOTO(out_mnt, rc);
7891
7892         rc = lu_site_init(&o->od_site, l);
7893         if (rc != 0)
7894                 GOTO(out_compat, rc);
7895         o->od_site.ls_bottom_dev = l;
7896
7897         rc = lu_site_init_finish(&o->od_site);
7898         if (rc != 0)
7899                 GOTO(out_site, rc);
7900
7901         INIT_LIST_HEAD(&o->od_ios_list);
7902         /* setup scrub, including OI files initialization */
7903         o->od_in_init = 1;
7904         rc = osd_scrub_setup(env, o);
7905         o->od_in_init = 0;
7906         if (rc < 0)
7907                 GOTO(out_site, rc);
7908
7909         rc = osd_procfs_init(o, o->od_svname);
7910         if (rc != 0) {
7911                 CERROR("%s: can't initialize procfs: rc = %d\n",
7912                        o->od_svname, rc);
7913                 GOTO(out_scrub, rc);
7914         }
7915
7916         LASSERT(l->ld_site->ls_linkage.next != NULL);
7917         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7918
7919         /* initialize quota slave instance */
7920         /* currently it's no need to prepare qsd_instance_md for OST */
7921         if (!o->od_is_ost) {
7922                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7923                                                 &o->od_dt_dev,
7924                                                 o->od_proc_entry, true);
7925                 if (IS_ERR(o->od_quota_slave_md)) {
7926                         rc = PTR_ERR(o->od_quota_slave_md);
7927                         o->od_quota_slave_md = NULL;
7928                         GOTO(out_procfs, rc);
7929                 }
7930         }
7931
7932         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7933                                         o->od_proc_entry, false);
7934
7935         if (IS_ERR(o->od_quota_slave_dt)) {
7936                 if (o->od_quota_slave_md != NULL) {
7937                         qsd_fini(env, o->od_quota_slave_md);
7938                         o->od_quota_slave_md = NULL;
7939                 }
7940
7941                 rc = PTR_ERR(o->od_quota_slave_dt);
7942                 o->od_quota_slave_dt = NULL;
7943                 GOTO(out_procfs, rc);
7944         }
7945
7946         RETURN(0);
7947
7948 out_procfs:
7949         osd_procfs_fini(o);
7950 out_scrub:
7951         osd_scrub_cleanup(env, o);
7952 out_site:
7953         lu_site_fini(&o->od_site);
7954 out_compat:
7955         osd_obj_map_fini(o);
7956 out_mnt:
7957         osd_umount(env, o);
7958 out:
7959         return rc;
7960 }
7961
7962 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7963                                           struct lu_device_type *t,
7964                                           struct lustre_cfg *cfg)
7965 {
7966         struct osd_device *o;
7967         int rc;
7968
7969         OBD_ALLOC_PTR(o);
7970         if (o == NULL)
7971                 return ERR_PTR(-ENOMEM);
7972
7973         rc = dt_device_init(&o->od_dt_dev, t);
7974         if (rc == 0) {
7975                 /*
7976                  * Because the ctx might be revived in dt_device_init,
7977                  * refill the env here
7978                  */
7979                 lu_env_refill((struct lu_env *)env);
7980                 rc = osd_device_init0(env, o, cfg);
7981                 if (rc)
7982                         dt_device_fini(&o->od_dt_dev);
7983         }
7984
7985         if (unlikely(rc != 0))
7986                 OBD_FREE_PTR(o);
7987
7988         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7989 }
7990
7991 static struct lu_device *osd_device_free(const struct lu_env *env,
7992                                          struct lu_device *d)
7993 {
7994         struct osd_device *o = osd_dev(d);
7995
7996         ENTRY;
7997
7998         /* XXX: make osd top device in order to release reference */
7999         d->ld_site->ls_top_dev = d;
8000         lu_site_purge(env, d->ld_site, -1);
8001         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8002                       D_ERROR, lu_cdebug_printer);
8003         lu_site_fini(&o->od_site);
8004         dt_device_fini(&o->od_dt_dev);
8005         OBD_FREE_PTR(o);
8006         RETURN(NULL);
8007 }
8008
8009 static int osd_process_config(const struct lu_env *env,
8010                               struct lu_device *d, struct lustre_cfg *cfg)
8011 {
8012         struct osd_device *o = osd_dev(d);
8013         ssize_t count;
8014         int rc;
8015
8016         ENTRY;
8017
8018         switch (cfg->lcfg_command) {
8019         case LCFG_SETUP:
8020                 rc = osd_mount(env, o, cfg);
8021                 break;
8022         case LCFG_CLEANUP:
8023                 /*
8024                  * For the case LCFG_PRE_CLEANUP is not called in advance,
8025                  * that may happend if hit failure during mount process.
8026                  */
8027                 osd_index_backup(env, o, false);
8028                 lu_dev_del_linkage(d->ld_site, d);
8029                 rc = osd_shutdown(env, o);
8030                 break;
8031         case LCFG_PARAM:
8032                 LASSERT(&o->od_dt_dev);
8033                 count  = class_modify_config(cfg, PARAM_OSD,
8034                                              &o->od_dt_dev.dd_kobj);
8035                 if (count < 0)
8036                         count = class_modify_config(cfg, PARAM_OST,
8037                                                     &o->od_dt_dev.dd_kobj);
8038                 rc = count > 0 ? 0 : count;
8039                 break;
8040         case LCFG_PRE_CLEANUP:
8041                 osd_scrub_stop(o);
8042                 osd_index_backup(env, o,
8043                                  o->od_index_backup_policy != LIBP_NONE);
8044                 rc = 0;
8045                 break;
8046         default:
8047                 rc = -ENOSYS;
8048         }
8049
8050         RETURN(rc);
8051 }
8052
8053 static int osd_recovery_complete(const struct lu_env *env,
8054                                  struct lu_device *d)
8055 {
8056         struct osd_device *osd = osd_dev(d);
8057         int rc = 0;
8058
8059         ENTRY;
8060
8061         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8062                 RETURN(0);
8063
8064         /*
8065          * start qsd instance on recovery completion, this notifies the quota
8066          * slave code that we are about to process new requests now
8067          */
8068         rc = qsd_start(env, osd->od_quota_slave_dt);
8069         if (rc == 0 && osd->od_quota_slave_md != NULL)
8070                 rc = qsd_start(env, osd->od_quota_slave_md);
8071
8072         RETURN(rc);
8073 }
8074
8075 /*
8076  * we use exports to track all osd users
8077  */
8078 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8079                            struct obd_device *obd, struct obd_uuid *cluuid,
8080                            struct obd_connect_data *data, void *localdata)
8081 {
8082         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8083         struct lustre_handle conn;
8084         int rc;
8085
8086         ENTRY;
8087
8088         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8089
8090         rc = class_connect(&conn, obd, cluuid);
8091         if (rc)
8092                 RETURN(rc);
8093
8094         *exp = class_conn2export(&conn);
8095
8096         spin_lock(&osd->od_osfs_lock);
8097         osd->od_connects++;
8098         spin_unlock(&osd->od_osfs_lock);
8099
8100         RETURN(0);
8101 }
8102
8103 /*
8104  * once last export (we don't count self-export) disappeared
8105  * osd can be released
8106  */
8107 static int osd_obd_disconnect(struct obd_export *exp)
8108 {
8109         struct obd_device *obd = exp->exp_obd;
8110         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8111         int rc, release = 0;
8112
8113         ENTRY;
8114
8115         /* Only disconnect the underlying layers on the final disconnect. */
8116         spin_lock(&osd->od_osfs_lock);
8117         osd->od_connects--;
8118         if (osd->od_connects == 0)
8119                 release = 1;
8120         spin_unlock(&osd->od_osfs_lock);
8121
8122         rc = class_disconnect(exp); /* bz 9811 */
8123
8124         if (rc == 0 && release)
8125                 class_manual_cleanup(obd);
8126         RETURN(rc);
8127 }
8128
8129 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8130                        struct lu_device *dev)
8131 {
8132         struct osd_device *osd = osd_dev(dev);
8133         struct lr_server_data *lsd =
8134                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8135         int result = 0;
8136
8137         ENTRY;
8138
8139         if (osd->od_quota_slave_md != NULL) {
8140                 /* set up quota slave objects for inode */
8141                 result = qsd_prepare(env, osd->od_quota_slave_md);
8142                 if (result != 0)
8143                         RETURN(result);
8144         }
8145
8146         if (osd->od_quota_slave_dt != NULL) {
8147                 /* set up quota slave objects for block */
8148                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8149                 if (result != 0)
8150                         RETURN(result);
8151         }
8152
8153
8154         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8155 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8156                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8157                         osd->od_index_in_idif = 1;
8158                 } else {
8159                         osd->od_index_in_idif = 0;
8160                         result = osd_register_proc_index_in_idif(osd);
8161                         if (result != 0)
8162                                 RETURN(result);
8163                 }
8164 #else
8165                 osd->od_index_in_idif = 1;
8166 #endif
8167         }
8168
8169         result = osd_fid_init(env, osd);
8170
8171         RETURN(result);
8172 }
8173
8174 /**
8175  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8176  *
8177  * Allocate FID.
8178  *
8179  * see include/lu_object.h for the details.
8180  */
8181 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8182                          struct lu_fid *fid, struct lu_object *parent,
8183                          const struct lu_name *name)
8184 {
8185         struct osd_device *osd = osd_dev(d);
8186
8187         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8188 }
8189
8190 static const struct lu_object_operations osd_lu_obj_ops = {
8191         .loo_object_init      = osd_object_init,
8192         .loo_object_delete    = osd_object_delete,
8193         .loo_object_release   = osd_object_release,
8194         .loo_object_free      = osd_object_free,
8195         .loo_object_print     = osd_object_print,
8196         .loo_object_invariant = osd_object_invariant
8197 };
8198
8199 const struct lu_device_operations osd_lu_ops = {
8200         .ldo_object_alloc      = osd_object_alloc,
8201         .ldo_process_config    = osd_process_config,
8202         .ldo_recovery_complete = osd_recovery_complete,
8203         .ldo_prepare           = osd_prepare,
8204         .ldo_fid_alloc         = osd_fid_alloc,
8205 };
8206
8207 static const struct lu_device_type_operations osd_device_type_ops = {
8208         .ldto_init = osd_type_init,
8209         .ldto_fini = osd_type_fini,
8210
8211         .ldto_start = osd_type_start,
8212         .ldto_stop  = osd_type_stop,
8213
8214         .ldto_device_alloc = osd_device_alloc,
8215         .ldto_device_free  = osd_device_free,
8216
8217         .ldto_device_init = osd_device_init,
8218         .ldto_device_fini = osd_device_fini
8219 };
8220
8221 static struct lu_device_type osd_device_type = {
8222         .ldt_tags     = LU_DEVICE_DT,
8223         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8224         .ldt_ops      = &osd_device_type_ops,
8225         .ldt_ctx_tags = LCT_LOCAL,
8226 };
8227
8228 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8229 {
8230         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8231         struct super_block *sb = osd_sb(osd);
8232
8233         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8234 }
8235
8236 /*
8237  * lprocfs legacy support.
8238  */
8239 static const struct obd_ops osd_obd_device_ops = {
8240         .o_owner = THIS_MODULE,
8241         .o_connect      = osd_obd_connect,
8242         .o_disconnect   = osd_obd_disconnect,
8243         .o_health_check = osd_health_check,
8244 };
8245
8246 static ssize_t track_declares_assert_show(struct kobject *kobj,
8247                                    struct attribute *attr,
8248                                    char *buf)
8249 {
8250         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8251 }
8252
8253 static ssize_t track_declares_assert_store(struct kobject *kobj,
8254                                            struct attribute *attr,
8255                                            const char *buffer, size_t count)
8256 {
8257         bool track_declares_assert;
8258         int rc;
8259
8260         rc = kstrtobool(buffer, &track_declares_assert);
8261         if (rc)
8262                 return rc;
8263
8264         ldiskfs_track_declares_assert = track_declares_assert;
8265
8266         return count;
8267 }
8268 LUSTRE_RW_ATTR(track_declares_assert);
8269
8270 static int __init osd_init(void)
8271 {
8272         struct kobject *kobj;
8273         int rc;
8274
8275         BUILD_BUG_ON(BH_DXLock >=
8276                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8277 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8278         /* please, try to keep osd_thread_info smaller than a page */
8279         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8280 #endif
8281
8282         osd_oi_mod_init();
8283
8284         rc = lu_kmem_init(ldiskfs_caches);
8285         if (rc)
8286                 return rc;
8287
8288 #ifdef CONFIG_KALLSYMS
8289         priv_security_file_alloc =
8290                 (void *)kallsyms_lookup_name("security_file_alloc");
8291 #endif
8292
8293         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8294                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8295         if (rc) {
8296                 lu_kmem_fini(ldiskfs_caches);
8297                 return rc;
8298         }
8299
8300         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8301         if (kobj) {
8302                 rc = sysfs_create_file(kobj,
8303                                        &lustre_attr_track_declares_assert.attr);
8304                 kobject_put(kobj);
8305                 if (rc) {
8306                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8307                         rc = 0;
8308                 }
8309         }
8310         return rc;
8311 }
8312
8313 static void __exit osd_exit(void)
8314 {
8315         struct kobject *kobj;
8316
8317         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8318         if (kobj) {
8319                 sysfs_remove_file(kobj,
8320                                   &lustre_attr_track_declares_assert.attr);
8321                 kobject_put(kobj);
8322         }
8323         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8324         lu_kmem_fini(ldiskfs_caches);
8325 }
8326
8327 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8328 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8329 MODULE_VERSION(LUSTRE_VERSION_STRING);
8330 MODULE_LICENSE("GPL");
8331
8332 module_init(osd_init);
8333 module_exit(osd_exit);