lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56 /* simple_mkdir() */
  57 #include <lvfs.h>
  58
  59 /*
  60  * struct OBD_{ALLOC,FREE}*()
  61  * OBD_FAIL_CHECK
  62  */
  63 #include <obd_support.h>
  64 /* struct ptlrpc_thread */
  65 #include <lustre_net.h>
  66
  67 /* fid_is_local() */
  68 #include <lustre_fid.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_igif.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75 #include <lustre_quota.h>
  76
  77 int ldiskfs_pdo = 1;
  78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  79                 "ldiskfs with parallel directory operations");
  80
  81 static const char dot[] = ".";
  82 static const char dotdot[] = "..";
  83 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  84
  85 static const struct lu_object_operations      osd_lu_obj_ops;
  86 static const struct dt_object_operations      osd_obj_ops;
  87 static const struct dt_object_operations      osd_obj_ea_ops;
  88 static const struct dt_object_operations      osd_obj_otable_it_ops;
  89 static const struct dt_index_operations       osd_index_iam_ops;
  90 static const struct dt_index_operations       osd_index_ea_ops;
  91
  92 #ifdef OSD_TRACK_DECLARES
  93 int osd_trans_declare_op2rb[] = {
  94         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
  95         [OSD_OT_PUNCH]          = OSD_OT_MAX,
  96         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
  97         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
  98         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
  99         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 100         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 101         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 102         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 103         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 104         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 105 };
 106 #endif
 107
 108 static int osd_has_index(const struct osd_object *obj)
 109 {
 110         return obj->oo_dt.do_index_ops != NULL;
 111 }
 112
 113 static int osd_object_invariant(const struct lu_object *l)
 114 {
 115         return osd_invariant(osd_obj(l));
 116 }
 117
 118 /*
 119  * Concurrency: doesn't matter
 120  */
 121 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 122 {
 123         return osd_oti_get(env)->oti_r_locks > 0;
 124 }
 125
 126 /*
 127  * Concurrency: doesn't matter
 128  */
 129 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 130 {
 131         struct osd_thread_info *oti = osd_oti_get(env);
 132         return oti->oti_w_locks > 0 && o->oo_owner == env;
 133 }
 134
 135 /*
 136  * Concurrency: doesn't access mutable data
 137  */
 138 static int osd_root_get(const struct lu_env *env,
 139                         struct dt_device *dev, struct lu_fid *f)
 140 {
 141         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 142         return 0;
 143 }
 144
 145 /*
 146  * OSD object methods.
 147  */
 148
 149 /*
 150  * Concurrency: no concurrent access is possible that early in object
 151  * life-cycle.
 152  */
 153 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 154                                           const struct lu_object_header *hdr,
 155                                           struct lu_device *d)
 156 {
 157         struct osd_object *mo;
 158
 159         OBD_ALLOC_PTR(mo);
 160         if (mo != NULL) {
 161                 struct lu_object *l;
 162
 163                 l = &mo->oo_dt.do_lu;
 164                 dt_object_init(&mo->oo_dt, NULL, d);
 165                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 166                 l->lo_ops = &osd_lu_obj_ops;
 167                 init_rwsem(&mo->oo_sem);
 168                 init_rwsem(&mo->oo_ext_idx_sem);
 169                 spin_lock_init(&mo->oo_guard);
 170                 return l;
 171         } else {
 172                 return NULL;
 173         }
 174 }
 175
 176 static int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 177                        struct dentry *dentry, struct lustre_mdt_attrs *lma)
 178 {
 179         int rc;
 180
 181         dentry->d_inode = inode;
 182         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
 183                                    sizeof(*lma));
 184         if (rc == -ERANGE) {
 185                 /* try with old lma size */
 186                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
 187                                            info->oti_mdt_attrs_old,
 188                                            LMA_OLD_SIZE);
 189                 if (rc > 0)
 190                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 191         }
 192         if (rc > 0) {
 193                 /* Check LMA compatibility */
 194                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 195                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 196                               "%lx/%#x\n",
 197                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 198                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 199                                                         ~LMA_INCOMPAT_SUPP);
 200                         rc = -ENOSYS;
 201                 } else {
 202                         lustre_lma_swab(lma);
 203                         rc = 0;
 204                 }
 205         } else if (rc == 0) {
 206                 rc = -ENODATA;
 207         }
 208
 209         return rc;
 210 }
 211
 212 /*
 213  * retrieve object from backend ext fs.
 214  **/
 215 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 216                        struct osd_inode_id *id)
 217 {
 218         struct inode *inode = NULL;
 219
 220         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 221         if (IS_ERR(inode)) {
 222                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 223                        id->oii_ino, PTR_ERR(inode));
 224         } else if (id->oii_gen != OSD_OII_NOGEN &&
 225                    inode->i_generation != id->oii_gen) {
 226                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 227                        "gen1 = %u\n",
 228                        id->oii_ino, id->oii_gen, inode->i_generation);
 229                 iput(inode);
 230                 inode = ERR_PTR(-ESTALE);
 231         } else if (inode->i_nlink == 0) {
 232                 /* due to parallel readdir and unlink,
 233                 * we can have dead inode here. */
 234                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 235                 make_bad_inode(inode);
 236                 iput(inode);
 237                 inode = ERR_PTR(-ESTALE);
 238         } else if (is_bad_inode(inode)) {
 239                 CWARN("%.16s: bad inode: ino = %u\n",
 240                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 241                 iput(inode);
 242                 inode = ERR_PTR(-ENOENT);
 243         } else {
 244                 if (id->oii_gen == OSD_OII_NOGEN)
 245                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 246
 247                 /* Do not update file c/mtime in ldiskfs.
 248                  * NB: we don't have any lock to protect this because we don't
 249                  * have reference on osd_object now, but contention with
 250                  * another lookup + attr_set can't happen in the tiny window
 251                  * between if (...) and set S_NOCMTIME. */
 252                 if (!(inode->i_flags & S_NOCMTIME))
 253                         inode->i_flags |= S_NOCMTIME;
 254         }
 255         return inode;
 256 }
 257
 258 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 259                            struct osd_inode_id *id, struct lu_fid *fid)
 260 {
 261         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 262         struct inode            *inode;
 263         int                      rc;
 264
 265         inode = osd_iget(info, dev, id);
 266         if (IS_ERR(inode))
 267                 return inode;
 268
 269         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 270         if (rc == 0) {
 271                 *fid = lma->lma_self_fid;
 272         } else if (rc == -ENODATA) {
 273                 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
 274         } else {
 275                 iput(inode);
 276                 inode = ERR_PTR(rc);
 277         }
 278         return inode;
 279 }
 280
 281 static struct inode *
 282 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 283                 struct osd_inode_id *id, const struct lu_fid *fid)
 284 {
 285         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 286         struct inode            *inode;
 287         int                      rc;
 288
 289         inode = osd_iget(info, dev, id);
 290         if (IS_ERR(inode))
 291                 return inode;
 292
 293         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 294         if (rc == -ENODATA)
 295                 return inode;
 296
 297         if (rc != 0) {
 298                 iput(inode);
 299                 return ERR_PTR(rc);
 300         }
 301
 302         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 303                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 304                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 305                 iput(inode);
 306                 return ERR_PTR(-EREMCHG);
 307         }
 308
 309         return inode;
 310 }
 311
 312 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 313                           const struct lu_fid *fid,
 314                           const struct lu_object_conf *conf)
 315 {
 316         struct osd_thread_info *info;
 317         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 318         struct osd_device      *dev;
 319         struct osd_idmap_cache *oic;
 320         struct osd_inode_id    *id;
 321         struct inode           *inode;
 322         struct osd_scrub       *scrub;
 323         struct scrub_file      *sf;
 324         int                     result;
 325         int                     verify = 0;
 326         ENTRY;
 327
 328         LINVRNT(osd_invariant(obj));
 329         LASSERT(obj->oo_inode == NULL);
 330         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 331
 332         dev = osd_dev(ldev);
 333         scrub = &dev->od_scrub;
 334         sf = &scrub->os_file;
 335         info = osd_oti_get(env);
 336         LASSERT(info);
 337         oic = &info->oti_cache;
 338
 339         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 340                 RETURN(-ENOENT);
 341
 342         /* Search order: 1. per-thread cache. */
 343         if (lu_fid_eq(fid, &oic->oic_fid)) {
 344                 id = &oic->oic_lid;
 345                 goto iget;
 346         }
 347
 348         id = &info->oti_id;
 349         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 350                 /* Search order: 2. OI scrub pending list. */
 351                 result = osd_oii_lookup(dev, fid, id);
 352                 if (result == 0)
 353                         goto iget;
 354         }
 355
 356         if (sf->sf_flags & SF_INCONSISTENT)
 357                 verify = 1;
 358
 359         /*
 360          * Objects are created as locking anchors or place holders for objects
 361          * yet to be created. No need to osd_oi_lookup() at here because FID
 362          * shouldn't never be re-used, if it's really a duplicate FID from
 363          * unexpected reason, we should be able to detect it later by calling
 364          * do_create->osd_oi_insert()
 365          */
 366         if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
 367                 GOTO(out, result = 0);
 368
 369         /* Search order: 3. OI files. */
 370         result = osd_oi_lookup(info, dev, fid, id);
 371         if (result == -ENOENT) {
 372                 if (!fid_is_norm(fid) ||
 373                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 374                                       sf->sf_oi_bitmap))
 375                         GOTO(out, result = 0);
 376
 377                 goto trigger;
 378         }
 379
 380         if (result != 0)
 381                 GOTO(out, result);
 382
 383 iget:
 384         if (verify == 0)
 385                 inode = osd_iget(info, dev, id);
 386         else
 387                 inode = osd_iget_verify(info, dev, id, fid);
 388         if (IS_ERR(inode)) {
 389                 result = PTR_ERR(inode);
 390                 if (result == -ENOENT || result == -ESTALE) {
 391                         fid_zero(&oic->oic_fid);
 392                         result = 0;
 393                 } else if (result == -EREMCHG) {
 394
 395 trigger:
 396                         if (thread_is_running(&scrub->os_thread)) {
 397                                 result = -EINPROGRESS;
 398                         } else if (!dev->od_noscrub) {
 399                                 result = osd_scrub_start(dev);
 400                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 401                                                "for "DFID", rc = %d [1]\n",
 402                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 403                                                s_volume_name,PFID(fid), result);
 404                                 if (result == 0 || result == -EALREADY)
 405                                         result = -EINPROGRESS;
 406                                 else
 407                                         result = -EREMCHG;
 408                         }
 409                 }
 410
 411                 GOTO(out, result);
 412         }
 413
 414         obj->oo_inode = inode;
 415         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 416
 417         obj->oo_compat_dot_created = 1;
 418         obj->oo_compat_dotdot_created = 1;
 419
 420         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 421                 GOTO(out, result = 0);
 422
 423         LASSERT(obj->oo_hl_head == NULL);
 424         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 425         if (obj->oo_hl_head == NULL) {
 426                 obj->oo_inode = NULL;
 427                 iput(inode);
 428                 GOTO(out, result = -ENOMEM);
 429         }
 430         GOTO(out, result = 0);
 431
 432 out:
 433         LINVRNT(osd_invariant(obj));
 434         return result;
 435 }
 436
 437 /*
 438  * Concurrency: shouldn't matter.
 439  */
 440 static void osd_object_init0(struct osd_object *obj)
 441 {
 442         LASSERT(obj->oo_inode != NULL);
 443         obj->oo_dt.do_body_ops = &osd_body_ops;
 444         obj->oo_dt.do_lu.lo_header->loh_attr |=
 445                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 446 }
 447
 448 /*
 449  * Concurrency: no concurrent access is possible that early in object
 450  * life-cycle.
 451  */
 452 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 453                            const struct lu_object_conf *conf)
 454 {
 455         struct osd_object *obj = osd_obj(l);
 456         int result;
 457
 458         LINVRNT(osd_invariant(obj));
 459
 460         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 461         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 462         if (result == 0) {
 463                 if (obj->oo_inode != NULL) {
 464                         osd_object_init0(obj);
 465                 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 466                         obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 467                         /* LFSCK iterator object is special without inode */
 468                         l->lo_header->loh_attr |= LOHA_EXISTS;
 469                 }
 470         }
 471         LINVRNT(osd_invariant(obj));
 472         return result;
 473 }
 474
 475 /*
 476  * Concurrency: no concurrent access is possible that late in object
 477  * life-cycle.
 478  */
 479 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 480 {
 481         struct osd_object *obj = osd_obj(l);
 482
 483         LINVRNT(osd_invariant(obj));
 484
 485         dt_object_fini(&obj->oo_dt);
 486         if (obj->oo_hl_head != NULL)
 487                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 488         OBD_FREE_PTR(obj);
 489 }
 490
 491 /*
 492  * Concurrency: no concurrent access is possible that late in object
 493  * life-cycle.
 494  */
 495 static void osd_index_fini(struct osd_object *o)
 496 {
 497         struct iam_container *bag;
 498
 499         if (o->oo_dir != NULL) {
 500                 bag = &o->oo_dir->od_container;
 501                 if (o->oo_inode != NULL) {
 502                         if (bag->ic_object == o->oo_inode)
 503                                 iam_container_fini(bag);
 504                 }
 505                 OBD_FREE_PTR(o->oo_dir);
 506                 o->oo_dir = NULL;
 507         }
 508 }
 509
 510 /*
 511  * Concurrency: no concurrent access is possible that late in object
 512  * life-cycle (for all existing callers, that is. New callers have to provide
 513  * their own locking.)
 514  */
 515 static int osd_inode_unlinked(const struct inode *inode)
 516 {
 517         return inode->i_nlink == 0;
 518 }
 519
 520 enum {
 521         OSD_TXN_OI_DELETE_CREDITS    = 20,
 522         OSD_TXN_INODE_DELETE_CREDITS = 20
 523 };
 524
 525 /*
 526  * Journal
 527  */
 528
 529 #if OSD_THANDLE_STATS
 530 /**
 531  * Set time when the handle is allocated
 532  */
 533 static void osd_th_alloced(struct osd_thandle *oth)
 534 {
 535         oth->oth_alloced = cfs_time_current();
 536 }
 537
 538 /**
 539  * Set time when the handle started
 540  */
 541 static void osd_th_started(struct osd_thandle *oth)
 542 {
 543         oth->oth_started = cfs_time_current();
 544 }
 545
 546 /**
 547  * Helper function to convert time interval to microseconds packed in
 548  * long int.
 549  */
 550 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 551 {
 552         struct timeval val;
 553
 554         cfs_duration_usec(cfs_time_sub(end, start), &val);
 555         return val.tv_sec * 1000000 + val.tv_usec;
 556 }
 557
 558 /**
 559  * Check whether the we deal with this handle for too long.
 560  */
 561 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 562                                 cfs_time_t alloced, cfs_time_t started,
 563                                 cfs_time_t closed)
 564 {
 565         cfs_time_t now = cfs_time_current();
 566
 567         LASSERT(dev != NULL);
 568
 569         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 570                             interval_to_usec(alloced, started));
 571         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 572                             interval_to_usec(started, closed));
 573         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 574                             interval_to_usec(closed, now));
 575
 576         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 577                 CWARN("transaction handle %p was open for too long: "
 578                       "now "CFS_TIME_T" ,"
 579                       "alloced "CFS_TIME_T" ,"
 580                       "started "CFS_TIME_T" ,"
 581                       "closed "CFS_TIME_T"\n",
 582                       oth, now, alloced, started, closed);
 583                 libcfs_debug_dumpstack(NULL);
 584         }
 585 }
 586
 587 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 588 {                                                                       \
 589         cfs_time_t __closed = cfs_time_current();                       \
 590         cfs_time_t __alloced = oth->oth_alloced;                        \
 591         cfs_time_t __started = oth->oth_started;                        \
 592                                                                         \
 593         expr;                                                           \
 594         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 595 }
 596
 597 #else /* OSD_THANDLE_STATS */
 598
 599 #define osd_th_alloced(h)                  do {} while(0)
 600 #define osd_th_started(h)                  do {} while(0)
 601 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 602
 603 #endif /* OSD_THANDLE_STATS */
 604
 605 /*
 606  * Concurrency: doesn't access mutable data.
 607  */
 608 static int osd_param_is_not_sane(const struct osd_device *dev,
 609                                  const struct thandle *th)
 610 {
 611         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 612
 613         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 614 }
 615
 616 /*
 617  * Concurrency: shouldn't matter.
 618  */
 619 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 620 static void osd_trans_commit_cb(struct super_block *sb,
 621                                 struct journal_callback *jcb, int error)
 622 #else
 623 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 624 #endif
 625 {
 626         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 627         struct thandle     *th  = &oh->ot_super;
 628         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 629         struct dt_txn_commit_cb *dcb, *tmp;
 630
 631         LASSERT(oh->ot_handle == NULL);
 632
 633         if (error)
 634                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 635
 636         dt_txn_hook_commit(th);
 637
 638         /* call per-transaction callbacks if any */
 639         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 640                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 641                          "commit callback entry: magic=%x name='%s'\n",
 642                          dcb->dcb_magic, dcb->dcb_name);
 643                 cfs_list_del_init(&dcb->dcb_linkage);
 644                 dcb->dcb_func(NULL, th, dcb, error);
 645         }
 646
 647         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 648         lu_device_put(lud);
 649         th->th_dev = NULL;
 650
 651         lu_context_exit(&th->th_ctx);
 652         lu_context_fini(&th->th_ctx);
 653         OBD_FREE_PTR(oh);
 654 }
 655
 656 static struct thandle *osd_trans_create(const struct lu_env *env,
 657                                         struct dt_device *d)
 658 {
 659         struct osd_thread_info *oti = osd_oti_get(env);
 660         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 661         struct osd_thandle     *oh;
 662         struct thandle         *th;
 663         ENTRY;
 664
 665         /* on pending IO in this thread should left from prev. request */
 666         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 667
 668         th = ERR_PTR(-ENOMEM);
 669         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 670         if (oh != NULL) {
 671                 oh->ot_quota_trans = &oti->oti_quota_trans;
 672                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 673                 th = &oh->ot_super;
 674                 th->th_dev = d;
 675                 th->th_result = 0;
 676                 th->th_tags = LCT_TX_HANDLE;
 677                 oh->ot_credits = 0;
 678                 oti->oti_dev = osd_dt_dev(d);
 679                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 680                 osd_th_alloced(oh);
 681
 682                 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
 683                 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
 684                 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
 685                 oti->oti_rollback = false;
 686         }
 687         RETURN(th);
 688 }
 689
 690 /*
 691  * Concurrency: shouldn't matter.
 692  */
 693 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 694                     struct thandle *th)
 695 {
 696         struct osd_thread_info *oti = osd_oti_get(env);
 697         struct osd_device  *dev = osd_dt_dev(d);
 698         handle_t           *jh;
 699         struct osd_thandle *oh;
 700         int rc;
 701
 702         ENTRY;
 703
 704         LASSERT(current->journal_info == NULL);
 705
 706         oh = container_of0(th, struct osd_thandle, ot_super);
 707         LASSERT(oh != NULL);
 708         LASSERT(oh->ot_handle == NULL);
 709
 710         rc = dt_txn_hook_start(env, d, th);
 711         if (rc != 0)
 712                 GOTO(out, rc);
 713
 714         if (unlikely(osd_param_is_not_sane(dev, th))) {
 715                 static unsigned long last_printed;
 716                 static int last_credits;
 717
 718                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 719                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 720                       oh->ot_credits,
 721                       osd_journal(dev)->j_max_transaction_buffers);
 722 #ifdef OSD_TRACK_DECLARES
 723                 CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
 724                       oti->oti_declare_ops[OSD_OT_CREATE],
 725                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
 726                       oti->oti_declare_ops[OSD_OT_DELETE],
 727                       oti->oti_declare_ops_cred[OSD_OT_DELETE],
 728                       oti->oti_declare_ops[OSD_OT_DESTROY],
 729                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 730                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
 731                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
 732                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
 733                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
 734                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
 735                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
 736                       oti->oti_declare_ops[OSD_OT_WRITE],
 737                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
 738                       oti->oti_declare_ops[OSD_OT_PUNCH],
 739                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
 740                       oti->oti_declare_ops[OSD_OT_QUOTA],
 741                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
 742                 CWARN("  insert: %u/%u, delete: %u/%u\n",
 743                       oti->oti_declare_ops[OSD_OT_INSERT],
 744                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
 745                       oti->oti_declare_ops[OSD_OT_DESTROY],
 746                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 747                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
 748                       oti->oti_declare_ops[OSD_OT_REF_ADD],
 749                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
 750                       oti->oti_declare_ops[OSD_OT_REF_DEL],
 751                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
 752
 753                 if (last_credits != oh->ot_credits &&
 754                     time_after(jiffies, last_printed + 60 * HZ)) {
 755                         libcfs_debug_dumpstack(NULL);
 756                         last_credits = oh->ot_credits;
 757                         last_printed = jiffies;
 758                 }
 759 #endif
 760                 /* XXX Limit the credits to 'max_transaction_buffers', and
 761                  *     let the underlying filesystem to catch the error if
 762                  *     we really need so many credits.
 763                  *
 764                  *     This should be removed when we can calculate the
 765                  *     credits precisely. */
 766                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 767         }
 768
 769         /*
 770          * XXX temporary stuff. Some abstraction layer should
 771          * be used.
 772          */
 773         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 774         osd_th_started(oh);
 775         if (!IS_ERR(jh)) {
 776                 oh->ot_handle = jh;
 777                 LASSERT(oti->oti_txns == 0);
 778                 lu_context_init(&th->th_ctx, th->th_tags);
 779                 lu_context_enter(&th->th_ctx);
 780
 781                 lu_device_get(&d->dd_lu_dev);
 782                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 783                                              "osd-tx", th);
 784                 oti->oti_txns++;
 785                 rc = 0;
 786         } else {
 787                 rc = PTR_ERR(jh);
 788         }
 789 out:
 790         RETURN(rc);
 791 }
 792
 793 /*
 794  * Concurrency: shouldn't matter.
 795  */
 796 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 797 {
 798         int                     rc = 0;
 799         struct osd_thandle     *oh;
 800         struct osd_thread_info *oti = osd_oti_get(env);
 801         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 802         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
 803         ENTRY;
 804
 805         oh = container_of0(th, struct osd_thandle, ot_super);
 806
 807         if (qsd != NULL)
 808                 /* inform the quota slave device that the transaction is
 809                  * stopping */
 810                 qsd_op_end(env, qsd, oh->ot_quota_trans);
 811         oh->ot_quota_trans = NULL;
 812
 813         if (oh->ot_handle != NULL) {
 814                 handle_t *hdl = oh->ot_handle;
 815
 816                 /*
 817                  * add commit callback
 818                  * notice we don't do this in osd_trans_start()
 819                  * as underlying transaction can change during truncate
 820                  */
 821                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 822                                          &oh->ot_jcb);
 823
 824                 LASSERT(oti->oti_txns == 1);
 825                 oti->oti_txns--;
 826                 rc = dt_txn_hook_stop(env, th);
 827                 if (rc != 0)
 828                         CERROR("Failure in transaction hook: %d\n", rc);
 829
 830                 /* hook functions might modify th_sync */
 831                 hdl->h_sync = th->th_sync;
 832
 833                 oh->ot_handle = NULL;
 834                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 835                                   rc = ldiskfs_journal_stop(hdl));
 836                 if (rc != 0)
 837                         CERROR("Failure to stop transaction: %d\n", rc);
 838         } else {
 839                 OBD_FREE_PTR(oh);
 840         }
 841
 842         /* as we want IO to journal and data IO be concurrent, we don't block
 843          * awaiting data IO completion in osd_do_bio(), instead we wait here
 844          * once transaction is submitted to the journal. all reqular requests
 845          * don't do direct IO (except read/write), thus this wait_event becomes
 846          * no-op for them.
 847          *
 848          * IMPORTANT: we have to wait till any IO submited by the thread is
 849          * completed otherwise iobuf may be corrupted by different request
 850          */
 851         cfs_wait_event(iobuf->dr_wait,
 852                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 853         if (!rc)
 854                 rc = iobuf->dr_error;
 855
 856         RETURN(rc);
 857 }
 858
 859 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 860 {
 861         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 862                                                ot_super);
 863
 864         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 865         LASSERT(&dcb->dcb_func != NULL);
 866         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 867
 868         return 0;
 869 }
 870
 871 /*
 872  * Called just before object is freed. Releases all resources except for
 873  * object itself (that is released by osd_object_free()).
 874  *
 875  * Concurrency: no concurrent access is possible that late in object
 876  * life-cycle.
 877  */
 878 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 879 {
 880         struct osd_object *obj   = osd_obj(l);
 881         struct inode      *inode = obj->oo_inode;
 882
 883         LINVRNT(osd_invariant(obj));
 884
 885         /*
 886          * If object is unlinked remove fid->ino mapping from object index.
 887          */
 888
 889         osd_index_fini(obj);
 890         if (inode != NULL) {
 891                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
 892                 qid_t                    uid = inode->i_uid;
 893                 qid_t                    gid = inode->i_gid;
 894
 895                 iput(inode);
 896                 obj->oo_inode = NULL;
 897
 898                 if (qsd != NULL) {
 899                         struct osd_thread_info  *info = osd_oti_get(env);
 900                         struct lquota_id_info   *qi = &info->oti_qi;
 901
 902                         /* Release granted quota to master if necessary */
 903                         qi->lqi_id.qid_uid = uid;
 904                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
 905
 906                         qi->lqi_id.qid_uid = gid;
 907                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
 908                 }
 909         }
 910 }
 911
 912 /*
 913  * Concurrency: ->loo_object_release() is called under site spin-lock.
 914  */
 915 static void osd_object_release(const struct lu_env *env,
 916                                struct lu_object *l)
 917 {
 918 }
 919
 920 /*
 921  * Concurrency: shouldn't matter.
 922  */
 923 static int osd_object_print(const struct lu_env *env, void *cookie,
 924                             lu_printer_t p, const struct lu_object *l)
 925 {
 926         struct osd_object *o = osd_obj(l);
 927         struct iam_descr  *d;
 928
 929         if (o->oo_dir != NULL)
 930                 d = o->oo_dir->od_container.ic_descr;
 931         else
 932                 d = NULL;
 933         return (*p)(env, cookie,
 934                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 935                     o, o->oo_inode,
 936                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 937                     o->oo_inode ? o->oo_inode->i_generation : 0,
 938                     d ? d->id_ops->id_name : "plain");
 939 }
 940
 941 /*
 942  * Concurrency: shouldn't matter.
 943  */
 944 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 945                struct obd_statfs *sfs)
 946 {
 947         struct osd_device  *osd = osd_dt_dev(d);
 948         struct super_block *sb = osd_sb(osd);
 949         struct kstatfs     *ksfs;
 950         int result = 0;
 951
 952         if (unlikely(osd->od_mnt == NULL))
 953                 return -EINPROGRESS;
 954
 955         /* osd_lproc.c call this without env, allocate ksfs for that case */
 956         if (unlikely(env == NULL)) {
 957                 OBD_ALLOC_PTR(ksfs);
 958                 if (ksfs == NULL)
 959                         return -ENOMEM;
 960         } else {
 961                 ksfs = &osd_oti_get(env)->oti_ksfs;
 962         }
 963
 964         spin_lock(&osd->od_osfs_lock);
 965         /* cache 1 second */
 966         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 967                 result = sb->s_op->statfs(sb->s_root, ksfs);
 968                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 969                         osd->od_osfs_age = cfs_time_current_64();
 970                         statfs_pack(&osd->od_statfs, ksfs);
 971                         if (sb->s_flags & MS_RDONLY)
 972                                 sfs->os_state = OS_STATE_READONLY;
 973                 }
 974         }
 975
 976         if (likely(result == 0))
 977                 *sfs = osd->od_statfs;
 978         spin_unlock(&osd->od_osfs_lock);
 979
 980         if (unlikely(env == NULL))
 981                 OBD_FREE_PTR(ksfs);
 982
 983         return result;
 984 }
 985
 986 /**
 987  * Estimate space needed for file creations. We assume the largest filename
 988  * which is 2^64 - 1, hence a filename of 20 chars.
 989  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
 990  */
 991 #ifdef __LDISKFS_DIR_REC_LEN
 992 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
 993 #else
 994 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
 995 #endif
 996
 997 /*
 998  * Concurrency: doesn't access mutable data.
 999  */
1000 static void osd_conf_get(const struct lu_env *env,
1001                          const struct dt_device *dev,
1002                          struct dt_device_param *param)
1003 {
1004         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1005
1006         /*
1007          * XXX should be taken from not-yet-existing fs abstraction layer.
1008          */
1009         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1010         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1011         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1012         param->ddp_block_shift  = sb->s_blocksize_bits;
1013         param->ddp_mount_type     = LDD_MT_LDISKFS;
1014         param->ddp_maxbytes       = sb->s_maxbytes;
1015         /* Overhead estimate should be fairly accurate, so we really take a tiny
1016          * error margin which also avoids fragmenting the filesystem too much */
1017         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1018         /* inode are statically allocated, so per-inode space consumption
1019          * is the space consumed by the directory entry */
1020         param->ddp_inodespace     = PER_OBJ_USAGE;
1021         /* per-fragment overhead to be used by the client code */
1022         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1023         param->ddp_mntopts      = 0;
1024         if (test_opt(sb, XATTR_USER))
1025                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1026         if (test_opt(sb, POSIX_ACL))
1027                 param->ddp_mntopts |= MNTOPT_ACL;
1028
1029 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1030         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1031                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1032         else
1033 #endif
1034                 param->ddp_max_ea_size = sb->s_blocksize;
1035
1036 }
1037
1038 /**
1039  * Helper function to get and fill the buffer with input values.
1040  */
1041 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1042 {
1043         struct lu_buf *buf;
1044
1045         buf = &osd_oti_get(env)->oti_buf;
1046         buf->lb_buf = area;
1047         buf->lb_len = len;
1048         return buf;
1049 }
1050
1051 /*
1052  * Concurrency: shouldn't matter.
1053  */
1054 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1055 {
1056         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1057         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1058 }
1059
1060 /**
1061  * Start commit for OSD device.
1062  *
1063  * An implementation of dt_commit_async method for OSD device.
1064  * Asychronously starts underlayng fs sync and thereby a transaction
1065  * commit.
1066  *
1067  * \param env environment
1068  * \param d dt device
1069  *
1070  * \see dt_device_operations
1071  */
1072 static int osd_commit_async(const struct lu_env *env,
1073                             struct dt_device *d)
1074 {
1075         struct super_block *s = osd_sb(osd_dt_dev(d));
1076         ENTRY;
1077
1078         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1079         RETURN(s->s_op->sync_fs(s, 0));
1080 }
1081
1082 /*
1083  * Concurrency: shouldn't matter.
1084  */
1085
1086 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1087 {
1088         struct super_block *sb = osd_sb(osd_dt_dev(d));
1089         int rc;
1090         ENTRY;
1091
1092         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1093
1094         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1095         RETURN(rc);
1096 }
1097
1098 /*
1099  * Concurrency: serialization provided by callers.
1100  */
1101 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1102                               int mode, unsigned long timeout, __u32 alg,
1103                               struct lustre_capa_key *keys)
1104 {
1105         struct osd_device *dev = osd_dt_dev(d);
1106         ENTRY;
1107
1108         dev->od_fl_capa = mode;
1109         dev->od_capa_timeout = timeout;
1110         dev->od_capa_alg = alg;
1111         dev->od_capa_keys = keys;
1112         RETURN(0);
1113 }
1114
1115 /**
1116  * Note: we do not count into QUOTA here.
1117  * If we mount with --data_journal we may need more.
1118  */
1119 const int osd_dto_credits_noquota[DTO_NR] = {
1120         /**
1121          * Insert/Delete.
1122          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1123          * SINGLEDATA_TRANS_BLOCKS(8)
1124          * XXX Note: maybe iam need more, since iam have more level than
1125          *           EXT3 htree.
1126          */
1127         [DTO_INDEX_INSERT]  = 16,
1128         [DTO_INDEX_DELETE]  = 16,
1129         /**
1130          * Used for OI scrub
1131          */
1132         [DTO_INDEX_UPDATE]  = 16,
1133         /**
1134          * Create a object. The same as create object in EXT3.
1135          * DATA_TRANS_BLOCKS(14) +
1136          * INDEX_EXTRA_BLOCKS(8) +
1137          * 3(inode bits, groups, GDT)
1138          */
1139         [DTO_OBJECT_CREATE] = 25,
1140         /**
1141          * XXX: real credits to be fixed
1142          */
1143         [DTO_OBJECT_DELETE] = 25,
1144         /**
1145          * Attr set credits (inode)
1146          */
1147         [DTO_ATTR_SET_BASE] = 1,
1148         /**
1149          * Xattr set. The same as xattr of EXT3.
1150          * DATA_TRANS_BLOCKS(14)
1151          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1152          * are also counted in. Do not know why?
1153          */
1154         [DTO_XATTR_SET]     = 14,
1155         [DTO_LOG_REC]       = 14,
1156         /**
1157          * credits for inode change during write.
1158          */
1159         [DTO_WRITE_BASE]    = 3,
1160         /**
1161          * credits for single block write.
1162          */
1163         [DTO_WRITE_BLOCK]   = 14,
1164         /**
1165          * Attr set credits for chown.
1166          * This is extra credits for setattr, and it is null without quota
1167          */
1168         [DTO_ATTR_SET_CHOWN]= 0
1169 };
1170
1171 static const struct dt_device_operations osd_dt_ops = {
1172         .dt_root_get       = osd_root_get,
1173         .dt_statfs         = osd_statfs,
1174         .dt_trans_create   = osd_trans_create,
1175         .dt_trans_start    = osd_trans_start,
1176         .dt_trans_stop     = osd_trans_stop,
1177         .dt_trans_cb_add   = osd_trans_cb_add,
1178         .dt_conf_get       = osd_conf_get,
1179         .dt_sync           = osd_sync,
1180         .dt_ro             = osd_ro,
1181         .dt_commit_async   = osd_commit_async,
1182         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1183 };
1184
1185 static void osd_object_read_lock(const struct lu_env *env,
1186                                  struct dt_object *dt, unsigned role)
1187 {
1188         struct osd_object *obj = osd_dt_obj(dt);
1189         struct osd_thread_info *oti = osd_oti_get(env);
1190
1191         LINVRNT(osd_invariant(obj));
1192
1193         LASSERT(obj->oo_owner != env);
1194         down_read_nested(&obj->oo_sem, role);
1195
1196         LASSERT(obj->oo_owner == NULL);
1197         oti->oti_r_locks++;
1198 }
1199
1200 static void osd_object_write_lock(const struct lu_env *env,
1201                                   struct dt_object *dt, unsigned role)
1202 {
1203         struct osd_object *obj = osd_dt_obj(dt);
1204         struct osd_thread_info *oti = osd_oti_get(env);
1205
1206         LINVRNT(osd_invariant(obj));
1207
1208         LASSERT(obj->oo_owner != env);
1209         down_write_nested(&obj->oo_sem, role);
1210
1211         LASSERT(obj->oo_owner == NULL);
1212         obj->oo_owner = env;
1213         oti->oti_w_locks++;
1214 }
1215
1216 static void osd_object_read_unlock(const struct lu_env *env,
1217                                    struct dt_object *dt)
1218 {
1219         struct osd_object *obj = osd_dt_obj(dt);
1220         struct osd_thread_info *oti = osd_oti_get(env);
1221
1222         LINVRNT(osd_invariant(obj));
1223
1224         LASSERT(oti->oti_r_locks > 0);
1225         oti->oti_r_locks--;
1226         up_read(&obj->oo_sem);
1227 }
1228
1229 static void osd_object_write_unlock(const struct lu_env *env,
1230                                     struct dt_object *dt)
1231 {
1232         struct osd_object *obj = osd_dt_obj(dt);
1233         struct osd_thread_info *oti = osd_oti_get(env);
1234
1235         LINVRNT(osd_invariant(obj));
1236
1237         LASSERT(obj->oo_owner == env);
1238         LASSERT(oti->oti_w_locks > 0);
1239         oti->oti_w_locks--;
1240         obj->oo_owner = NULL;
1241         up_write(&obj->oo_sem);
1242 }
1243
1244 static int osd_object_write_locked(const struct lu_env *env,
1245                                    struct dt_object *dt)
1246 {
1247         struct osd_object *obj = osd_dt_obj(dt);
1248
1249         LINVRNT(osd_invariant(obj));
1250
1251         return obj->oo_owner == env;
1252 }
1253
1254 static int capa_is_sane(const struct lu_env *env,
1255                         struct osd_device *dev,
1256                         struct lustre_capa *capa,
1257                         struct lustre_capa_key *keys)
1258 {
1259         struct osd_thread_info *oti = osd_oti_get(env);
1260         struct lustre_capa *tcapa = &oti->oti_capa;
1261         struct obd_capa *oc;
1262         int i, rc = 0;
1263         ENTRY;
1264
1265         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1266         if (oc) {
1267                 if (capa_is_expired(oc)) {
1268                         DEBUG_CAPA(D_ERROR, capa, "expired");
1269                         rc = -ESTALE;
1270                 }
1271                 capa_put(oc);
1272                 RETURN(rc);
1273         }
1274
1275         if (capa_is_expired_sec(capa)) {
1276                 DEBUG_CAPA(D_ERROR, capa, "expired");
1277                 RETURN(-ESTALE);
1278         }
1279
1280         spin_lock(&capa_lock);
1281         for (i = 0; i < 2; i++) {
1282                 if (keys[i].lk_keyid == capa->lc_keyid) {
1283                         oti->oti_capa_key = keys[i];
1284                         break;
1285                 }
1286         }
1287         spin_unlock(&capa_lock);
1288
1289         if (i == 2) {
1290                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1291                 RETURN(-ESTALE);
1292         }
1293
1294         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1295         if (rc)
1296                 RETURN(rc);
1297
1298         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1299                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1300                 RETURN(-EACCES);
1301         }
1302
1303         oc = capa_add(dev->od_capa_hash, capa);
1304         capa_put(oc);
1305
1306         RETURN(0);
1307 }
1308
1309 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1310                     struct lustre_capa *capa, __u64 opc)
1311 {
1312         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1313         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1314         struct md_capainfo *ci;
1315         int rc;
1316
1317         if (!dev->od_fl_capa)
1318                 return 0;
1319
1320         if (capa == BYPASS_CAPA)
1321                 return 0;
1322
1323         ci = md_capainfo(env);
1324         if (unlikely(!ci))
1325                 return 0;
1326
1327         if (ci->mc_auth == LC_ID_NONE)
1328                 return 0;
1329
1330         if (!capa) {
1331                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1332                 return -EACCES;
1333         }
1334
1335         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1336                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1337                            PFID(fid));
1338                 return -EACCES;
1339         }
1340
1341         if (!capa_opc_supported(capa, opc)) {
1342                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1343                 return -EACCES;
1344         }
1345
1346         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1347                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1348                 return -EACCES;
1349         }
1350
1351         return 0;
1352 }
1353
1354 static struct timespec *osd_inode_time(const struct lu_env *env,
1355                                        struct inode *inode, __u64 seconds)
1356 {
1357         struct osd_thread_info  *oti = osd_oti_get(env);
1358         struct timespec         *t   = &oti->oti_time;
1359
1360         t->tv_sec = seconds;
1361         t->tv_nsec = 0;
1362         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1363         return t;
1364 }
1365
1366
1367 static void osd_inode_getattr(const struct lu_env *env,
1368                               struct inode *inode, struct lu_attr *attr)
1369 {
1370         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1371                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1372                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1373
1374         attr->la_atime      = LTIME_S(inode->i_atime);
1375         attr->la_mtime      = LTIME_S(inode->i_mtime);
1376         attr->la_ctime      = LTIME_S(inode->i_ctime);
1377         attr->la_mode       = inode->i_mode;
1378         attr->la_size       = i_size_read(inode);
1379         attr->la_blocks     = inode->i_blocks;
1380         attr->la_uid        = inode->i_uid;
1381         attr->la_gid        = inode->i_gid;
1382         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1383         attr->la_nlink      = inode->i_nlink;
1384         attr->la_rdev       = inode->i_rdev;
1385         attr->la_blksize    = 1 << inode->i_blkbits;
1386         attr->la_blkbits    = inode->i_blkbits;
1387 }
1388
1389 static int osd_attr_get(const struct lu_env *env,
1390                         struct dt_object *dt,
1391                         struct lu_attr *attr,
1392                         struct lustre_capa *capa)
1393 {
1394         struct osd_object *obj = osd_dt_obj(dt);
1395
1396         LASSERT(dt_object_exists(dt));
1397         LINVRNT(osd_invariant(obj));
1398
1399         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1400                 return -EACCES;
1401
1402         spin_lock(&obj->oo_guard);
1403         osd_inode_getattr(env, obj->oo_inode, attr);
1404         spin_unlock(&obj->oo_guard);
1405         return 0;
1406 }
1407
1408 static int osd_declare_attr_set(const struct lu_env *env,
1409                                 struct dt_object *dt,
1410                                 const struct lu_attr *attr,
1411                                 struct thandle *handle)
1412 {
1413         struct osd_thandle     *oh;
1414         struct osd_object      *obj;
1415         struct osd_thread_info *info = osd_oti_get(env);
1416         struct lquota_id_info  *qi = &info->oti_qi;
1417         long long               bspace;
1418         int                     rc = 0;
1419         bool                    allocated;
1420         ENTRY;
1421
1422         LASSERT(dt != NULL);
1423         LASSERT(handle != NULL);
1424
1425         obj = osd_dt_obj(dt);
1426         LASSERT(osd_invariant(obj));
1427
1428         oh = container_of0(handle, struct osd_thandle, ot_super);
1429         LASSERT(oh->ot_handle == NULL);
1430
1431         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1432                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1433
1434         if (attr == NULL || obj->oo_inode == NULL)
1435                 RETURN(rc);
1436
1437         bspace   = obj->oo_inode->i_blocks;
1438         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1439         bspace   = toqb(bspace);
1440
1441         /* Changing ownership is always preformed by super user, it should not
1442          * fail with EDQUOT.
1443          *
1444          * We still need to call the osd_declare_qid() to calculate the journal
1445          * credits for updating quota accounting files and to trigger quota
1446          * space adjustment once the operation is completed.*/
1447         if ((attr->la_valid & LA_UID) != 0 &&
1448              attr->la_uid != obj->oo_inode->i_uid) {
1449                 qi->lqi_type = USRQUOTA;
1450
1451                 /* inode accounting */
1452                 qi->lqi_is_blk = false;
1453
1454                 /* one more inode for the new owner ... */
1455                 qi->lqi_id.qid_uid = attr->la_uid;
1456                 qi->lqi_space      = 1;
1457                 allocated = (attr->la_uid == 0) ? true : false;
1458                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1459                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1460                         rc = 0;
1461                 if (rc)
1462                         RETURN(rc);
1463
1464                 /* and one less inode for the current uid */
1465                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1466                 qi->lqi_space      = -1;
1467                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1468                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1469                         rc = 0;
1470                 if (rc)
1471                         RETURN(rc);
1472
1473                 /* block accounting */
1474                 qi->lqi_is_blk = true;
1475
1476                 /* more blocks for the new owner ... */
1477                 qi->lqi_id.qid_uid = attr->la_uid;
1478                 qi->lqi_space      = bspace;
1479                 allocated = (attr->la_uid == 0) ? true : false;
1480                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1481                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1482                         rc = 0;
1483                 if (rc)
1484                         RETURN(rc);
1485
1486                 /* and finally less blocks for the current owner */
1487                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1488                 qi->lqi_space      = -bspace;
1489                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1490                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1491                         rc = 0;
1492                 if (rc)
1493                         RETURN(rc);
1494         }
1495
1496         if (attr->la_valid & LA_GID &&
1497             attr->la_gid != obj->oo_inode->i_gid) {
1498                 qi->lqi_type = GRPQUOTA;
1499
1500                 /* inode accounting */
1501                 qi->lqi_is_blk = false;
1502
1503                 /* one more inode for the new group owner ... */
1504                 qi->lqi_id.qid_gid = attr->la_gid;
1505                 qi->lqi_space      = 1;
1506                 allocated = (attr->la_gid == 0) ? true : false;
1507                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1508                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1509                         rc = 0;
1510                 if (rc)
1511                         RETURN(rc);
1512
1513                 /* and one less inode for the current gid */
1514                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1515                 qi->lqi_space      = -1;
1516                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1517                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1518                         rc = 0;
1519                 if (rc)
1520                         RETURN(rc);
1521
1522                 /* block accounting */
1523                 qi->lqi_is_blk = true;
1524
1525                 /* more blocks for the new owner ... */
1526                 qi->lqi_id.qid_gid = attr->la_gid;
1527                 qi->lqi_space      = bspace;
1528                 allocated = (attr->la_gid == 0) ? true : false;
1529                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1530                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1531                         rc = 0;
1532                 if (rc)
1533                         RETURN(rc);
1534
1535                 /* and finally less blocks for the current owner */
1536                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1537                 qi->lqi_space      = -bspace;
1538                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1539                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1540                         rc = 0;
1541                 if (rc)
1542                         RETURN(rc);
1543         }
1544
1545         RETURN(rc);
1546 }
1547
1548 static int osd_inode_setattr(const struct lu_env *env,
1549                              struct inode *inode, const struct lu_attr *attr)
1550 {
1551         __u64 bits;
1552
1553         bits = attr->la_valid;
1554
1555         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1556
1557         if (bits & LA_ATIME)
1558                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1559         if (bits & LA_CTIME)
1560                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1561         if (bits & LA_MTIME)
1562                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1563         if (bits & LA_SIZE) {
1564                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1565                 i_size_write(inode, attr->la_size);
1566         }
1567
1568 #if 0
1569         /* OSD should not change "i_blocks" which is used by quota.
1570          * "i_blocks" should be changed by ldiskfs only. */
1571         if (bits & LA_BLOCKS)
1572                 inode->i_blocks = attr->la_blocks;
1573 #endif
1574         if (bits & LA_MODE)
1575                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1576                         (attr->la_mode & ~S_IFMT);
1577         if (bits & LA_UID)
1578                 inode->i_uid    = attr->la_uid;
1579         if (bits & LA_GID)
1580                 inode->i_gid    = attr->la_gid;
1581         if (bits & LA_NLINK)
1582                 set_nlink(inode, attr->la_nlink);
1583         if (bits & LA_RDEV)
1584                 inode->i_rdev   = attr->la_rdev;
1585
1586         if (bits & LA_FLAGS) {
1587                 /* always keep S_NOCMTIME */
1588                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1589                                  S_NOCMTIME;
1590         }
1591         return 0;
1592 }
1593
1594 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1595 {
1596         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1597             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1598                 struct iattr    iattr;
1599                 int             rc;
1600
1601                 iattr.ia_valid = 0;
1602                 if (attr->la_valid & LA_UID)
1603                         iattr.ia_valid |= ATTR_UID;
1604                 if (attr->la_valid & LA_GID)
1605                         iattr.ia_valid |= ATTR_GID;
1606                 iattr.ia_uid = attr->la_uid;
1607                 iattr.ia_gid = attr->la_gid;
1608
1609                 rc = ll_vfs_dq_transfer(inode, &iattr);
1610                 if (rc) {
1611                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1612                                "enforcement enabled on the ldiskfs filesystem?",
1613                                inode->i_sb->s_id, rc);
1614                         return rc;
1615                 }
1616         }
1617         return 0;
1618 }
1619
1620 static int osd_attr_set(const struct lu_env *env,
1621                         struct dt_object *dt,
1622                         const struct lu_attr *attr,
1623                         struct thandle *handle,
1624                         struct lustre_capa *capa)
1625 {
1626         struct osd_object *obj = osd_dt_obj(dt);
1627         struct inode      *inode;
1628         int rc;
1629
1630         LASSERT(handle != NULL);
1631         LASSERT(dt_object_exists(dt));
1632         LASSERT(osd_invariant(obj));
1633
1634         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1635                 return -EACCES;
1636
1637         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1638
1639         inode = obj->oo_inode;
1640         ll_vfs_dq_init(inode);
1641
1642         rc = osd_quota_transfer(inode, attr);
1643         if (rc)
1644                 return rc;
1645
1646         spin_lock(&obj->oo_guard);
1647         rc = osd_inode_setattr(env, inode, attr);
1648         spin_unlock(&obj->oo_guard);
1649
1650         if (!rc)
1651                 inode->i_sb->s_op->dirty_inode(inode);
1652         return rc;
1653 }
1654
1655 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1656                                     struct osd_object *obj,
1657                                     const char *name, const int namelen)
1658 {
1659         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1660 }
1661
1662 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1663                       cfs_umode_t mode,
1664                       struct dt_allocation_hint *hint,
1665                       struct thandle *th)
1666 {
1667         int result;
1668         struct osd_device  *osd = osd_obj2dev(obj);
1669         struct osd_thandle *oth;
1670         struct dt_object   *parent = NULL;
1671         struct inode       *inode;
1672
1673         LINVRNT(osd_invariant(obj));
1674         LASSERT(obj->oo_inode == NULL);
1675         LASSERT(obj->oo_hl_head == NULL);
1676
1677         if (S_ISDIR(mode) && ldiskfs_pdo) {
1678                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1679                 if (obj->oo_hl_head == NULL)
1680                         return -ENOMEM;
1681         }
1682
1683         oth = container_of(th, struct osd_thandle, ot_super);
1684         LASSERT(oth->ot_handle->h_transaction != NULL);
1685
1686         if (hint && hint->dah_parent)
1687                 parent = hint->dah_parent;
1688
1689         inode = ldiskfs_create_inode(oth->ot_handle,
1690                                      parent ? osd_dt_obj(parent)->oo_inode :
1691                                               osd_sb(osd)->s_root->d_inode,
1692                                      mode);
1693         if (!IS_ERR(inode)) {
1694                 /* Do not update file c/mtime in ldiskfs.
1695                  * NB: don't need any lock because no contention at this
1696                  * early stage */
1697                 inode->i_flags |= S_NOCMTIME;
1698                 inode->i_state |= I_LUSTRE_NOSCRUB;
1699                 obj->oo_inode = inode;
1700                 result = 0;
1701         } else {
1702                 if (obj->oo_hl_head != NULL) {
1703                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1704                         obj->oo_hl_head = NULL;
1705                 }
1706                 result = PTR_ERR(inode);
1707         }
1708         LINVRNT(osd_invariant(obj));
1709         return result;
1710 }
1711
1712 enum {
1713         OSD_NAME_LEN = 255
1714 };
1715
1716 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1717                      struct lu_attr *attr,
1718                      struct dt_allocation_hint *hint,
1719                      struct dt_object_format *dof,
1720                      struct thandle *th)
1721 {
1722         int result;
1723         struct osd_thandle *oth;
1724         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1725
1726         LASSERT(S_ISDIR(attr->la_mode));
1727
1728         oth = container_of(th, struct osd_thandle, ot_super);
1729         LASSERT(oth->ot_handle->h_transaction != NULL);
1730         result = osd_mkfile(info, obj, mode, hint, th);
1731
1732         return result;
1733 }
1734
1735 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1736                         struct lu_attr *attr,
1737                         struct dt_allocation_hint *hint,
1738                         struct dt_object_format *dof,
1739                         struct thandle *th)
1740 {
1741         int result;
1742         struct osd_thandle *oth;
1743         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1744
1745         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1746
1747         LASSERT(S_ISREG(attr->la_mode));
1748
1749         oth = container_of(th, struct osd_thandle, ot_super);
1750         LASSERT(oth->ot_handle->h_transaction != NULL);
1751
1752         result = osd_mkfile(info, obj, mode, hint, th);
1753         if (result == 0) {
1754                 LASSERT(obj->oo_inode != NULL);
1755                 if (feat->dif_flags & DT_IND_VARKEY)
1756                         result = iam_lvar_create(obj->oo_inode,
1757                                                  feat->dif_keysize_max,
1758                                                  feat->dif_ptrsize,
1759                                                  feat->dif_recsize_max,
1760                                                  oth->ot_handle);
1761                 else
1762                         result = iam_lfix_create(obj->oo_inode,
1763                                                  feat->dif_keysize_max,
1764                                                  feat->dif_ptrsize,
1765                                                  feat->dif_recsize_max,
1766                                                  oth->ot_handle);
1767
1768         }
1769         return result;
1770 }
1771
1772 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1773                      struct lu_attr *attr,
1774                      struct dt_allocation_hint *hint,
1775                      struct dt_object_format *dof,
1776                      struct thandle *th)
1777 {
1778         LASSERT(S_ISREG(attr->la_mode));
1779         return osd_mkfile(info, obj, (attr->la_mode &
1780                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1781 }
1782
1783 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1784                      struct lu_attr *attr,
1785                      struct dt_allocation_hint *hint,
1786                      struct dt_object_format *dof,
1787                      struct thandle *th)
1788 {
1789         LASSERT(S_ISLNK(attr->la_mode));
1790         return osd_mkfile(info, obj, (attr->la_mode &
1791                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1792 }
1793
1794 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1795                      struct lu_attr *attr,
1796                      struct dt_allocation_hint *hint,
1797                      struct dt_object_format *dof,
1798                      struct thandle *th)
1799 {
1800         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1801         int result;
1802
1803         LINVRNT(osd_invariant(obj));
1804         LASSERT(obj->oo_inode == NULL);
1805         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1806                 S_ISFIFO(mode) || S_ISSOCK(mode));
1807
1808         result = osd_mkfile(info, obj, mode, hint, th);
1809         if (result == 0) {
1810                 LASSERT(obj->oo_inode != NULL);
1811                 /*
1812                  * This inode should be marked dirty for i_rdev.  Currently
1813                  * that is done in the osd_attr_init().
1814                  */
1815                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1816                                    attr->la_rdev);
1817         }
1818         LINVRNT(osd_invariant(obj));
1819         return result;
1820 }
1821
1822 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1823                               struct lu_attr *,
1824                               struct dt_allocation_hint *hint,
1825                               struct dt_object_format *dof,
1826                               struct thandle *);
1827
1828 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1829 {
1830         osd_obj_type_f result;
1831
1832         switch (type) {
1833         case DFT_DIR:
1834                 result = osd_mkdir;
1835                 break;
1836         case DFT_REGULAR:
1837                 result = osd_mkreg;
1838                 break;
1839         case DFT_SYM:
1840                 result = osd_mksym;
1841                 break;
1842         case DFT_NODE:
1843                 result = osd_mknod;
1844                 break;
1845         case DFT_INDEX:
1846                 result = osd_mk_index;
1847                 break;
1848
1849         default:
1850                 LBUG();
1851                 break;
1852         }
1853         return result;
1854 }
1855
1856
1857 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1858                         struct dt_object *parent, struct dt_object *child,
1859                         cfs_umode_t child_mode)
1860 {
1861         LASSERT(ah);
1862
1863         memset(ah, 0, sizeof(*ah));
1864         ah->dah_parent = parent;
1865         ah->dah_mode = child_mode;
1866 }
1867
1868 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1869                           struct lu_attr *attr, struct dt_object_format *dof)
1870 {
1871         struct inode   *inode = obj->oo_inode;
1872         __u64           valid = attr->la_valid;
1873         int             result;
1874
1875         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1876
1877         if (dof->dof_type != DFT_NODE)
1878                 attr->la_valid &= ~LA_RDEV;
1879         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1880                 attr->la_valid &= ~LA_ATIME;
1881         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1882                 attr->la_valid &= ~LA_CTIME;
1883         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1884                 attr->la_valid &= ~LA_MTIME;
1885
1886         result = osd_quota_transfer(inode, attr);
1887         if (result)
1888                 return;
1889
1890         if (attr->la_valid != 0) {
1891                 result = osd_inode_setattr(info->oti_env, inode, attr);
1892                 /*
1893                  * The osd_inode_setattr() should always succeed here.  The
1894                  * only error that could be returned is EDQUOT when we are
1895                  * trying to change the UID or GID of the inode. However, this
1896                  * should not happen since quota enforcement is no longer
1897                  * enabled on ldiskfs (lquota takes care of it).
1898                  */
1899                 LASSERTF(result == 0, "%d", result);
1900                 inode->i_sb->s_op->dirty_inode(inode);
1901         }
1902
1903         attr->la_valid = valid;
1904 }
1905
1906 /**
1907  * Helper function for osd_object_create()
1908  *
1909  * \retval 0, on success
1910  */
1911 static int __osd_object_create(struct osd_thread_info *info,
1912                                struct osd_object *obj, struct lu_attr *attr,
1913                                struct dt_allocation_hint *hint,
1914                                struct dt_object_format *dof,
1915                                struct thandle *th)
1916 {
1917         int     result;
1918
1919         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1920                                                   th);
1921         if (result == 0) {
1922                 osd_attr_init(info, obj, attr, dof);
1923                 osd_object_init0(obj);
1924                 /* bz 24037 */
1925                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1926                         unlock_new_inode(obj->oo_inode);
1927         }
1928
1929         return result;
1930 }
1931
1932 /**
1933  * Helper function for osd_object_create()
1934  *
1935  * \retval 0, on success
1936  */
1937 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1938                            const struct lu_fid *fid, struct thandle *th)
1939 {
1940         struct osd_thread_info *info = osd_oti_get(env);
1941         struct osd_inode_id    *id   = &info->oti_id;
1942         struct osd_device      *osd  = osd_obj2dev(obj);
1943
1944         LASSERT(obj->oo_inode != NULL);
1945
1946         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1947         return osd_oi_insert(info, osd, fid, id, th);
1948 }
1949
1950 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1951                    const struct lu_fid *fid, struct lu_seq_range *range)
1952 {
1953         struct seq_server_site  *ss = osd_seq_site(osd);
1954         int                     rc;
1955
1956         if (fid_is_igif(fid)) {
1957                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1958                 range->lsr_index = 0;
1959                 return 0;
1960         }
1961
1962         if (fid_is_idif(fid)) {
1963                 range->lsr_flags = LU_SEQ_RANGE_OST;
1964                 range->lsr_index = fid_idif_ost_idx(fid);
1965                 return 0;
1966         }
1967
1968         if (!fid_is_norm(fid)) {
1969                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1970                 if (ss != NULL)
1971                         /* FIXME: If ss is NULL, it suppose not get lsr_index
1972                          * at all */
1973                         range->lsr_index = ss->ss_node_id;
1974                 return 0;
1975         }
1976
1977         LASSERT(ss != NULL);
1978         range->lsr_flags = -1;
1979         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1980         if (rc != 0) {
1981                 CERROR("%s can not find "DFID": rc = %d\n",
1982                        osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc);
1983         }
1984         return rc;
1985 }
1986
1987
1988 static int osd_declare_object_create(const struct lu_env *env,
1989                                      struct dt_object *dt,
1990                                      struct lu_attr *attr,
1991                                      struct dt_allocation_hint *hint,
1992                                      struct dt_object_format *dof,
1993                                      struct thandle *handle)
1994 {
1995         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1996         struct osd_thandle      *oh;
1997         int                      rc;
1998         ENTRY;
1999
2000         LASSERT(handle != NULL);
2001
2002         oh = container_of0(handle, struct osd_thandle, ot_super);
2003         LASSERT(oh->ot_handle == NULL);
2004
2005         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2006                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2007         /* XXX: So far, only normal fid needs be inserted into the oi,
2008          *      things could be changed later. Revise following code then. */
2009         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2010             !fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2011                            lu_object_fid(&dt->do_lu))) {
2012                 /* Reuse idle OI block may cause additional one OI block
2013                  * to be changed. */
2014                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2015                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2016         }
2017         /* If this is directory, then we expect . and .. to be inserted as
2018          * well. The one directory block always needs to be created for the
2019          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2020          * block), there is no danger of needing a tree for the first block.
2021          */
2022         if (attr && S_ISDIR(attr->la_mode)) {
2023                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2024                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2025                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2026         }
2027
2028         if (!attr)
2029                 RETURN(0);
2030
2031         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2032                                    false, false, NULL, false);
2033         if (rc != 0)
2034                 RETURN(rc);
2035
2036         /* It does fld look up inside declare, and the result will be
2037          * added to fld cache, so the following fld lookup inside insert
2038          * does not need send RPC anymore, so avoid send rpc with holding
2039          * transaction */
2040         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2041                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2042                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2043                                lu_object_fid(&dt->do_lu), range);
2044
2045
2046         RETURN(rc);
2047 }
2048
2049 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2050                              struct lu_attr *attr,
2051                              struct dt_allocation_hint *hint,
2052                              struct dt_object_format *dof,
2053                              struct thandle *th)
2054 {
2055         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2056         struct osd_object      *obj    = osd_dt_obj(dt);
2057         struct osd_thread_info *info   = osd_oti_get(env);
2058         int result;
2059
2060         ENTRY;
2061
2062         LINVRNT(osd_invariant(obj));
2063         LASSERT(!dt_object_exists(dt));
2064         LASSERT(osd_write_locked(env, obj));
2065         LASSERT(th != NULL);
2066
2067         if (unlikely(fid_is_acct(fid)))
2068                 /* Quota files can't be created from the kernel any more,
2069                  * 'tune2fs -O quota' will take care of creating them */
2070                 RETURN(-EPERM);
2071
2072         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2073         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2074
2075         result = __osd_object_create(info, obj, attr, hint, dof, th);
2076         if (result == 0)
2077                 result = __osd_oi_insert(env, obj, fid, th);
2078
2079         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2080         LASSERT(osd_invariant(obj));
2081         RETURN(result);
2082 }
2083
2084 /**
2085  * Called to destroy on-disk representation of the object
2086  *
2087  * Concurrency: must be locked
2088  */
2089 static int osd_declare_object_destroy(const struct lu_env *env,
2090                                       struct dt_object *dt,
2091                                       struct thandle *th)
2092 {
2093         struct osd_object  *obj = osd_dt_obj(dt);
2094         struct inode       *inode = obj->oo_inode;
2095         struct osd_thandle *oh;
2096         int                 rc;
2097         ENTRY;
2098
2099         oh = container_of0(th, struct osd_thandle, ot_super);
2100         LASSERT(oh->ot_handle == NULL);
2101         LASSERT(inode);
2102
2103         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2104                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2105         /* XXX: So far, only normal fid needs to be inserted into the OI,
2106          *      so only normal fid needs to be removed from the OI also.
2107          * Recycle idle OI leaf may cause additional three OI blocks
2108          * to be changed. */
2109         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2110                              fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2111                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0);
2112
2113         /* one less inode */
2114         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2115                                    false, true, NULL, false);
2116         if (rc)
2117                 RETURN(rc);
2118         /* data to be truncated */
2119         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2120                                    true, true, NULL, false);
2121         RETURN(rc);
2122 }
2123
2124 static int osd_object_destroy(const struct lu_env *env,
2125                               struct dt_object *dt,
2126                               struct thandle *th)
2127 {
2128         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2129         struct osd_object      *obj = osd_dt_obj(dt);
2130         struct inode           *inode = obj->oo_inode;
2131         struct osd_device      *osd = osd_obj2dev(obj);
2132         struct osd_thandle     *oh;
2133         int                     result;
2134         ENTRY;
2135
2136         oh = container_of0(th, struct osd_thandle, ot_super);
2137         LASSERT(oh->ot_handle);
2138         LASSERT(inode);
2139         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2140
2141         if (unlikely(fid_is_acct(fid)))
2142                 RETURN(-EPERM);
2143
2144         /* Parallel control for OI scrub. For most of cases, there is no
2145          * lock contention. So it will not affect unlink performance. */
2146         mutex_lock(&inode->i_mutex);
2147         if (S_ISDIR(inode->i_mode)) {
2148                 LASSERT(osd_inode_unlinked(inode) ||
2149                         inode->i_nlink == 1);
2150                 spin_lock(&obj->oo_guard);
2151                 clear_nlink(inode);
2152                 spin_unlock(&obj->oo_guard);
2153                 inode->i_sb->s_op->dirty_inode(inode);
2154         }
2155
2156         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2157
2158         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2159         mutex_unlock(&inode->i_mutex);
2160
2161         /* XXX: add to ext3 orphan list */
2162         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2163
2164         /* not needed in the cache anymore */
2165         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2166
2167         RETURN(0);
2168 }
2169
2170 /**
2171  * Helper function for osd_xattr_set()
2172  */
2173 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2174                            const struct lu_buf *buf, const char *name, int fl)
2175 {
2176         struct osd_object      *obj      = osd_dt_obj(dt);
2177         struct inode           *inode    = obj->oo_inode;
2178         struct osd_thread_info *info     = osd_oti_get(env);
2179         struct dentry          *dentry   = &info->oti_child_dentry;
2180         int                     fs_flags = 0;
2181         int                     rc;
2182
2183         LASSERT(dt_object_exists(dt));
2184         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2185
2186         if (fl & LU_XATTR_REPLACE)
2187                 fs_flags |= XATTR_REPLACE;
2188
2189         if (fl & LU_XATTR_CREATE)
2190                 fs_flags |= XATTR_CREATE;
2191
2192         ll_vfs_dq_init(inode);
2193         dentry->d_inode = inode;
2194         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2195                                    buf->lb_len, fs_flags);
2196         return rc;
2197 }
2198
2199 /**
2200  * Put the fid into lustre_mdt_attrs, and then place the structure
2201  * inode's ea. This fid should not be altered during the life time
2202  * of the inode.
2203  *
2204  * \retval +ve, on success
2205  * \retval -ve, on error
2206  *
2207  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2208  */
2209 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2210                           const struct lu_fid *fid)
2211 {
2212         struct osd_thread_info  *info      = osd_oti_get(env);
2213         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2214
2215         lustre_lma_init(mdt_attrs, fid);
2216         lustre_lma_swab(mdt_attrs);
2217         return __osd_xattr_set(env, dt,
2218                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2219                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2220
2221 }
2222
2223 /**
2224  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2225  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2226  * To have compatilibility with 1.8 ldiskfs driver we need to have
2227  * magic number at start of fid data.
2228  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2229  * its inmemory API.
2230  */
2231 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2232                                   const struct dt_rec *fid)
2233 {
2234         param->edp_magic = LDISKFS_LUFID_MAGIC;
2235         param->edp_len =  sizeof(struct lu_fid) + 1;
2236
2237         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2238                       (struct lu_fid *)fid);
2239 }
2240
2241 /**
2242  * Try to read the fid from inode ea into dt_rec, if return value
2243  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2244  *
2245  * \param fid object fid.
2246  *
2247  * \retval 0 on success
2248  */
2249 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2250                           __u32 ino, struct lu_fid *fid,
2251                           struct osd_inode_id *id)
2252 {
2253         struct osd_thread_info *info  = osd_oti_get(env);
2254         struct inode           *inode;
2255         ENTRY;
2256
2257         osd_id_gen(id, ino, OSD_OII_NOGEN);
2258         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2259         if (IS_ERR(inode))
2260                 RETURN(PTR_ERR(inode));
2261
2262         iput(inode);
2263         RETURN(0);
2264 }
2265
2266 /**
2267  * OSD layer object create function for interoperability mode (b11826).
2268  * This is mostly similar to osd_object_create(). Only difference being, fid is
2269  * inserted into inode ea here.
2270  *
2271  * \retval   0, on success
2272  * \retval -ve, on error
2273  */
2274 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2275                                 struct lu_attr *attr,
2276                                 struct dt_allocation_hint *hint,
2277                                 struct dt_object_format *dof,
2278                                 struct thandle *th)
2279 {
2280         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2281         struct osd_object      *obj    = osd_dt_obj(dt);
2282         struct osd_thread_info *info   = osd_oti_get(env);
2283         int                     result;
2284
2285         ENTRY;
2286
2287         LASSERT(osd_invariant(obj));
2288         LASSERT(!dt_object_exists(dt));
2289         LASSERT(osd_write_locked(env, obj));
2290         LASSERT(th != NULL);
2291
2292         if (unlikely(fid_is_acct(fid)))
2293                 /* Quota files can't be created from the kernel any more,
2294                  * 'tune2fs -O quota' will take care of creating them */
2295                 RETURN(-EPERM);
2296
2297         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2298         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2299
2300         result = __osd_object_create(info, obj, attr, hint, dof, th);
2301         /* objects under osd root shld have igif fid, so dont add fid EA */
2302         /* For ost object, the fid will be stored during first write */
2303         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL &&
2304             !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid))
2305                 result = osd_ea_fid_set(env, dt, fid);
2306
2307         if (result == 0)
2308                 result = __osd_oi_insert(env, obj, fid, th);
2309
2310         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2311         LINVRNT(osd_invariant(obj));
2312         RETURN(result);
2313 }
2314
2315 static int osd_declare_object_ref_add(const struct lu_env *env,
2316                                       struct dt_object *dt,
2317                                       struct thandle *handle)
2318 {
2319         struct osd_thandle       *oh;
2320
2321         /* it's possible that object doesn't exist yet */
2322         LASSERT(handle != NULL);
2323
2324         oh = container_of0(handle, struct osd_thandle, ot_super);
2325         LASSERT(oh->ot_handle == NULL);
2326
2327         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2328                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2329
2330         return 0;
2331 }
2332
2333 /*
2334  * Concurrency: @dt is write locked.
2335  */
2336 static int osd_object_ref_add(const struct lu_env *env,
2337                               struct dt_object *dt, struct thandle *th)
2338 {
2339         struct osd_object *obj = osd_dt_obj(dt);
2340         struct inode      *inode = obj->oo_inode;
2341
2342         LINVRNT(osd_invariant(obj));
2343         LASSERT(dt_object_exists(dt));
2344         LASSERT(osd_write_locked(env, obj));
2345         LASSERT(th != NULL);
2346
2347         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2348
2349         /*
2350          * DIR_NLINK feature is set for compatibility reasons if:
2351          * 1) nlinks > LDISKFS_LINK_MAX, or
2352          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2353          *
2354          * It is easier to always set this flag (rather than check and set),
2355          * since it has less overhead, and the superblock will be dirtied
2356          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2357          * do not actually care whether this flag is set or not.
2358          */
2359         spin_lock(&obj->oo_guard);
2360         /* inc_nlink from 0 may cause WARN_ON */
2361         if(inode->i_nlink == 0)
2362                 set_nlink(inode, 1);
2363         else
2364                 inc_nlink(inode);
2365         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2366                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2367                     inode->i_nlink == 2)
2368                         set_nlink(inode, 1);
2369         }
2370         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2371         spin_unlock(&obj->oo_guard);
2372         inode->i_sb->s_op->dirty_inode(inode);
2373         LINVRNT(osd_invariant(obj));
2374
2375         return 0;
2376 }
2377
2378 static int osd_declare_object_ref_del(const struct lu_env *env,
2379                                       struct dt_object *dt,
2380                                       struct thandle *handle)
2381 {
2382         struct osd_thandle *oh;
2383
2384         LASSERT(dt_object_exists(dt));
2385         LASSERT(handle != NULL);
2386
2387         oh = container_of0(handle, struct osd_thandle, ot_super);
2388         LASSERT(oh->ot_handle == NULL);
2389
2390         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2391                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2392
2393         return 0;
2394 }
2395
2396 /*
2397  * Concurrency: @dt is write locked.
2398  */
2399 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2400                               struct thandle *th)
2401 {
2402         struct osd_object *obj = osd_dt_obj(dt);
2403         struct inode      *inode = obj->oo_inode;
2404
2405         LINVRNT(osd_invariant(obj));
2406         LASSERT(dt_object_exists(dt));
2407         LASSERT(osd_write_locked(env, obj));
2408         LASSERT(th != NULL);
2409
2410         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2411
2412         spin_lock(&obj->oo_guard);
2413         LASSERT(inode->i_nlink > 0);
2414         drop_nlink(inode);
2415         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2416          * then the nlink count is 1. Don't let it be set to 0 or the directory
2417          * inode will be deleted incorrectly. */
2418         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2419                 set_nlink(inode, 1);
2420         spin_unlock(&obj->oo_guard);
2421         inode->i_sb->s_op->dirty_inode(inode);
2422         LINVRNT(osd_invariant(obj));
2423
2424         return 0;
2425 }
2426
2427 /*
2428  * Get the 64-bit version for an inode.
2429  */
2430 static int osd_object_version_get(const struct lu_env *env,
2431                                   struct dt_object *dt, dt_obj_version_t *ver)
2432 {
2433         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2434
2435         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2436                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2437         *ver = LDISKFS_I(inode)->i_fs_version;
2438         return 0;
2439 }
2440
2441 /*
2442  * Concurrency: @dt is read locked.
2443  */
2444 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2445                          struct lu_buf *buf, const char *name,
2446                          struct lustre_capa *capa)
2447 {
2448         struct osd_object      *obj    = osd_dt_obj(dt);
2449         struct inode           *inode  = obj->oo_inode;
2450         struct osd_thread_info *info   = osd_oti_get(env);
2451         struct dentry          *dentry = &info->oti_obj_dentry;
2452
2453         /* version get is not real XATTR but uses xattr API */
2454         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2455                 /* for version we are just using xattr API but change inode
2456                  * field instead */
2457                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2458                 osd_object_version_get(env, dt, buf->lb_buf);
2459                 return sizeof(dt_obj_version_t);
2460         }
2461
2462         LASSERT(dt_object_exists(dt));
2463         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2464
2465         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2466                 return -EACCES;
2467
2468         dentry->d_inode = inode;
2469         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2470 }
2471
2472
2473 static int osd_declare_xattr_set(const struct lu_env *env,
2474                                  struct dt_object *dt,
2475                                  const struct lu_buf *buf, const char *name,
2476                                  int fl, struct thandle *handle)
2477 {
2478         struct osd_thandle *oh;
2479
2480         LASSERT(handle != NULL);
2481
2482         oh = container_of0(handle, struct osd_thandle, ot_super);
2483         LASSERT(oh->ot_handle == NULL);
2484
2485         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2486                              strcmp(name, XATTR_NAME_VERSION) == 0 ?
2487                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2488                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2489
2490         return 0;
2491 }
2492
2493 /*
2494  * Set the 64-bit version for object
2495  */
2496 static void osd_object_version_set(const struct lu_env *env,
2497                                    struct dt_object *dt,
2498                                    dt_obj_version_t *new_version)
2499 {
2500         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2501
2502         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2503                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2504
2505         LDISKFS_I(inode)->i_fs_version = *new_version;
2506         /** Version is set after all inode operations are finished,
2507          *  so we should mark it dirty here */
2508         inode->i_sb->s_op->dirty_inode(inode);
2509 }
2510
2511 /*
2512  * Concurrency: @dt is write locked.
2513  */
2514 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2515                          const struct lu_buf *buf, const char *name, int fl,
2516                          struct thandle *handle, struct lustre_capa *capa)
2517 {
2518         LASSERT(handle != NULL);
2519
2520         /* version set is not real XATTR */
2521         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2522                 /* for version we are just using xattr API but change inode
2523                  * field instead */
2524                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2525                 osd_object_version_set(env, dt, buf->lb_buf);
2526                 return sizeof(dt_obj_version_t);
2527         }
2528
2529         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2530                 return -EACCES;
2531
2532         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2533         return __osd_xattr_set(env, dt, buf, name, fl);
2534 }
2535
2536 /*
2537  * Concurrency: @dt is read locked.
2538  */
2539 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2540                           struct lu_buf *buf, struct lustre_capa *capa)
2541 {
2542         struct osd_object      *obj    = osd_dt_obj(dt);
2543         struct inode           *inode  = obj->oo_inode;
2544         struct osd_thread_info *info   = osd_oti_get(env);
2545         struct dentry          *dentry = &info->oti_obj_dentry;
2546
2547         LASSERT(dt_object_exists(dt));
2548         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2549         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2550
2551         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2552                 return -EACCES;
2553
2554         dentry->d_inode = inode;
2555         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2556 }
2557
2558 static int osd_declare_xattr_del(const struct lu_env *env,
2559                                  struct dt_object *dt, const char *name,
2560                                  struct thandle *handle)
2561 {
2562         struct osd_thandle *oh;
2563
2564         LASSERT(dt_object_exists(dt));
2565         LASSERT(handle != NULL);
2566
2567         oh = container_of0(handle, struct osd_thandle, ot_super);
2568         LASSERT(oh->ot_handle == NULL);
2569
2570         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2571                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2572
2573         return 0;
2574 }
2575
2576 /*
2577  * Concurrency: @dt is write locked.
2578  */
2579 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2580                          const char *name, struct thandle *handle,
2581                          struct lustre_capa *capa)
2582 {
2583         struct osd_object      *obj    = osd_dt_obj(dt);
2584         struct inode           *inode  = obj->oo_inode;
2585         struct osd_thread_info *info   = osd_oti_get(env);
2586         struct dentry          *dentry = &info->oti_obj_dentry;
2587         int                     rc;
2588
2589         LASSERT(dt_object_exists(dt));
2590         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2591         LASSERT(osd_write_locked(env, obj));
2592         LASSERT(handle != NULL);
2593
2594         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2595                 return -EACCES;
2596
2597         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2598
2599         ll_vfs_dq_init(inode);
2600         dentry->d_inode = inode;
2601         rc = inode->i_op->removexattr(dentry, name);
2602         return rc;
2603 }
2604
2605 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2606                                      struct dt_object *dt,
2607                                      struct lustre_capa *old,
2608                                      __u64 opc)
2609 {
2610         struct osd_thread_info *info = osd_oti_get(env);
2611         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2612         struct osd_object *obj = osd_dt_obj(dt);
2613         struct osd_device *dev = osd_obj2dev(obj);
2614         struct lustre_capa_key *key = &info->oti_capa_key;
2615         struct lustre_capa *capa = &info->oti_capa;
2616         struct obd_capa *oc;
2617         struct md_capainfo *ci;
2618         int rc;
2619         ENTRY;
2620
2621         if (!dev->od_fl_capa)
2622                 RETURN(ERR_PTR(-ENOENT));
2623
2624         LASSERT(dt_object_exists(dt));
2625         LINVRNT(osd_invariant(obj));
2626
2627         /* renewal sanity check */
2628         if (old && osd_object_auth(env, dt, old, opc))
2629                 RETURN(ERR_PTR(-EACCES));
2630
2631         ci = md_capainfo(env);
2632         if (unlikely(!ci))
2633                 RETURN(ERR_PTR(-ENOENT));
2634
2635         switch (ci->mc_auth) {
2636         case LC_ID_NONE:
2637                 RETURN(NULL);
2638         case LC_ID_PLAIN:
2639                 capa->lc_uid = obj->oo_inode->i_uid;
2640                 capa->lc_gid = obj->oo_inode->i_gid;
2641                 capa->lc_flags = LC_ID_PLAIN;
2642                 break;
2643         case LC_ID_CONVERT: {
2644                 __u32 d[4], s[4];
2645
2646                 s[0] = obj->oo_inode->i_uid;
2647                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2648                 s[2] = obj->oo_inode->i_gid;
2649                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2650                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2651                 if (unlikely(rc))
2652                         RETURN(ERR_PTR(rc));
2653
2654                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2655                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2656                 capa->lc_flags = LC_ID_CONVERT;
2657                 break;
2658         }
2659         default:
2660                 RETURN(ERR_PTR(-EINVAL));
2661         }
2662
2663         capa->lc_fid = *fid;
2664         capa->lc_opc = opc;
2665         capa->lc_flags |= dev->od_capa_alg << 24;
2666         capa->lc_timeout = dev->od_capa_timeout;
2667         capa->lc_expiry = 0;
2668
2669         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2670         if (oc) {
2671                 LASSERT(!capa_is_expired(oc));
2672                 RETURN(oc);
2673         }
2674
2675         spin_lock(&capa_lock);
2676         *key = dev->od_capa_keys[1];
2677         spin_unlock(&capa_lock);
2678
2679         capa->lc_keyid = key->lk_keyid;
2680         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2681
2682         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2683         if (rc) {
2684                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2685                 RETURN(ERR_PTR(rc));
2686         }
2687
2688         oc = capa_add(dev->od_capa_hash, capa);
2689         RETURN(oc);
2690 }
2691
2692 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2693 {
2694         struct osd_object       *obj    = osd_dt_obj(dt);
2695         struct inode            *inode  = obj->oo_inode;
2696         struct osd_thread_info  *info   = osd_oti_get(env);
2697         struct dentry           *dentry = &info->oti_obj_dentry;
2698         struct file             *file   = &info->oti_file;
2699         int                     rc;
2700
2701         ENTRY;
2702
2703         dentry->d_inode = inode;
2704         file->f_dentry = dentry;
2705         file->f_mapping = inode->i_mapping;
2706         file->f_op = inode->i_fop;
2707         mutex_lock(&inode->i_mutex);
2708         rc = file->f_op->fsync(file, dentry, 0);
2709         mutex_unlock(&inode->i_mutex);
2710         RETURN(rc);
2711 }
2712
2713 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2714                         void **data)
2715 {
2716         struct osd_object *obj = osd_dt_obj(dt);
2717         ENTRY;
2718
2719         *data = (void *)obj->oo_inode;
2720         RETURN(0);
2721 }
2722
2723 /*
2724  * Index operations.
2725  */
2726
2727 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2728                            const struct dt_index_features *feat)
2729 {
2730         struct iam_descr *descr;
2731
2732         if (osd_object_is_root(o))
2733                 return feat == &dt_directory_features;
2734
2735         LASSERT(o->oo_dir != NULL);
2736
2737         descr = o->oo_dir->od_container.ic_descr;
2738         if (feat == &dt_directory_features) {
2739                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2740                         return 1;
2741                 else
2742                         return 0;
2743         } else {
2744                 return
2745                         feat->dif_keysize_min <= descr->id_key_size &&
2746                         descr->id_key_size <= feat->dif_keysize_max &&
2747                         feat->dif_recsize_min <= descr->id_rec_size &&
2748                         descr->id_rec_size <= feat->dif_recsize_max &&
2749                         !(feat->dif_flags & (DT_IND_VARKEY |
2750                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2751                         ergo(feat->dif_flags & DT_IND_UPDATE,
2752                              1 /* XXX check that object (and file system) is
2753                                 * writable */);
2754         }
2755 }
2756
2757 static int osd_iam_container_init(const struct lu_env *env,
2758                                   struct osd_object *obj,
2759                                   struct osd_directory *dir)
2760 {
2761         struct iam_container *bag = &dir->od_container;
2762         int result;
2763
2764         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2765         if (result != 0)
2766                 return result;
2767
2768         result = iam_container_setup(bag);
2769         if (result == 0)
2770                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2771         else
2772                 iam_container_fini(bag);
2773
2774         return result;
2775 }
2776
2777
2778 /*
2779  * Concurrency: no external locking is necessary.
2780  */
2781 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2782                          const struct dt_index_features *feat)
2783 {
2784         int                      result;
2785         int                      skip_iam = 0;
2786         struct osd_object       *obj = osd_dt_obj(dt);
2787
2788         LINVRNT(osd_invariant(obj));
2789         LASSERT(dt_object_exists(dt));
2790
2791         if (osd_object_is_root(obj)) {
2792                 dt->do_index_ops = &osd_index_ea_ops;
2793                 result = 0;
2794         } else if (feat == &dt_directory_features) {
2795                 dt->do_index_ops = &osd_index_ea_ops;
2796                 if (S_ISDIR(obj->oo_inode->i_mode))
2797                         result = 0;
2798                 else
2799                         result = -ENOTDIR;
2800                 skip_iam = 1;
2801         } else if (unlikely(feat == &dt_otable_features)) {
2802                 dt->do_index_ops = &osd_otable_ops;
2803                 return 0;
2804         } else if (feat == &dt_acct_features) {
2805                 dt->do_index_ops = &osd_acct_index_ops;
2806                 result = 0;
2807                 skip_iam = 1;
2808         } else if (!osd_has_index(obj)) {
2809                 struct osd_directory *dir;
2810
2811                 OBD_ALLOC_PTR(dir);
2812                 if (dir != NULL) {
2813
2814                         spin_lock(&obj->oo_guard);
2815                         if (obj->oo_dir == NULL)
2816                                 obj->oo_dir = dir;
2817                         else
2818                                 /*
2819                                  * Concurrent thread allocated container data.
2820                                  */
2821                                 OBD_FREE_PTR(dir);
2822                         spin_unlock(&obj->oo_guard);
2823                         /*
2824                          * Now, that we have container data, serialize its
2825                          * initialization.
2826                          */
2827                         down_write(&obj->oo_ext_idx_sem);
2828                         /*
2829                          * recheck under lock.
2830                          */
2831                         if (!osd_has_index(obj))
2832                                 result = osd_iam_container_init(env, obj, dir);
2833                         else
2834                                 result = 0;
2835                         up_write(&obj->oo_ext_idx_sem);
2836                 } else {
2837                         result = -ENOMEM;
2838                 }
2839         } else {
2840                 result = 0;
2841         }
2842
2843         if (result == 0 && skip_iam == 0) {
2844                 if (!osd_iam_index_probe(env, obj, feat))
2845                         result = -ENOTDIR;
2846         }
2847         LINVRNT(osd_invariant(obj));
2848
2849         if (is_quota_glb_feat(feat))
2850                 result = osd_quota_migration(env, dt, feat);
2851
2852         return result;
2853 }
2854
2855 static int osd_otable_it_attr_get(const struct lu_env *env,
2856                                  struct dt_object *dt,
2857                                  struct lu_attr *attr,
2858                                  struct lustre_capa *capa)
2859 {
2860         attr->la_valid = 0;
2861         return 0;
2862 }
2863
2864 static const struct dt_object_operations osd_obj_ops = {
2865         .do_read_lock         = osd_object_read_lock,
2866         .do_write_lock        = osd_object_write_lock,
2867         .do_read_unlock       = osd_object_read_unlock,
2868         .do_write_unlock      = osd_object_write_unlock,
2869         .do_write_locked      = osd_object_write_locked,
2870         .do_attr_get          = osd_attr_get,
2871         .do_declare_attr_set  = osd_declare_attr_set,
2872         .do_attr_set          = osd_attr_set,
2873         .do_ah_init           = osd_ah_init,
2874         .do_declare_create    = osd_declare_object_create,
2875         .do_create            = osd_object_create,
2876         .do_declare_destroy   = osd_declare_object_destroy,
2877         .do_destroy           = osd_object_destroy,
2878         .do_index_try         = osd_index_try,
2879         .do_declare_ref_add   = osd_declare_object_ref_add,
2880         .do_ref_add           = osd_object_ref_add,
2881         .do_declare_ref_del   = osd_declare_object_ref_del,
2882         .do_ref_del           = osd_object_ref_del,
2883         .do_xattr_get         = osd_xattr_get,
2884         .do_declare_xattr_set = osd_declare_xattr_set,
2885         .do_xattr_set         = osd_xattr_set,
2886         .do_declare_xattr_del = osd_declare_xattr_del,
2887         .do_xattr_del         = osd_xattr_del,
2888         .do_xattr_list        = osd_xattr_list,
2889         .do_capa_get          = osd_capa_get,
2890         .do_object_sync       = osd_object_sync,
2891         .do_data_get          = osd_data_get,
2892 };
2893
2894 /**
2895  * dt_object_operations for interoperability mode
2896  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2897  */
2898 static const struct dt_object_operations osd_obj_ea_ops = {
2899         .do_read_lock         = osd_object_read_lock,
2900         .do_write_lock        = osd_object_write_lock,
2901         .do_read_unlock       = osd_object_read_unlock,
2902         .do_write_unlock      = osd_object_write_unlock,
2903         .do_write_locked      = osd_object_write_locked,
2904         .do_attr_get          = osd_attr_get,
2905         .do_declare_attr_set  = osd_declare_attr_set,
2906         .do_attr_set          = osd_attr_set,
2907         .do_ah_init           = osd_ah_init,
2908         .do_declare_create    = osd_declare_object_create,
2909         .do_create            = osd_object_ea_create,
2910         .do_declare_destroy   = osd_declare_object_destroy,
2911         .do_destroy           = osd_object_destroy,
2912         .do_index_try         = osd_index_try,
2913         .do_declare_ref_add   = osd_declare_object_ref_add,
2914         .do_ref_add           = osd_object_ref_add,
2915         .do_declare_ref_del   = osd_declare_object_ref_del,
2916         .do_ref_del           = osd_object_ref_del,
2917         .do_xattr_get         = osd_xattr_get,
2918         .do_declare_xattr_set = osd_declare_xattr_set,
2919         .do_xattr_set         = osd_xattr_set,
2920         .do_declare_xattr_del = osd_declare_xattr_del,
2921         .do_xattr_del         = osd_xattr_del,
2922         .do_xattr_list        = osd_xattr_list,
2923         .do_capa_get          = osd_capa_get,
2924         .do_object_sync       = osd_object_sync,
2925         .do_data_get          = osd_data_get,
2926 };
2927
2928 static const struct dt_object_operations osd_obj_otable_it_ops = {
2929         .do_attr_get    = osd_otable_it_attr_get,
2930         .do_index_try   = osd_index_try,
2931 };
2932
2933 static int osd_index_declare_iam_delete(const struct lu_env *env,
2934                                         struct dt_object *dt,
2935                                         const struct dt_key *key,
2936                                         struct thandle *handle)
2937 {
2938         struct osd_thandle    *oh;
2939
2940         oh = container_of0(handle, struct osd_thandle, ot_super);
2941         LASSERT(oh->ot_handle == NULL);
2942
2943         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2944                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2945
2946         return 0;
2947 }
2948
2949 /**
2950  *      delete a (key, value) pair from index \a dt specified by \a key
2951  *
2952  *      \param  dt      osd index object
2953  *      \param  key     key for index
2954  *      \param  rec     record reference
2955  *      \param  handle  transaction handler
2956  *
2957  *      \retval  0  success
2958  *      \retval -ve   failure
2959  */
2960
2961 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2962                                 const struct dt_key *key,
2963                                 struct thandle *handle,
2964                                 struct lustre_capa *capa)
2965 {
2966         struct osd_thread_info *oti = osd_oti_get(env);
2967         struct osd_object      *obj = osd_dt_obj(dt);
2968         struct osd_thandle     *oh;
2969         struct iam_path_descr  *ipd;
2970         struct iam_container   *bag = &obj->oo_dir->od_container;
2971         int                     rc;
2972
2973         ENTRY;
2974
2975         LINVRNT(osd_invariant(obj));
2976         LASSERT(dt_object_exists(dt));
2977         LASSERT(bag->ic_object == obj->oo_inode);
2978         LASSERT(handle != NULL);
2979
2980         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2981                 RETURN(-EACCES);
2982
2983         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
2984
2985         ipd = osd_idx_ipd_get(env, bag);
2986         if (unlikely(ipd == NULL))
2987                 RETURN(-ENOMEM);
2988
2989         oh = container_of0(handle, struct osd_thandle, ot_super);
2990         LASSERT(oh->ot_handle != NULL);
2991         LASSERT(oh->ot_handle->h_transaction != NULL);
2992
2993         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
2994                 /* swab quota uid/gid provided by caller */
2995                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
2996                 key = (const struct dt_key *)&oti->oti_quota_id;
2997         }
2998
2999         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3000         osd_ipd_put(env, bag, ipd);
3001         LINVRNT(osd_invariant(obj));
3002         RETURN(rc);
3003 }
3004
3005 static int osd_index_declare_ea_delete(const struct lu_env *env,
3006                                        struct dt_object *dt,
3007                                        const struct dt_key *key,
3008                                        struct thandle *handle)
3009 {
3010         struct osd_thandle *oh;
3011         struct inode       *inode;
3012         int                 rc;
3013         ENTRY;
3014
3015         LASSERT(dt_object_exists(dt));
3016         LASSERT(handle != NULL);
3017
3018         oh = container_of0(handle, struct osd_thandle, ot_super);
3019         LASSERT(oh->ot_handle == NULL);
3020
3021         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3022                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3023
3024         inode = osd_dt_obj(dt)->oo_inode;
3025         LASSERT(inode);
3026
3027         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3028                                    true, true, NULL, false);
3029         RETURN(rc);
3030 }
3031
3032 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3033                                           struct dt_rec *fid)
3034 {
3035         struct osd_fid_pack *rec;
3036         int                  rc = -ENODATA;
3037
3038         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3039                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3040                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3041         }
3042         RETURN(rc);
3043 }
3044
3045 /**
3046  * Index delete function for interoperability mode (b11826).
3047  * It will remove the directory entry added by osd_index_ea_insert().
3048  * This entry is needed to maintain name->fid mapping.
3049  *
3050  * \param key,  key i.e. file entry to be deleted
3051  *
3052  * \retval   0, on success
3053  * \retval -ve, on error
3054  */
3055 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3056                                const struct dt_key *key,
3057                                struct thandle *handle,
3058                                struct lustre_capa *capa)
3059 {
3060         struct osd_object          *obj    = osd_dt_obj(dt);
3061         struct inode               *dir    = obj->oo_inode;
3062         struct dentry              *dentry;
3063         struct osd_thandle         *oh;
3064         struct ldiskfs_dir_entry_2 *de;
3065         struct buffer_head         *bh;
3066         struct htree_lock          *hlock = NULL;
3067         int                         rc;
3068
3069         ENTRY;
3070
3071         LINVRNT(osd_invariant(obj));
3072         LASSERT(dt_object_exists(dt));
3073         LASSERT(handle != NULL);
3074
3075         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3076
3077         oh = container_of(handle, struct osd_thandle, ot_super);
3078         LASSERT(oh->ot_handle != NULL);
3079         LASSERT(oh->ot_handle->h_transaction != NULL);
3080
3081         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3082                 RETURN(-EACCES);
3083
3084         ll_vfs_dq_init(dir);
3085         dentry = osd_child_dentry_get(env, obj,
3086                                       (char *)key, strlen((char *)key));
3087
3088         if (obj->oo_hl_head != NULL) {
3089                 hlock = osd_oti_get(env)->oti_hlock;
3090                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3091                                    dir, LDISKFS_HLOCK_DEL);
3092         } else {
3093                 down_write(&obj->oo_ext_idx_sem);
3094         }
3095
3096         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3097         if (bh) {
3098                 rc = ldiskfs_delete_entry(oh->ot_handle,
3099                                           dir, de, bh);
3100                 brelse(bh);
3101         } else {
3102                 rc = -ENOENT;
3103         }
3104         if (hlock != NULL)
3105                 ldiskfs_htree_unlock(hlock);
3106         else
3107                 up_write(&obj->oo_ext_idx_sem);
3108
3109         LASSERT(osd_invariant(obj));
3110         RETURN(rc);
3111 }
3112
3113 /**
3114  *      Lookup index for \a key and copy record to \a rec.
3115  *
3116  *      \param  dt      osd index object
3117  *      \param  key     key for index
3118  *      \param  rec     record reference
3119  *
3120  *      \retval  +ve  success : exact mach
3121  *      \retval  0    return record with key not greater than \a key
3122  *      \retval -ve   failure
3123  */
3124 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3125                                 struct dt_rec *rec, const struct dt_key *key,
3126                                 struct lustre_capa *capa)
3127 {
3128         struct osd_object      *obj = osd_dt_obj(dt);
3129         struct iam_path_descr  *ipd;
3130         struct iam_container   *bag = &obj->oo_dir->od_container;
3131         struct osd_thread_info *oti = osd_oti_get(env);
3132         struct iam_iterator    *it = &oti->oti_idx_it;
3133         struct iam_rec         *iam_rec;
3134         int                     rc;
3135
3136         ENTRY;
3137
3138         LASSERT(osd_invariant(obj));
3139         LASSERT(dt_object_exists(dt));
3140         LASSERT(bag->ic_object == obj->oo_inode);
3141
3142         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3143                 RETURN(-EACCES);
3144
3145         ipd = osd_idx_ipd_get(env, bag);
3146         if (IS_ERR(ipd))
3147                 RETURN(-ENOMEM);
3148
3149         /* got ipd now we can start iterator. */
3150         iam_it_init(it, bag, 0, ipd);
3151
3152         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3153                 /* swab quota uid/gid provided by caller */
3154                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3155                 key = (const struct dt_key *)&oti->oti_quota_id;
3156         }
3157
3158         rc = iam_it_get(it, (struct iam_key *)key);
3159         if (rc >= 0) {
3160                 if (S_ISDIR(obj->oo_inode->i_mode))
3161                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3162                 else
3163                         iam_rec = (struct iam_rec *) rec;
3164
3165                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3166
3167                 if (S_ISDIR(obj->oo_inode->i_mode))
3168                         osd_fid_unpack((struct lu_fid *) rec,
3169                                        (struct osd_fid_pack *)iam_rec);
3170                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3171                         osd_quota_unpack(obj, rec);
3172         }
3173
3174         iam_it_put(it);
3175         iam_it_fini(it);
3176         osd_ipd_put(env, bag, ipd);
3177
3178         LINVRNT(osd_invariant(obj));
3179
3180         RETURN(rc);
3181 }
3182
3183 static int osd_index_declare_iam_insert(const struct lu_env *env,
3184                                         struct dt_object *dt,
3185                                         const struct dt_rec *rec,
3186                                         const struct dt_key *key,
3187                                         struct thandle *handle)
3188 {
3189         struct osd_thandle *oh;
3190
3191         LASSERT(dt_object_exists(dt));
3192         LASSERT(handle != NULL);
3193
3194         oh = container_of0(handle, struct osd_thandle, ot_super);
3195         LASSERT(oh->ot_handle == NULL);
3196
3197         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3198                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3199
3200         return 0;
3201 }
3202
3203 /**
3204  *      Inserts (key, value) pair in \a dt index object.
3205  *
3206  *      \param  dt      osd index object
3207  *      \param  key     key for index
3208  *      \param  rec     record reference
3209  *      \param  th      transaction handler
3210  *
3211  *      \retval  0  success
3212  *      \retval -ve failure
3213  */
3214 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3215                                 const struct dt_rec *rec,
3216                                 const struct dt_key *key, struct thandle *th,
3217                                 struct lustre_capa *capa, int ignore_quota)
3218 {
3219         struct osd_object     *obj = osd_dt_obj(dt);
3220         struct iam_path_descr *ipd;
3221         struct osd_thandle    *oh;
3222         struct iam_container  *bag = &obj->oo_dir->od_container;
3223         struct osd_thread_info *oti = osd_oti_get(env);
3224         struct iam_rec         *iam_rec;
3225         int                     rc;
3226
3227         ENTRY;
3228
3229         LINVRNT(osd_invariant(obj));
3230         LASSERT(dt_object_exists(dt));
3231         LASSERT(bag->ic_object == obj->oo_inode);
3232         LASSERT(th != NULL);
3233
3234         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3235                 RETURN(-EACCES);
3236
3237         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3238
3239         ipd = osd_idx_ipd_get(env, bag);
3240         if (unlikely(ipd == NULL))
3241                 RETURN(-ENOMEM);
3242
3243         oh = container_of0(th, struct osd_thandle, ot_super);
3244         LASSERT(oh->ot_handle != NULL);
3245         LASSERT(oh->ot_handle->h_transaction != NULL);
3246         if (S_ISDIR(obj->oo_inode->i_mode)) {
3247                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3248                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3249         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3250                 /* pack quota uid/gid */
3251                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3252                 key = (const struct dt_key *)&oti->oti_quota_id;
3253                 /* pack quota record */
3254                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3255                 iam_rec = (struct iam_rec *)rec;
3256         } else {
3257                 iam_rec = (struct iam_rec *)rec;
3258         }
3259
3260         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3261                         iam_rec, ipd);
3262         osd_ipd_put(env, bag, ipd);
3263         LINVRNT(osd_invariant(obj));
3264         RETURN(rc);
3265 }
3266
3267 /**
3268  * Calls ldiskfs_add_entry() to add directory entry
3269  * into the directory. This is required for
3270  * interoperability mode (b11826)
3271  *
3272  * \retval   0, on success
3273  * \retval -ve, on error
3274  */
3275 static int __osd_ea_add_rec(struct osd_thread_info *info,
3276                             struct osd_object *pobj, struct inode  *cinode,
3277                             const char *name, const struct dt_rec *fid,
3278                             struct htree_lock *hlock, struct thandle *th)
3279 {
3280         struct ldiskfs_dentry_param *ldp;
3281         struct dentry               *child;
3282         struct osd_thandle          *oth;
3283         int                          rc;
3284
3285         oth = container_of(th, struct osd_thandle, ot_super);
3286         LASSERT(oth->ot_handle != NULL);
3287         LASSERT(oth->ot_handle->h_transaction != NULL);
3288
3289         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3290
3291         /* XXX: remove fid_is_igif() check here.
3292          * IGIF check is just to handle insertion of .. when it is 'ROOT',
3293          * it is IGIF now but needs FID in dir entry as well for readdir
3294          * to work.
3295          * LU-838 should fix that and remove fid_is_igif() check */
3296         if (fid_is_igif((struct lu_fid *)fid) ||
3297             fid_is_norm((struct lu_fid *)fid)) {
3298                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3299                 osd_get_ldiskfs_dirent_param(ldp, fid);
3300                 child->d_fsdata = (void *)ldp;
3301         } else {
3302                 child->d_fsdata = NULL;
3303         }
3304         LASSERT(pobj->oo_inode);
3305         ll_vfs_dq_init(pobj->oo_inode);
3306         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3307
3308         RETURN(rc);
3309 }
3310
3311 /**
3312  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3313  * into the directory.Also sets flags into osd object to
3314  * indicate dot and dotdot are created. This is required for
3315  * interoperability mode (b11826)
3316  *
3317  * \param dir   directory for dot and dotdot fixup.
3318  * \param obj   child object for linking
3319  *
3320  * \retval   0, on success
3321  * \retval -ve, on error
3322  */
3323 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3324                               struct osd_object *dir,
3325                               struct inode  *parent_dir, const char *name,
3326                               const struct dt_rec *dot_fid,
3327                               const struct dt_rec *dot_dot_fid,
3328                               struct thandle *th)
3329 {
3330         struct inode                *inode = dir->oo_inode;
3331         struct ldiskfs_dentry_param *dot_ldp;
3332         struct ldiskfs_dentry_param *dot_dot_ldp;
3333         struct osd_thandle          *oth;
3334         int result = 0;
3335
3336         oth = container_of(th, struct osd_thandle, ot_super);
3337         LASSERT(oth->ot_handle->h_transaction != NULL);
3338         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3339
3340         if (strcmp(name, dot) == 0) {
3341                 if (dir->oo_compat_dot_created) {
3342                         result = -EEXIST;
3343                 } else {
3344                         LASSERT(inode == parent_dir);
3345                         dir->oo_compat_dot_created = 1;
3346                         result = 0;
3347                 }
3348         } else if(strcmp(name, dotdot) == 0) {
3349                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3350                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3351
3352                 if (!dir->oo_compat_dot_created)
3353                         return -EINVAL;
3354                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3355                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3356                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3357                 } else {
3358                         dot_ldp = NULL;
3359                         dot_dot_ldp = NULL;
3360                 }
3361                 /* in case of rename, dotdot is already created */
3362                 if (dir->oo_compat_dotdot_created) {
3363                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3364                                                 dot_dot_fid, NULL, th);
3365                 }
3366
3367                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3368                                                 inode, dot_ldp, dot_dot_ldp);
3369                 if (result == 0)
3370                        dir->oo_compat_dotdot_created = 1;
3371         }
3372
3373         return result;
3374 }
3375
3376
3377 /**
3378  * It will call the appropriate osd_add* function and return the
3379  * value, return by respective functions.
3380  */
3381 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3382                           struct inode *cinode, const char *name,
3383                           const struct dt_rec *fid, struct thandle *th)
3384 {
3385         struct osd_thread_info *info   = osd_oti_get(env);
3386         struct htree_lock      *hlock;
3387         int                     rc;
3388
3389         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3390
3391         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3392                                                    name[2] =='\0'))) {
3393                 if (hlock != NULL) {
3394                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3395                                            pobj->oo_inode, 0);
3396                 } else {
3397                         down_write(&pobj->oo_ext_idx_sem);
3398                 }
3399                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3400                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3401                                         fid, th);
3402         } else {
3403                 if (hlock != NULL) {
3404                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3405                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3406                 } else {
3407                         down_write(&pobj->oo_ext_idx_sem);
3408                 }
3409
3410                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3411                                       hlock, th);
3412         }
3413         if (hlock != NULL)
3414                 ldiskfs_htree_unlock(hlock);
3415         else
3416                 up_write(&pobj->oo_ext_idx_sem);
3417
3418         return rc;
3419 }
3420
3421 static void
3422 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3423                       struct osd_idmap_cache *oic)
3424 {
3425         struct osd_scrub    *scrub = &dev->od_scrub;
3426         struct lu_fid       *fid   = &oic->oic_fid;
3427         struct osd_inode_id *id    = &oti->oti_id;
3428         int                  once  = 0;
3429         int                  rc;
3430         ENTRY;
3431
3432         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3433                 RETURN_EXIT;
3434
3435 again:
3436         rc = osd_oi_lookup(oti, dev, fid, id);
3437         if (rc != 0 && rc != -ENOENT)
3438                 RETURN_EXIT;
3439
3440         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3441                 RETURN_EXIT;
3442
3443         if (thread_is_running(&scrub->os_thread)) {
3444                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3445                 /* There is race condition between osd_oi_lookup and OI scrub.
3446                  * The OI scrub finished just after osd_oi_lookup() failure.
3447                  * Under such case, it is unnecessary to trigger OI scrub again,
3448                  * but try to call osd_oi_lookup() again. */
3449                 if (unlikely(rc == -EAGAIN))
3450                         goto again;
3451
3452                 RETURN_EXIT;
3453         }
3454
3455         if (!dev->od_noscrub && ++once == 1) {
3456                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3457                        PFID(fid));
3458                 rc = osd_scrub_start(dev);
3459                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3460                                ", rc = %d [2]\n",
3461                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3462                                PFID(fid), rc);
3463                 if (rc == 0)
3464                         goto again;
3465         }
3466
3467         EXIT;
3468 }
3469
3470 /**
3471  * Calls ->lookup() to find dentry. From dentry get inode and
3472  * read inode's ea to get fid. This is required for  interoperability
3473  * mode (b11826)
3474  *
3475  * \retval   0, on success
3476  * \retval -ve, on error
3477  */
3478 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3479                              struct dt_rec *rec, const struct dt_key *key)
3480 {
3481         struct inode               *dir    = obj->oo_inode;
3482         struct dentry              *dentry;
3483         struct ldiskfs_dir_entry_2 *de;
3484         struct buffer_head         *bh;
3485         struct lu_fid              *fid = (struct lu_fid *) rec;
3486         struct htree_lock          *hlock = NULL;
3487         int                         ino;
3488         int                         rc;
3489
3490         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3491
3492         dentry = osd_child_dentry_get(env, obj,
3493                                       (char *)key, strlen((char *)key));
3494
3495         if (obj->oo_hl_head != NULL) {
3496                 hlock = osd_oti_get(env)->oti_hlock;
3497                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3498                                    dir, LDISKFS_HLOCK_LOOKUP);
3499         } else {
3500                 down_read(&obj->oo_ext_idx_sem);
3501         }
3502
3503         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3504         if (bh) {
3505                 struct osd_thread_info *oti = osd_oti_get(env);
3506                 struct osd_idmap_cache *oic = &oti->oti_cache;
3507                 struct osd_device *dev = osd_obj2dev(obj);
3508                 struct osd_scrub *scrub = &dev->od_scrub;
3509                 struct scrub_file *sf = &scrub->os_file;
3510
3511                 ino = le32_to_cpu(de->inode);
3512                 rc = osd_get_fid_from_dentry(de, rec);
3513
3514                 /* done with de, release bh */
3515                 brelse(bh);
3516                 if (rc != 0)
3517                         rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3518                 else
3519                         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3520                 if (rc != 0) {
3521                         fid_zero(&oic->oic_fid);
3522                         GOTO(out, rc);
3523                 }
3524
3525                 oic->oic_fid = *fid;
3526                 if ((scrub->os_pos_current <= ino) &&
3527                     (sf->sf_flags & SF_INCONSISTENT ||
3528                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3529                                       sf->sf_oi_bitmap)))
3530                         osd_consistency_check(oti, dev, oic);
3531         } else {
3532                 rc = -ENOENT;
3533         }
3534
3535         GOTO(out, rc);
3536
3537 out:
3538         if (hlock != NULL)
3539                 ldiskfs_htree_unlock(hlock);
3540         else
3541                 up_read(&obj->oo_ext_idx_sem);
3542         return rc;
3543 }
3544
3545 /**
3546  * Find the osd object for given fid.
3547  *
3548  * \param fid need to find the osd object having this fid
3549  *
3550  * \retval osd_object on success
3551  * \retval        -ve on error
3552  */
3553 struct osd_object *osd_object_find(const struct lu_env *env,
3554                                    struct dt_object *dt,
3555                                    const struct lu_fid *fid)
3556 {
3557         struct lu_device  *ludev = dt->do_lu.lo_dev;
3558         struct osd_object *child = NULL;
3559         struct lu_object  *luch;
3560         struct lu_object  *lo;
3561
3562         /*
3563          * at this point topdev might not exist yet
3564          * (i.e. MGS is preparing profiles). so we can
3565          * not rely on topdev and instead lookup with
3566          * our device passed as topdev. this can't work
3567          * if the object isn't cached yet (as osd doesn't
3568          * allocate lu_header). IOW, the object must be
3569          * in the cache, otherwise lu_object_alloc() crashes
3570          * -bzzz
3571          */
3572         luch = lu_object_find_at(env, ludev, fid, NULL);
3573         if (!IS_ERR(luch)) {
3574                 if (lu_object_exists(luch)) {
3575                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3576                         if (lo != NULL)
3577                                 child = osd_obj(lo);
3578                         else
3579                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3580                                                 "lu_object can't be located"
3581                                                 DFID"\n", PFID(fid));
3582
3583                         if (child == NULL) {
3584                                 lu_object_put(env, luch);
3585                                 CERROR("Unable to get osd_object\n");
3586                                 child = ERR_PTR(-ENOENT);
3587                         }
3588                 } else {
3589                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3590                                         "lu_object does not exists "DFID"\n",
3591                                         PFID(fid));
3592                         lu_object_put(env, luch);
3593                         child = ERR_PTR(-ENOENT);
3594                 }
3595         } else
3596                 child = (void *)luch;
3597
3598         return child;
3599 }
3600
3601 /**
3602  * Put the osd object once done with it.
3603  *
3604  * \param obj osd object that needs to be put
3605  */
3606 static inline void osd_object_put(const struct lu_env *env,
3607                                   struct osd_object *obj)
3608 {
3609         lu_object_put(env, &obj->oo_dt.do_lu);
3610 }
3611
3612 static int osd_index_declare_ea_insert(const struct lu_env *env,
3613                                        struct dt_object *dt,
3614                                        const struct dt_rec *rec,
3615                                        const struct dt_key *key,
3616                                        struct thandle *handle)
3617 {
3618         struct osd_thandle      *oh;
3619         struct inode            *inode;
3620         struct lu_fid           *fid = (struct lu_fid *)rec;
3621         int                     rc;
3622         ENTRY;
3623
3624         LASSERT(dt_object_exists(dt));
3625         LASSERT(handle != NULL);
3626
3627         oh = container_of0(handle, struct osd_thandle, ot_super);
3628         LASSERT(oh->ot_handle == NULL);
3629
3630         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3631                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3632
3633         inode = osd_dt_obj(dt)->oo_inode;
3634         LASSERT(inode);
3635
3636         /* We ignore block quota on meta pool (MDTs), so needn't
3637          * calculate how many blocks will be consumed by this index
3638          * insert */
3639         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3640                                    true, true, NULL, false);
3641         if (fid == NULL)
3642                 RETURN(0);
3643
3644         /* It does fld look up inside declare, and the result will be
3645         * added to fld cache, so the following fld lookup inside insert
3646         * does not need send RPC anymore, so avoid send rpc with holding
3647         * transaction */
3648         LASSERTF(fid_is_sane(fid), "fid is insane"DFID"\n", PFID(fid));
3649         osd_fld_lookup(env, osd_dt_dev(handle->th_dev), fid,
3650                         &osd_oti_get(env)->oti_seq_range);
3651
3652         RETURN(rc);
3653 }
3654
3655 /**
3656  * Index add function for interoperability mode (b11826).
3657  * It will add the directory entry.This entry is needed to
3658  * maintain name->fid mapping.
3659  *
3660  * \param key it is key i.e. file entry to be inserted
3661  * \param rec it is value of given key i.e. fid
3662  *
3663  * \retval   0, on success
3664  * \retval -ve, on error
3665  */
3666 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3667                                const struct dt_rec *rec,
3668                                const struct dt_key *key, struct thandle *th,
3669                                struct lustre_capa *capa, int ignore_quota)
3670 {
3671         struct osd_object *obj   = osd_dt_obj(dt);
3672         struct lu_fid     *fid   = (struct lu_fid *) rec;
3673         const char        *name  = (const char *)key;
3674         struct osd_object *child;
3675         int                rc;
3676
3677         ENTRY;
3678
3679         LASSERT(osd_invariant(obj));
3680         LASSERT(dt_object_exists(dt));
3681         LASSERT(th != NULL);
3682
3683         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3684
3685         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3686                 RETURN(-EACCES);
3687
3688         child = osd_object_find(env, dt, fid);
3689         if (!IS_ERR(child)) {
3690                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3691                 osd_object_put(env, child);
3692         } else {
3693                 rc = PTR_ERR(child);
3694         }
3695
3696         LASSERT(osd_invariant(obj));
3697         RETURN(rc);
3698 }
3699
3700 /**
3701  *  Initialize osd Iterator for given osd index object.
3702  *
3703  *  \param  dt      osd index object
3704  */
3705
3706 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3707                                      struct dt_object *dt,
3708                                      __u32 unused,
3709                                      struct lustre_capa *capa)
3710 {
3711         struct osd_it_iam      *it;
3712         struct osd_thread_info *oti = osd_oti_get(env);
3713         struct osd_object      *obj = osd_dt_obj(dt);
3714         struct lu_object       *lo  = &dt->do_lu;
3715         struct iam_path_descr  *ipd;
3716         struct iam_container   *bag = &obj->oo_dir->od_container;
3717
3718         LASSERT(lu_object_exists(lo));
3719
3720         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3721                 return ERR_PTR(-EACCES);
3722
3723         it = &oti->oti_it;
3724         ipd = osd_it_ipd_get(env, bag);
3725         if (likely(ipd != NULL)) {
3726                 it->oi_obj = obj;
3727                 it->oi_ipd = ipd;
3728                 lu_object_get(lo);
3729                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3730                 return (struct dt_it *)it;
3731         }
3732         return ERR_PTR(-ENOMEM);
3733 }
3734
3735 /**
3736  * free given Iterator.
3737  */
3738
3739 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3740 {
3741         struct osd_it_iam *it = (struct osd_it_iam *)di;
3742         struct osd_object *obj = it->oi_obj;
3743
3744         iam_it_fini(&it->oi_it);
3745         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3746         lu_object_put(env, &obj->oo_dt.do_lu);
3747 }
3748
3749 /**
3750  *  Move Iterator to record specified by \a key
3751  *
3752  *  \param  di      osd iterator
3753  *  \param  key     key for index
3754  *
3755  *  \retval +ve  di points to record with least key not larger than key
3756  *  \retval  0   di points to exact matched key
3757  *  \retval -ve  failure
3758  */
3759
3760 static int osd_it_iam_get(const struct lu_env *env,
3761                           struct dt_it *di, const struct dt_key *key)
3762 {
3763         struct osd_thread_info  *oti = osd_oti_get(env);
3764         struct osd_it_iam       *it = (struct osd_it_iam *)di;
3765
3766         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3767                 /* swab quota uid/gid */
3768                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3769                 key = (struct dt_key *)&oti->oti_quota_id;
3770         }
3771
3772         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3773 }
3774
3775 /**
3776  *  Release Iterator
3777  *
3778  *  \param  di      osd iterator
3779  */
3780
3781 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3782 {
3783         struct osd_it_iam *it = (struct osd_it_iam *)di;
3784
3785         iam_it_put(&it->oi_it);
3786 }
3787
3788 /**
3789  *  Move iterator by one record
3790  *
3791  *  \param  di      osd iterator
3792  *
3793  *  \retval +1   end of container reached
3794  *  \retval  0   success
3795  *  \retval -ve  failure
3796  */
3797
3798 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3799 {
3800         struct osd_it_iam *it = (struct osd_it_iam *)di;
3801
3802         return iam_it_next(&it->oi_it);
3803 }
3804
3805 /**
3806  * Return pointer to the key under iterator.
3807  */
3808
3809 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3810                                  const struct dt_it *di)
3811 {
3812         struct osd_thread_info *oti = osd_oti_get(env);
3813         struct osd_it_iam      *it = (struct osd_it_iam *)di;
3814         struct osd_object      *obj = it->oi_obj;
3815         struct dt_key          *key;
3816
3817         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3818
3819         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3820                 /* swab quota uid/gid */
3821                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3822                 key = (struct dt_key *)&oti->oti_quota_id;
3823         }
3824
3825         return key;
3826 }
3827
3828 /**
3829  * Return size of key under iterator (in bytes)
3830  */
3831
3832 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3833 {
3834         struct osd_it_iam *it = (struct osd_it_iam *)di;
3835
3836         return iam_it_key_size(&it->oi_it);
3837 }
3838
3839 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3840                                        int len, __u16 type)
3841 {
3842         struct luda_type *lt;
3843         const unsigned    align = sizeof(struct luda_type) - 1;
3844
3845         /* check if file type is required */
3846         if (attr & LUDA_TYPE) {
3847                         len = (len + align) & ~align;
3848
3849                         lt = (void *) ent->lde_name + len;
3850                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3851                         ent->lde_attrs |= LUDA_TYPE;
3852         }
3853
3854         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3855 }
3856
3857 /**
3858  * build lu direct from backend fs dirent.
3859  */
3860
3861 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3862                                       struct lu_fid *fid, __u64 offset,
3863                                       char *name, __u16 namelen,
3864                                       __u16 type, __u32 attr)
3865 {
3866         fid_cpu_to_le(&ent->lde_fid, fid);
3867         ent->lde_attrs = LUDA_FID;
3868
3869         ent->lde_hash = cpu_to_le64(offset);
3870         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3871
3872         strncpy(ent->lde_name, name, namelen);
3873         ent->lde_namelen = cpu_to_le16(namelen);
3874
3875         /* append lustre attributes */
3876         osd_it_append_attrs(ent, attr, namelen, type);
3877 }
3878
3879 /**
3880  * Return pointer to the record under iterator.
3881  */
3882 static int osd_it_iam_rec(const struct lu_env *env,
3883                           const struct dt_it *di,
3884                           struct dt_rec *dtrec, __u32 attr)
3885 {
3886         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
3887         struct osd_thread_info *info = osd_oti_get(env);
3888         ENTRY;
3889
3890         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3891                 const struct osd_fid_pack *rec;
3892                 struct lu_fid             *fid = &info->oti_fid;
3893                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
3894                 char                      *name;
3895                 int                        namelen;
3896                 __u64                      hash;
3897                 int                        rc;
3898
3899                 name = (char *)iam_it_key_get(&it->oi_it);
3900                 if (IS_ERR(name))
3901                         RETURN(PTR_ERR(name));
3902
3903                 namelen = iam_it_key_size(&it->oi_it);
3904
3905                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3906                 if (IS_ERR(rec))
3907                         RETURN(PTR_ERR(rec));
3908
3909                 rc = osd_fid_unpack(fid, rec);
3910                 if (rc)
3911                         RETURN(rc);
3912
3913                 hash = iam_it_store(&it->oi_it);
3914
3915                 /* IAM does not store object type in IAM index (dir) */
3916                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3917                                    0, LUDA_FID);
3918         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3919                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3920                            (struct iam_rec *)dtrec);
3921                 osd_quota_unpack(it->oi_obj, dtrec);
3922         } else {
3923                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3924                            (struct iam_rec *)dtrec);
3925         }
3926
3927         RETURN(0);
3928 }
3929
3930 /**
3931  * Returns cookie for current Iterator position.
3932  */
3933 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3934 {
3935         struct osd_it_iam *it = (struct osd_it_iam *)di;
3936
3937         return iam_it_store(&it->oi_it);
3938 }
3939
3940 /**
3941  * Restore iterator from cookie.
3942  *
3943  * \param  di      osd iterator
3944  * \param  hash    Iterator location cookie
3945  *
3946  * \retval +ve  di points to record with least key not larger than key.
3947  * \retval  0   di points to exact matched key
3948  * \retval -ve  failure
3949  */
3950
3951 static int osd_it_iam_load(const struct lu_env *env,
3952                            const struct dt_it *di, __u64 hash)
3953 {
3954         struct osd_it_iam *it = (struct osd_it_iam *)di;
3955
3956         return iam_it_load(&it->oi_it, hash);
3957 }
3958
3959 static const struct dt_index_operations osd_index_iam_ops = {
3960         .dio_lookup         = osd_index_iam_lookup,
3961         .dio_declare_insert = osd_index_declare_iam_insert,
3962         .dio_insert         = osd_index_iam_insert,
3963         .dio_declare_delete = osd_index_declare_iam_delete,
3964         .dio_delete         = osd_index_iam_delete,
3965         .dio_it     = {
3966                 .init     = osd_it_iam_init,
3967                 .fini     = osd_it_iam_fini,
3968                 .get      = osd_it_iam_get,
3969                 .put      = osd_it_iam_put,
3970                 .next     = osd_it_iam_next,
3971                 .key      = osd_it_iam_key,
3972                 .key_size = osd_it_iam_key_size,
3973                 .rec      = osd_it_iam_rec,
3974                 .store    = osd_it_iam_store,
3975                 .load     = osd_it_iam_load
3976         }
3977 };
3978
3979
3980 /**
3981  * Creates or initializes iterator context.
3982  *
3983  * \retval struct osd_it_ea, iterator structure on success
3984  *
3985  */
3986 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3987                                     struct dt_object *dt,
3988                                     __u32 attr,
3989                                     struct lustre_capa *capa)
3990 {
3991         struct osd_object       *obj  = osd_dt_obj(dt);
3992         struct osd_thread_info  *info = osd_oti_get(env);
3993         struct osd_it_ea        *it   = &info->oti_it_ea;
3994         struct lu_object        *lo   = &dt->do_lu;
3995         struct dentry           *obj_dentry = &info->oti_it_dentry;
3996         ENTRY;
3997         LASSERT(lu_object_exists(lo));
3998
3999         obj_dentry->d_inode = obj->oo_inode;
4000         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4001         obj_dentry->d_name.hash = 0;
4002
4003         it->oie_rd_dirent       = 0;
4004         it->oie_it_dirent       = 0;
4005         it->oie_dirent          = NULL;
4006         it->oie_buf             = info->oti_it_ea_buf;
4007         it->oie_obj             = obj;
4008         it->oie_file.f_pos      = 0;
4009         it->oie_file.f_dentry   = obj_dentry;
4010         if (attr & LUDA_64BITHASH)
4011                 it->oie_file.f_mode |= FMODE_64BITHASH;
4012         else
4013                 it->oie_file.f_mode |= FMODE_32BITHASH;
4014         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4015         it->oie_file.f_op         = obj->oo_inode->i_fop;
4016         it->oie_file.private_data = NULL;
4017         lu_object_get(lo);
4018         RETURN((struct dt_it *) it);
4019 }
4020
4021 /**
4022  * Destroy or finishes iterator context.
4023  *
4024  * \param di iterator structure to be destroyed
4025  */
4026 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4027 {
4028         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4029         struct osd_object    *obj  = it->oie_obj;
4030         struct inode       *inode  = obj->oo_inode;
4031
4032         ENTRY;
4033         it->oie_file.f_op->release(inode, &it->oie_file);
4034         lu_object_put(env, &obj->oo_dt.do_lu);
4035         EXIT;
4036 }
4037
4038 /**
4039  * It position the iterator at given key, so that next lookup continues from
4040  * that key Or it is similar to dio_it->load() but based on a key,
4041  * rather than file position.
4042  *
4043  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4044  * to the beginning.
4045  *
4046  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4047  */
4048 static int osd_it_ea_get(const struct lu_env *env,
4049                          struct dt_it *di, const struct dt_key *key)
4050 {
4051         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4052
4053         ENTRY;
4054         LASSERT(((const char *)key)[0] == '\0');
4055         it->oie_file.f_pos      = 0;
4056         it->oie_rd_dirent       = 0;
4057         it->oie_it_dirent       = 0;
4058         it->oie_dirent          = NULL;
4059
4060         RETURN(+1);
4061 }
4062
4063 /**
4064  * Does nothing
4065  */
4066 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4067 {
4068 }
4069
4070 /**
4071  * It is called internally by ->readdir(). It fills the
4072  * iterator's in-memory data structure with required
4073  * information i.e. name, namelen, rec_size etc.
4074  *
4075  * \param buf in which information to be filled in.
4076  * \param name name of the file in given dir
4077  *
4078  * \retval 0 on success
4079  * \retval 1 on buffer full
4080  */
4081 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4082                                loff_t offset, __u64 ino,
4083                                unsigned d_type)
4084 {
4085         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4086         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4087         struct lu_fid           *fid  = &ent->oied_fid;
4088         struct osd_fid_pack     *rec;
4089         ENTRY;
4090
4091         /* this should never happen */
4092         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4093                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4094                 RETURN(-EIO);
4095         }
4096
4097         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4098             OSD_IT_EA_BUFSIZE)
4099                 RETURN(1);
4100
4101         if (d_type & LDISKFS_DIRENT_LUFID) {
4102                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4103
4104                 if (osd_fid_unpack(fid, rec) != 0)
4105                         fid_zero(fid);
4106
4107                 d_type &= ~LDISKFS_DIRENT_LUFID;
4108         } else {
4109                 fid_zero(fid);
4110         }
4111
4112         ent->oied_ino     = ino;
4113         ent->oied_off     = offset;
4114         ent->oied_namelen = namelen;
4115         ent->oied_type    = d_type;
4116
4117         memcpy(ent->oied_name, name, namelen);
4118
4119         it->oie_rd_dirent++;
4120         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4121         RETURN(0);
4122 }
4123
4124 /**
4125  * Calls ->readdir() to load a directory entry at a time
4126  * and stored it in iterator's in-memory data structure.
4127  *
4128  * \param di iterator's in memory structure
4129  *
4130  * \retval   0 on success
4131  * \retval -ve on error
4132  */
4133 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4134                                const struct dt_it *di)
4135 {
4136         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4137         struct osd_object  *obj   = it->oie_obj;
4138         struct inode       *inode = obj->oo_inode;
4139         struct htree_lock  *hlock = NULL;
4140         int                 result = 0;
4141
4142         ENTRY;
4143         it->oie_dirent = it->oie_buf;
4144         it->oie_rd_dirent = 0;
4145
4146         if (obj->oo_hl_head != NULL) {
4147                 hlock = osd_oti_get(env)->oti_hlock;
4148                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4149                                    inode, LDISKFS_HLOCK_READDIR);
4150         } else {
4151                 down_read(&obj->oo_ext_idx_sem);
4152         }
4153
4154         result = inode->i_fop->readdir(&it->oie_file, it,
4155                                        (filldir_t) osd_ldiskfs_filldir);
4156
4157         if (hlock != NULL)
4158                 ldiskfs_htree_unlock(hlock);
4159         else
4160                 up_read(&obj->oo_ext_idx_sem);
4161
4162         if (it->oie_rd_dirent == 0) {
4163                 result = -EIO;
4164         } else {
4165                 it->oie_dirent = it->oie_buf;
4166                 it->oie_it_dirent = 1;
4167         }
4168
4169         RETURN(result);
4170 }
4171
4172 /**
4173  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4174  * to load a directory entry at a time and stored it in
4175  * iterator's in-memory data structure.
4176  *
4177  * \param di iterator's in memory structure
4178  *
4179  * \retval +ve iterator reached to end
4180  * \retval   0 iterator not reached to end
4181  * \retval -ve on error
4182  */
4183 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4184 {
4185         struct osd_it_ea *it = (struct osd_it_ea *)di;
4186         int rc;
4187
4188         ENTRY;
4189
4190         if (it->oie_it_dirent < it->oie_rd_dirent) {
4191                 it->oie_dirent =
4192                         (void *) it->oie_dirent +
4193                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4194                                        it->oie_dirent->oied_namelen);
4195                 it->oie_it_dirent++;
4196                 RETURN(0);
4197         } else {
4198                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4199                         rc = +1;
4200                 else
4201                         rc = osd_ldiskfs_it_fill(env, di);
4202         }
4203
4204         RETURN(rc);
4205 }
4206
4207 /**
4208  * Returns the key at current position from iterator's in memory structure.
4209  *
4210  * \param di iterator's in memory structure
4211  *
4212  * \retval key i.e. struct dt_key on success
4213  */
4214 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4215                                     const struct dt_it *di)
4216 {
4217         struct osd_it_ea *it = (struct osd_it_ea *)di;
4218
4219         return (struct dt_key *)it->oie_dirent->oied_name;
4220 }
4221
4222 /**
4223  * Returns the key's size at current position from iterator's in memory structure.
4224  *
4225  * \param di iterator's in memory structure
4226  *
4227  * \retval key_size i.e. struct dt_key on success
4228  */
4229 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4230 {
4231         struct osd_it_ea *it = (struct osd_it_ea *)di;
4232
4233         return it->oie_dirent->oied_namelen;
4234 }
4235
4236
4237 /**
4238  * Returns the value (i.e. fid/igif) at current position from iterator's
4239  * in memory structure.
4240  *
4241  * \param di struct osd_it_ea, iterator's in memory structure
4242  * \param attr attr requested for dirent.
4243  * \param lde lustre dirent
4244  *
4245  * \retval   0 no error and \param lde has correct lustre dirent.
4246  * \retval -ve on error
4247  */
4248 static inline int osd_it_ea_rec(const struct lu_env *env,
4249                                 const struct dt_it *di,
4250                                 struct dt_rec *dtrec, __u32 attr)
4251 {
4252         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4253         struct osd_object      *obj   = it->oie_obj;
4254         struct osd_device      *dev   = osd_obj2dev(obj);
4255         struct osd_scrub       *scrub = &dev->od_scrub;
4256         struct scrub_file      *sf    = &scrub->os_file;
4257         struct osd_thread_info *oti   = osd_oti_get(env);
4258         struct osd_idmap_cache *oic   = &oti->oti_cache;
4259         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4260         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4261         __u32                   ino   = it->oie_dirent->oied_ino;
4262         int                     rc    = 0;
4263         ENTRY;
4264
4265         if (!fid_is_sane(fid)) {
4266                 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4267                 if (rc != 0) {
4268                         fid_zero(&oic->oic_fid);
4269                         RETURN(rc);
4270                 }
4271         } else {
4272                 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4273         }
4274
4275         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4276                            it->oie_dirent->oied_name,
4277                            it->oie_dirent->oied_namelen,
4278                            it->oie_dirent->oied_type, attr);
4279         oic->oic_fid = *fid;
4280         if ((scrub->os_pos_current <= ino) &&
4281             (sf->sf_flags & SF_INCONSISTENT ||
4282              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4283                 osd_consistency_check(oti, dev, oic);
4284
4285         RETURN(rc);
4286 }
4287
4288 /**
4289  * Returns a cookie for current position of the iterator head, so that
4290  * user can use this cookie to load/start the iterator next time.
4291  *
4292  * \param di iterator's in memory structure
4293  *
4294  * \retval cookie for current position, on success
4295  */
4296 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4297 {
4298         struct osd_it_ea *it = (struct osd_it_ea *)di;
4299
4300         return it->oie_dirent->oied_off;
4301 }
4302
4303 /**
4304  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4305  * to load a directory entry at a time and stored it i inn,
4306  * in iterator's in-memory data structure.
4307  *
4308  * \param di struct osd_it_ea, iterator's in memory structure
4309  *
4310  * \retval +ve on success
4311  * \retval -ve on error
4312  */
4313 static int osd_it_ea_load(const struct lu_env *env,
4314                           const struct dt_it *di, __u64 hash)
4315 {
4316         struct osd_it_ea *it = (struct osd_it_ea *)di;
4317         int rc;
4318
4319         ENTRY;
4320         it->oie_file.f_pos = hash;
4321
4322         rc =  osd_ldiskfs_it_fill(env, di);
4323         if (rc == 0)
4324                 rc = +1;
4325
4326         RETURN(rc);
4327 }
4328
4329 /**
4330  * Index lookup function for interoperability mode (b11826).
4331  *
4332  * \param key,  key i.e. file name to be searched
4333  *
4334  * \retval +ve, on success
4335  * \retval -ve, on error
4336  */
4337 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4338                                struct dt_rec *rec, const struct dt_key *key,
4339                                struct lustre_capa *capa)
4340 {
4341         struct osd_object *obj = osd_dt_obj(dt);
4342         int rc = 0;
4343
4344         ENTRY;
4345
4346         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4347         LINVRNT(osd_invariant(obj));
4348
4349         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4350                 return -EACCES;
4351
4352         rc = osd_ea_lookup_rec(env, obj, rec, key);
4353         if (rc == 0)
4354                 rc = +1;
4355         RETURN(rc);
4356 }
4357
4358 /**
4359  * Index and Iterator operations for interoperability
4360  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4361  */
4362 static const struct dt_index_operations osd_index_ea_ops = {
4363         .dio_lookup         = osd_index_ea_lookup,
4364         .dio_declare_insert = osd_index_declare_ea_insert,
4365         .dio_insert         = osd_index_ea_insert,
4366         .dio_declare_delete = osd_index_declare_ea_delete,
4367         .dio_delete         = osd_index_ea_delete,
4368         .dio_it     = {
4369                 .init     = osd_it_ea_init,
4370                 .fini     = osd_it_ea_fini,
4371                 .get      = osd_it_ea_get,
4372                 .put      = osd_it_ea_put,
4373                 .next     = osd_it_ea_next,
4374                 .key      = osd_it_ea_key,
4375                 .key_size = osd_it_ea_key_size,
4376                 .rec      = osd_it_ea_rec,
4377                 .store    = osd_it_ea_store,
4378                 .load     = osd_it_ea_load
4379         }
4380 };
4381
4382 static void *osd_key_init(const struct lu_context *ctx,
4383                           struct lu_context_key *key)
4384 {
4385         struct osd_thread_info *info;
4386
4387         OBD_ALLOC_PTR(info);
4388         if (info == NULL)
4389                 return ERR_PTR(-ENOMEM);
4390
4391         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4392         if (info->oti_it_ea_buf == NULL)
4393                 goto out_free_info;
4394
4395         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4396
4397         info->oti_hlock = ldiskfs_htree_lock_alloc();
4398         if (info->oti_hlock == NULL)
4399                 goto out_free_ea;
4400
4401         return info;
4402
4403  out_free_ea:
4404         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4405  out_free_info:
4406         OBD_FREE_PTR(info);
4407         return ERR_PTR(-ENOMEM);
4408 }
4409
4410 static void osd_key_fini(const struct lu_context *ctx,
4411                          struct lu_context_key *key, void* data)
4412 {
4413         struct osd_thread_info *info = data;
4414
4415         if (info->oti_hlock != NULL)
4416                 ldiskfs_htree_lock_free(info->oti_hlock);
4417         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4418         OBD_FREE_PTR(info);
4419 }
4420
4421 static void osd_key_exit(const struct lu_context *ctx,
4422                          struct lu_context_key *key, void *data)
4423 {
4424         struct osd_thread_info *info = data;
4425
4426         LASSERT(info->oti_r_locks == 0);
4427         LASSERT(info->oti_w_locks == 0);
4428         LASSERT(info->oti_txns    == 0);
4429 }
4430
4431 /* type constructor/destructor: osd_type_init, osd_type_fini */
4432 LU_TYPE_INIT_FINI(osd, &osd_key);
4433
4434 struct lu_context_key osd_key = {
4435         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4436         .lct_init = osd_key_init,
4437         .lct_fini = osd_key_fini,
4438         .lct_exit = osd_key_exit
4439 };
4440
4441
4442 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4443                            const char *name, struct lu_device *next)
4444 {
4445         struct osd_device *osd = osd_dev(d);
4446
4447         strncpy(osd->od_svname, name, MAX_OBD_NAME);
4448         return osd_procfs_init(osd, name);
4449 }
4450
4451 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4452 {
4453         ENTRY;
4454
4455         osd_scrub_cleanup(env, o);
4456
4457         if (o->od_fsops) {
4458                 fsfilt_put_ops(o->od_fsops);
4459                 o->od_fsops = NULL;
4460         }
4461
4462         /* shutdown quota slave instance associated with the device */
4463         if (o->od_quota_slave != NULL) {
4464                 qsd_fini(env, o->od_quota_slave);
4465                 o->od_quota_slave = NULL;
4466         }
4467
4468         RETURN(0);
4469 }
4470
4471 static int osd_mount(const struct lu_env *env,
4472                      struct osd_device *o, struct lustre_cfg *cfg)
4473 {
4474         const char              *name  = lustre_cfg_string(cfg, 0);
4475         const char              *dev  = lustre_cfg_string(cfg, 1);
4476         const char              *opts;
4477         unsigned long            page, s_flags, lmd_flags = 0;
4478         struct page             *__page;
4479         struct file_system_type *type;
4480         char                    *options = NULL;
4481         char                    *str;
4482         int                       rc = 0;
4483         ENTRY;
4484
4485         if (o->od_mnt != NULL)
4486                 RETURN(0);
4487
4488         if (strlen(dev) >= sizeof(o->od_mntdev))
4489                 RETURN(-E2BIG);
4490         strcpy(o->od_mntdev, dev);
4491
4492         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4493         if (o->od_fsops == NULL) {
4494                 CERROR("Can't find fsfilt_ldiskfs\n");
4495                 RETURN(-ENOTSUPP);
4496         }
4497
4498         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4499         if (__page == NULL)
4500                 GOTO(out, rc = -ENOMEM);
4501
4502         str = lustre_cfg_string(cfg, 2);
4503         s_flags = simple_strtoul(str, NULL, 0);
4504         str = strstr(str, ":");
4505         if (str)
4506                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4507         opts = lustre_cfg_string(cfg, 3);
4508         page = (unsigned long)cfs_page_address(__page);
4509         options = (char *)page;
4510         *options = '\0';
4511         if (opts == NULL)
4512                 strcat(options, "user_xattr,acl");
4513         else
4514                 strcat(options, opts);
4515
4516         /* Glom up mount options */
4517         if (*options != '\0')
4518                 strcat(options, ",");
4519         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4520
4521         type = get_fs_type("ldiskfs");
4522         if (!type) {
4523                 CERROR("%s: cannot find ldiskfs module\n", name);
4524                 GOTO(out, rc = -ENODEV);
4525         }
4526
4527         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4528         cfs_module_put(type->owner);
4529
4530         if (IS_ERR(o->od_mnt)) {
4531                 rc = PTR_ERR(o->od_mnt);
4532                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4533                 o->od_mnt = NULL;
4534                 GOTO(out, rc);
4535         }
4536
4537         if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4538                 CERROR("%s: underlying device %s is marked as read-only. "
4539                        "Setup failed\n", name, dev);
4540                 mntput(o->od_mnt);
4541                 o->od_mnt = NULL;
4542                 GOTO(out, rc = -EROFS);
4543         }
4544
4545         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4546             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4547                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4548                 mntput(o->od_mnt);
4549                 o->od_mnt = NULL;
4550                 GOTO(out, rc = -EINVAL);
4551         }
4552
4553         if (lmd_flags & LMD_FLG_NOSCRUB)
4554                 o->od_noscrub = 1;
4555
4556 out:
4557         if (__page)
4558                 OBD_PAGE_FREE(__page);
4559         if (rc)
4560                 fsfilt_put_ops(o->od_fsops);
4561
4562         RETURN(rc);
4563 }
4564
4565 static struct lu_device *osd_device_fini(const struct lu_env *env,
4566                                          struct lu_device *d)
4567 {
4568         int rc;
4569         ENTRY;
4570
4571         rc = osd_shutdown(env, osd_dev(d));
4572
4573         osd_obj_map_fini(osd_dev(d));
4574
4575         shrink_dcache_sb(osd_sb(osd_dev(d)));
4576         osd_sync(env, lu2dt_dev(d));
4577
4578         rc = osd_procfs_fini(osd_dev(d));
4579         if (rc) {
4580                 CERROR("proc fini error %d \n", rc);
4581                 RETURN (ERR_PTR(rc));
4582         }
4583
4584         if (osd_dev(d)->od_mnt) {
4585                 mntput(osd_dev(d)->od_mnt);
4586                 osd_dev(d)->od_mnt = NULL;
4587         }
4588
4589         RETURN(NULL);
4590 }
4591
4592 static int osd_device_init0(const struct lu_env *env,
4593                             struct osd_device *o,
4594                             struct lustre_cfg *cfg)
4595 {
4596         struct lu_device        *l = osd2lu_dev(o);
4597         struct osd_thread_info *info;
4598         int                     rc;
4599
4600         /* if the module was re-loaded, env can loose its keys */
4601         rc = lu_env_refill((struct lu_env *) env);
4602         if (rc)
4603                 GOTO(out, rc);
4604         info = osd_oti_get(env);
4605         LASSERT(info);
4606
4607         l->ld_ops = &osd_lu_ops;
4608         o->od_dt_dev.dd_ops = &osd_dt_ops;
4609
4610         spin_lock_init(&o->od_osfs_lock);
4611         mutex_init(&o->od_otable_mutex);
4612         o->od_osfs_age = cfs_time_shift_64(-1000);
4613
4614         o->od_capa_hash = init_capa_hash();
4615         if (o->od_capa_hash == NULL)
4616                 GOTO(out, rc = -ENOMEM);
4617
4618         o->od_read_cache = 1;
4619         o->od_writethrough_cache = 1;
4620         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
4621
4622         rc = osd_mount(env, o, cfg);
4623         if (rc)
4624                 GOTO(out_capa, rc);
4625
4626         /* setup scrub, including OI files initialization */
4627         rc = osd_scrub_setup(env, o);
4628         if (rc < 0)
4629                 GOTO(out_mnt, rc);
4630
4631         strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4632                         sizeof(o->od_svname) - 1);
4633
4634         rc = osd_obj_map_init(o);
4635         if (rc != 0)
4636                 GOTO(out_scrub, rc);
4637
4638         rc = lu_site_init(&o->od_site, l);
4639         if (rc)
4640                 GOTO(out_compat, rc);
4641         o->od_site.ls_bottom_dev = l;
4642
4643         rc = lu_site_init_finish(&o->od_site);
4644         if (rc)
4645                 GOTO(out_site, rc);
4646
4647         rc = osd_procfs_init(o, o->od_svname);
4648         if (rc != 0) {
4649                 CERROR("%s: can't initialize procfs: rc = %d\n",
4650                        o->od_svname, rc);
4651                 GOTO(out_site, rc);
4652         }
4653
4654         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4655
4656         /* initialize quota slave instance */
4657         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
4658                                      o->od_proc_entry);
4659         if (IS_ERR(o->od_quota_slave)) {
4660                 rc = PTR_ERR(o->od_quota_slave);
4661                 o->od_quota_slave = NULL;
4662                 GOTO(out_procfs, rc);
4663         }
4664
4665         RETURN(0);
4666 out_procfs:
4667         osd_procfs_fini(o);
4668 out_site:
4669         lu_site_fini(&o->od_site);
4670 out_compat:
4671         osd_obj_map_fini(o);
4672 out_scrub:
4673         osd_scrub_cleanup(env, o);
4674 out_mnt:
4675         osd_oi_fini(info, o);
4676         osd_shutdown(env, o);
4677         mntput(o->od_mnt);
4678         o->od_mnt = NULL;
4679 out_capa:
4680         cleanup_capa_hash(o->od_capa_hash);
4681 out:
4682         RETURN(rc);
4683 }
4684
4685 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4686                                           struct lu_device_type *t,
4687                                           struct lustre_cfg *cfg)
4688 {
4689         struct osd_device *o;
4690         int                rc;
4691
4692         OBD_ALLOC_PTR(o);
4693         if (o == NULL)
4694                 return ERR_PTR(-ENOMEM);
4695
4696         rc = dt_device_init(&o->od_dt_dev, t);
4697         if (rc == 0) {
4698                 /* Because the ctx might be revived in dt_device_init,
4699                  * refill the env here */
4700                 lu_env_refill((struct lu_env *)env);
4701                 rc = osd_device_init0(env, o, cfg);
4702                 if (rc)
4703                         dt_device_fini(&o->od_dt_dev);
4704         }
4705
4706         if (unlikely(rc != 0))
4707                 OBD_FREE_PTR(o);
4708
4709         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4710 }
4711
4712 static struct lu_device *osd_device_free(const struct lu_env *env,
4713                                          struct lu_device *d)
4714 {
4715         struct osd_device *o = osd_dev(d);
4716         ENTRY;
4717
4718         cleanup_capa_hash(o->od_capa_hash);
4719         /* XXX: make osd top device in order to release reference */
4720         d->ld_site->ls_top_dev = d;
4721         lu_site_purge(env, d->ld_site, -1);
4722         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4723                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4724                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4725         }
4726         lu_site_fini(&o->od_site);
4727         dt_device_fini(&o->od_dt_dev);
4728         OBD_FREE_PTR(o);
4729         RETURN(NULL);
4730 }
4731
4732 static int osd_process_config(const struct lu_env *env,
4733                               struct lu_device *d, struct lustre_cfg *cfg)
4734 {
4735         struct osd_device *o = osd_dev(d);
4736         int err;
4737         ENTRY;
4738
4739         switch(cfg->lcfg_command) {
4740         case LCFG_SETUP:
4741                 err = osd_mount(env, o, cfg);
4742                 break;
4743         case LCFG_CLEANUP:
4744                 lu_dev_del_linkage(d->ld_site, d);
4745                 err = osd_shutdown(env, o);
4746                 break;
4747         default:
4748                 err = -ENOSYS;
4749         }
4750
4751         RETURN(err);
4752 }
4753
4754 static int osd_recovery_complete(const struct lu_env *env,
4755                                  struct lu_device *d)
4756 {
4757         struct osd_device       *osd = osd_dev(d);
4758         int                      rc = 0;
4759         ENTRY;
4760
4761         if (osd->od_quota_slave == NULL)
4762                 RETURN(0);
4763
4764         /* start qsd instance on recovery completion, this notifies the quota
4765          * slave code that we are about to process new requests now */
4766         rc = qsd_start(env, osd->od_quota_slave);
4767         RETURN(rc);
4768 }
4769
4770 /*
4771  * we use exports to track all osd users
4772  */
4773 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4774                            struct obd_device *obd, struct obd_uuid *cluuid,
4775                            struct obd_connect_data *data, void *localdata)
4776 {
4777         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
4778         struct lustre_handle  conn;
4779         int                   rc;
4780         ENTRY;
4781
4782         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4783
4784         rc = class_connect(&conn, obd, cluuid);
4785         if (rc)
4786                 RETURN(rc);
4787
4788         *exp = class_conn2export(&conn);
4789
4790         spin_lock(&osd->od_osfs_lock);
4791         osd->od_connects++;
4792         spin_unlock(&osd->od_osfs_lock);
4793
4794         RETURN(0);
4795 }
4796
4797 /*
4798  * once last export (we don't count self-export) disappeared
4799  * osd can be released
4800  */
4801 static int osd_obd_disconnect(struct obd_export *exp)
4802 {
4803         struct obd_device *obd = exp->exp_obd;
4804         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4805         int                rc, release = 0;
4806         ENTRY;
4807
4808         /* Only disconnect the underlying layers on the final disconnect. */
4809         spin_lock(&osd->od_osfs_lock);
4810         osd->od_connects--;
4811         if (osd->od_connects == 0)
4812                 release = 1;
4813         spin_unlock(&osd->od_osfs_lock);
4814
4815         rc = class_disconnect(exp); /* bz 9811 */
4816
4817         if (rc == 0 && release)
4818                 class_manual_cleanup(obd);
4819         RETURN(rc);
4820 }
4821
4822 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4823                        struct lu_device *dev)
4824 {
4825         struct osd_device *osd = osd_dev(dev);
4826         int                result = 0;
4827         ENTRY;
4828
4829         if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
4830                 /* MDT/MDD still use old infrastructure to create
4831                  * special files */
4832                 result = llo_local_objects_setup(env, lu2md_dev(pdev),
4833                                                  lu2dt_dev(dev));
4834                 if (result)
4835                         RETURN(result);
4836         }
4837
4838         if (osd->od_quota_slave != NULL)
4839                 /* set up quota slave objects */
4840                 result = qsd_prepare(env, osd->od_quota_slave);
4841
4842         RETURN(result);
4843 }
4844
4845 static const struct lu_object_operations osd_lu_obj_ops = {
4846         .loo_object_init      = osd_object_init,
4847         .loo_object_delete    = osd_object_delete,
4848         .loo_object_release   = osd_object_release,
4849         .loo_object_free      = osd_object_free,
4850         .loo_object_print     = osd_object_print,
4851         .loo_object_invariant = osd_object_invariant
4852 };
4853
4854 const struct lu_device_operations osd_lu_ops = {
4855         .ldo_object_alloc      = osd_object_alloc,
4856         .ldo_process_config    = osd_process_config,
4857         .ldo_recovery_complete = osd_recovery_complete,
4858         .ldo_prepare           = osd_prepare,
4859 };
4860
4861 static const struct lu_device_type_operations osd_device_type_ops = {
4862         .ldto_init = osd_type_init,
4863         .ldto_fini = osd_type_fini,
4864
4865         .ldto_start = osd_type_start,
4866         .ldto_stop  = osd_type_stop,
4867
4868         .ldto_device_alloc = osd_device_alloc,
4869         .ldto_device_free  = osd_device_free,
4870
4871         .ldto_device_init    = osd_device_init,
4872         .ldto_device_fini    = osd_device_fini
4873 };
4874
4875 struct lu_device_type osd_device_type = {
4876         .ldt_tags     = LU_DEVICE_DT,
4877         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
4878         .ldt_ops      = &osd_device_type_ops,
4879         .ldt_ctx_tags = LCT_LOCAL,
4880 };
4881
4882 /*
4883  * lprocfs legacy support.
4884  */
4885 static struct obd_ops osd_obd_device_ops = {
4886         .o_owner = THIS_MODULE,
4887         .o_connect      = osd_obd_connect,
4888         .o_disconnect   = osd_obd_disconnect
4889 };
4890
4891 static int __init osd_mod_init(void)
4892 {
4893         struct lprocfs_static_vars lvars;
4894
4895         osd_oi_mod_init();
4896         lprocfs_osd_init_vars(&lvars);
4897         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4898                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4899 }
4900
4901 static void __exit osd_mod_exit(void)
4902 {
4903         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4904 }
4905
4906 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4907 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4908 MODULE_LICENSE("GPL");
4909
4910 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);