lustre/osd-ldiskfs/osd_handler.c

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