lustre/osd-ldiskfs/osd_handler.c

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