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