lustre/osd-ldiskfs/osd_handler.c

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