lustre/osd-ldiskfs/osd_handler.c

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