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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56 /* simple_mkdir() */
  57 #include <lvfs.h>
  58
  59 /*
  60  * struct OBD_{ALLOC,FREE}*()
  61  * OBD_FAIL_CHECK
  62  */
  63 #include <obd_support.h>
  64 /* struct ptlrpc_thread */
  65 #include <lustre_net.h>
  66
  67 /* fid_is_local() */
  68 #include <lustre_fid.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_igif.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75 #include <lustre_quota.h>
  76
  77 int ldiskfs_pdo = 1;
  78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  79                 "ldiskfs with parallel directory operations");
  80
  81 static const char dot[] = ".";
  82 static const char dotdot[] = "..";
  83 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  84
  85 static const struct lu_object_operations      osd_lu_obj_ops;
  86 static const struct dt_object_operations      osd_obj_ops;
  87 static const struct dt_object_operations      osd_obj_ea_ops;
  88 static const struct dt_object_operations      osd_obj_otable_it_ops;
  89 static const struct dt_index_operations       osd_index_iam_ops;
  90 static const struct dt_index_operations       osd_index_ea_ops;
  91
  92 static int osd_has_index(const struct osd_object *obj)
  93 {
  94         return obj->oo_dt.do_index_ops != NULL;
  95 }
  96
  97 static int osd_object_invariant(const struct lu_object *l)
  98 {
  99         return osd_invariant(osd_obj(l));
 100 }
 101
 102 /*
 103  * Concurrency: doesn't matter
 104  */
 105 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 106 {
 107         return osd_oti_get(env)->oti_r_locks > 0;
 108 }
 109
 110 /*
 111  * Concurrency: doesn't matter
 112  */
 113 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 114 {
 115         struct osd_thread_info *oti = osd_oti_get(env);
 116         return oti->oti_w_locks > 0 && o->oo_owner == env;
 117 }
 118
 119 /*
 120  * Concurrency: doesn't access mutable data
 121  */
 122 static int osd_root_get(const struct lu_env *env,
 123                         struct dt_device *dev, struct lu_fid *f)
 124 {
 125         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 126         return 0;
 127 }
 128
 129 /*
 130  * OSD object methods.
 131  */
 132
 133 /*
 134  * Concurrency: no concurrent access is possible that early in object
 135  * life-cycle.
 136  */
 137 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 138                                           const struct lu_object_header *hdr,
 139                                           struct lu_device *d)
 140 {
 141         struct osd_object *mo;
 142
 143         OBD_ALLOC_PTR(mo);
 144         if (mo != NULL) {
 145                 struct lu_object *l;
 146
 147                 l = &mo->oo_dt.do_lu;
 148                 dt_object_init(&mo->oo_dt, NULL, d);
 149                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 150                 l->lo_ops = &osd_lu_obj_ops;
 151                 init_rwsem(&mo->oo_sem);
 152                 init_rwsem(&mo->oo_ext_idx_sem);
 153                 spin_lock_init(&mo->oo_guard);
 154                 return l;
 155         } else {
 156                 return NULL;
 157         }
 158 }
 159
 160 static int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 161                        struct dentry *dentry, struct lustre_mdt_attrs *lma)
 162 {
 163         int rc;
 164
 165         dentry->d_inode = inode;
 166         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
 167                                    sizeof(*lma));
 168         if (rc == -ERANGE) {
 169                 /* try with old lma size */
 170                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
 171                                            info->oti_mdt_attrs_old,
 172                                            LMA_OLD_SIZE);
 173                 if (rc > 0)
 174                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 175         }
 176         if (rc > 0) {
 177                 /* Check LMA compatibility */
 178                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 179                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 180                               "%lx/%#x\n",
 181                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 182                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 183                                                         ~LMA_INCOMPAT_SUPP);
 184                         rc = -ENOSYS;
 185                 } else {
 186                         lustre_lma_swab(lma);
 187                         rc = 0;
 188                 }
 189         } else if (rc == 0) {
 190                 rc = -ENODATA;
 191         }
 192
 193         return rc;
 194 }
 195
 196 /*
 197  * retrieve object from backend ext fs.
 198  **/
 199 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 200                        struct osd_inode_id *id)
 201 {
 202         struct inode *inode = NULL;
 203
 204         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 205         if (IS_ERR(inode)) {
 206                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 207                        id->oii_ino, PTR_ERR(inode));
 208         } else if (id->oii_gen != OSD_OII_NOGEN &&
 209                    inode->i_generation != id->oii_gen) {
 210                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 211                        "gen1 = %u\n",
 212                        id->oii_ino, id->oii_gen, inode->i_generation);
 213                 iput(inode);
 214                 inode = ERR_PTR(-ESTALE);
 215         } else if (inode->i_nlink == 0) {
 216                 /* due to parallel readdir and unlink,
 217                 * we can have dead inode here. */
 218                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 219                 make_bad_inode(inode);
 220                 iput(inode);
 221                 inode = ERR_PTR(-ESTALE);
 222         } else if (is_bad_inode(inode)) {
 223                 CWARN("%.16s: bad inode: ino = %u\n",
 224                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 225                 iput(inode);
 226                 inode = ERR_PTR(-ENOENT);
 227         } else {
 228                 if (id->oii_gen == OSD_OII_NOGEN)
 229                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 230
 231                 /* Do not update file c/mtime in ldiskfs.
 232                  * NB: we don't have any lock to protect this because we don't
 233                  * have reference on osd_object now, but contention with
 234                  * another lookup + attr_set can't happen in the tiny window
 235                  * between if (...) and set S_NOCMTIME. */
 236                 if (!(inode->i_flags & S_NOCMTIME))
 237                         inode->i_flags |= S_NOCMTIME;
 238         }
 239         return inode;
 240 }
 241
 242 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 243                            struct osd_inode_id *id, struct lu_fid *fid)
 244 {
 245         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 246         struct inode            *inode;
 247         int                      rc;
 248
 249         inode = osd_iget(info, dev, id);
 250         if (IS_ERR(inode))
 251                 return inode;
 252
 253         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 254         if (rc == 0) {
 255                 *fid = lma->lma_self_fid;
 256         } else if (rc == -ENODATA) {
 257                 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
 258         } else {
 259                 iput(inode);
 260                 inode = ERR_PTR(rc);
 261         }
 262         return inode;
 263 }
 264
 265 static struct inode *
 266 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 267                 struct osd_inode_id *id, const struct lu_fid *fid)
 268 {
 269         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 270         struct inode            *inode;
 271         int                      rc;
 272
 273         inode = osd_iget(info, dev, id);
 274         if (IS_ERR(inode))
 275                 return inode;
 276
 277         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 278         if (rc == -ENODATA)
 279                 return inode;
 280
 281         if (rc != 0) {
 282                 iput(inode);
 283                 return ERR_PTR(rc);
 284         }
 285
 286         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 287                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 288                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 289                 iput(inode);
 290                 return ERR_PTR(-EREMCHG);
 291         }
 292
 293         return inode;
 294 }
 295
 296 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 297                           const struct lu_fid *fid,
 298                           const struct lu_object_conf *conf)
 299 {
 300         struct osd_thread_info *info;
 301         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 302         struct osd_device      *dev;
 303         struct osd_idmap_cache *oic;
 304         struct osd_inode_id    *id;
 305         struct inode           *inode;
 306         struct osd_scrub       *scrub;
 307         struct scrub_file      *sf;
 308         int                     result;
 309         int                     verify = 0;
 310         ENTRY;
 311
 312         LINVRNT(osd_invariant(obj));
 313         LASSERT(obj->oo_inode == NULL);
 314         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 315
 316         dev = osd_dev(ldev);
 317         scrub = &dev->od_scrub;
 318         sf = &scrub->os_file;
 319         info = osd_oti_get(env);
 320         LASSERT(info);
 321         oic = &info->oti_cache;
 322
 323         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 324                 RETURN(-ENOENT);
 325
 326         /* Search order: 1. per-thread cache. */
 327         if (lu_fid_eq(fid, &oic->oic_fid)) {
 328                 id = &oic->oic_lid;
 329                 goto iget;
 330         }
 331
 332         id = &info->oti_id;
 333         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 334                 /* Search order: 2. OI scrub pending list. */
 335                 result = osd_oii_lookup(dev, fid, id);
 336                 if (result == 0)
 337                         goto iget;
 338         }
 339
 340         if (sf->sf_flags & SF_INCONSISTENT)
 341                 verify = 1;
 342
 343         /*
 344          * Objects are created as locking anchors or place holders for objects
 345          * yet to be created. No need to osd_oi_lookup() at here because FID
 346          * shouldn't never be re-used, if it's really a duplicate FID from
 347          * unexpected reason, we should be able to detect it later by calling
 348          * do_create->osd_oi_insert()
 349          */
 350         if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
 351                 GOTO(out, result = 0);
 352
 353         /* Search order: 3. OI files. */
 354         result = osd_oi_lookup(info, dev, fid, id);
 355         if (result == -ENOENT) {
 356                 if (!fid_is_norm(fid) ||
 357                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 358                                       sf->sf_oi_bitmap))
 359                         GOTO(out, result = 0);
 360
 361                 goto trigger;
 362         }
 363
 364         if (result != 0)
 365                 GOTO(out, result);
 366
 367 iget:
 368         if (verify == 0)
 369                 inode = osd_iget(info, dev, id);
 370         else
 371                 inode = osd_iget_verify(info, dev, id, fid);
 372         if (IS_ERR(inode)) {
 373                 result = PTR_ERR(inode);
 374                 if (result == -ENOENT || result == -ESTALE) {
 375                         fid_zero(&oic->oic_fid);
 376                         result = 0;
 377                 } else if (result == -EREMCHG) {
 378
 379 trigger:
 380                         if (thread_is_running(&scrub->os_thread)) {
 381                                 result = -EINPROGRESS;
 382                         } else if (!dev->od_noscrub) {
 383                                 result = osd_scrub_start(dev);
 384                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 385                                                "for "DFID", rc = %d [1]\n",
 386                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 387                                                s_volume_name,PFID(fid), result);
 388                                 if (result == 0 || result == -EALREADY)
 389                                         result = -EINPROGRESS;
 390                                 else
 391                                         result = -EREMCHG;
 392                         }
 393                 }
 394
 395                 GOTO(out, result);
 396         }
 397
 398         obj->oo_inode = inode;
 399         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 400
 401         obj->oo_compat_dot_created = 1;
 402         obj->oo_compat_dotdot_created = 1;
 403
 404         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 405                 GOTO(out, result = 0);
 406
 407         LASSERT(obj->oo_hl_head == NULL);
 408         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 409         if (obj->oo_hl_head == NULL) {
 410                 obj->oo_inode = NULL;
 411                 iput(inode);
 412                 GOTO(out, result = -ENOMEM);
 413         }
 414         GOTO(out, result = 0);
 415
 416 out:
 417         LINVRNT(osd_invariant(obj));
 418         return result;
 419 }
 420
 421 /*
 422  * Concurrency: shouldn't matter.
 423  */
 424 static void osd_object_init0(struct osd_object *obj)
 425 {
 426         LASSERT(obj->oo_inode != NULL);
 427         obj->oo_dt.do_body_ops = &osd_body_ops;
 428         obj->oo_dt.do_lu.lo_header->loh_attr |=
 429                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 430 }
 431
 432 /*
 433  * Concurrency: no concurrent access is possible that early in object
 434  * life-cycle.
 435  */
 436 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 437                            const struct lu_object_conf *conf)
 438 {
 439         struct osd_object *obj = osd_obj(l);
 440         int result;
 441
 442         LINVRNT(osd_invariant(obj));
 443
 444         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 445         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 446         if (result == 0) {
 447                 if (obj->oo_inode != NULL) {
 448                         osd_object_init0(obj);
 449                 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 450                         obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 451                         /* LFSCK iterator object is special without inode */
 452                         l->lo_header->loh_attr |= LOHA_EXISTS;
 453                 }
 454         }
 455         LINVRNT(osd_invariant(obj));
 456         return result;
 457 }
 458
 459 /*
 460  * Concurrency: no concurrent access is possible that late in object
 461  * life-cycle.
 462  */
 463 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 464 {
 465         struct osd_object *obj = osd_obj(l);
 466
 467         LINVRNT(osd_invariant(obj));
 468
 469         dt_object_fini(&obj->oo_dt);
 470         if (obj->oo_hl_head != NULL)
 471                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 472         OBD_FREE_PTR(obj);
 473 }
 474
 475 /*
 476  * Concurrency: no concurrent access is possible that late in object
 477  * life-cycle.
 478  */
 479 static void osd_index_fini(struct osd_object *o)
 480 {
 481         struct iam_container *bag;
 482
 483         if (o->oo_dir != NULL) {
 484                 bag = &o->oo_dir->od_container;
 485                 if (o->oo_inode != NULL) {
 486                         if (bag->ic_object == o->oo_inode)
 487                                 iam_container_fini(bag);
 488                 }
 489                 OBD_FREE_PTR(o->oo_dir);
 490                 o->oo_dir = NULL;
 491         }
 492 }
 493
 494 /*
 495  * Concurrency: no concurrent access is possible that late in object
 496  * life-cycle (for all existing callers, that is. New callers have to provide
 497  * their own locking.)
 498  */
 499 static int osd_inode_unlinked(const struct inode *inode)
 500 {
 501         return inode->i_nlink == 0;
 502 }
 503
 504 enum {
 505         OSD_TXN_OI_DELETE_CREDITS    = 20,
 506         OSD_TXN_INODE_DELETE_CREDITS = 20
 507 };
 508
 509 /*
 510  * Journal
 511  */
 512
 513 #if OSD_THANDLE_STATS
 514 /**
 515  * Set time when the handle is allocated
 516  */
 517 static void osd_th_alloced(struct osd_thandle *oth)
 518 {
 519         oth->oth_alloced = cfs_time_current();
 520 }
 521
 522 /**
 523  * Set time when the handle started
 524  */
 525 static void osd_th_started(struct osd_thandle *oth)
 526 {
 527         oth->oth_started = cfs_time_current();
 528 }
 529
 530 /**
 531  * Helper function to convert time interval to microseconds packed in
 532  * long int.
 533  */
 534 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 535 {
 536         struct timeval val;
 537
 538         cfs_duration_usec(cfs_time_sub(end, start), &val);
 539         return val.tv_sec * 1000000 + val.tv_usec;
 540 }
 541
 542 /**
 543  * Check whether the we deal with this handle for too long.
 544  */
 545 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 546                                 cfs_time_t alloced, cfs_time_t started,
 547                                 cfs_time_t closed)
 548 {
 549         cfs_time_t now = cfs_time_current();
 550
 551         LASSERT(dev != NULL);
 552
 553         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 554                             interval_to_usec(alloced, started));
 555         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 556                             interval_to_usec(started, closed));
 557         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 558                             interval_to_usec(closed, now));
 559
 560         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 561                 CWARN("transaction handle %p was open for too long: "
 562                       "now "CFS_TIME_T" ,"
 563                       "alloced "CFS_TIME_T" ,"
 564                       "started "CFS_TIME_T" ,"
 565                       "closed "CFS_TIME_T"\n",
 566                       oth, now, alloced, started, closed);
 567                 libcfs_debug_dumpstack(NULL);
 568         }
 569 }
 570
 571 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 572 {                                                                       \
 573         cfs_time_t __closed = cfs_time_current();                       \
 574         cfs_time_t __alloced = oth->oth_alloced;                        \
 575         cfs_time_t __started = oth->oth_started;                        \
 576                                                                         \
 577         expr;                                                           \
 578         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 579 }
 580
 581 #else /* OSD_THANDLE_STATS */
 582
 583 #define osd_th_alloced(h)                  do {} while(0)
 584 #define osd_th_started(h)                  do {} while(0)
 585 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 586
 587 #endif /* OSD_THANDLE_STATS */
 588
 589 /*
 590  * Concurrency: doesn't access mutable data.
 591  */
 592 static int osd_param_is_not_sane(const struct osd_device *dev,
 593                                  const struct thandle *th)
 594 {
 595         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 596
 597         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 598 }
 599
 600 /*
 601  * Concurrency: shouldn't matter.
 602  */
 603 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 604 static void osd_trans_commit_cb(struct super_block *sb,
 605                                 struct journal_callback *jcb, int error)
 606 #else
 607 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 608 #endif
 609 {
 610         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 611         struct thandle     *th  = &oh->ot_super;
 612         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 613         struct dt_txn_commit_cb *dcb, *tmp;
 614
 615         LASSERT(oh->ot_handle == NULL);
 616
 617         if (error)
 618                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 619
 620         dt_txn_hook_commit(th);
 621
 622         /* call per-transaction callbacks if any */
 623         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 624                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 625                          "commit callback entry: magic=%x name='%s'\n",
 626                          dcb->dcb_magic, dcb->dcb_name);
 627                 cfs_list_del_init(&dcb->dcb_linkage);
 628                 dcb->dcb_func(NULL, th, dcb, error);
 629         }
 630
 631         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 632         lu_device_put(lud);
 633         th->th_dev = NULL;
 634
 635         lu_context_exit(&th->th_ctx);
 636         lu_context_fini(&th->th_ctx);
 637         OBD_FREE_PTR(oh);
 638 }
 639
 640 static struct thandle *osd_trans_create(const struct lu_env *env,
 641                                         struct dt_device *d)
 642 {
 643         struct osd_thread_info *oti = osd_oti_get(env);
 644         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 645         struct osd_thandle     *oh;
 646         struct thandle         *th;
 647         ENTRY;
 648
 649         /* on pending IO in this thread should left from prev. request */
 650         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 651
 652         th = ERR_PTR(-ENOMEM);
 653         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 654         if (oh != NULL) {
 655                 oh->ot_quota_trans = &oti->oti_quota_trans;
 656                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 657                 th = &oh->ot_super;
 658                 th->th_dev = d;
 659                 th->th_result = 0;
 660                 th->th_tags = LCT_TX_HANDLE;
 661                 oh->ot_credits = 0;
 662                 oti->oti_dev = osd_dt_dev(d);
 663                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 664                 osd_th_alloced(oh);
 665         }
 666         RETURN(th);
 667 }
 668
 669 /*
 670  * Concurrency: shouldn't matter.
 671  */
 672 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 673                     struct thandle *th)
 674 {
 675         struct osd_thread_info *oti = osd_oti_get(env);
 676         struct osd_device  *dev = osd_dt_dev(d);
 677         handle_t           *jh;
 678         struct osd_thandle *oh;
 679         int rc;
 680
 681         ENTRY;
 682
 683         LASSERT(current->journal_info == NULL);
 684
 685         oh = container_of0(th, struct osd_thandle, ot_super);
 686         LASSERT(oh != NULL);
 687         LASSERT(oh->ot_handle == NULL);
 688
 689         rc = dt_txn_hook_start(env, d, th);
 690         if (rc != 0)
 691                 GOTO(out, rc);
 692
 693         if (unlikely(osd_param_is_not_sane(dev, th))) {
 694                 static unsigned long last_printed;
 695                 static int last_credits;
 696
 697                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 698                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 699                       oh->ot_credits,
 700                       osd_journal(dev)->j_max_transaction_buffers);
 701 #ifdef OSD_TRACK_DECLARES
 702                 CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
 703                       oh->ot_declare_create, oh->ot_declare_create_cred,
 704                       oh->ot_declare_delete, oh->ot_declare_delete_cred,
 705                       oh->ot_declare_destroy, oh->ot_declare_destroy_cred);
 706                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
 707                       oh->ot_declare_attr_set, oh->ot_declare_attr_set_cred,
 708                       oh->ot_declare_xattr_set, oh->ot_declare_xattr_set_cred);
 709                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
 710                       oh->ot_declare_write, oh->ot_declare_write_cred,
 711                       oh->ot_declare_punch, oh->ot_declare_punch_cred,
 712                       oh->ot_declare_quota, oh->ot_declare_quota_cred);
 713                 CWARN("  insert: %u/%u, delete: %u/%u\n",
 714                       oh->ot_declare_insert, oh->ot_declare_insert_cred,
 715                       oh->ot_declare_delete, oh->ot_declare_destroy_cred);
 716                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
 717                       oh->ot_declare_ref_add, oh->ot_declare_ref_add_cred,
 718                       oh->ot_declare_ref_del, oh->ot_declare_ref_del_cred);
 719
 720                 if (last_credits != oh->ot_credits &&
 721                     time_after(jiffies, last_printed + 60 * HZ)) {
 722                         libcfs_debug_dumpstack(NULL);
 723                         last_credits = oh->ot_credits;
 724                         last_printed = jiffies;
 725                 }
 726 #endif
 727                 /* XXX Limit the credits to 'max_transaction_buffers', and
 728                  *     let the underlying filesystem to catch the error if
 729                  *     we really need so many credits.
 730                  *
 731                  *     This should be removed when we can calculate the
 732                  *     credits precisely. */
 733                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 734         }
 735
 736         /*
 737          * XXX temporary stuff. Some abstraction layer should
 738          * be used.
 739          */
 740         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 741         osd_th_started(oh);
 742         if (!IS_ERR(jh)) {
 743                 oh->ot_handle = jh;
 744                 LASSERT(oti->oti_txns == 0);
 745                 lu_context_init(&th->th_ctx, th->th_tags);
 746                 lu_context_enter(&th->th_ctx);
 747
 748                 lu_device_get(&d->dd_lu_dev);
 749                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 750                                              "osd-tx", th);
 751                 oti->oti_txns++;
 752                 rc = 0;
 753         } else {
 754                 rc = PTR_ERR(jh);
 755         }
 756 out:
 757         RETURN(rc);
 758 }
 759
 760 /*
 761  * Concurrency: shouldn't matter.
 762  */
 763 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 764 {
 765         int                     rc = 0;
 766         struct osd_thandle     *oh;
 767         struct osd_thread_info *oti = osd_oti_get(env);
 768         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 769         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
 770         ENTRY;
 771
 772         oh = container_of0(th, struct osd_thandle, ot_super);
 773
 774         if (qsd != NULL)
 775                 /* inform the quota slave device that the transaction is
 776                  * stopping */
 777                 qsd_op_end(env, qsd, oh->ot_quota_trans);
 778         oh->ot_quota_trans = NULL;
 779
 780         if (oh->ot_handle != NULL) {
 781                 handle_t *hdl = oh->ot_handle;
 782
 783                 /*
 784                  * add commit callback
 785                  * notice we don't do this in osd_trans_start()
 786                  * as underlying transaction can change during truncate
 787                  */
 788                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 789                                          &oh->ot_jcb);
 790
 791                 LASSERT(oti->oti_txns == 1);
 792                 oti->oti_txns--;
 793                 rc = dt_txn_hook_stop(env, th);
 794                 if (rc != 0)
 795                         CERROR("Failure in transaction hook: %d\n", rc);
 796
 797                 /* hook functions might modify th_sync */
 798                 hdl->h_sync = th->th_sync;
 799
 800                 oh->ot_handle = NULL;
 801                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 802                                   rc = ldiskfs_journal_stop(hdl));
 803                 if (rc != 0)
 804                         CERROR("Failure to stop transaction: %d\n", rc);
 805         } else {
 806                 OBD_FREE_PTR(oh);
 807         }
 808
 809         /* as we want IO to journal and data IO be concurrent, we don't block
 810          * awaiting data IO completion in osd_do_bio(), instead we wait here
 811          * once transaction is submitted to the journal. all reqular requests
 812          * don't do direct IO (except read/write), thus this wait_event becomes
 813          * no-op for them.
 814          *
 815          * IMPORTANT: we have to wait till any IO submited by the thread is
 816          * completed otherwise iobuf may be corrupted by different request
 817          */
 818         cfs_wait_event(iobuf->dr_wait,
 819                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 820         if (!rc)
 821                 rc = iobuf->dr_error;
 822
 823         RETURN(rc);
 824 }
 825
 826 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 827 {
 828         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 829                                                ot_super);
 830
 831         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 832         LASSERT(&dcb->dcb_func != NULL);
 833         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 834
 835         return 0;
 836 }
 837
 838 /*
 839  * Called just before object is freed. Releases all resources except for
 840  * object itself (that is released by osd_object_free()).
 841  *
 842  * Concurrency: no concurrent access is possible that late in object
 843  * life-cycle.
 844  */
 845 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 846 {
 847         struct osd_object *obj   = osd_obj(l);
 848         struct inode      *inode = obj->oo_inode;
 849
 850         LINVRNT(osd_invariant(obj));
 851
 852         /*
 853          * If object is unlinked remove fid->ino mapping from object index.
 854          */
 855
 856         osd_index_fini(obj);
 857         if (inode != NULL) {
 858                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
 859                 qid_t                    uid = inode->i_uid;
 860                 qid_t                    gid = inode->i_gid;
 861
 862                 iput(inode);
 863                 obj->oo_inode = NULL;
 864
 865                 if (qsd != NULL) {
 866                         struct osd_thread_info  *info = osd_oti_get(env);
 867                         struct lquota_id_info   *qi = &info->oti_qi;
 868
 869                         /* Release granted quota to master if necessary */
 870                         qi->lqi_id.qid_uid = uid;
 871                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
 872
 873                         qi->lqi_id.qid_uid = gid;
 874                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
 875                 }
 876         }
 877 }
 878
 879 /*
 880  * Concurrency: ->loo_object_release() is called under site spin-lock.
 881  */
 882 static void osd_object_release(const struct lu_env *env,
 883                                struct lu_object *l)
 884 {
 885 }
 886
 887 /*
 888  * Concurrency: shouldn't matter.
 889  */
 890 static int osd_object_print(const struct lu_env *env, void *cookie,
 891                             lu_printer_t p, const struct lu_object *l)
 892 {
 893         struct osd_object *o = osd_obj(l);
 894         struct iam_descr  *d;
 895
 896         if (o->oo_dir != NULL)
 897                 d = o->oo_dir->od_container.ic_descr;
 898         else
 899                 d = NULL;
 900         return (*p)(env, cookie,
 901                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 902                     o, o->oo_inode,
 903                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 904                     o->oo_inode ? o->oo_inode->i_generation : 0,
 905                     d ? d->id_ops->id_name : "plain");
 906 }
 907
 908 /*
 909  * Concurrency: shouldn't matter.
 910  */
 911 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 912                struct obd_statfs *sfs)
 913 {
 914         struct osd_device  *osd = osd_dt_dev(d);
 915         struct super_block *sb = osd_sb(osd);
 916         struct kstatfs     *ksfs;
 917         int result = 0;
 918
 919         if (unlikely(osd->od_mnt == NULL))
 920                 return -EINPROGRESS;
 921
 922         /* osd_lproc.c call this without env, allocate ksfs for that case */
 923         if (unlikely(env == NULL)) {
 924                 OBD_ALLOC_PTR(ksfs);
 925                 if (ksfs == NULL)
 926                         return -ENOMEM;
 927         } else {
 928                 ksfs = &osd_oti_get(env)->oti_ksfs;
 929         }
 930
 931         spin_lock(&osd->od_osfs_lock);
 932         /* cache 1 second */
 933         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 934                 result = sb->s_op->statfs(sb->s_root, ksfs);
 935                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 936                         osd->od_osfs_age = cfs_time_current_64();
 937                         statfs_pack(&osd->od_statfs, ksfs);
 938                         if (sb->s_flags & MS_RDONLY)
 939                                 sfs->os_state = OS_STATE_READONLY;
 940                 }
 941         }
 942
 943         if (likely(result == 0))
 944                 *sfs = osd->od_statfs;
 945         spin_unlock(&osd->od_osfs_lock);
 946
 947         if (unlikely(env == NULL))
 948                 OBD_FREE_PTR(ksfs);
 949
 950         return result;
 951 }
 952
 953 /**
 954  * Estimate space needed for file creations. We assume the largest filename
 955  * which is 2^64 - 1, hence a filename of 20 chars.
 956  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
 957  */
 958 #ifdef __LDISKFS_DIR_REC_LEN
 959 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
 960 #else
 961 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
 962 #endif
 963
 964 /*
 965  * Concurrency: doesn't access mutable data.
 966  */
 967 static void osd_conf_get(const struct lu_env *env,
 968                          const struct dt_device *dev,
 969                          struct dt_device_param *param)
 970 {
 971         struct super_block *sb = osd_sb(osd_dt_dev(dev));
 972
 973         /*
 974          * XXX should be taken from not-yet-existing fs abstraction layer.
 975          */
 976         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
 977         param->ddp_max_name_len = LDISKFS_NAME_LEN;
 978         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
 979         param->ddp_block_shift  = sb->s_blocksize_bits;
 980         param->ddp_mount_type     = LDD_MT_LDISKFS;
 981         param->ddp_maxbytes       = sb->s_maxbytes;
 982         /* Overhead estimate should be fairly accurate, so we really take a tiny
 983          * error margin which also avoids fragmenting the filesystem too much */
 984         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
 985         /* inode are statically allocated, so per-inode space consumption
 986          * is the space consumed by the directory entry */
 987         param->ddp_inodespace     = PER_OBJ_USAGE;
 988         /* per-fragment overhead to be used by the client code */
 989         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
 990         param->ddp_mntopts      = 0;
 991         if (test_opt(sb, XATTR_USER))
 992                 param->ddp_mntopts |= MNTOPT_USERXATTR;
 993         if (test_opt(sb, POSIX_ACL))
 994                 param->ddp_mntopts |= MNTOPT_ACL;
 995
 996 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
 997         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
 998                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
 999         else
1000 #endif
1001                 param->ddp_max_ea_size = sb->s_blocksize;
1002
1003 }
1004
1005 /**
1006  * Helper function to get and fill the buffer with input values.
1007  */
1008 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1009 {
1010         struct lu_buf *buf;
1011
1012         buf = &osd_oti_get(env)->oti_buf;
1013         buf->lb_buf = area;
1014         buf->lb_len = len;
1015         return buf;
1016 }
1017
1018 /*
1019  * Concurrency: shouldn't matter.
1020  */
1021 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1022 {
1023         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1024         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1025 }
1026
1027 /**
1028  * Start commit for OSD device.
1029  *
1030  * An implementation of dt_commit_async method for OSD device.
1031  * Asychronously starts underlayng fs sync and thereby a transaction
1032  * commit.
1033  *
1034  * \param env environment
1035  * \param d dt device
1036  *
1037  * \see dt_device_operations
1038  */
1039 static int osd_commit_async(const struct lu_env *env,
1040                             struct dt_device *d)
1041 {
1042         struct super_block *s = osd_sb(osd_dt_dev(d));
1043         ENTRY;
1044
1045         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1046         RETURN(s->s_op->sync_fs(s, 0));
1047 }
1048
1049 /*
1050  * Concurrency: shouldn't matter.
1051  */
1052
1053 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1054 {
1055         struct super_block *sb = osd_sb(osd_dt_dev(d));
1056         int rc;
1057         ENTRY;
1058
1059         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1060
1061         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1062         RETURN(rc);
1063 }
1064
1065 /*
1066  * Concurrency: serialization provided by callers.
1067  */
1068 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1069                               int mode, unsigned long timeout, __u32 alg,
1070                               struct lustre_capa_key *keys)
1071 {
1072         struct osd_device *dev = osd_dt_dev(d);
1073         ENTRY;
1074
1075         dev->od_fl_capa = mode;
1076         dev->od_capa_timeout = timeout;
1077         dev->od_capa_alg = alg;
1078         dev->od_capa_keys = keys;
1079         RETURN(0);
1080 }
1081
1082 /**
1083  * Note: we do not count into QUOTA here.
1084  * If we mount with --data_journal we may need more.
1085  */
1086 const int osd_dto_credits_noquota[DTO_NR] = {
1087         /**
1088          * Insert/Delete.
1089          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1090          * SINGLEDATA_TRANS_BLOCKS(8)
1091          * XXX Note: maybe iam need more, since iam have more level than
1092          *           EXT3 htree.
1093          */
1094         [DTO_INDEX_INSERT]  = 16,
1095         [DTO_INDEX_DELETE]  = 16,
1096         /**
1097          * Used for OI scrub
1098          */
1099         [DTO_INDEX_UPDATE]  = 16,
1100         /**
1101          * Create a object. The same as create object in EXT3.
1102          * DATA_TRANS_BLOCKS(14) +
1103          * INDEX_EXTRA_BLOCKS(8) +
1104          * 3(inode bits, groups, GDT)
1105          */
1106         [DTO_OBJECT_CREATE] = 25,
1107         /**
1108          * XXX: real credits to be fixed
1109          */
1110         [DTO_OBJECT_DELETE] = 25,
1111         /**
1112          * Attr set credits (inode)
1113          */
1114         [DTO_ATTR_SET_BASE] = 1,
1115         /**
1116          * Xattr set. The same as xattr of EXT3.
1117          * DATA_TRANS_BLOCKS(14)
1118          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1119          * are also counted in. Do not know why?
1120          */
1121         [DTO_XATTR_SET]     = 14,
1122         [DTO_LOG_REC]       = 14,
1123         /**
1124          * credits for inode change during write.
1125          */
1126         [DTO_WRITE_BASE]    = 3,
1127         /**
1128          * credits for single block write.
1129          */
1130         [DTO_WRITE_BLOCK]   = 14,
1131         /**
1132          * Attr set credits for chown.
1133          * This is extra credits for setattr, and it is null without quota
1134          */
1135         [DTO_ATTR_SET_CHOWN]= 0
1136 };
1137
1138 static const struct dt_device_operations osd_dt_ops = {
1139         .dt_root_get       = osd_root_get,
1140         .dt_statfs         = osd_statfs,
1141         .dt_trans_create   = osd_trans_create,
1142         .dt_trans_start    = osd_trans_start,
1143         .dt_trans_stop     = osd_trans_stop,
1144         .dt_trans_cb_add   = osd_trans_cb_add,
1145         .dt_conf_get       = osd_conf_get,
1146         .dt_sync           = osd_sync,
1147         .dt_ro             = osd_ro,
1148         .dt_commit_async   = osd_commit_async,
1149         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1150 };
1151
1152 static void osd_object_read_lock(const struct lu_env *env,
1153                                  struct dt_object *dt, unsigned role)
1154 {
1155         struct osd_object *obj = osd_dt_obj(dt);
1156         struct osd_thread_info *oti = osd_oti_get(env);
1157
1158         LINVRNT(osd_invariant(obj));
1159
1160         LASSERT(obj->oo_owner != env);
1161         down_read_nested(&obj->oo_sem, role);
1162
1163         LASSERT(obj->oo_owner == NULL);
1164         oti->oti_r_locks++;
1165 }
1166
1167 static void osd_object_write_lock(const struct lu_env *env,
1168                                   struct dt_object *dt, unsigned role)
1169 {
1170         struct osd_object *obj = osd_dt_obj(dt);
1171         struct osd_thread_info *oti = osd_oti_get(env);
1172
1173         LINVRNT(osd_invariant(obj));
1174
1175         LASSERT(obj->oo_owner != env);
1176         down_write_nested(&obj->oo_sem, role);
1177
1178         LASSERT(obj->oo_owner == NULL);
1179         obj->oo_owner = env;
1180         oti->oti_w_locks++;
1181 }
1182
1183 static void osd_object_read_unlock(const struct lu_env *env,
1184                                    struct dt_object *dt)
1185 {
1186         struct osd_object *obj = osd_dt_obj(dt);
1187         struct osd_thread_info *oti = osd_oti_get(env);
1188
1189         LINVRNT(osd_invariant(obj));
1190
1191         LASSERT(oti->oti_r_locks > 0);
1192         oti->oti_r_locks--;
1193         up_read(&obj->oo_sem);
1194 }
1195
1196 static void osd_object_write_unlock(const struct lu_env *env,
1197                                     struct dt_object *dt)
1198 {
1199         struct osd_object *obj = osd_dt_obj(dt);
1200         struct osd_thread_info *oti = osd_oti_get(env);
1201
1202         LINVRNT(osd_invariant(obj));
1203
1204         LASSERT(obj->oo_owner == env);
1205         LASSERT(oti->oti_w_locks > 0);
1206         oti->oti_w_locks--;
1207         obj->oo_owner = NULL;
1208         up_write(&obj->oo_sem);
1209 }
1210
1211 static int osd_object_write_locked(const struct lu_env *env,
1212                                    struct dt_object *dt)
1213 {
1214         struct osd_object *obj = osd_dt_obj(dt);
1215
1216         LINVRNT(osd_invariant(obj));
1217
1218         return obj->oo_owner == env;
1219 }
1220
1221 static int capa_is_sane(const struct lu_env *env,
1222                         struct osd_device *dev,
1223                         struct lustre_capa *capa,
1224                         struct lustre_capa_key *keys)
1225 {
1226         struct osd_thread_info *oti = osd_oti_get(env);
1227         struct lustre_capa *tcapa = &oti->oti_capa;
1228         struct obd_capa *oc;
1229         int i, rc = 0;
1230         ENTRY;
1231
1232         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1233         if (oc) {
1234                 if (capa_is_expired(oc)) {
1235                         DEBUG_CAPA(D_ERROR, capa, "expired");
1236                         rc = -ESTALE;
1237                 }
1238                 capa_put(oc);
1239                 RETURN(rc);
1240         }
1241
1242         if (capa_is_expired_sec(capa)) {
1243                 DEBUG_CAPA(D_ERROR, capa, "expired");
1244                 RETURN(-ESTALE);
1245         }
1246
1247         spin_lock(&capa_lock);
1248         for (i = 0; i < 2; i++) {
1249                 if (keys[i].lk_keyid == capa->lc_keyid) {
1250                         oti->oti_capa_key = keys[i];
1251                         break;
1252                 }
1253         }
1254         spin_unlock(&capa_lock);
1255
1256         if (i == 2) {
1257                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1258                 RETURN(-ESTALE);
1259         }
1260
1261         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1262         if (rc)
1263                 RETURN(rc);
1264
1265         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1266                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1267                 RETURN(-EACCES);
1268         }
1269
1270         oc = capa_add(dev->od_capa_hash, capa);
1271         capa_put(oc);
1272
1273         RETURN(0);
1274 }
1275
1276 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1277                     struct lustre_capa *capa, __u64 opc)
1278 {
1279         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1280         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1281         struct md_capainfo *ci;
1282         int rc;
1283
1284         if (!dev->od_fl_capa)
1285                 return 0;
1286
1287         if (capa == BYPASS_CAPA)
1288                 return 0;
1289
1290         ci = md_capainfo(env);
1291         if (unlikely(!ci))
1292                 return 0;
1293
1294         if (ci->mc_auth == LC_ID_NONE)
1295                 return 0;
1296
1297         if (!capa) {
1298                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1299                 return -EACCES;
1300         }
1301
1302         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1303                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1304                            PFID(fid));
1305                 return -EACCES;
1306         }
1307
1308         if (!capa_opc_supported(capa, opc)) {
1309                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1310                 return -EACCES;
1311         }
1312
1313         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1314                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1315                 return -EACCES;
1316         }
1317
1318         return 0;
1319 }
1320
1321 static struct timespec *osd_inode_time(const struct lu_env *env,
1322                                        struct inode *inode, __u64 seconds)
1323 {
1324         struct osd_thread_info  *oti = osd_oti_get(env);
1325         struct timespec         *t   = &oti->oti_time;
1326
1327         t->tv_sec = seconds;
1328         t->tv_nsec = 0;
1329         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1330         return t;
1331 }
1332
1333
1334 static void osd_inode_getattr(const struct lu_env *env,
1335                               struct inode *inode, struct lu_attr *attr)
1336 {
1337         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1338                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1339                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1340
1341         attr->la_atime      = LTIME_S(inode->i_atime);
1342         attr->la_mtime      = LTIME_S(inode->i_mtime);
1343         attr->la_ctime      = LTIME_S(inode->i_ctime);
1344         attr->la_mode       = inode->i_mode;
1345         attr->la_size       = i_size_read(inode);
1346         attr->la_blocks     = inode->i_blocks;
1347         attr->la_uid        = inode->i_uid;
1348         attr->la_gid        = inode->i_gid;
1349         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1350         attr->la_nlink      = inode->i_nlink;
1351         attr->la_rdev       = inode->i_rdev;
1352         attr->la_blksize    = 1 << inode->i_blkbits;
1353         attr->la_blkbits    = inode->i_blkbits;
1354 }
1355
1356 static int osd_attr_get(const struct lu_env *env,
1357                         struct dt_object *dt,
1358                         struct lu_attr *attr,
1359                         struct lustre_capa *capa)
1360 {
1361         struct osd_object *obj = osd_dt_obj(dt);
1362
1363         LASSERT(dt_object_exists(dt));
1364         LINVRNT(osd_invariant(obj));
1365
1366         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1367                 return -EACCES;
1368
1369         spin_lock(&obj->oo_guard);
1370         osd_inode_getattr(env, obj->oo_inode, attr);
1371         spin_unlock(&obj->oo_guard);
1372         return 0;
1373 }
1374
1375 static int osd_declare_attr_set(const struct lu_env *env,
1376                                 struct dt_object *dt,
1377                                 const struct lu_attr *attr,
1378                                 struct thandle *handle)
1379 {
1380         struct osd_thandle     *oh;
1381         struct osd_object      *obj;
1382         struct osd_thread_info *info = osd_oti_get(env);
1383         struct lquota_id_info  *qi = &info->oti_qi;
1384         long long               bspace;
1385         int                     rc = 0;
1386         bool                    allocated;
1387         ENTRY;
1388
1389         LASSERT(dt != NULL);
1390         LASSERT(handle != NULL);
1391
1392         obj = osd_dt_obj(dt);
1393         LASSERT(osd_invariant(obj));
1394
1395         oh = container_of0(handle, struct osd_thandle, ot_super);
1396         LASSERT(oh->ot_handle == NULL);
1397
1398         OSD_DECLARE_OP(oh, attr_set,
1399                        osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1400
1401         if (attr == NULL || obj->oo_inode == NULL)
1402                 RETURN(rc);
1403
1404         bspace   = obj->oo_inode->i_blocks;
1405         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1406         bspace   = toqb(bspace);
1407
1408         /* Changing ownership is always preformed by super user, it should not
1409          * fail with EDQUOT.
1410          *
1411          * We still need to call the osd_declare_qid() to calculate the journal
1412          * credits for updating quota accounting files and to trigger quota
1413          * space adjustment once the operation is completed.*/
1414         if ((attr->la_valid & LA_UID) != 0 &&
1415              attr->la_uid != obj->oo_inode->i_uid) {
1416                 qi->lqi_type = USRQUOTA;
1417
1418                 /* inode accounting */
1419                 qi->lqi_is_blk = false;
1420
1421                 /* one more inode for the new owner ... */
1422                 qi->lqi_id.qid_uid = attr->la_uid;
1423                 qi->lqi_space      = 1;
1424                 allocated = (attr->la_uid == 0) ? true : false;
1425                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1426                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1427                         rc = 0;
1428                 if (rc)
1429                         RETURN(rc);
1430
1431                 /* and one less inode for the current uid */
1432                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1433                 qi->lqi_space      = -1;
1434                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1435                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1436                         rc = 0;
1437                 if (rc)
1438                         RETURN(rc);
1439
1440                 /* block accounting */
1441                 qi->lqi_is_blk = true;
1442
1443                 /* more blocks for the new owner ... */
1444                 qi->lqi_id.qid_uid = attr->la_uid;
1445                 qi->lqi_space      = bspace;
1446                 allocated = (attr->la_uid == 0) ? true : false;
1447                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1448                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1449                         rc = 0;
1450                 if (rc)
1451                         RETURN(rc);
1452
1453                 /* and finally less blocks for the current owner */
1454                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1455                 qi->lqi_space      = -bspace;
1456                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1457                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1458                         rc = 0;
1459                 if (rc)
1460                         RETURN(rc);
1461         }
1462
1463         if (attr->la_valid & LA_GID &&
1464             attr->la_gid != obj->oo_inode->i_gid) {
1465                 qi->lqi_type = GRPQUOTA;
1466
1467                 /* inode accounting */
1468                 qi->lqi_is_blk = false;
1469
1470                 /* one more inode for the new group owner ... */
1471                 qi->lqi_id.qid_gid = attr->la_gid;
1472                 qi->lqi_space      = 1;
1473                 allocated = (attr->la_gid == 0) ? true : false;
1474                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1475                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1476                         rc = 0;
1477                 if (rc)
1478                         RETURN(rc);
1479
1480                 /* and one less inode for the current gid */
1481                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1482                 qi->lqi_space      = -1;
1483                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1484                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1485                         rc = 0;
1486                 if (rc)
1487                         RETURN(rc);
1488
1489                 /* block accounting */
1490                 qi->lqi_is_blk = true;
1491
1492                 /* more blocks for the new owner ... */
1493                 qi->lqi_id.qid_gid = attr->la_gid;
1494                 qi->lqi_space      = bspace;
1495                 allocated = (attr->la_gid == 0) ? true : false;
1496                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1497                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1498                         rc = 0;
1499                 if (rc)
1500                         RETURN(rc);
1501
1502                 /* and finally less blocks for the current owner */
1503                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1504                 qi->lqi_space      = -bspace;
1505                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1506                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1507                         rc = 0;
1508                 if (rc)
1509                         RETURN(rc);
1510         }
1511
1512         RETURN(rc);
1513 }
1514
1515 static int osd_inode_setattr(const struct lu_env *env,
1516                              struct inode *inode, const struct lu_attr *attr)
1517 {
1518         __u64 bits;
1519
1520         bits = attr->la_valid;
1521
1522         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1523
1524         if (bits & LA_ATIME)
1525                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1526         if (bits & LA_CTIME)
1527                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1528         if (bits & LA_MTIME)
1529                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1530         if (bits & LA_SIZE) {
1531                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1532                 i_size_write(inode, attr->la_size);
1533         }
1534
1535 #if 0
1536         /* OSD should not change "i_blocks" which is used by quota.
1537          * "i_blocks" should be changed by ldiskfs only. */
1538         if (bits & LA_BLOCKS)
1539                 inode->i_blocks = attr->la_blocks;
1540 #endif
1541         if (bits & LA_MODE)
1542                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1543                         (attr->la_mode & ~S_IFMT);
1544         if (bits & LA_UID)
1545                 inode->i_uid    = attr->la_uid;
1546         if (bits & LA_GID)
1547                 inode->i_gid    = attr->la_gid;
1548         if (bits & LA_NLINK)
1549                 set_nlink(inode, attr->la_nlink);
1550         if (bits & LA_RDEV)
1551                 inode->i_rdev   = attr->la_rdev;
1552
1553         if (bits & LA_FLAGS) {
1554                 /* always keep S_NOCMTIME */
1555                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1556                                  S_NOCMTIME;
1557         }
1558         return 0;
1559 }
1560
1561 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1562 {
1563         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1564             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1565                 struct iattr    iattr;
1566                 int             rc;
1567
1568                 iattr.ia_valid = 0;
1569                 if (attr->la_valid & LA_UID)
1570                         iattr.ia_valid |= ATTR_UID;
1571                 if (attr->la_valid & LA_GID)
1572                         iattr.ia_valid |= ATTR_GID;
1573                 iattr.ia_uid = attr->la_uid;
1574                 iattr.ia_gid = attr->la_gid;
1575
1576                 rc = ll_vfs_dq_transfer(inode, &iattr);
1577                 if (rc) {
1578                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1579                                "enforcement enabled on the ldiskfs filesystem?",
1580                                inode->i_sb->s_id, rc);
1581                         return rc;
1582                 }
1583         }
1584         return 0;
1585 }
1586
1587 static int osd_attr_set(const struct lu_env *env,
1588                         struct dt_object *dt,
1589                         const struct lu_attr *attr,
1590                         struct thandle *handle,
1591                         struct lustre_capa *capa)
1592 {
1593         struct osd_object *obj = osd_dt_obj(dt);
1594         struct inode      *inode;
1595         int rc;
1596
1597         LASSERT(handle != NULL);
1598         LASSERT(dt_object_exists(dt));
1599         LASSERT(osd_invariant(obj));
1600
1601         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1602                 return -EACCES;
1603
1604         OSD_EXEC_OP(handle, attr_set);
1605
1606         inode = obj->oo_inode;
1607         ll_vfs_dq_init(inode);
1608
1609         rc = osd_quota_transfer(inode, attr);
1610         if (rc)
1611                 return rc;
1612
1613         spin_lock(&obj->oo_guard);
1614         rc = osd_inode_setattr(env, inode, attr);
1615         spin_unlock(&obj->oo_guard);
1616
1617         if (!rc)
1618                 inode->i_sb->s_op->dirty_inode(inode);
1619         return rc;
1620 }
1621
1622 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1623                                     struct osd_object *obj,
1624                                     const char *name, const int namelen)
1625 {
1626         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1627 }
1628
1629 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1630                       cfs_umode_t mode,
1631                       struct dt_allocation_hint *hint,
1632                       struct thandle *th)
1633 {
1634         int result;
1635         struct osd_device  *osd = osd_obj2dev(obj);
1636         struct osd_thandle *oth;
1637         struct dt_object   *parent = NULL;
1638         struct inode       *inode;
1639
1640         LINVRNT(osd_invariant(obj));
1641         LASSERT(obj->oo_inode == NULL);
1642         LASSERT(obj->oo_hl_head == NULL);
1643
1644         if (S_ISDIR(mode) && ldiskfs_pdo) {
1645                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1646                 if (obj->oo_hl_head == NULL)
1647                         return -ENOMEM;
1648         }
1649
1650         oth = container_of(th, struct osd_thandle, ot_super);
1651         LASSERT(oth->ot_handle->h_transaction != NULL);
1652
1653         if (hint && hint->dah_parent)
1654                 parent = hint->dah_parent;
1655
1656         inode = ldiskfs_create_inode(oth->ot_handle,
1657                                      parent ? osd_dt_obj(parent)->oo_inode :
1658                                               osd_sb(osd)->s_root->d_inode,
1659                                      mode);
1660         if (!IS_ERR(inode)) {
1661                 /* Do not update file c/mtime in ldiskfs.
1662                  * NB: don't need any lock because no contention at this
1663                  * early stage */
1664                 inode->i_flags |= S_NOCMTIME;
1665                 inode->i_state |= I_LUSTRE_NOSCRUB;
1666                 obj->oo_inode = inode;
1667                 result = 0;
1668         } else {
1669                 if (obj->oo_hl_head != NULL) {
1670                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1671                         obj->oo_hl_head = NULL;
1672                 }
1673                 result = PTR_ERR(inode);
1674         }
1675         LINVRNT(osd_invariant(obj));
1676         return result;
1677 }
1678
1679 enum {
1680         OSD_NAME_LEN = 255
1681 };
1682
1683 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1684                      struct lu_attr *attr,
1685                      struct dt_allocation_hint *hint,
1686                      struct dt_object_format *dof,
1687                      struct thandle *th)
1688 {
1689         int result;
1690         struct osd_thandle *oth;
1691         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1692
1693         LASSERT(S_ISDIR(attr->la_mode));
1694
1695         oth = container_of(th, struct osd_thandle, ot_super);
1696         LASSERT(oth->ot_handle->h_transaction != NULL);
1697         result = osd_mkfile(info, obj, mode, hint, th);
1698
1699         return result;
1700 }
1701
1702 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1703                         struct lu_attr *attr,
1704                         struct dt_allocation_hint *hint,
1705                         struct dt_object_format *dof,
1706                         struct thandle *th)
1707 {
1708         int result;
1709         struct osd_thandle *oth;
1710         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1711
1712         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1713
1714         LASSERT(S_ISREG(attr->la_mode));
1715
1716         oth = container_of(th, struct osd_thandle, ot_super);
1717         LASSERT(oth->ot_handle->h_transaction != NULL);
1718
1719         result = osd_mkfile(info, obj, mode, hint, th);
1720         if (result == 0) {
1721                 LASSERT(obj->oo_inode != NULL);
1722                 if (feat->dif_flags & DT_IND_VARKEY)
1723                         result = iam_lvar_create(obj->oo_inode,
1724                                                  feat->dif_keysize_max,
1725                                                  feat->dif_ptrsize,
1726                                                  feat->dif_recsize_max,
1727                                                  oth->ot_handle);
1728                 else
1729                         result = iam_lfix_create(obj->oo_inode,
1730                                                  feat->dif_keysize_max,
1731                                                  feat->dif_ptrsize,
1732                                                  feat->dif_recsize_max,
1733                                                  oth->ot_handle);
1734
1735         }
1736         return result;
1737 }
1738
1739 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1740                      struct lu_attr *attr,
1741                      struct dt_allocation_hint *hint,
1742                      struct dt_object_format *dof,
1743                      struct thandle *th)
1744 {
1745         LASSERT(S_ISREG(attr->la_mode));
1746         return osd_mkfile(info, obj, (attr->la_mode &
1747                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1748 }
1749
1750 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1751                      struct lu_attr *attr,
1752                      struct dt_allocation_hint *hint,
1753                      struct dt_object_format *dof,
1754                      struct thandle *th)
1755 {
1756         LASSERT(S_ISLNK(attr->la_mode));
1757         return osd_mkfile(info, obj, (attr->la_mode &
1758                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1759 }
1760
1761 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1762                      struct lu_attr *attr,
1763                      struct dt_allocation_hint *hint,
1764                      struct dt_object_format *dof,
1765                      struct thandle *th)
1766 {
1767         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1768         int result;
1769
1770         LINVRNT(osd_invariant(obj));
1771         LASSERT(obj->oo_inode == NULL);
1772         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1773                 S_ISFIFO(mode) || S_ISSOCK(mode));
1774
1775         result = osd_mkfile(info, obj, mode, hint, th);
1776         if (result == 0) {
1777                 LASSERT(obj->oo_inode != NULL);
1778                 /*
1779                  * This inode should be marked dirty for i_rdev.  Currently
1780                  * that is done in the osd_attr_init().
1781                  */
1782                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1783                                    attr->la_rdev);
1784         }
1785         LINVRNT(osd_invariant(obj));
1786         return result;
1787 }
1788
1789 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1790                               struct lu_attr *,
1791                               struct dt_allocation_hint *hint,
1792                               struct dt_object_format *dof,
1793                               struct thandle *);
1794
1795 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1796 {
1797         osd_obj_type_f result;
1798
1799         switch (type) {
1800         case DFT_DIR:
1801                 result = osd_mkdir;
1802                 break;
1803         case DFT_REGULAR:
1804                 result = osd_mkreg;
1805                 break;
1806         case DFT_SYM:
1807                 result = osd_mksym;
1808                 break;
1809         case DFT_NODE:
1810                 result = osd_mknod;
1811                 break;
1812         case DFT_INDEX:
1813                 result = osd_mk_index;
1814                 break;
1815
1816         default:
1817                 LBUG();
1818                 break;
1819         }
1820         return result;
1821 }
1822
1823
1824 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1825                         struct dt_object *parent, struct dt_object *child,
1826                         cfs_umode_t child_mode)
1827 {
1828         LASSERT(ah);
1829
1830         memset(ah, 0, sizeof(*ah));
1831         ah->dah_parent = parent;
1832         ah->dah_mode = child_mode;
1833 }
1834
1835 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1836                           struct lu_attr *attr, struct dt_object_format *dof)
1837 {
1838         struct inode   *inode = obj->oo_inode;
1839         __u64           valid = attr->la_valid;
1840         int             result;
1841
1842         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1843
1844         if (dof->dof_type != DFT_NODE)
1845                 attr->la_valid &= ~LA_RDEV;
1846         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1847                 attr->la_valid &= ~LA_ATIME;
1848         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1849                 attr->la_valid &= ~LA_CTIME;
1850         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1851                 attr->la_valid &= ~LA_MTIME;
1852
1853         result = osd_quota_transfer(inode, attr);
1854         if (result)
1855                 return;
1856
1857         if (attr->la_valid != 0) {
1858                 result = osd_inode_setattr(info->oti_env, inode, attr);
1859                 /*
1860                  * The osd_inode_setattr() should always succeed here.  The
1861                  * only error that could be returned is EDQUOT when we are
1862                  * trying to change the UID or GID of the inode. However, this
1863                  * should not happen since quota enforcement is no longer
1864                  * enabled on ldiskfs (lquota takes care of it).
1865                  */
1866                 LASSERTF(result == 0, "%d", result);
1867                 inode->i_sb->s_op->dirty_inode(inode);
1868         }
1869
1870         attr->la_valid = valid;
1871 }
1872
1873 /**
1874  * Helper function for osd_object_create()
1875  *
1876  * \retval 0, on success
1877  */
1878 static int __osd_object_create(struct osd_thread_info *info,
1879                                struct osd_object *obj, struct lu_attr *attr,
1880                                struct dt_allocation_hint *hint,
1881                                struct dt_object_format *dof,
1882                                struct thandle *th)
1883 {
1884         int     result;
1885
1886         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1887                                                   th);
1888         if (result == 0) {
1889                 osd_attr_init(info, obj, attr, dof);
1890                 osd_object_init0(obj);
1891                 /* bz 24037 */
1892                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1893                         unlock_new_inode(obj->oo_inode);
1894         }
1895
1896         return result;
1897 }
1898
1899 /**
1900  * Helper function for osd_object_create()
1901  *
1902  * \retval 0, on success
1903  */
1904 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1905                            const struct lu_fid *fid, struct thandle *th)
1906 {
1907         struct osd_thread_info *info = osd_oti_get(env);
1908         struct osd_inode_id    *id   = &info->oti_id;
1909         struct osd_device      *osd  = osd_obj2dev(obj);
1910
1911         LASSERT(obj->oo_inode != NULL);
1912
1913         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1914         return osd_oi_insert(info, osd, fid, id, th);
1915 }
1916
1917 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1918                    const struct lu_fid *fid, struct lu_seq_range *range)
1919 {
1920         struct seq_server_site  *ss = osd_seq_site(osd);
1921         int                     rc;
1922
1923         if (fid_is_igif(fid)) {
1924                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1925                 range->lsr_index = 0;
1926                 return 0;
1927         }
1928
1929         if (fid_is_idif(fid)) {
1930                 range->lsr_flags = LU_SEQ_RANGE_OST;
1931                 range->lsr_index = fid_idif_ost_idx(fid);
1932                 return 0;
1933         }
1934
1935         if (!fid_is_norm(fid)) {
1936                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1937                 if (ss != NULL)
1938                         /* FIXME: If ss is NULL, it suppose not get lsr_index
1939                          * at all */
1940                         range->lsr_index = ss->ss_node_id;
1941                 return 0;
1942         }
1943
1944         LASSERT(ss != NULL);
1945         range->lsr_flags = -1;
1946         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1947         if (rc != 0) {
1948                 CERROR("%s can not find "DFID": rc = %d\n",
1949                        osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc);
1950         }
1951         return rc;
1952 }
1953
1954
1955 static int osd_declare_object_create(const struct lu_env *env,
1956                                      struct dt_object *dt,
1957                                      struct lu_attr *attr,
1958                                      struct dt_allocation_hint *hint,
1959                                      struct dt_object_format *dof,
1960                                      struct thandle *handle)
1961 {
1962         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1963         struct osd_thandle      *oh;
1964         int                      rc;
1965         ENTRY;
1966
1967         LASSERT(handle != NULL);
1968
1969         oh = container_of0(handle, struct osd_thandle, ot_super);
1970         LASSERT(oh->ot_handle == NULL);
1971
1972         OSD_DECLARE_OP(oh, create, osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
1973         /* XXX: So far, only normal fid needs be inserted into the oi,
1974          *      things could be changed later. Revise following code then. */
1975         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
1976             !fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
1977                            lu_object_fid(&dt->do_lu))) {
1978                 /* Reuse idle OI block may cause additional one OI block
1979                  * to be changed. */
1980                 OSD_DECLARE_OP(oh, insert,
1981                                osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
1982         }
1983         /* If this is directory, then we expect . and .. to be inserted as
1984          * well. The one directory block always needs to be created for the
1985          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1986          * block), there is no danger of needing a tree for the first block.
1987          */
1988         if (attr && S_ISDIR(attr->la_mode)) {
1989                 OSD_DECLARE_OP(oh, insert,
1990                                osd_dto_credits_noquota[DTO_WRITE_BASE]);
1991                 OSD_DECLARE_OP(oh, insert, 0);
1992         }
1993
1994         if (!attr)
1995                 RETURN(0);
1996
1997         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
1998                                    false, false, NULL, false);
1999         if (rc != 0)
2000                 RETURN(rc);
2001
2002         /* It does fld look up inside declare, and the result will be
2003          * added to fld cache, so the following fld lookup inside insert
2004          * does not need send RPC anymore, so avoid send rpc with holding
2005          * transaction */
2006         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2007                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2008                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2009                                lu_object_fid(&dt->do_lu), range);
2010
2011
2012         RETURN(rc);
2013 }
2014
2015 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2016                              struct lu_attr *attr,
2017                              struct dt_allocation_hint *hint,
2018                              struct dt_object_format *dof,
2019                              struct thandle *th)
2020 {
2021         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2022         struct osd_object      *obj    = osd_dt_obj(dt);
2023         struct osd_thread_info *info   = osd_oti_get(env);
2024         int result;
2025
2026         ENTRY;
2027
2028         LINVRNT(osd_invariant(obj));
2029         LASSERT(!dt_object_exists(dt));
2030         LASSERT(osd_write_locked(env, obj));
2031         LASSERT(th != NULL);
2032
2033         if (unlikely(fid_is_acct(fid)))
2034                 /* Quota files can't be created from the kernel any more,
2035                  * 'tune2fs -O quota' will take care of creating them */
2036                 RETURN(-EPERM);
2037
2038         OSD_EXEC_OP(th, create);
2039
2040         result = __osd_object_create(info, obj, attr, hint, dof, th);
2041         if (result == 0)
2042                 result = __osd_oi_insert(env, obj, fid, th);
2043
2044         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2045         LASSERT(osd_invariant(obj));
2046         RETURN(result);
2047 }
2048
2049 /**
2050  * Called to destroy on-disk representation of the object
2051  *
2052  * Concurrency: must be locked
2053  */
2054 static int osd_declare_object_destroy(const struct lu_env *env,
2055                                       struct dt_object *dt,
2056                                       struct thandle *th)
2057 {
2058         struct osd_object  *obj = osd_dt_obj(dt);
2059         struct inode       *inode = obj->oo_inode;
2060         struct osd_thandle *oh;
2061         int                 rc;
2062         ENTRY;
2063
2064         oh = container_of0(th, struct osd_thandle, ot_super);
2065         LASSERT(oh->ot_handle == NULL);
2066         LASSERT(inode);
2067
2068         OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2069         /* XXX: So far, only normal fid needs to be inserted into the OI,
2070          *      so only normal fid needs to be removed from the OI also.
2071          * Recycle idle OI leaf may cause additional three OI blocks
2072          * to be changed. */
2073         OSD_DECLARE_OP(oh, destroy, fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2074                         osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0);
2075
2076         /* one less inode */
2077         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2078                                    false, true, NULL, false);
2079         if (rc)
2080                 RETURN(rc);
2081         /* data to be truncated */
2082         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2083                                    true, true, NULL, false);
2084         RETURN(rc);
2085 }
2086
2087 static int osd_object_destroy(const struct lu_env *env,
2088                               struct dt_object *dt,
2089                               struct thandle *th)
2090 {
2091         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2092         struct osd_object      *obj = osd_dt_obj(dt);
2093         struct inode           *inode = obj->oo_inode;
2094         struct osd_device      *osd = osd_obj2dev(obj);
2095         struct osd_thandle     *oh;
2096         int                     result;
2097         ENTRY;
2098
2099         oh = container_of0(th, struct osd_thandle, ot_super);
2100         LASSERT(oh->ot_handle);
2101         LASSERT(inode);
2102         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2103
2104         if (unlikely(fid_is_acct(fid)))
2105                 RETURN(-EPERM);
2106
2107         /* Parallel control for OI scrub. For most of cases, there is no
2108          * lock contention. So it will not affect unlink performance. */
2109         mutex_lock(&inode->i_mutex);
2110         if (S_ISDIR(inode->i_mode)) {
2111                 LASSERT(osd_inode_unlinked(inode) ||
2112                         inode->i_nlink == 1);
2113                 spin_lock(&obj->oo_guard);
2114                 clear_nlink(inode);
2115                 spin_unlock(&obj->oo_guard);
2116                 inode->i_sb->s_op->dirty_inode(inode);
2117         } else {
2118                 LASSERT(osd_inode_unlinked(inode));
2119         }
2120
2121         OSD_EXEC_OP(th, destroy);
2122
2123         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2124         mutex_unlock(&inode->i_mutex);
2125
2126         /* XXX: add to ext3 orphan list */
2127         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2128
2129         /* not needed in the cache anymore */
2130         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2131
2132         RETURN(0);
2133 }
2134
2135 /**
2136  * Helper function for osd_xattr_set()
2137  */
2138 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2139                            const struct lu_buf *buf, const char *name, int fl)
2140 {
2141         struct osd_object      *obj      = osd_dt_obj(dt);
2142         struct inode           *inode    = obj->oo_inode;
2143         struct osd_thread_info *info     = osd_oti_get(env);
2144         struct dentry          *dentry   = &info->oti_child_dentry;
2145         int                     fs_flags = 0;
2146         int                     rc;
2147
2148         LASSERT(dt_object_exists(dt));
2149         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2150
2151         if (fl & LU_XATTR_REPLACE)
2152                 fs_flags |= XATTR_REPLACE;
2153
2154         if (fl & LU_XATTR_CREATE)
2155                 fs_flags |= XATTR_CREATE;
2156
2157         ll_vfs_dq_init(inode);
2158         dentry->d_inode = inode;
2159         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2160                                    buf->lb_len, fs_flags);
2161         return rc;
2162 }
2163
2164 /**
2165  * Put the fid into lustre_mdt_attrs, and then place the structure
2166  * inode's ea. This fid should not be altered during the life time
2167  * of the inode.
2168  *
2169  * \retval +ve, on success
2170  * \retval -ve, on error
2171  *
2172  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2173  */
2174 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2175                           const struct lu_fid *fid)
2176 {
2177         struct osd_thread_info  *info      = osd_oti_get(env);
2178         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2179
2180         lustre_lma_init(mdt_attrs, fid);
2181         lustre_lma_swab(mdt_attrs);
2182         return __osd_xattr_set(env, dt,
2183                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2184                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2185
2186 }
2187
2188 /**
2189  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2190  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2191  * To have compatilibility with 1.8 ldiskfs driver we need to have
2192  * magic number at start of fid data.
2193  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2194  * its inmemory API.
2195  */
2196 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2197                                   const struct dt_rec *fid)
2198 {
2199         param->edp_magic = LDISKFS_LUFID_MAGIC;
2200         param->edp_len =  sizeof(struct lu_fid) + 1;
2201
2202         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2203                       (struct lu_fid *)fid);
2204 }
2205
2206 /**
2207  * Try to read the fid from inode ea into dt_rec, if return value
2208  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2209  *
2210  * \param fid object fid.
2211  *
2212  * \retval 0 on success
2213  */
2214 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2215                           __u32 ino, struct lu_fid *fid,
2216                           struct osd_inode_id *id)
2217 {
2218         struct osd_thread_info *info  = osd_oti_get(env);
2219         struct inode           *inode;
2220         ENTRY;
2221
2222         osd_id_gen(id, ino, OSD_OII_NOGEN);
2223         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2224         if (IS_ERR(inode))
2225                 RETURN(PTR_ERR(inode));
2226
2227         iput(inode);
2228         RETURN(0);
2229 }
2230
2231 /**
2232  * OSD layer object create function for interoperability mode (b11826).
2233  * This is mostly similar to osd_object_create(). Only difference being, fid is
2234  * inserted into inode ea here.
2235  *
2236  * \retval   0, on success
2237  * \retval -ve, on error
2238  */
2239 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2240                                 struct lu_attr *attr,
2241                                 struct dt_allocation_hint *hint,
2242                                 struct dt_object_format *dof,
2243                                 struct thandle *th)
2244 {
2245         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2246         struct osd_object      *obj    = osd_dt_obj(dt);
2247         struct osd_thread_info *info   = osd_oti_get(env);
2248         int                     result;
2249
2250         ENTRY;
2251
2252         LASSERT(osd_invariant(obj));
2253         LASSERT(!dt_object_exists(dt));
2254         LASSERT(osd_write_locked(env, obj));
2255         LASSERT(th != NULL);
2256
2257         if (unlikely(fid_is_acct(fid)))
2258                 /* Quota files can't be created from the kernel any more,
2259                  * 'tune2fs -O quota' will take care of creating them */
2260                 RETURN(-EPERM);
2261
2262         OSD_EXEC_OP(th, create);
2263
2264         result = __osd_object_create(info, obj, attr, hint, dof, th);
2265         /* objects under osd root shld have igif fid, so dont add fid EA */
2266         /* For ost object, the fid will be stored during first write */
2267         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL &&
2268             !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid))
2269                 result = osd_ea_fid_set(env, dt, fid);
2270
2271         if (result == 0)
2272                 result = __osd_oi_insert(env, obj, fid, th);
2273
2274         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2275         LINVRNT(osd_invariant(obj));
2276         RETURN(result);
2277 }
2278
2279 static int osd_declare_object_ref_add(const struct lu_env *env,
2280                                       struct dt_object *dt,
2281                                       struct thandle *handle)
2282 {
2283         struct osd_thandle       *oh;
2284
2285         /* it's possible that object doesn't exist yet */
2286         LASSERT(handle != NULL);
2287
2288         oh = container_of0(handle, struct osd_thandle, ot_super);
2289         LASSERT(oh->ot_handle == NULL);
2290
2291         OSD_DECLARE_OP(oh, ref_add, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2292
2293         return 0;
2294 }
2295
2296 /*
2297  * Concurrency: @dt is write locked.
2298  */
2299 static int osd_object_ref_add(const struct lu_env *env,
2300                               struct dt_object *dt, struct thandle *th)
2301 {
2302         struct osd_object *obj = osd_dt_obj(dt);
2303         struct inode      *inode = obj->oo_inode;
2304
2305         LINVRNT(osd_invariant(obj));
2306         LASSERT(dt_object_exists(dt));
2307         LASSERT(osd_write_locked(env, obj));
2308         LASSERT(th != NULL);
2309
2310         OSD_EXEC_OP(th, ref_add);
2311
2312         /*
2313          * DIR_NLINK feature is set for compatibility reasons if:
2314          * 1) nlinks > LDISKFS_LINK_MAX, or
2315          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2316          *
2317          * It is easier to always set this flag (rather than check and set),
2318          * since it has less overhead, and the superblock will be dirtied
2319          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2320          * do not actually care whether this flag is set or not.
2321          */
2322         spin_lock(&obj->oo_guard);
2323         /* inc_nlink from 0 may cause WARN_ON */
2324         if(inode->i_nlink == 0)
2325                 set_nlink(inode, 1);
2326         else
2327                 inc_nlink(inode);
2328         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2329                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2330                     inode->i_nlink == 2)
2331                         set_nlink(inode, 1);
2332         }
2333         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2334         spin_unlock(&obj->oo_guard);
2335         inode->i_sb->s_op->dirty_inode(inode);
2336         LINVRNT(osd_invariant(obj));
2337
2338         return 0;
2339 }
2340
2341 static int osd_declare_object_ref_del(const struct lu_env *env,
2342                                       struct dt_object *dt,
2343                                       struct thandle *handle)
2344 {
2345         struct osd_thandle *oh;
2346
2347         LASSERT(dt_object_exists(dt));
2348         LASSERT(handle != NULL);
2349
2350         oh = container_of0(handle, struct osd_thandle, ot_super);
2351         LASSERT(oh->ot_handle == NULL);
2352
2353         OSD_DECLARE_OP(oh, ref_del, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2354
2355         return 0;
2356 }
2357
2358 /*
2359  * Concurrency: @dt is write locked.
2360  */
2361 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2362                               struct thandle *th)
2363 {
2364         struct osd_object *obj = osd_dt_obj(dt);
2365         struct inode      *inode = obj->oo_inode;
2366
2367         LINVRNT(osd_invariant(obj));
2368         LASSERT(dt_object_exists(dt));
2369         LASSERT(osd_write_locked(env, obj));
2370         LASSERT(th != NULL);
2371
2372         OSD_EXEC_OP(th, ref_del);
2373
2374         spin_lock(&obj->oo_guard);
2375         LASSERT(inode->i_nlink > 0);
2376         drop_nlink(inode);
2377         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2378          * then the nlink count is 1. Don't let it be set to 0 or the directory
2379          * inode will be deleted incorrectly. */
2380         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2381                 set_nlink(inode, 1);
2382         spin_unlock(&obj->oo_guard);
2383         inode->i_sb->s_op->dirty_inode(inode);
2384         LINVRNT(osd_invariant(obj));
2385
2386         return 0;
2387 }
2388
2389 /*
2390  * Get the 64-bit version for an inode.
2391  */
2392 static int osd_object_version_get(const struct lu_env *env,
2393                                   struct dt_object *dt, dt_obj_version_t *ver)
2394 {
2395         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2396
2397         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2398                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2399         *ver = LDISKFS_I(inode)->i_fs_version;
2400         return 0;
2401 }
2402
2403 /*
2404  * Concurrency: @dt is read locked.
2405  */
2406 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2407                          struct lu_buf *buf, const char *name,
2408                          struct lustre_capa *capa)
2409 {
2410         struct osd_object      *obj    = osd_dt_obj(dt);
2411         struct inode           *inode  = obj->oo_inode;
2412         struct osd_thread_info *info   = osd_oti_get(env);
2413         struct dentry          *dentry = &info->oti_obj_dentry;
2414
2415         /* version get is not real XATTR but uses xattr API */
2416         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2417                 /* for version we are just using xattr API but change inode
2418                  * field instead */
2419                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2420                 osd_object_version_get(env, dt, buf->lb_buf);
2421                 return sizeof(dt_obj_version_t);
2422         }
2423
2424         LASSERT(dt_object_exists(dt));
2425         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2426
2427         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2428                 return -EACCES;
2429
2430         dentry->d_inode = inode;
2431         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2432 }
2433
2434
2435 static int osd_declare_xattr_set(const struct lu_env *env,
2436                                  struct dt_object *dt,
2437                                  const struct lu_buf *buf, const char *name,
2438                                  int fl, struct thandle *handle)
2439 {
2440         struct osd_thandle *oh;
2441
2442         LASSERT(handle != NULL);
2443
2444         oh = container_of0(handle, struct osd_thandle, ot_super);
2445         LASSERT(oh->ot_handle == NULL);
2446
2447         OSD_DECLARE_OP(oh, xattr_set, strcmp(name, XATTR_NAME_VERSION) == 0 ?
2448                        osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2449                        osd_dto_credits_noquota[DTO_XATTR_SET]);
2450
2451         return 0;
2452 }
2453
2454 /*
2455  * Set the 64-bit version for object
2456  */
2457 static void osd_object_version_set(const struct lu_env *env,
2458                                    struct dt_object *dt,
2459                                    dt_obj_version_t *new_version)
2460 {
2461         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2462
2463         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2464                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2465
2466         LDISKFS_I(inode)->i_fs_version = *new_version;
2467         /** Version is set after all inode operations are finished,
2468          *  so we should mark it dirty here */
2469         inode->i_sb->s_op->dirty_inode(inode);
2470 }
2471
2472 /*
2473  * Concurrency: @dt is write locked.
2474  */
2475 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2476                          const struct lu_buf *buf, const char *name, int fl,
2477                          struct thandle *handle, struct lustre_capa *capa)
2478 {
2479         LASSERT(handle != NULL);
2480
2481         /* version set is not real XATTR */
2482         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2483                 /* for version we are just using xattr API but change inode
2484                  * field instead */
2485                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2486                 osd_object_version_set(env, dt, buf->lb_buf);
2487                 return sizeof(dt_obj_version_t);
2488         }
2489
2490         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2491                 return -EACCES;
2492
2493         OSD_EXEC_OP(handle, xattr_set);
2494         return __osd_xattr_set(env, dt, buf, name, fl);
2495 }
2496
2497 /*
2498  * Concurrency: @dt is read locked.
2499  */
2500 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2501                           struct lu_buf *buf, struct lustre_capa *capa)
2502 {
2503         struct osd_object      *obj    = osd_dt_obj(dt);
2504         struct inode           *inode  = obj->oo_inode;
2505         struct osd_thread_info *info   = osd_oti_get(env);
2506         struct dentry          *dentry = &info->oti_obj_dentry;
2507
2508         LASSERT(dt_object_exists(dt));
2509         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2510         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2511
2512         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2513                 return -EACCES;
2514
2515         dentry->d_inode = inode;
2516         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2517 }
2518
2519 static int osd_declare_xattr_del(const struct lu_env *env,
2520                                  struct dt_object *dt, const char *name,
2521                                  struct thandle *handle)
2522 {
2523         struct osd_thandle *oh;
2524
2525         LASSERT(dt_object_exists(dt));
2526         LASSERT(handle != NULL);
2527
2528         oh = container_of0(handle, struct osd_thandle, ot_super);
2529         LASSERT(oh->ot_handle == NULL);
2530
2531         OSD_DECLARE_OP(oh, xattr_set, osd_dto_credits_noquota[DTO_XATTR_SET]);
2532
2533         return 0;
2534 }
2535
2536 /*
2537  * Concurrency: @dt is write locked.
2538  */
2539 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2540                          const char *name, struct thandle *handle,
2541                          struct lustre_capa *capa)
2542 {
2543         struct osd_object      *obj    = osd_dt_obj(dt);
2544         struct inode           *inode  = obj->oo_inode;
2545         struct osd_thread_info *info   = osd_oti_get(env);
2546         struct dentry          *dentry = &info->oti_obj_dentry;
2547         int                     rc;
2548
2549         LASSERT(dt_object_exists(dt));
2550         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2551         LASSERT(osd_write_locked(env, obj));
2552         LASSERT(handle != NULL);
2553
2554         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2555                 return -EACCES;
2556
2557         OSD_EXEC_OP(handle, xattr_set);
2558
2559         ll_vfs_dq_init(inode);
2560         dentry->d_inode = inode;
2561         rc = inode->i_op->removexattr(dentry, name);
2562         return rc;
2563 }
2564
2565 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2566                                      struct dt_object *dt,
2567                                      struct lustre_capa *old,
2568                                      __u64 opc)
2569 {
2570         struct osd_thread_info *info = osd_oti_get(env);
2571         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2572         struct osd_object *obj = osd_dt_obj(dt);
2573         struct osd_device *dev = osd_obj2dev(obj);
2574         struct lustre_capa_key *key = &info->oti_capa_key;
2575         struct lustre_capa *capa = &info->oti_capa;
2576         struct obd_capa *oc;
2577         struct md_capainfo *ci;
2578         int rc;
2579         ENTRY;
2580
2581         if (!dev->od_fl_capa)
2582                 RETURN(ERR_PTR(-ENOENT));
2583
2584         LASSERT(dt_object_exists(dt));
2585         LINVRNT(osd_invariant(obj));
2586
2587         /* renewal sanity check */
2588         if (old && osd_object_auth(env, dt, old, opc))
2589                 RETURN(ERR_PTR(-EACCES));
2590
2591         ci = md_capainfo(env);
2592         if (unlikely(!ci))
2593                 RETURN(ERR_PTR(-ENOENT));
2594
2595         switch (ci->mc_auth) {
2596         case LC_ID_NONE:
2597                 RETURN(NULL);
2598         case LC_ID_PLAIN:
2599                 capa->lc_uid = obj->oo_inode->i_uid;
2600                 capa->lc_gid = obj->oo_inode->i_gid;
2601                 capa->lc_flags = LC_ID_PLAIN;
2602                 break;
2603         case LC_ID_CONVERT: {
2604                 __u32 d[4], s[4];
2605
2606                 s[0] = obj->oo_inode->i_uid;
2607                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2608                 s[2] = obj->oo_inode->i_gid;
2609                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2610                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2611                 if (unlikely(rc))
2612                         RETURN(ERR_PTR(rc));
2613
2614                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2615                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2616                 capa->lc_flags = LC_ID_CONVERT;
2617                 break;
2618         }
2619         default:
2620                 RETURN(ERR_PTR(-EINVAL));
2621         }
2622
2623         capa->lc_fid = *fid;
2624         capa->lc_opc = opc;
2625         capa->lc_flags |= dev->od_capa_alg << 24;
2626         capa->lc_timeout = dev->od_capa_timeout;
2627         capa->lc_expiry = 0;
2628
2629         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2630         if (oc) {
2631                 LASSERT(!capa_is_expired(oc));
2632                 RETURN(oc);
2633         }
2634
2635         spin_lock(&capa_lock);
2636         *key = dev->od_capa_keys[1];
2637         spin_unlock(&capa_lock);
2638
2639         capa->lc_keyid = key->lk_keyid;
2640         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2641
2642         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2643         if (rc) {
2644                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2645                 RETURN(ERR_PTR(rc));
2646         }
2647
2648         oc = capa_add(dev->od_capa_hash, capa);
2649         RETURN(oc);
2650 }
2651
2652 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2653 {
2654         struct osd_object       *obj    = osd_dt_obj(dt);
2655         struct inode            *inode  = obj->oo_inode;
2656         struct osd_thread_info  *info   = osd_oti_get(env);
2657         struct dentry           *dentry = &info->oti_obj_dentry;
2658         struct file             *file   = &info->oti_file;
2659         int                     rc;
2660
2661         ENTRY;
2662
2663         dentry->d_inode = inode;
2664         file->f_dentry = dentry;
2665         file->f_mapping = inode->i_mapping;
2666         file->f_op = inode->i_fop;
2667         mutex_lock(&inode->i_mutex);
2668         rc = file->f_op->fsync(file, dentry, 0);
2669         mutex_unlock(&inode->i_mutex);
2670         RETURN(rc);
2671 }
2672
2673 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2674                         void **data)
2675 {
2676         struct osd_object *obj = osd_dt_obj(dt);
2677         ENTRY;
2678
2679         *data = (void *)obj->oo_inode;
2680         RETURN(0);
2681 }
2682
2683 /*
2684  * Index operations.
2685  */
2686
2687 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2688                            const struct dt_index_features *feat)
2689 {
2690         struct iam_descr *descr;
2691
2692         if (osd_object_is_root(o))
2693                 return feat == &dt_directory_features;
2694
2695         LASSERT(o->oo_dir != NULL);
2696
2697         descr = o->oo_dir->od_container.ic_descr;
2698         if (feat == &dt_directory_features) {
2699                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2700                         return 1;
2701                 else
2702                         return 0;
2703         } else {
2704                 return
2705                         feat->dif_keysize_min <= descr->id_key_size &&
2706                         descr->id_key_size <= feat->dif_keysize_max &&
2707                         feat->dif_recsize_min <= descr->id_rec_size &&
2708                         descr->id_rec_size <= feat->dif_recsize_max &&
2709                         !(feat->dif_flags & (DT_IND_VARKEY |
2710                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2711                         ergo(feat->dif_flags & DT_IND_UPDATE,
2712                              1 /* XXX check that object (and file system) is
2713                                 * writable */);
2714         }
2715 }
2716
2717 static int osd_iam_container_init(const struct lu_env *env,
2718                                   struct osd_object *obj,
2719                                   struct osd_directory *dir)
2720 {
2721         struct iam_container *bag = &dir->od_container;
2722         int result;
2723
2724         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2725         if (result != 0)
2726                 return result;
2727
2728         result = iam_container_setup(bag);
2729         if (result == 0)
2730                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2731         else
2732                 iam_container_fini(bag);
2733
2734         return result;
2735 }
2736
2737
2738 /*
2739  * Concurrency: no external locking is necessary.
2740  */
2741 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2742                          const struct dt_index_features *feat)
2743 {
2744         int                      result;
2745         int                      skip_iam = 0;
2746         struct osd_object       *obj = osd_dt_obj(dt);
2747
2748         LINVRNT(osd_invariant(obj));
2749         LASSERT(dt_object_exists(dt));
2750
2751         if (osd_object_is_root(obj)) {
2752                 dt->do_index_ops = &osd_index_ea_ops;
2753                 result = 0;
2754         } else if (feat == &dt_directory_features) {
2755                 dt->do_index_ops = &osd_index_ea_ops;
2756                 if (S_ISDIR(obj->oo_inode->i_mode))
2757                         result = 0;
2758                 else
2759                         result = -ENOTDIR;
2760                 skip_iam = 1;
2761         } else if (unlikely(feat == &dt_otable_features)) {
2762                 dt->do_index_ops = &osd_otable_ops;
2763                 return 0;
2764         } else if (feat == &dt_acct_features) {
2765                 dt->do_index_ops = &osd_acct_index_ops;
2766                 result = 0;
2767                 skip_iam = 1;
2768         } else if (!osd_has_index(obj)) {
2769                 struct osd_directory *dir;
2770
2771                 OBD_ALLOC_PTR(dir);
2772                 if (dir != NULL) {
2773
2774                         spin_lock(&obj->oo_guard);
2775                         if (obj->oo_dir == NULL)
2776                                 obj->oo_dir = dir;
2777                         else
2778                                 /*
2779                                  * Concurrent thread allocated container data.
2780                                  */
2781                                 OBD_FREE_PTR(dir);
2782                         spin_unlock(&obj->oo_guard);
2783                         /*
2784                          * Now, that we have container data, serialize its
2785                          * initialization.
2786                          */
2787                         down_write(&obj->oo_ext_idx_sem);
2788                         /*
2789                          * recheck under lock.
2790                          */
2791                         if (!osd_has_index(obj))
2792                                 result = osd_iam_container_init(env, obj, dir);
2793                         else
2794                                 result = 0;
2795                         up_write(&obj->oo_ext_idx_sem);
2796                 } else {
2797                         result = -ENOMEM;
2798                 }
2799         } else {
2800                 result = 0;
2801         }
2802
2803         if (result == 0 && skip_iam == 0) {
2804                 if (!osd_iam_index_probe(env, obj, feat))
2805                         result = -ENOTDIR;
2806         }
2807         LINVRNT(osd_invariant(obj));
2808
2809         if (is_quota_glb_feat(feat))
2810                 result = osd_quota_migration(env, dt, feat);
2811
2812         return result;
2813 }
2814
2815 static int osd_otable_it_attr_get(const struct lu_env *env,
2816                                  struct dt_object *dt,
2817                                  struct lu_attr *attr,
2818                                  struct lustre_capa *capa)
2819 {
2820         attr->la_valid = 0;
2821         return 0;
2822 }
2823
2824 static const struct dt_object_operations osd_obj_ops = {
2825         .do_read_lock         = osd_object_read_lock,
2826         .do_write_lock        = osd_object_write_lock,
2827         .do_read_unlock       = osd_object_read_unlock,
2828         .do_write_unlock      = osd_object_write_unlock,
2829         .do_write_locked      = osd_object_write_locked,
2830         .do_attr_get          = osd_attr_get,
2831         .do_declare_attr_set  = osd_declare_attr_set,
2832         .do_attr_set          = osd_attr_set,
2833         .do_ah_init           = osd_ah_init,
2834         .do_declare_create    = osd_declare_object_create,
2835         .do_create            = osd_object_create,
2836         .do_declare_destroy   = osd_declare_object_destroy,
2837         .do_destroy           = osd_object_destroy,
2838         .do_index_try         = osd_index_try,
2839         .do_declare_ref_add   = osd_declare_object_ref_add,
2840         .do_ref_add           = osd_object_ref_add,
2841         .do_declare_ref_del   = osd_declare_object_ref_del,
2842         .do_ref_del           = osd_object_ref_del,
2843         .do_xattr_get         = osd_xattr_get,
2844         .do_declare_xattr_set = osd_declare_xattr_set,
2845         .do_xattr_set         = osd_xattr_set,
2846         .do_declare_xattr_del = osd_declare_xattr_del,
2847         .do_xattr_del         = osd_xattr_del,
2848         .do_xattr_list        = osd_xattr_list,
2849         .do_capa_get          = osd_capa_get,
2850         .do_object_sync       = osd_object_sync,
2851         .do_data_get          = osd_data_get,
2852 };
2853
2854 /**
2855  * dt_object_operations for interoperability mode
2856  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2857  */
2858 static const struct dt_object_operations osd_obj_ea_ops = {
2859         .do_read_lock         = osd_object_read_lock,
2860         .do_write_lock        = osd_object_write_lock,
2861         .do_read_unlock       = osd_object_read_unlock,
2862         .do_write_unlock      = osd_object_write_unlock,
2863         .do_write_locked      = osd_object_write_locked,
2864         .do_attr_get          = osd_attr_get,
2865         .do_declare_attr_set  = osd_declare_attr_set,
2866         .do_attr_set          = osd_attr_set,
2867         .do_ah_init           = osd_ah_init,
2868         .do_declare_create    = osd_declare_object_create,
2869         .do_create            = osd_object_ea_create,
2870         .do_declare_destroy   = osd_declare_object_destroy,
2871         .do_destroy           = osd_object_destroy,
2872         .do_index_try         = osd_index_try,
2873         .do_declare_ref_add   = osd_declare_object_ref_add,
2874         .do_ref_add           = osd_object_ref_add,
2875         .do_declare_ref_del   = osd_declare_object_ref_del,
2876         .do_ref_del           = osd_object_ref_del,
2877         .do_xattr_get         = osd_xattr_get,
2878         .do_declare_xattr_set = osd_declare_xattr_set,
2879         .do_xattr_set         = osd_xattr_set,
2880         .do_declare_xattr_del = osd_declare_xattr_del,
2881         .do_xattr_del         = osd_xattr_del,
2882         .do_xattr_list        = osd_xattr_list,
2883         .do_capa_get          = osd_capa_get,
2884         .do_object_sync       = osd_object_sync,
2885         .do_data_get          = osd_data_get,
2886 };
2887
2888 static const struct dt_object_operations osd_obj_otable_it_ops = {
2889         .do_attr_get    = osd_otable_it_attr_get,
2890         .do_index_try   = osd_index_try,
2891 };
2892
2893 static int osd_index_declare_iam_delete(const struct lu_env *env,
2894                                         struct dt_object *dt,
2895                                         const struct dt_key *key,
2896                                         struct thandle *handle)
2897 {
2898         struct osd_thandle    *oh;
2899
2900         oh = container_of0(handle, struct osd_thandle, ot_super);
2901         LASSERT(oh->ot_handle == NULL);
2902
2903         OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2904
2905         return 0;
2906 }
2907
2908 /**
2909  *      delete a (key, value) pair from index \a dt specified by \a key
2910  *
2911  *      \param  dt      osd index object
2912  *      \param  key     key for index
2913  *      \param  rec     record reference
2914  *      \param  handle  transaction handler
2915  *
2916  *      \retval  0  success
2917  *      \retval -ve   failure
2918  */
2919
2920 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2921                                 const struct dt_key *key,
2922                                 struct thandle *handle,
2923                                 struct lustre_capa *capa)
2924 {
2925         struct osd_thread_info *oti = osd_oti_get(env);
2926         struct osd_object      *obj = osd_dt_obj(dt);
2927         struct osd_thandle     *oh;
2928         struct iam_path_descr  *ipd;
2929         struct iam_container   *bag = &obj->oo_dir->od_container;
2930         int                     rc;
2931
2932         ENTRY;
2933
2934         LINVRNT(osd_invariant(obj));
2935         LASSERT(dt_object_exists(dt));
2936         LASSERT(bag->ic_object == obj->oo_inode);
2937         LASSERT(handle != NULL);
2938
2939         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2940                 RETURN(-EACCES);
2941
2942         OSD_EXEC_OP(handle, delete);
2943
2944         ipd = osd_idx_ipd_get(env, bag);
2945         if (unlikely(ipd == NULL))
2946                 RETURN(-ENOMEM);
2947
2948         oh = container_of0(handle, struct osd_thandle, ot_super);
2949         LASSERT(oh->ot_handle != NULL);
2950         LASSERT(oh->ot_handle->h_transaction != NULL);
2951
2952         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
2953                 /* swab quota uid/gid provided by caller */
2954                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
2955                 key = (const struct dt_key *)&oti->oti_quota_id;
2956         }
2957
2958         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2959         osd_ipd_put(env, bag, ipd);
2960         LINVRNT(osd_invariant(obj));
2961         RETURN(rc);
2962 }
2963
2964 static int osd_index_declare_ea_delete(const struct lu_env *env,
2965                                        struct dt_object *dt,
2966                                        const struct dt_key *key,
2967                                        struct thandle *handle)
2968 {
2969         struct osd_thandle *oh;
2970         struct inode       *inode;
2971         int                 rc;
2972         ENTRY;
2973
2974         LASSERT(dt_object_exists(dt));
2975         LASSERT(handle != NULL);
2976
2977         oh = container_of0(handle, struct osd_thandle, ot_super);
2978         LASSERT(oh->ot_handle == NULL);
2979
2980         OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2981
2982         inode = osd_dt_obj(dt)->oo_inode;
2983         LASSERT(inode);
2984
2985         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2986                                    true, true, NULL, false);
2987         RETURN(rc);
2988 }
2989
2990 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2991                                           struct dt_rec *fid)
2992 {
2993         struct osd_fid_pack *rec;
2994         int                  rc = -ENODATA;
2995
2996         if (de->file_type & LDISKFS_DIRENT_LUFID) {
2997                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2998                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2999         }
3000         RETURN(rc);
3001 }
3002
3003 /**
3004  * Index delete function for interoperability mode (b11826).
3005  * It will remove the directory entry added by osd_index_ea_insert().
3006  * This entry is needed to maintain name->fid mapping.
3007  *
3008  * \param key,  key i.e. file entry to be deleted
3009  *
3010  * \retval   0, on success
3011  * \retval -ve, on error
3012  */
3013 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3014                                const struct dt_key *key,
3015                                struct thandle *handle,
3016                                struct lustre_capa *capa)
3017 {
3018         struct osd_object          *obj    = osd_dt_obj(dt);
3019         struct inode               *dir    = obj->oo_inode;
3020         struct dentry              *dentry;
3021         struct osd_thandle         *oh;
3022         struct ldiskfs_dir_entry_2 *de;
3023         struct buffer_head         *bh;
3024         struct htree_lock          *hlock = NULL;
3025         int                         rc;
3026
3027         ENTRY;
3028
3029         LINVRNT(osd_invariant(obj));
3030         LASSERT(dt_object_exists(dt));
3031         LASSERT(handle != NULL);
3032
3033         OSD_EXEC_OP(handle, delete);
3034
3035         oh = container_of(handle, struct osd_thandle, ot_super);
3036         LASSERT(oh->ot_handle != NULL);
3037         LASSERT(oh->ot_handle->h_transaction != NULL);
3038
3039         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3040                 RETURN(-EACCES);
3041
3042         ll_vfs_dq_init(dir);
3043         dentry = osd_child_dentry_get(env, obj,
3044                                       (char *)key, strlen((char *)key));
3045
3046         if (obj->oo_hl_head != NULL) {
3047                 hlock = osd_oti_get(env)->oti_hlock;
3048                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3049                                    dir, LDISKFS_HLOCK_DEL);
3050         } else {
3051                 down_write(&obj->oo_ext_idx_sem);
3052         }
3053
3054         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3055         if (bh) {
3056                 rc = ldiskfs_delete_entry(oh->ot_handle,
3057                                           dir, de, bh);
3058                 brelse(bh);
3059         } else {
3060                 rc = -ENOENT;
3061         }
3062         if (hlock != NULL)
3063                 ldiskfs_htree_unlock(hlock);
3064         else
3065                 up_write(&obj->oo_ext_idx_sem);
3066
3067         LASSERT(osd_invariant(obj));
3068         RETURN(rc);
3069 }
3070
3071 /**
3072  *      Lookup index for \a key and copy record to \a rec.
3073  *
3074  *      \param  dt      osd index object
3075  *      \param  key     key for index
3076  *      \param  rec     record reference
3077  *
3078  *      \retval  +ve  success : exact mach
3079  *      \retval  0    return record with key not greater than \a key
3080  *      \retval -ve   failure
3081  */
3082 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3083                                 struct dt_rec *rec, const struct dt_key *key,
3084                                 struct lustre_capa *capa)
3085 {
3086         struct osd_object      *obj = osd_dt_obj(dt);
3087         struct iam_path_descr  *ipd;
3088         struct iam_container   *bag = &obj->oo_dir->od_container;
3089         struct osd_thread_info *oti = osd_oti_get(env);
3090         struct iam_iterator    *it = &oti->oti_idx_it;
3091         struct iam_rec         *iam_rec;
3092         int                     rc;
3093
3094         ENTRY;
3095
3096         LASSERT(osd_invariant(obj));
3097         LASSERT(dt_object_exists(dt));
3098         LASSERT(bag->ic_object == obj->oo_inode);
3099
3100         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3101                 RETURN(-EACCES);
3102
3103         ipd = osd_idx_ipd_get(env, bag);
3104         if (IS_ERR(ipd))
3105                 RETURN(-ENOMEM);
3106
3107         /* got ipd now we can start iterator. */
3108         iam_it_init(it, bag, 0, ipd);
3109
3110         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3111                 /* swab quota uid/gid provided by caller */
3112                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3113                 key = (const struct dt_key *)&oti->oti_quota_id;
3114         }
3115
3116         rc = iam_it_get(it, (struct iam_key *)key);
3117         if (rc >= 0) {
3118                 if (S_ISDIR(obj->oo_inode->i_mode))
3119                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3120                 else
3121                         iam_rec = (struct iam_rec *) rec;
3122
3123                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3124
3125                 if (S_ISDIR(obj->oo_inode->i_mode))
3126                         osd_fid_unpack((struct lu_fid *) rec,
3127                                        (struct osd_fid_pack *)iam_rec);
3128                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3129                         osd_quota_unpack(obj, rec);
3130         }
3131
3132         iam_it_put(it);
3133         iam_it_fini(it);
3134         osd_ipd_put(env, bag, ipd);
3135
3136         LINVRNT(osd_invariant(obj));
3137
3138         RETURN(rc);
3139 }
3140
3141 static int osd_index_declare_iam_insert(const struct lu_env *env,
3142                                         struct dt_object *dt,
3143                                         const struct dt_rec *rec,
3144                                         const struct dt_key *key,
3145                                         struct thandle *handle)
3146 {
3147         struct osd_thandle *oh;
3148
3149         LASSERT(dt_object_exists(dt));
3150         LASSERT(handle != NULL);
3151
3152         oh = container_of0(handle, struct osd_thandle, ot_super);
3153         LASSERT(oh->ot_handle == NULL);
3154
3155         OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3156
3157         return 0;
3158 }
3159
3160 /**
3161  *      Inserts (key, value) pair in \a dt index object.
3162  *
3163  *      \param  dt      osd index object
3164  *      \param  key     key for index
3165  *      \param  rec     record reference
3166  *      \param  th      transaction handler
3167  *
3168  *      \retval  0  success
3169  *      \retval -ve failure
3170  */
3171 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3172                                 const struct dt_rec *rec,
3173                                 const struct dt_key *key, struct thandle *th,
3174                                 struct lustre_capa *capa, int ignore_quota)
3175 {
3176         struct osd_object     *obj = osd_dt_obj(dt);
3177         struct iam_path_descr *ipd;
3178         struct osd_thandle    *oh;
3179         struct iam_container  *bag = &obj->oo_dir->od_container;
3180         struct osd_thread_info *oti = osd_oti_get(env);
3181         struct iam_rec         *iam_rec;
3182         int                     rc;
3183
3184         ENTRY;
3185
3186         LINVRNT(osd_invariant(obj));
3187         LASSERT(dt_object_exists(dt));
3188         LASSERT(bag->ic_object == obj->oo_inode);
3189         LASSERT(th != NULL);
3190
3191         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3192                 RETURN(-EACCES);
3193
3194         OSD_EXEC_OP(th, insert);
3195
3196         ipd = osd_idx_ipd_get(env, bag);
3197         if (unlikely(ipd == NULL))
3198                 RETURN(-ENOMEM);
3199
3200         oh = container_of0(th, struct osd_thandle, ot_super);
3201         LASSERT(oh->ot_handle != NULL);
3202         LASSERT(oh->ot_handle->h_transaction != NULL);
3203         if (S_ISDIR(obj->oo_inode->i_mode)) {
3204                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3205                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3206         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3207                 /* pack quota uid/gid */
3208                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3209                 key = (const struct dt_key *)&oti->oti_quota_id;
3210                 /* pack quota record */
3211                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3212                 iam_rec = (struct iam_rec *)rec;
3213         } else {
3214                 iam_rec = (struct iam_rec *)rec;
3215         }
3216
3217         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3218                         iam_rec, ipd);
3219         osd_ipd_put(env, bag, ipd);
3220         LINVRNT(osd_invariant(obj));
3221         RETURN(rc);
3222 }
3223
3224 /**
3225  * Calls ldiskfs_add_entry() to add directory entry
3226  * into the directory. This is required for
3227  * interoperability mode (b11826)
3228  *
3229  * \retval   0, on success
3230  * \retval -ve, on error
3231  */
3232 static int __osd_ea_add_rec(struct osd_thread_info *info,
3233                             struct osd_object *pobj, struct inode  *cinode,
3234                             const char *name, const struct dt_rec *fid,
3235                             struct htree_lock *hlock, struct thandle *th)
3236 {
3237         struct ldiskfs_dentry_param *ldp;
3238         struct dentry               *child;
3239         struct osd_thandle          *oth;
3240         int                          rc;
3241
3242         oth = container_of(th, struct osd_thandle, ot_super);
3243         LASSERT(oth->ot_handle != NULL);
3244         LASSERT(oth->ot_handle->h_transaction != NULL);
3245
3246         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3247
3248         /* XXX: remove fid_is_igif() check here.
3249          * IGIF check is just to handle insertion of .. when it is 'ROOT',
3250          * it is IGIF now but needs FID in dir entry as well for readdir
3251          * to work.
3252          * LU-838 should fix that and remove fid_is_igif() check */
3253         if (fid_is_igif((struct lu_fid *)fid) ||
3254             fid_is_norm((struct lu_fid *)fid)) {
3255                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3256                 osd_get_ldiskfs_dirent_param(ldp, fid);
3257                 child->d_fsdata = (void *)ldp;
3258         } else {
3259                 child->d_fsdata = NULL;
3260         }
3261         LASSERT(pobj->oo_inode);
3262         ll_vfs_dq_init(pobj->oo_inode);
3263         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3264
3265         RETURN(rc);
3266 }
3267
3268 /**
3269  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3270  * into the directory.Also sets flags into osd object to
3271  * indicate dot and dotdot are created. This is required for
3272  * interoperability mode (b11826)
3273  *
3274  * \param dir   directory for dot and dotdot fixup.
3275  * \param obj   child object for linking
3276  *
3277  * \retval   0, on success
3278  * \retval -ve, on error
3279  */
3280 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3281                               struct osd_object *dir,
3282                               struct inode  *parent_dir, const char *name,
3283                               const struct dt_rec *dot_fid,
3284                               const struct dt_rec *dot_dot_fid,
3285                               struct thandle *th)
3286 {
3287         struct inode                *inode = dir->oo_inode;
3288         struct ldiskfs_dentry_param *dot_ldp;
3289         struct ldiskfs_dentry_param *dot_dot_ldp;
3290         struct osd_thandle          *oth;
3291         int result = 0;
3292
3293         oth = container_of(th, struct osd_thandle, ot_super);
3294         LASSERT(oth->ot_handle->h_transaction != NULL);
3295         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3296
3297         if (strcmp(name, dot) == 0) {
3298                 if (dir->oo_compat_dot_created) {
3299                         result = -EEXIST;
3300                 } else {
3301                         LASSERT(inode == parent_dir);
3302                         dir->oo_compat_dot_created = 1;
3303                         result = 0;
3304                 }
3305         } else if(strcmp(name, dotdot) == 0) {
3306                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3307                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3308
3309                 if (!dir->oo_compat_dot_created)
3310                         return -EINVAL;
3311                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3312                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3313                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3314                 } else {
3315                         dot_ldp = NULL;
3316                         dot_dot_ldp = NULL;
3317                 }
3318                 /* in case of rename, dotdot is already created */
3319                 if (dir->oo_compat_dotdot_created) {
3320                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3321                                                 dot_dot_fid, NULL, th);
3322                 }
3323
3324                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3325                                                 inode, dot_ldp, dot_dot_ldp);
3326                 if (result == 0)
3327                        dir->oo_compat_dotdot_created = 1;
3328         }
3329
3330         return result;
3331 }
3332
3333
3334 /**
3335  * It will call the appropriate osd_add* function and return the
3336  * value, return by respective functions.
3337  */
3338 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3339                           struct inode *cinode, const char *name,
3340                           const struct dt_rec *fid, struct thandle *th)
3341 {
3342         struct osd_thread_info *info   = osd_oti_get(env);
3343         struct htree_lock      *hlock;
3344         int                     rc;
3345
3346         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3347
3348         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3349                                                    name[2] =='\0'))) {
3350                 if (hlock != NULL) {
3351                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3352                                            pobj->oo_inode, 0);
3353                 } else {
3354                         down_write(&pobj->oo_ext_idx_sem);
3355                 }
3356                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3357                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3358                                         fid, th);
3359         } else {
3360                 if (hlock != NULL) {
3361                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3362                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3363                 } else {
3364                         down_write(&pobj->oo_ext_idx_sem);
3365                 }
3366
3367                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3368                                       hlock, th);
3369         }
3370         if (hlock != NULL)
3371                 ldiskfs_htree_unlock(hlock);
3372         else
3373                 up_write(&pobj->oo_ext_idx_sem);
3374
3375         return rc;
3376 }
3377
3378 static void
3379 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3380                       struct osd_idmap_cache *oic)
3381 {
3382         struct osd_scrub    *scrub = &dev->od_scrub;
3383         struct lu_fid       *fid   = &oic->oic_fid;
3384         struct osd_inode_id *id    = &oti->oti_id;
3385         int                  once  = 0;
3386         int                  rc;
3387         ENTRY;
3388
3389         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3390                 RETURN_EXIT;
3391
3392 again:
3393         rc = osd_oi_lookup(oti, dev, fid, id);
3394         if (rc != 0 && rc != -ENOENT)
3395                 RETURN_EXIT;
3396
3397         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3398                 RETURN_EXIT;
3399
3400         if (thread_is_running(&scrub->os_thread)) {
3401                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3402                 /* There is race condition between osd_oi_lookup and OI scrub.
3403                  * The OI scrub finished just after osd_oi_lookup() failure.
3404                  * Under such case, it is unnecessary to trigger OI scrub again,
3405                  * but try to call osd_oi_lookup() again. */
3406                 if (unlikely(rc == -EAGAIN))
3407                         goto again;
3408
3409                 RETURN_EXIT;
3410         }
3411
3412         if (!dev->od_noscrub && ++once == 1) {
3413                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3414                        PFID(fid));
3415                 rc = osd_scrub_start(dev);
3416                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3417                                ", rc = %d [2]\n",
3418                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3419                                PFID(fid), rc);
3420                 if (rc == 0)
3421                         goto again;
3422         }
3423
3424         EXIT;
3425 }
3426
3427 /**
3428  * Calls ->lookup() to find dentry. From dentry get inode and
3429  * read inode's ea to get fid. This is required for  interoperability
3430  * mode (b11826)
3431  *
3432  * \retval   0, on success
3433  * \retval -ve, on error
3434  */
3435 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3436                              struct dt_rec *rec, const struct dt_key *key)
3437 {
3438         struct inode               *dir    = obj->oo_inode;
3439         struct dentry              *dentry;
3440         struct ldiskfs_dir_entry_2 *de;
3441         struct buffer_head         *bh;
3442         struct lu_fid              *fid = (struct lu_fid *) rec;
3443         struct htree_lock          *hlock = NULL;
3444         int                         ino;
3445         int                         rc;
3446
3447         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3448
3449         dentry = osd_child_dentry_get(env, obj,
3450                                       (char *)key, strlen((char *)key));
3451
3452         if (obj->oo_hl_head != NULL) {
3453                 hlock = osd_oti_get(env)->oti_hlock;
3454                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3455                                    dir, LDISKFS_HLOCK_LOOKUP);
3456         } else {
3457                 down_read(&obj->oo_ext_idx_sem);
3458         }
3459
3460         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3461         if (bh) {
3462                 struct osd_thread_info *oti = osd_oti_get(env);
3463                 struct osd_idmap_cache *oic = &oti->oti_cache;
3464                 struct osd_device *dev = osd_obj2dev(obj);
3465                 struct osd_scrub *scrub = &dev->od_scrub;
3466                 struct scrub_file *sf = &scrub->os_file;
3467
3468                 ino = le32_to_cpu(de->inode);
3469                 rc = osd_get_fid_from_dentry(de, rec);
3470
3471                 /* done with de, release bh */
3472                 brelse(bh);
3473                 if (rc != 0)
3474                         rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3475                 else
3476                         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3477                 if (rc != 0) {
3478                         fid_zero(&oic->oic_fid);
3479                         GOTO(out, rc);
3480                 }
3481
3482                 oic->oic_fid = *fid;
3483                 if ((scrub->os_pos_current <= ino) &&
3484                     (sf->sf_flags & SF_INCONSISTENT ||
3485                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3486                                       sf->sf_oi_bitmap)))
3487                         osd_consistency_check(oti, dev, oic);
3488         } else {
3489                 rc = -ENOENT;
3490         }
3491
3492         GOTO(out, rc);
3493
3494 out:
3495         if (hlock != NULL)
3496                 ldiskfs_htree_unlock(hlock);
3497         else
3498                 up_read(&obj->oo_ext_idx_sem);
3499         return rc;
3500 }
3501
3502 /**
3503  * Find the osd object for given fid.
3504  *
3505  * \param fid need to find the osd object having this fid
3506  *
3507  * \retval osd_object on success
3508  * \retval        -ve on error
3509  */
3510 struct osd_object *osd_object_find(const struct lu_env *env,
3511                                    struct dt_object *dt,
3512                                    const struct lu_fid *fid)
3513 {
3514         struct lu_device  *ludev = dt->do_lu.lo_dev;
3515         struct osd_object *child = NULL;
3516         struct lu_object  *luch;
3517         struct lu_object  *lo;
3518
3519         /*
3520          * at this point topdev might not exist yet
3521          * (i.e. MGS is preparing profiles). so we can
3522          * not rely on topdev and instead lookup with
3523          * our device passed as topdev. this can't work
3524          * if the object isn't cached yet (as osd doesn't
3525          * allocate lu_header). IOW, the object must be
3526          * in the cache, otherwise lu_object_alloc() crashes
3527          * -bzzz
3528          */
3529         luch = lu_object_find_at(env, ludev, fid, NULL);
3530         if (!IS_ERR(luch)) {
3531                 if (lu_object_exists(luch)) {
3532                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3533                         if (lo != NULL)
3534                                 child = osd_obj(lo);
3535                         else
3536                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3537                                                 "lu_object can't be located"
3538                                                 DFID"\n", PFID(fid));
3539
3540                         if (child == NULL) {
3541                                 lu_object_put(env, luch);
3542                                 CERROR("Unable to get osd_object\n");
3543                                 child = ERR_PTR(-ENOENT);
3544                         }
3545                 } else {
3546                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3547                                         "lu_object does not exists "DFID"\n",
3548                                         PFID(fid));
3549                         lu_object_put(env, luch);
3550                         child = ERR_PTR(-ENOENT);
3551                 }
3552         } else
3553                 child = (void *)luch;
3554
3555         return child;
3556 }
3557
3558 /**
3559  * Put the osd object once done with it.
3560  *
3561  * \param obj osd object that needs to be put
3562  */
3563 static inline void osd_object_put(const struct lu_env *env,
3564                                   struct osd_object *obj)
3565 {
3566         lu_object_put(env, &obj->oo_dt.do_lu);
3567 }
3568
3569 static int osd_index_declare_ea_insert(const struct lu_env *env,
3570                                        struct dt_object *dt,
3571                                        const struct dt_rec *rec,
3572                                        const struct dt_key *key,
3573                                        struct thandle *handle)
3574 {
3575         struct osd_thandle      *oh;
3576         struct inode            *inode;
3577         struct lu_fid           *fid = (struct lu_fid *)rec;
3578         int                     rc;
3579         ENTRY;
3580
3581         LASSERT(dt_object_exists(dt));
3582         LASSERT(handle != NULL);
3583
3584         oh = container_of0(handle, struct osd_thandle, ot_super);
3585         LASSERT(oh->ot_handle == NULL);
3586
3587         OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3588
3589         inode = osd_dt_obj(dt)->oo_inode;
3590         LASSERT(inode);
3591
3592         /* We ignore block quota on meta pool (MDTs), so needn't
3593          * calculate how many blocks will be consumed by this index
3594          * insert */
3595         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3596                                    true, true, NULL, false);
3597         if (fid == NULL)
3598                 RETURN(0);
3599
3600         /* It does fld look up inside declare, and the result will be
3601         * added to fld cache, so the following fld lookup inside insert
3602         * does not need send RPC anymore, so avoid send rpc with holding
3603         * transaction */
3604         LASSERTF(fid_is_sane(fid), "fid is insane"DFID"\n", PFID(fid));
3605         osd_fld_lookup(env, osd_dt_dev(handle->th_dev), fid,
3606                         &osd_oti_get(env)->oti_seq_range);
3607
3608         RETURN(rc);
3609 }
3610
3611 /**
3612  * Index add function for interoperability mode (b11826).
3613  * It will add the directory entry.This entry is needed to
3614  * maintain name->fid mapping.
3615  *
3616  * \param key it is key i.e. file entry to be inserted
3617  * \param rec it is value of given key i.e. fid
3618  *
3619  * \retval   0, on success
3620  * \retval -ve, on error
3621  */
3622 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3623                                const struct dt_rec *rec,
3624                                const struct dt_key *key, struct thandle *th,
3625                                struct lustre_capa *capa, int ignore_quota)
3626 {
3627         struct osd_object *obj   = osd_dt_obj(dt);
3628         struct lu_fid     *fid   = (struct lu_fid *) rec;
3629         const char        *name  = (const char *)key;
3630         struct osd_object *child;
3631         int                rc;
3632
3633         ENTRY;
3634
3635         LASSERT(osd_invariant(obj));
3636         LASSERT(dt_object_exists(dt));
3637         LASSERT(th != NULL);
3638
3639         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3640                 RETURN(-EACCES);
3641
3642         child = osd_object_find(env, dt, fid);
3643         if (!IS_ERR(child)) {
3644                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3645                 osd_object_put(env, child);
3646         } else {
3647                 rc = PTR_ERR(child);
3648         }
3649
3650         LASSERT(osd_invariant(obj));
3651         RETURN(rc);
3652 }
3653
3654 /**
3655  *  Initialize osd Iterator for given osd index object.
3656  *
3657  *  \param  dt      osd index object
3658  */
3659
3660 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3661                                      struct dt_object *dt,
3662                                      __u32 unused,
3663                                      struct lustre_capa *capa)
3664 {
3665         struct osd_it_iam      *it;
3666         struct osd_thread_info *oti = osd_oti_get(env);
3667         struct osd_object      *obj = osd_dt_obj(dt);
3668         struct lu_object       *lo  = &dt->do_lu;
3669         struct iam_path_descr  *ipd;
3670         struct iam_container   *bag = &obj->oo_dir->od_container;
3671
3672         LASSERT(lu_object_exists(lo));
3673
3674         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3675                 return ERR_PTR(-EACCES);
3676
3677         it = &oti->oti_it;
3678         ipd = osd_it_ipd_get(env, bag);
3679         if (likely(ipd != NULL)) {
3680                 it->oi_obj = obj;
3681                 it->oi_ipd = ipd;
3682                 lu_object_get(lo);
3683                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3684                 return (struct dt_it *)it;
3685         }
3686         return ERR_PTR(-ENOMEM);
3687 }
3688
3689 /**
3690  * free given Iterator.
3691  */
3692
3693 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3694 {
3695         struct osd_it_iam *it = (struct osd_it_iam *)di;
3696         struct osd_object *obj = it->oi_obj;
3697
3698         iam_it_fini(&it->oi_it);
3699         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3700         lu_object_put(env, &obj->oo_dt.do_lu);
3701 }
3702
3703 /**
3704  *  Move Iterator to record specified by \a key
3705  *
3706  *  \param  di      osd iterator
3707  *  \param  key     key for index
3708  *
3709  *  \retval +ve  di points to record with least key not larger than key
3710  *  \retval  0   di points to exact matched key
3711  *  \retval -ve  failure
3712  */
3713
3714 static int osd_it_iam_get(const struct lu_env *env,
3715                           struct dt_it *di, const struct dt_key *key)
3716 {
3717         struct osd_thread_info  *oti = osd_oti_get(env);
3718         struct osd_it_iam       *it = (struct osd_it_iam *)di;
3719
3720         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3721                 /* swab quota uid/gid */
3722                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3723                 key = (struct dt_key *)&oti->oti_quota_id;
3724         }
3725
3726         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3727 }
3728
3729 /**
3730  *  Release Iterator
3731  *
3732  *  \param  di      osd iterator
3733  */
3734
3735 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3736 {
3737         struct osd_it_iam *it = (struct osd_it_iam *)di;
3738
3739         iam_it_put(&it->oi_it);
3740 }
3741
3742 /**
3743  *  Move iterator by one record
3744  *
3745  *  \param  di      osd iterator
3746  *
3747  *  \retval +1   end of container reached
3748  *  \retval  0   success
3749  *  \retval -ve  failure
3750  */
3751
3752 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3753 {
3754         struct osd_it_iam *it = (struct osd_it_iam *)di;
3755
3756         return iam_it_next(&it->oi_it);
3757 }
3758
3759 /**
3760  * Return pointer to the key under iterator.
3761  */
3762
3763 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3764                                  const struct dt_it *di)
3765 {
3766         struct osd_thread_info *oti = osd_oti_get(env);
3767         struct osd_it_iam      *it = (struct osd_it_iam *)di;
3768         struct osd_object      *obj = it->oi_obj;
3769         struct dt_key          *key;
3770
3771         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3772
3773         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3774                 /* swab quota uid/gid */
3775                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3776                 key = (struct dt_key *)&oti->oti_quota_id;
3777         }
3778
3779         return key;
3780 }
3781
3782 /**
3783  * Return size of key under iterator (in bytes)
3784  */
3785
3786 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3787 {
3788         struct osd_it_iam *it = (struct osd_it_iam *)di;
3789
3790         return iam_it_key_size(&it->oi_it);
3791 }
3792
3793 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3794                                        int len, __u16 type)
3795 {
3796         struct luda_type *lt;
3797         const unsigned    align = sizeof(struct luda_type) - 1;
3798
3799         /* check if file type is required */
3800         if (attr & LUDA_TYPE) {
3801                         len = (len + align) & ~align;
3802
3803                         lt = (void *) ent->lde_name + len;
3804                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3805                         ent->lde_attrs |= LUDA_TYPE;
3806         }
3807
3808         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3809 }
3810
3811 /**
3812  * build lu direct from backend fs dirent.
3813  */
3814
3815 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3816                                       struct lu_fid *fid, __u64 offset,
3817                                       char *name, __u16 namelen,
3818                                       __u16 type, __u32 attr)
3819 {
3820         fid_cpu_to_le(&ent->lde_fid, fid);
3821         ent->lde_attrs = LUDA_FID;
3822
3823         ent->lde_hash = cpu_to_le64(offset);
3824         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3825
3826         strncpy(ent->lde_name, name, namelen);
3827         ent->lde_namelen = cpu_to_le16(namelen);
3828
3829         /* append lustre attributes */
3830         osd_it_append_attrs(ent, attr, namelen, type);
3831 }
3832
3833 /**
3834  * Return pointer to the record under iterator.
3835  */
3836 static int osd_it_iam_rec(const struct lu_env *env,
3837                           const struct dt_it *di,
3838                           struct dt_rec *dtrec, __u32 attr)
3839 {
3840         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
3841         struct osd_thread_info *info = osd_oti_get(env);
3842         ENTRY;
3843
3844         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3845                 const struct osd_fid_pack *rec;
3846                 struct lu_fid             *fid = &info->oti_fid;
3847                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
3848                 char                      *name;
3849                 int                        namelen;
3850                 __u64                      hash;
3851                 int                        rc;
3852
3853                 name = (char *)iam_it_key_get(&it->oi_it);
3854                 if (IS_ERR(name))
3855                         RETURN(PTR_ERR(name));
3856
3857                 namelen = iam_it_key_size(&it->oi_it);
3858
3859                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3860                 if (IS_ERR(rec))
3861                         RETURN(PTR_ERR(rec));
3862
3863                 rc = osd_fid_unpack(fid, rec);
3864                 if (rc)
3865                         RETURN(rc);
3866
3867                 hash = iam_it_store(&it->oi_it);
3868
3869                 /* IAM does not store object type in IAM index (dir) */
3870                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3871                                    0, LUDA_FID);
3872         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3873                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3874                            (struct iam_rec *)dtrec);
3875                 osd_quota_unpack(it->oi_obj, dtrec);
3876         } else {
3877                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3878                            (struct iam_rec *)dtrec);
3879         }
3880
3881         RETURN(0);
3882 }
3883
3884 /**
3885  * Returns cookie for current Iterator position.
3886  */
3887 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3888 {
3889         struct osd_it_iam *it = (struct osd_it_iam *)di;
3890
3891         return iam_it_store(&it->oi_it);
3892 }
3893
3894 /**
3895  * Restore iterator from cookie.
3896  *
3897  * \param  di      osd iterator
3898  * \param  hash    Iterator location cookie
3899  *
3900  * \retval +ve  di points to record with least key not larger than key.
3901  * \retval  0   di points to exact matched key
3902  * \retval -ve  failure
3903  */
3904
3905 static int osd_it_iam_load(const struct lu_env *env,
3906                            const struct dt_it *di, __u64 hash)
3907 {
3908         struct osd_it_iam *it = (struct osd_it_iam *)di;
3909
3910         return iam_it_load(&it->oi_it, hash);
3911 }
3912
3913 static const struct dt_index_operations osd_index_iam_ops = {
3914         .dio_lookup         = osd_index_iam_lookup,
3915         .dio_declare_insert = osd_index_declare_iam_insert,
3916         .dio_insert         = osd_index_iam_insert,
3917         .dio_declare_delete = osd_index_declare_iam_delete,
3918         .dio_delete         = osd_index_iam_delete,
3919         .dio_it     = {
3920                 .init     = osd_it_iam_init,
3921                 .fini     = osd_it_iam_fini,
3922                 .get      = osd_it_iam_get,
3923                 .put      = osd_it_iam_put,
3924                 .next     = osd_it_iam_next,
3925                 .key      = osd_it_iam_key,
3926                 .key_size = osd_it_iam_key_size,
3927                 .rec      = osd_it_iam_rec,
3928                 .store    = osd_it_iam_store,
3929                 .load     = osd_it_iam_load
3930         }
3931 };
3932
3933
3934 /**
3935  * Creates or initializes iterator context.
3936  *
3937  * \retval struct osd_it_ea, iterator structure on success
3938  *
3939  */
3940 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3941                                     struct dt_object *dt,
3942                                     __u32 attr,
3943                                     struct lustre_capa *capa)
3944 {
3945         struct osd_object       *obj  = osd_dt_obj(dt);
3946         struct osd_thread_info  *info = osd_oti_get(env);
3947         struct osd_it_ea        *it   = &info->oti_it_ea;
3948         struct lu_object        *lo   = &dt->do_lu;
3949         struct dentry           *obj_dentry = &info->oti_it_dentry;
3950         ENTRY;
3951         LASSERT(lu_object_exists(lo));
3952
3953         obj_dentry->d_inode = obj->oo_inode;
3954         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3955         obj_dentry->d_name.hash = 0;
3956
3957         it->oie_rd_dirent       = 0;
3958         it->oie_it_dirent       = 0;
3959         it->oie_dirent          = NULL;
3960         it->oie_buf             = info->oti_it_ea_buf;
3961         it->oie_obj             = obj;
3962         it->oie_file.f_pos      = 0;
3963         it->oie_file.f_dentry   = obj_dentry;
3964         if (attr & LUDA_64BITHASH)
3965                 it->oie_file.f_mode |= FMODE_64BITHASH;
3966         else
3967                 it->oie_file.f_mode |= FMODE_32BITHASH;
3968         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
3969         it->oie_file.f_op         = obj->oo_inode->i_fop;
3970         it->oie_file.private_data = NULL;
3971         lu_object_get(lo);
3972         RETURN((struct dt_it *) it);
3973 }
3974
3975 /**
3976  * Destroy or finishes iterator context.
3977  *
3978  * \param di iterator structure to be destroyed
3979  */
3980 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3981 {
3982         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3983         struct osd_object    *obj  = it->oie_obj;
3984         struct inode       *inode  = obj->oo_inode;
3985
3986         ENTRY;
3987         it->oie_file.f_op->release(inode, &it->oie_file);
3988         lu_object_put(env, &obj->oo_dt.do_lu);
3989         EXIT;
3990 }
3991
3992 /**
3993  * It position the iterator at given key, so that next lookup continues from
3994  * that key Or it is similar to dio_it->load() but based on a key,
3995  * rather than file position.
3996  *
3997  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3998  * to the beginning.
3999  *
4000  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4001  */
4002 static int osd_it_ea_get(const struct lu_env *env,
4003                          struct dt_it *di, const struct dt_key *key)
4004 {
4005         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4006
4007         ENTRY;
4008         LASSERT(((const char *)key)[0] == '\0');
4009         it->oie_file.f_pos      = 0;
4010         it->oie_rd_dirent       = 0;
4011         it->oie_it_dirent       = 0;
4012         it->oie_dirent          = NULL;
4013
4014         RETURN(+1);
4015 }
4016
4017 /**
4018  * Does nothing
4019  */
4020 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4021 {
4022 }
4023
4024 /**
4025  * It is called internally by ->readdir(). It fills the
4026  * iterator's in-memory data structure with required
4027  * information i.e. name, namelen, rec_size etc.
4028  *
4029  * \param buf in which information to be filled in.
4030  * \param name name of the file in given dir
4031  *
4032  * \retval 0 on success
4033  * \retval 1 on buffer full
4034  */
4035 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4036                                loff_t offset, __u64 ino,
4037                                unsigned d_type)
4038 {
4039         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4040         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4041         struct lu_fid           *fid  = &ent->oied_fid;
4042         struct osd_fid_pack     *rec;
4043         ENTRY;
4044
4045         /* this should never happen */
4046         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4047                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4048                 RETURN(-EIO);
4049         }
4050
4051         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4052             OSD_IT_EA_BUFSIZE)
4053                 RETURN(1);
4054
4055         if (d_type & LDISKFS_DIRENT_LUFID) {
4056                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4057
4058                 if (osd_fid_unpack(fid, rec) != 0)
4059                         fid_zero(fid);
4060
4061                 d_type &= ~LDISKFS_DIRENT_LUFID;
4062         } else {
4063                 fid_zero(fid);
4064         }
4065
4066         ent->oied_ino     = ino;
4067         ent->oied_off     = offset;
4068         ent->oied_namelen = namelen;
4069         ent->oied_type    = d_type;
4070
4071         memcpy(ent->oied_name, name, namelen);
4072
4073         it->oie_rd_dirent++;
4074         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4075         RETURN(0);
4076 }
4077
4078 /**
4079  * Calls ->readdir() to load a directory entry at a time
4080  * and stored it in iterator's in-memory data structure.
4081  *
4082  * \param di iterator's in memory structure
4083  *
4084  * \retval   0 on success
4085  * \retval -ve on error
4086  */
4087 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4088                                const struct dt_it *di)
4089 {
4090         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4091         struct osd_object  *obj   = it->oie_obj;
4092         struct inode       *inode = obj->oo_inode;
4093         struct htree_lock  *hlock = NULL;
4094         int                 result = 0;
4095
4096         ENTRY;
4097         it->oie_dirent = it->oie_buf;
4098         it->oie_rd_dirent = 0;
4099
4100         if (obj->oo_hl_head != NULL) {
4101                 hlock = osd_oti_get(env)->oti_hlock;
4102                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4103                                    inode, LDISKFS_HLOCK_READDIR);
4104         } else {
4105                 down_read(&obj->oo_ext_idx_sem);
4106         }
4107
4108         result = inode->i_fop->readdir(&it->oie_file, it,
4109                                        (filldir_t) osd_ldiskfs_filldir);
4110
4111         if (hlock != NULL)
4112                 ldiskfs_htree_unlock(hlock);
4113         else
4114                 up_read(&obj->oo_ext_idx_sem);
4115
4116         if (it->oie_rd_dirent == 0) {
4117                 result = -EIO;
4118         } else {
4119                 it->oie_dirent = it->oie_buf;
4120                 it->oie_it_dirent = 1;
4121         }
4122
4123         RETURN(result);
4124 }
4125
4126 /**
4127  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4128  * to load a directory entry at a time and stored it in
4129  * iterator's in-memory data structure.
4130  *
4131  * \param di iterator's in memory structure
4132  *
4133  * \retval +ve iterator reached to end
4134  * \retval   0 iterator not reached to end
4135  * \retval -ve on error
4136  */
4137 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4138 {
4139         struct osd_it_ea *it = (struct osd_it_ea *)di;
4140         int rc;
4141
4142         ENTRY;
4143
4144         if (it->oie_it_dirent < it->oie_rd_dirent) {
4145                 it->oie_dirent =
4146                         (void *) it->oie_dirent +
4147                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4148                                        it->oie_dirent->oied_namelen);
4149                 it->oie_it_dirent++;
4150                 RETURN(0);
4151         } else {
4152                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4153                         rc = +1;
4154                 else
4155                         rc = osd_ldiskfs_it_fill(env, di);
4156         }
4157
4158         RETURN(rc);
4159 }
4160
4161 /**
4162  * Returns the key at current position from iterator's in memory structure.
4163  *
4164  * \param di iterator's in memory structure
4165  *
4166  * \retval key i.e. struct dt_key on success
4167  */
4168 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4169                                     const struct dt_it *di)
4170 {
4171         struct osd_it_ea *it = (struct osd_it_ea *)di;
4172
4173         return (struct dt_key *)it->oie_dirent->oied_name;
4174 }
4175
4176 /**
4177  * Returns the key's size at current position from iterator's in memory structure.
4178  *
4179  * \param di iterator's in memory structure
4180  *
4181  * \retval key_size i.e. struct dt_key on success
4182  */
4183 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4184 {
4185         struct osd_it_ea *it = (struct osd_it_ea *)di;
4186
4187         return it->oie_dirent->oied_namelen;
4188 }
4189
4190
4191 /**
4192  * Returns the value (i.e. fid/igif) at current position from iterator's
4193  * in memory structure.
4194  *
4195  * \param di struct osd_it_ea, iterator's in memory structure
4196  * \param attr attr requested for dirent.
4197  * \param lde lustre dirent
4198  *
4199  * \retval   0 no error and \param lde has correct lustre dirent.
4200  * \retval -ve on error
4201  */
4202 static inline int osd_it_ea_rec(const struct lu_env *env,
4203                                 const struct dt_it *di,
4204                                 struct dt_rec *dtrec, __u32 attr)
4205 {
4206         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4207         struct osd_object      *obj   = it->oie_obj;
4208         struct osd_device      *dev   = osd_obj2dev(obj);
4209         struct osd_scrub       *scrub = &dev->od_scrub;
4210         struct scrub_file      *sf    = &scrub->os_file;
4211         struct osd_thread_info *oti   = osd_oti_get(env);
4212         struct osd_idmap_cache *oic   = &oti->oti_cache;
4213         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4214         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4215         __u32                   ino   = it->oie_dirent->oied_ino;
4216         int                     rc    = 0;
4217         ENTRY;
4218
4219         if (!fid_is_sane(fid)) {
4220                 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4221                 if (rc != 0) {
4222                         fid_zero(&oic->oic_fid);
4223                         RETURN(rc);
4224                 }
4225         } else {
4226                 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4227         }
4228
4229         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4230                            it->oie_dirent->oied_name,
4231                            it->oie_dirent->oied_namelen,
4232                            it->oie_dirent->oied_type, attr);
4233         oic->oic_fid = *fid;
4234         if ((scrub->os_pos_current <= ino) &&
4235             (sf->sf_flags & SF_INCONSISTENT ||
4236              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4237                 osd_consistency_check(oti, dev, oic);
4238
4239         RETURN(rc);
4240 }
4241
4242 /**
4243  * Returns a cookie for current position of the iterator head, so that
4244  * user can use this cookie to load/start the iterator next time.
4245  *
4246  * \param di iterator's in memory structure
4247  *
4248  * \retval cookie for current position, on success
4249  */
4250 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4251 {
4252         struct osd_it_ea *it = (struct osd_it_ea *)di;
4253
4254         return it->oie_dirent->oied_off;
4255 }
4256
4257 /**
4258  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4259  * to load a directory entry at a time and stored it i inn,
4260  * in iterator's in-memory data structure.
4261  *
4262  * \param di struct osd_it_ea, iterator's in memory structure
4263  *
4264  * \retval +ve on success
4265  * \retval -ve on error
4266  */
4267 static int osd_it_ea_load(const struct lu_env *env,
4268                           const struct dt_it *di, __u64 hash)
4269 {
4270         struct osd_it_ea *it = (struct osd_it_ea *)di;
4271         int rc;
4272
4273         ENTRY;
4274         it->oie_file.f_pos = hash;
4275
4276         rc =  osd_ldiskfs_it_fill(env, di);
4277         if (rc == 0)
4278                 rc = +1;
4279
4280         RETURN(rc);
4281 }
4282
4283 /**
4284  * Index lookup function for interoperability mode (b11826).
4285  *
4286  * \param key,  key i.e. file name to be searched
4287  *
4288  * \retval +ve, on success
4289  * \retval -ve, on error
4290  */
4291 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4292                                struct dt_rec *rec, const struct dt_key *key,
4293                                struct lustre_capa *capa)
4294 {
4295         struct osd_object *obj = osd_dt_obj(dt);
4296         int rc = 0;
4297
4298         ENTRY;
4299
4300         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4301         LINVRNT(osd_invariant(obj));
4302
4303         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4304                 return -EACCES;
4305
4306         rc = osd_ea_lookup_rec(env, obj, rec, key);
4307         if (rc == 0)
4308                 rc = +1;
4309         RETURN(rc);
4310 }
4311
4312 /**
4313  * Index and Iterator operations for interoperability
4314  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4315  */
4316 static const struct dt_index_operations osd_index_ea_ops = {
4317         .dio_lookup         = osd_index_ea_lookup,
4318         .dio_declare_insert = osd_index_declare_ea_insert,
4319         .dio_insert         = osd_index_ea_insert,
4320         .dio_declare_delete = osd_index_declare_ea_delete,
4321         .dio_delete         = osd_index_ea_delete,
4322         .dio_it     = {
4323                 .init     = osd_it_ea_init,
4324                 .fini     = osd_it_ea_fini,
4325                 .get      = osd_it_ea_get,
4326                 .put      = osd_it_ea_put,
4327                 .next     = osd_it_ea_next,
4328                 .key      = osd_it_ea_key,
4329                 .key_size = osd_it_ea_key_size,
4330                 .rec      = osd_it_ea_rec,
4331                 .store    = osd_it_ea_store,
4332                 .load     = osd_it_ea_load
4333         }
4334 };
4335
4336 static void *osd_key_init(const struct lu_context *ctx,
4337                           struct lu_context_key *key)
4338 {
4339         struct osd_thread_info *info;
4340
4341         OBD_ALLOC_PTR(info);
4342         if (info == NULL)
4343                 return ERR_PTR(-ENOMEM);
4344
4345         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4346         if (info->oti_it_ea_buf == NULL)
4347                 goto out_free_info;
4348
4349         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4350
4351         info->oti_hlock = ldiskfs_htree_lock_alloc();
4352         if (info->oti_hlock == NULL)
4353                 goto out_free_ea;
4354
4355         return info;
4356
4357  out_free_ea:
4358         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4359  out_free_info:
4360         OBD_FREE_PTR(info);
4361         return ERR_PTR(-ENOMEM);
4362 }
4363
4364 static void osd_key_fini(const struct lu_context *ctx,
4365                          struct lu_context_key *key, void* data)
4366 {
4367         struct osd_thread_info *info = data;
4368
4369         if (info->oti_hlock != NULL)
4370                 ldiskfs_htree_lock_free(info->oti_hlock);
4371         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4372         OBD_FREE_PTR(info);
4373 }
4374
4375 static void osd_key_exit(const struct lu_context *ctx,
4376                          struct lu_context_key *key, void *data)
4377 {
4378         struct osd_thread_info *info = data;
4379
4380         LASSERT(info->oti_r_locks == 0);
4381         LASSERT(info->oti_w_locks == 0);
4382         LASSERT(info->oti_txns    == 0);
4383 }
4384
4385 /* type constructor/destructor: osd_type_init, osd_type_fini */
4386 LU_TYPE_INIT_FINI(osd, &osd_key);
4387
4388 struct lu_context_key osd_key = {
4389         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4390         .lct_init = osd_key_init,
4391         .lct_fini = osd_key_fini,
4392         .lct_exit = osd_key_exit
4393 };
4394
4395
4396 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4397                            const char *name, struct lu_device *next)
4398 {
4399         struct osd_device *osd = osd_dev(d);
4400
4401         strncpy(osd->od_svname, name, MAX_OBD_NAME);
4402         return osd_procfs_init(osd, name);
4403 }
4404
4405 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4406 {
4407         ENTRY;
4408
4409         osd_scrub_cleanup(env, o);
4410
4411         if (o->od_fsops) {
4412                 fsfilt_put_ops(o->od_fsops);
4413                 o->od_fsops = NULL;
4414         }
4415
4416         /* shutdown quota slave instance associated with the device */
4417         if (o->od_quota_slave != NULL) {
4418                 qsd_fini(env, o->od_quota_slave);
4419                 o->od_quota_slave = NULL;
4420         }
4421
4422         RETURN(0);
4423 }
4424
4425 static int osd_mount(const struct lu_env *env,
4426                      struct osd_device *o, struct lustre_cfg *cfg)
4427 {
4428         const char              *name  = lustre_cfg_string(cfg, 0);
4429         const char              *dev  = lustre_cfg_string(cfg, 1);
4430         const char              *opts;
4431         unsigned long            page, s_flags, lmd_flags = 0;
4432         struct page             *__page;
4433         struct file_system_type *type;
4434         char                    *options = NULL;
4435         char                    *str;
4436         int                       rc = 0;
4437         ENTRY;
4438
4439         if (o->od_mnt != NULL)
4440                 RETURN(0);
4441
4442         if (strlen(dev) >= sizeof(o->od_mntdev))
4443                 RETURN(-E2BIG);
4444         strcpy(o->od_mntdev, dev);
4445
4446         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4447         if (o->od_fsops == NULL) {
4448                 CERROR("Can't find fsfilt_ldiskfs\n");
4449                 RETURN(-ENOTSUPP);
4450         }
4451
4452         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4453         if (__page == NULL)
4454                 GOTO(out, rc = -ENOMEM);
4455
4456         str = lustre_cfg_string(cfg, 2);
4457         s_flags = simple_strtoul(str, NULL, 0);
4458         str = strstr(str, ":");
4459         if (str)
4460                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4461         opts = lustre_cfg_string(cfg, 3);
4462         page = (unsigned long)cfs_page_address(__page);
4463         options = (char *)page;
4464         *options = '\0';
4465         if (opts == NULL)
4466                 strcat(options, "user_xattr,acl");
4467         else
4468                 strcat(options, opts);
4469
4470         /* Glom up mount options */
4471         if (*options != '\0')
4472                 strcat(options, ",");
4473         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4474
4475         type = get_fs_type("ldiskfs");
4476         if (!type) {
4477                 CERROR("%s: cannot find ldiskfs module\n", name);
4478                 GOTO(out, rc = -ENODEV);
4479         }
4480
4481         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4482         cfs_module_put(type->owner);
4483
4484         if (IS_ERR(o->od_mnt)) {
4485                 rc = PTR_ERR(o->od_mnt);
4486                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4487                 o->od_mnt = NULL;
4488                 GOTO(out, rc);
4489         }
4490
4491         if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4492                 CERROR("%s: underlying device %s is marked as read-only. "
4493                        "Setup failed\n", name, dev);
4494                 mntput(o->od_mnt);
4495                 o->od_mnt = NULL;
4496                 GOTO(out, rc = -EROFS);
4497         }
4498
4499         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4500             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4501                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4502                 mntput(o->od_mnt);
4503                 o->od_mnt = NULL;
4504                 GOTO(out, rc = -EINVAL);
4505         }
4506
4507         if (lmd_flags & LMD_FLG_NOSCRUB)
4508                 o->od_noscrub = 1;
4509
4510 out:
4511         if (__page)
4512                 OBD_PAGE_FREE(__page);
4513         if (rc)
4514                 fsfilt_put_ops(o->od_fsops);
4515
4516         RETURN(rc);
4517 }
4518
4519 static struct lu_device *osd_device_fini(const struct lu_env *env,
4520                                          struct lu_device *d)
4521 {
4522         int rc;
4523         ENTRY;
4524
4525         rc = osd_shutdown(env, osd_dev(d));
4526
4527         osd_obj_map_fini(osd_dev(d));
4528
4529         shrink_dcache_sb(osd_sb(osd_dev(d)));
4530         osd_sync(env, lu2dt_dev(d));
4531
4532         rc = osd_procfs_fini(osd_dev(d));
4533         if (rc) {
4534                 CERROR("proc fini error %d \n", rc);
4535                 RETURN (ERR_PTR(rc));
4536         }
4537
4538         if (osd_dev(d)->od_mnt) {
4539                 mntput(osd_dev(d)->od_mnt);
4540                 osd_dev(d)->od_mnt = NULL;
4541         }
4542
4543         RETURN(NULL);
4544 }
4545
4546 static int osd_device_init0(const struct lu_env *env,
4547                             struct osd_device *o,
4548                             struct lustre_cfg *cfg)
4549 {
4550         struct lu_device        *l = osd2lu_dev(o);
4551         struct osd_thread_info *info;
4552         int                     rc;
4553
4554         /* if the module was re-loaded, env can loose its keys */
4555         rc = lu_env_refill((struct lu_env *) env);
4556         if (rc)
4557                 GOTO(out, rc);
4558         info = osd_oti_get(env);
4559         LASSERT(info);
4560
4561         l->ld_ops = &osd_lu_ops;
4562         o->od_dt_dev.dd_ops = &osd_dt_ops;
4563
4564         spin_lock_init(&o->od_osfs_lock);
4565         mutex_init(&o->od_otable_mutex);
4566         o->od_osfs_age = cfs_time_shift_64(-1000);
4567
4568         o->od_capa_hash = init_capa_hash();
4569         if (o->od_capa_hash == NULL)
4570                 GOTO(out, rc = -ENOMEM);
4571
4572         o->od_read_cache = 1;
4573         o->od_writethrough_cache = 1;
4574         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
4575
4576         rc = osd_mount(env, o, cfg);
4577         if (rc)
4578                 GOTO(out_capa, rc);
4579
4580         /* setup scrub, including OI files initialization */
4581         rc = osd_scrub_setup(env, o);
4582         if (rc < 0)
4583                 GOTO(out_mnt, rc);
4584
4585         strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4586                         sizeof(o->od_svname) - 1);
4587
4588         rc = osd_obj_map_init(o);
4589         if (rc != 0)
4590                 GOTO(out_scrub, rc);
4591
4592         rc = lu_site_init(&o->od_site, l);
4593         if (rc)
4594                 GOTO(out_compat, rc);
4595         o->od_site.ls_bottom_dev = l;
4596
4597         rc = lu_site_init_finish(&o->od_site);
4598         if (rc)
4599                 GOTO(out_site, rc);
4600
4601         rc = osd_procfs_init(o, o->od_svname);
4602         if (rc != 0) {
4603                 CERROR("%s: can't initialize procfs: rc = %d\n",
4604                        o->od_svname, rc);
4605                 GOTO(out_site, rc);
4606         }
4607
4608         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4609
4610         /* initialize quota slave instance */
4611         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
4612                                      o->od_proc_entry);
4613         if (IS_ERR(o->od_quota_slave)) {
4614                 rc = PTR_ERR(o->od_quota_slave);
4615                 o->od_quota_slave = NULL;
4616                 GOTO(out_procfs, rc);
4617         }
4618
4619         RETURN(0);
4620 out_procfs:
4621         osd_procfs_fini(o);
4622 out_site:
4623         lu_site_fini(&o->od_site);
4624 out_compat:
4625         osd_obj_map_fini(o);
4626 out_scrub:
4627         osd_scrub_cleanup(env, o);
4628 out_mnt:
4629         osd_oi_fini(info, o);
4630         osd_shutdown(env, o);
4631         mntput(o->od_mnt);
4632         o->od_mnt = NULL;
4633 out_capa:
4634         cleanup_capa_hash(o->od_capa_hash);
4635 out:
4636         RETURN(rc);
4637 }
4638
4639 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4640                                           struct lu_device_type *t,
4641                                           struct lustre_cfg *cfg)
4642 {
4643         struct osd_device *o;
4644         int                rc;
4645
4646         OBD_ALLOC_PTR(o);
4647         if (o == NULL)
4648                 return ERR_PTR(-ENOMEM);
4649
4650         rc = dt_device_init(&o->od_dt_dev, t);
4651         if (rc == 0) {
4652                 /* Because the ctx might be revived in dt_device_init,
4653                  * refill the env here */
4654                 lu_env_refill((struct lu_env *)env);
4655                 rc = osd_device_init0(env, o, cfg);
4656                 if (rc)
4657                         dt_device_fini(&o->od_dt_dev);
4658         }
4659
4660         if (unlikely(rc != 0))
4661                 OBD_FREE_PTR(o);
4662
4663         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4664 }
4665
4666 static struct lu_device *osd_device_free(const struct lu_env *env,
4667                                          struct lu_device *d)
4668 {
4669         struct osd_device *o = osd_dev(d);
4670         ENTRY;
4671
4672         cleanup_capa_hash(o->od_capa_hash);
4673         /* XXX: make osd top device in order to release reference */
4674         d->ld_site->ls_top_dev = d;
4675         lu_site_purge(env, d->ld_site, -1);
4676         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4677                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4678                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4679         }
4680         lu_site_fini(&o->od_site);
4681         dt_device_fini(&o->od_dt_dev);
4682         OBD_FREE_PTR(o);
4683         RETURN(NULL);
4684 }
4685
4686 static int osd_process_config(const struct lu_env *env,
4687                               struct lu_device *d, struct lustre_cfg *cfg)
4688 {
4689         struct osd_device *o = osd_dev(d);
4690         int err;
4691         ENTRY;
4692
4693         switch(cfg->lcfg_command) {
4694         case LCFG_SETUP:
4695                 err = osd_mount(env, o, cfg);
4696                 break;
4697         case LCFG_CLEANUP:
4698                 lu_dev_del_linkage(d->ld_site, d);
4699                 err = osd_shutdown(env, o);
4700                 break;
4701         default:
4702                 err = -ENOSYS;
4703         }
4704
4705         RETURN(err);
4706 }
4707
4708 static int osd_recovery_complete(const struct lu_env *env,
4709                                  struct lu_device *d)
4710 {
4711         struct osd_device       *osd = osd_dev(d);
4712         int                      rc = 0;
4713         ENTRY;
4714
4715         if (osd->od_quota_slave == NULL)
4716                 RETURN(0);
4717
4718         /* start qsd instance on recovery completion, this notifies the quota
4719          * slave code that we are about to process new requests now */
4720         rc = qsd_start(env, osd->od_quota_slave);
4721         RETURN(rc);
4722 }
4723
4724 /*
4725  * we use exports to track all osd users
4726  */
4727 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4728                            struct obd_device *obd, struct obd_uuid *cluuid,
4729                            struct obd_connect_data *data, void *localdata)
4730 {
4731         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
4732         struct lustre_handle  conn;
4733         int                   rc;
4734         ENTRY;
4735
4736         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4737
4738         rc = class_connect(&conn, obd, cluuid);
4739         if (rc)
4740                 RETURN(rc);
4741
4742         *exp = class_conn2export(&conn);
4743
4744         spin_lock(&osd->od_osfs_lock);
4745         osd->od_connects++;
4746         spin_unlock(&osd->od_osfs_lock);
4747
4748         RETURN(0);
4749 }
4750
4751 /*
4752  * once last export (we don't count self-export) disappeared
4753  * osd can be released
4754  */
4755 static int osd_obd_disconnect(struct obd_export *exp)
4756 {
4757         struct obd_device *obd = exp->exp_obd;
4758         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4759         int                rc, release = 0;
4760         ENTRY;
4761
4762         /* Only disconnect the underlying layers on the final disconnect. */
4763         spin_lock(&osd->od_osfs_lock);
4764         osd->od_connects--;
4765         if (osd->od_connects == 0)
4766                 release = 1;
4767         spin_unlock(&osd->od_osfs_lock);
4768
4769         rc = class_disconnect(exp); /* bz 9811 */
4770
4771         if (rc == 0 && release)
4772                 class_manual_cleanup(obd);
4773         RETURN(rc);
4774 }
4775
4776 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4777                        struct lu_device *dev)
4778 {
4779         struct osd_device *osd = osd_dev(dev);
4780         int                result = 0;
4781         ENTRY;
4782
4783         if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
4784                 /* MDT/MDD still use old infrastructure to create
4785                  * special files */
4786                 result = llo_local_objects_setup(env, lu2md_dev(pdev),
4787                                                  lu2dt_dev(dev));
4788                 if (result)
4789                         RETURN(result);
4790         }
4791
4792         if (osd->od_quota_slave != NULL)
4793                 /* set up quota slave objects */
4794                 result = qsd_prepare(env, osd->od_quota_slave);
4795
4796         RETURN(result);
4797 }
4798
4799 static const struct lu_object_operations osd_lu_obj_ops = {
4800         .loo_object_init      = osd_object_init,
4801         .loo_object_delete    = osd_object_delete,
4802         .loo_object_release   = osd_object_release,
4803         .loo_object_free      = osd_object_free,
4804         .loo_object_print     = osd_object_print,
4805         .loo_object_invariant = osd_object_invariant
4806 };
4807
4808 const struct lu_device_operations osd_lu_ops = {
4809         .ldo_object_alloc      = osd_object_alloc,
4810         .ldo_process_config    = osd_process_config,
4811         .ldo_recovery_complete = osd_recovery_complete,
4812         .ldo_prepare           = osd_prepare,
4813 };
4814
4815 static const struct lu_device_type_operations osd_device_type_ops = {
4816         .ldto_init = osd_type_init,
4817         .ldto_fini = osd_type_fini,
4818
4819         .ldto_start = osd_type_start,
4820         .ldto_stop  = osd_type_stop,
4821
4822         .ldto_device_alloc = osd_device_alloc,
4823         .ldto_device_free  = osd_device_free,
4824
4825         .ldto_device_init    = osd_device_init,
4826         .ldto_device_fini    = osd_device_fini
4827 };
4828
4829 struct lu_device_type osd_device_type = {
4830         .ldt_tags     = LU_DEVICE_DT,
4831         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
4832         .ldt_ops      = &osd_device_type_ops,
4833         .ldt_ctx_tags = LCT_LOCAL,
4834 };
4835
4836 /*
4837  * lprocfs legacy support.
4838  */
4839 static struct obd_ops osd_obd_device_ops = {
4840         .o_owner = THIS_MODULE,
4841         .o_connect      = osd_obd_connect,
4842         .o_disconnect   = osd_obd_disconnect
4843 };
4844
4845 static int __init osd_mod_init(void)
4846 {
4847         struct lprocfs_static_vars lvars;
4848
4849         osd_oi_mod_init();
4850         lprocfs_osd_init_vars(&lvars);
4851         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4852                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4853 }
4854
4855 static void __exit osd_mod_exit(void)
4856 {
4857         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4858 }
4859
4860 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4861 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4862 MODULE_LICENSE("GPL");
4863
4864 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);