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