lustre/osd-ldiskfs/osd_handler.c

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