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 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_MDS
  48
  49 #include <linux/module.h>
  50
  51 /* LUSTRE_VERSION_CODE */
  52 #include <lustre_ver.h>
  53 /* prerequisite for linux/xattr.h */
  54 #include <linux/types.h>
  55 /* prerequisite for linux/xattr.h */
  56 #include <linux/fs.h>
  57 /* XATTR_{REPLACE,CREATE} */
  58 #include <linux/xattr.h>
  59 /* simple_mkdir() */
  60 #include <lvfs.h>
  61
  62 /*
  63  * struct OBD_{ALLOC,FREE}*()
  64  * OBD_FAIL_CHECK
  65  */
  66 #include <obd_support.h>
  67 /* struct ptlrpc_thread */
  68 #include <lustre_net.h>
  69
  70 /* fid_is_local() */
  71 #include <lustre_fid.h>
  72
  73 #include "osd_internal.h"
  74 #include "osd_igif.h"
  75
  76 /* llo_* api support */
  77 #include <md_object.h>
  78
  79 #ifdef HAVE_LDISKFS_PDO
  80 int ldiskfs_pdo = 1;
  81 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  82                 "ldiskfs with parallel directory operations");
  83 #else
  84 int ldiskfs_pdo = 0;
  85 #endif
  86
  87 static const char dot[] = ".";
  88 static const char dotdot[] = "..";
  89 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  90
  91 static const struct lu_object_operations      osd_lu_obj_ops;
  92 static const struct dt_object_operations      osd_obj_ops;
  93 static const struct dt_object_operations      osd_obj_ea_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)
 110 {
 111         struct md_ucred *uc = md_ucred(env);
 112         struct cred     *tc;
 113
 114         LASSERT(uc != NULL);
 115
 116         save->oc_uid = current_fsuid();
 117         save->oc_gid = current_fsgid();
 118         save->oc_cap = current_cap();
 119         if ((tc = prepare_creds())) {
 120                 tc->fsuid         = uc->mu_fsuid;
 121                 tc->fsgid         = uc->mu_fsgid;
 122                 commit_creds(tc);
 123         }
 124         /* XXX not suboptimal */
 125         cfs_curproc_cap_unpack(uc->mu_cap);
 126 }
 127
 128 static inline void
 129 osd_pop_ctxt(struct osd_ctxt *save)
 130 {
 131         struct cred *tc;
 132
 133         if ((tc = prepare_creds())) {
 134                 tc->fsuid         = save->oc_uid;
 135                 tc->fsgid         = save->oc_gid;
 136                 tc->cap_effective = save->oc_cap;
 137                 commit_creds(tc);
 138         }
 139 }
 140 #endif
 141
 142 /*
 143  * Concurrency: doesn't matter
 144  */
 145 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 146 {
 147         return osd_oti_get(env)->oti_r_locks > 0;
 148 }
 149
 150 /*
 151  * Concurrency: doesn't matter
 152  */
 153 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 154 {
 155         struct osd_thread_info *oti = osd_oti_get(env);
 156         return oti->oti_w_locks > 0 && o->oo_owner == env;
 157 }
 158
 159 /*
 160  * Concurrency: doesn't access mutable data
 161  */
 162 static int osd_root_get(const struct lu_env *env,
 163                         struct dt_device *dev, struct lu_fid *f)
 164 {
 165         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 166         return 0;
 167 }
 168
 169 static inline int osd_qid_type(struct osd_thandle *oh, int i)
 170 {
 171         return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
 172 }
 173
 174 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
 175 {
 176         oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
 177 }
 178
 179 void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
 180                      int type, uid_t id, struct inode *inode)
 181 {
 182 #ifdef CONFIG_QUOTA
 183         int i, allocated = 0;
 184         struct osd_object *obj;
 185
 186         LASSERT(dt != NULL);
 187         LASSERT(oh != NULL);
 188         LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
 189                  oh->ot_id_cnt);
 190
 191         /* id entry is allocated in the quota file */
 192         if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
 193                 allocated = 1;
 194
 195         for (i = 0; i < oh->ot_id_cnt; i++) {
 196                 if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
 197                         return;
 198         }
 199
 200         if (unlikely(i >= OSD_MAX_UGID_CNT)) {
 201                 CERROR("more than %d uid/gids for a transaction?\n", i);
 202                 return;
 203         }
 204
 205         oh->ot_id_array[i] = id;
 206         osd_qid_set_type(oh, i, type);
 207         oh->ot_id_cnt++;
 208         obj = osd_dt_obj(dt);
 209         oh->ot_credits += (allocated || id == 0) ?
 210                 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
 211 #endif
 212 }
 213
 214 /*
 215  * OSD object methods.
 216  */
 217
 218 /*
 219  * Concurrency: no concurrent access is possible that early in object
 220  * life-cycle.
 221  */
 222 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 223                                           const struct lu_object_header *hdr,
 224                                           struct lu_device *d)
 225 {
 226         struct osd_object *mo;
 227
 228         OBD_ALLOC_PTR(mo);
 229         if (mo != NULL) {
 230                 struct lu_object *l;
 231
 232                 l = &mo->oo_dt.do_lu;
 233                 dt_object_init(&mo->oo_dt, NULL, d);
 234                 if (osd_dev(d)->od_iop_mode)
 235                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 236                 else
 237                         mo->oo_dt.do_ops = &osd_obj_ops;
 238
 239                 l->lo_ops = &osd_lu_obj_ops;
 240                 cfs_init_rwsem(&mo->oo_sem);
 241                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 242                 cfs_spin_lock_init(&mo->oo_guard);
 243                 return l;
 244         } else {
 245                 return NULL;
 246         }
 247 }
 248
 249 /*
 250  * retrieve object from backend ext fs.
 251  **/
 252 struct inode *osd_iget(struct osd_thread_info *info,
 253                        struct osd_device *dev,
 254                        const struct osd_inode_id *id)
 255 {
 256         struct inode *inode = NULL;
 257
 258         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 259         if (IS_ERR(inode)) {
 260                 CERROR("Cannot get inode, rc = %li\n", PTR_ERR(inode));
 261         } else if (id->oii_gen != OSD_OII_NOGEN &&
 262                    inode->i_generation != id->oii_gen) {
 263                 iput(inode);
 264                 inode = ERR_PTR(-ESTALE);
 265         } else if (inode->i_nlink == 0) {
 266                 /* due to parallel readdir and unlink,
 267                 * we can have dead inode here. */
 268                 CWARN("stale inode\n");
 269                 make_bad_inode(inode);
 270                 iput(inode);
 271                 inode = ERR_PTR(-ESTALE);
 272         } else if (is_bad_inode(inode)) {
 273                 CERROR("bad inode %lx\n",inode->i_ino);
 274                 iput(inode);
 275                 inode = ERR_PTR(-ENOENT);
 276         } else {
 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 static int osd_fid_lookup(const struct lu_env *env,
 289                           struct osd_object *obj, const struct lu_fid *fid)
 290 {
 291         struct osd_thread_info *info;
 292         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
 293         struct osd_device      *dev;
 294         struct osd_inode_id    *id;
 295         struct inode           *inode;
 296         int                     result;
 297
 298         LINVRNT(osd_invariant(obj));
 299         LASSERT(obj->oo_inode == NULL);
 300         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 301         /*
 302          * This assertion checks that osd layer sees only local
 303          * fids. Unfortunately it is somewhat expensive (does a
 304          * cache-lookup). Disabling it for production/acceptance-testing.
 305          */
 306         LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
 307
 308         ENTRY;
 309
 310         info = osd_oti_get(env);
 311         LASSERT(info);
 312         dev  = osd_dev(ldev);
 313         id   = &info->oti_id;
 314
 315         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 316                 RETURN(-ENOENT);
 317
 318         result = osd_oi_lookup(info, dev, fid, id);
 319         if (result != 0) {
 320                 if (result == -ENOENT)
 321                         result = 0;
 322                 GOTO(out, result);
 323         }
 324
 325         inode = osd_iget(info, dev, id);
 326         if (IS_ERR(inode)) {
 327                 /*
 328                  * If fid wasn't found in oi, inode-less object is
 329                  * created, for which lu_object_exists() returns
 330                  * false. This is used in a (frequent) case when
 331                  * objects are created as locking anchors or
 332                  * place holders for objects yet to be created.
 333                  */
 334                 result = PTR_ERR(inode);
 335                 GOTO(out, result);
 336         }
 337
 338         obj->oo_inode = inode;
 339         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 340         if (dev->od_iop_mode) {
 341                 obj->oo_compat_dot_created = 1;
 342                 obj->oo_compat_dotdot_created = 1;
 343         }
 344
 345         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 346                 goto out;
 347
 348         LASSERT(obj->oo_hl_head == NULL);
 349         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 350         if (obj->oo_hl_head == NULL) {
 351                 obj->oo_inode = NULL;
 352                 iput(inode);
 353                 result = -ENOMEM;
 354         }
 355 out:
 356         LINVRNT(osd_invariant(obj));
 357         RETURN(result);
 358 }
 359
 360 /*
 361  * Concurrency: shouldn't matter.
 362  */
 363 static void osd_object_init0(struct osd_object *obj)
 364 {
 365         LASSERT(obj->oo_inode != NULL);
 366         obj->oo_dt.do_body_ops = &osd_body_ops;
 367         obj->oo_dt.do_lu.lo_header->loh_attr |=
 368                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 369 }
 370
 371 /*
 372  * Concurrency: no concurrent access is possible that early in object
 373  * life-cycle.
 374  */
 375 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 376                            const struct lu_object_conf *unused)
 377 {
 378         struct osd_object *obj = osd_obj(l);
 379         int result;
 380
 381         LINVRNT(osd_invariant(obj));
 382
 383         result = osd_fid_lookup(env, obj, lu_object_fid(l));
 384         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 385         if (result == 0) {
 386                 if (obj->oo_inode != NULL)
 387                         osd_object_init0(obj);
 388         }
 389         LINVRNT(osd_invariant(obj));
 390         return result;
 391 }
 392
 393 /*
 394  * Concurrency: no concurrent access is possible that late in object
 395  * life-cycle.
 396  */
 397 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 398 {
 399         struct osd_object *obj = osd_obj(l);
 400
 401         LINVRNT(osd_invariant(obj));
 402
 403         dt_object_fini(&obj->oo_dt);
 404         if (obj->oo_hl_head != NULL)
 405                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 406         OBD_FREE_PTR(obj);
 407 }
 408
 409 /*
 410  * Concurrency: no concurrent access is possible that late in object
 411  * life-cycle.
 412  */
 413 static void osd_index_fini(struct osd_object *o)
 414 {
 415         struct iam_container *bag;
 416
 417         if (o->oo_dir != NULL) {
 418                 bag = &o->oo_dir->od_container;
 419                 if (o->oo_inode != NULL) {
 420                         if (bag->ic_object == o->oo_inode)
 421                                 iam_container_fini(bag);
 422                 }
 423                 OBD_FREE_PTR(o->oo_dir);
 424                 o->oo_dir = NULL;
 425         }
 426 }
 427
 428 /*
 429  * Concurrency: no concurrent access is possible that late in object
 430  * life-cycle (for all existing callers, that is. New callers have to provide
 431  * their own locking.)
 432  */
 433 static int osd_inode_unlinked(const struct inode *inode)
 434 {
 435         return inode->i_nlink == 0;
 436 }
 437
 438 enum {
 439         OSD_TXN_OI_DELETE_CREDITS    = 20,
 440         OSD_TXN_INODE_DELETE_CREDITS = 20
 441 };
 442
 443 /*
 444  * Journal
 445  */
 446
 447 #if OSD_THANDLE_STATS
 448 /**
 449  * Set time when the handle is allocated
 450  */
 451 static void osd_th_alloced(struct osd_thandle *oth)
 452 {
 453         oth->oth_alloced = cfs_time_current();
 454 }
 455
 456 /**
 457  * Set time when the handle started
 458  */
 459 static void osd_th_started(struct osd_thandle *oth)
 460 {
 461         oth->oth_started = cfs_time_current();
 462 }
 463
 464 /**
 465  * Helper function to convert time interval to microseconds packed in
 466  * long int (default time units for the counter in "stats" initialized
 467  * by lu_time_init() )
 468  */
 469 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 470 {
 471         struct timeval val;
 472
 473         cfs_duration_usec(cfs_time_sub(end, start), &val);
 474         return val.tv_sec * 1000000 + val.tv_usec;
 475 }
 476
 477 /**
 478  * Check whether the we deal with this handle for too long.
 479  */
 480 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 481                                 cfs_time_t alloced, cfs_time_t started,
 482                                 cfs_time_t closed)
 483 {
 484         cfs_time_t now = cfs_time_current();
 485
 486         LASSERT(dev != NULL);
 487
 488         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 489                             interval_to_usec(alloced, started));
 490         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 491                             interval_to_usec(started, closed));
 492         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 493                             interval_to_usec(closed, now));
 494
 495         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 496                 CWARN("transaction handle %p was open for too long: "
 497                       "now "CFS_TIME_T" ,"
 498                       "alloced "CFS_TIME_T" ,"
 499                       "started "CFS_TIME_T" ,"
 500                       "closed "CFS_TIME_T"\n",
 501                       oth, now, alloced, started, closed);
 502                 libcfs_debug_dumpstack(NULL);
 503         }
 504 }
 505
 506 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 507 {                                                                       \
 508         cfs_time_t __closed = cfs_time_current();                       \
 509         cfs_time_t __alloced = oth->oth_alloced;                        \
 510         cfs_time_t __started = oth->oth_started;                        \
 511                                                                         \
 512         expr;                                                           \
 513         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 514 }
 515
 516 #else /* OSD_THANDLE_STATS */
 517
 518 #define osd_th_alloced(h)                  do {} while(0)
 519 #define osd_th_started(h)                  do {} while(0)
 520 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 521
 522 #endif /* OSD_THANDLE_STATS */
 523
 524 /*
 525  * Concurrency: doesn't access mutable data.
 526  */
 527 static int osd_param_is_sane(const struct osd_device *dev,
 528                              const struct thandle *th)
 529 {
 530         struct osd_thandle *oh;
 531         oh = container_of0(th, struct osd_thandle, ot_super);
 532         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 533 }
 534
 535 /*
 536  * Concurrency: shouldn't matter.
 537  */
 538 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 539 static void osd_trans_commit_cb(struct super_block *sb,
 540                                 struct journal_callback *jcb, int error)
 541 #else
 542 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 543 #endif
 544 {
 545         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 546         struct thandle     *th  = &oh->ot_super;
 547         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 548         struct dt_txn_commit_cb *dcb, *tmp;
 549
 550         LASSERT(oh->ot_handle == NULL);
 551
 552         if (error)
 553                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 554
 555         dt_txn_hook_commit(th);
 556
 557         /* call per-transaction callbacks if any */
 558         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage)
 559                 dcb->dcb_func(NULL, th, dcb, error);
 560
 561         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 562         lu_device_put(lud);
 563         th->th_dev = NULL;
 564
 565         lu_context_exit(&th->th_ctx);
 566         lu_context_fini(&th->th_ctx);
 567         OBD_FREE_PTR(oh);
 568 }
 569
 570 static struct thandle *osd_trans_create(const struct lu_env *env,
 571                                         struct dt_device *d)
 572 {
 573         struct osd_thread_info *oti = osd_oti_get(env);
 574         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 575         struct osd_thandle     *oh;
 576         struct thandle         *th;
 577         ENTRY;
 578
 579         /* on pending IO in this thread should left from prev. request */
 580         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 581
 582         th = ERR_PTR(-ENOMEM);
 583         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 584         if (oh != NULL) {
 585                 th = &oh->ot_super;
 586                 th->th_dev = d;
 587                 th->th_result = 0;
 588                 th->th_tags = LCT_TX_HANDLE;
 589                 oh->ot_credits = 0;
 590                 oti->oti_dev = osd_dt_dev(d);
 591                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 592                 osd_th_alloced(oh);
 593         }
 594         RETURN(th);
 595 }
 596
 597 /*
 598  * Concurrency: shouldn't matter.
 599  */
 600 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 601                     struct thandle *th)
 602 {
 603         struct osd_thread_info *oti = osd_oti_get(env);
 604         struct osd_device  *dev = osd_dt_dev(d);
 605         handle_t           *jh;
 606         struct osd_thandle *oh;
 607         int rc;
 608
 609         ENTRY;
 610
 611         LASSERT(current->journal_info == NULL);
 612
 613         oh = container_of0(th, struct osd_thandle, ot_super);
 614         LASSERT(oh != NULL);
 615         LASSERT(oh->ot_handle == NULL);
 616
 617         rc = dt_txn_hook_start(env, d, th);
 618         if (rc != 0)
 619                 GOTO(out, rc);
 620
 621         if (!osd_param_is_sane(dev, th)) {
 622                 CWARN("%s: too many transaction credits (%d > %d)\n",
 623                       d->dd_lu_dev.ld_obd->obd_name, oh->ot_credits,
 624                       osd_journal(dev)->j_max_transaction_buffers);
 625                 /* XXX Limit the credits to 'max_transaction_buffers', and
 626                  *     let the underlying filesystem to catch the error if
 627                  *     we really need so many credits.
 628                  *
 629                  *     This should be removed when we can calculate the
 630                  *     credits precisely. */
 631                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 632 #ifdef OSD_TRACK_DECLARES
 633                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 634                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 635                        oh->ot_declare_xattr_set);
 636                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 637                        oh->ot_declare_create, oh->ot_declare_ref_add,
 638                        oh->ot_declare_ref_del, oh->ot_declare_write);
 639                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 640                        oh->ot_declare_insert, oh->ot_declare_delete,
 641                        oh->ot_declare_destroy);
 642 #endif
 643         }
 644
 645         /*
 646          * XXX temporary stuff. Some abstraction layer should
 647          * be used.
 648          */
 649         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 650         osd_th_started(oh);
 651         if (!IS_ERR(jh)) {
 652                 oh->ot_handle = jh;
 653                 LASSERT(oti->oti_txns == 0);
 654                 lu_context_init(&th->th_ctx, th->th_tags);
 655                 lu_context_enter(&th->th_ctx);
 656
 657                 lu_device_get(&d->dd_lu_dev);
 658                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 659                                              "osd-tx", th);
 660
 661                 /*
 662                  * XXX: current rule is that we first start tx,
 663                  *      then lock object(s), but we can't use
 664                  *      this rule for data (due to locking specifics
 665                  *      in ldiskfs). also in long-term we'd like to
 666                  *      use usually-used (locks;tx) ordering. so,
 667                  *      UGLY thing is that we'll use one ordering for
 668                  *      data (ofd) and reverse ordering for metadata
 669                  *      (mdd). then at some point we'll fix the latter
 670                  */
 671                 if (lu_device_is_md(&d->dd_lu_dev)) {
 672                         LASSERT(oti->oti_r_locks == 0);
 673                         LASSERT(oti->oti_w_locks == 0);
 674                 }
 675
 676                 oti->oti_txns++;
 677                 rc = 0;
 678         } else {
 679                 rc = PTR_ERR(jh);
 680         }
 681 out:
 682         RETURN(rc);
 683 }
 684
 685 /*
 686  * Concurrency: shouldn't matter.
 687  */
 688 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 689 {
 690         int                     rc = 0;
 691         struct osd_thandle     *oh;
 692         struct osd_thread_info *oti = osd_oti_get(env);
 693         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 694
 695         ENTRY;
 696
 697         oh = container_of0(th, struct osd_thandle, ot_super);
 698
 699         if (oh->ot_handle != NULL) {
 700                 handle_t *hdl = oh->ot_handle;
 701
 702                 hdl->h_sync = th->th_sync;
 703
 704                 /*
 705                  * add commit callback
 706                  * notice we don't do this in osd_trans_start()
 707                  * as underlying transaction can change during truncate
 708                  */
 709                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 710                                          &oh->ot_jcb);
 711
 712                 LASSERT(oti->oti_txns == 1);
 713                 oti->oti_txns--;
 714                 /*
 715                  * XXX: current rule is that we first start tx,
 716                  *      then lock object(s), but we can't use
 717                  *      this rule for data (due to locking specifics
 718                  *      in ldiskfs). also in long-term we'd like to
 719                  *      use usually-used (locks;tx) ordering. so,
 720                  *      UGLY thing is that we'll use one ordering for
 721                  *      data (ofd) and reverse ordering for metadata
 722                  *      (mdd). then at some point we'll fix the latter
 723                  */
 724                 if (lu_device_is_md(&th->th_dev->dd_lu_dev)) {
 725                         LASSERT(oti->oti_r_locks == 0);
 726                         LASSERT(oti->oti_w_locks == 0);
 727                 }
 728                 rc = dt_txn_hook_stop(env, th);
 729                 if (rc != 0)
 730                         CERROR("Failure in transaction hook: %d\n", rc);
 731                 oh->ot_handle = NULL;
 732                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 733                                   rc = ldiskfs_journal_stop(hdl));
 734                 if (rc != 0)
 735                         CERROR("Failure to stop transaction: %d\n", rc);
 736         } else {
 737                 OBD_FREE_PTR(oh);
 738         }
 739
 740         /* as we want IO to journal and data IO be concurrent, we don't block
 741          * awaiting data IO completion in osd_do_bio(), instead we wait here
 742          * once transaction is submitted to the journal. all reqular requests
 743          * don't do direct IO (except read/write), thus this wait_event becomes
 744          * no-op for them.
 745          *
 746          * IMPORTANT: we have to wait till any IO submited by the thread is
 747          * completed otherwise iobuf may be corrupted by different request
 748          */
 749         cfs_wait_event(iobuf->dr_wait,
 750                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 751         if (!rc)
 752                 rc = iobuf->dr_error;
 753
 754         RETURN(rc);
 755 }
 756
 757 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 758 {
 759         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 760                                                ot_super);
 761
 762         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 763
 764         return 0;
 765 }
 766
 767 /*
 768  * Called just before object is freed. Releases all resources except for
 769  * object itself (that is released by osd_object_free()).
 770  *
 771  * Concurrency: no concurrent access is possible that late in object
 772  * life-cycle.
 773  */
 774 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 775 {
 776         struct osd_object *obj   = osd_obj(l);
 777         struct inode      *inode = obj->oo_inode;
 778
 779         LINVRNT(osd_invariant(obj));
 780
 781         /*
 782          * If object is unlinked remove fid->ino mapping from object index.
 783          */
 784
 785         osd_index_fini(obj);
 786         if (inode != NULL) {
 787                 iput(inode);
 788                 obj->oo_inode = NULL;
 789         }
 790 }
 791
 792 /*
 793  * Concurrency: ->loo_object_release() is called under site spin-lock.
 794  */
 795 static void osd_object_release(const struct lu_env *env,
 796                                struct lu_object *l)
 797 {
 798 }
 799
 800 /*
 801  * Concurrency: shouldn't matter.
 802  */
 803 static int osd_object_print(const struct lu_env *env, void *cookie,
 804                             lu_printer_t p, const struct lu_object *l)
 805 {
 806         struct osd_object *o = osd_obj(l);
 807         struct iam_descr  *d;
 808
 809         if (o->oo_dir != NULL)
 810                 d = o->oo_dir->od_container.ic_descr;
 811         else
 812                 d = NULL;
 813         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 814                     o, o->oo_inode,
 815                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 816                     o->oo_inode ? o->oo_inode->i_generation : 0,
 817                     d ? d->id_ops->id_name : "plain");
 818 }
 819
 820 /*
 821  * Concurrency: shouldn't matter.
 822  */
 823 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 824                struct obd_statfs *sfs)
 825 {
 826         struct osd_device  *osd = osd_dt_dev(d);
 827         struct super_block *sb = osd_sb(osd);
 828         struct kstatfs     *ksfs;
 829         int result = 0;
 830
 831         /* osd_lproc.c call this without env, allocate ksfs for that case */
 832         if (unlikely(env == NULL)) {
 833                 OBD_ALLOC_PTR(ksfs);
 834                 if (ksfs == NULL)
 835                         return -ENOMEM;
 836         } else {
 837                 ksfs = &osd_oti_get(env)->oti_ksfs;
 838         }
 839
 840         cfs_spin_lock(&osd->od_osfs_lock);
 841         /* cache 1 second */
 842         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 843                 result = ll_do_statfs(sb, ksfs);
 844                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 845                         osd->od_osfs_age = cfs_time_current_64();
 846                         statfs_pack(&osd->od_statfs, ksfs);
 847                 }
 848         }
 849
 850         if (likely(result == 0))
 851                 *sfs = osd->od_statfs;
 852         cfs_spin_unlock(&osd->od_osfs_lock);
 853
 854         if (unlikely(env == NULL))
 855                 OBD_FREE_PTR(ksfs);
 856
 857         return result;
 858 }
 859
 860 /*
 861  * Concurrency: doesn't access mutable data.
 862  */
 863 static void osd_conf_get(const struct lu_env *env,
 864                          const struct dt_device *dev,
 865                          struct dt_device_param *param)
 866 {
 867         struct super_block *sb = osd_sb(osd_dt_dev(dev));
 868
 869         /*
 870          * XXX should be taken from not-yet-existing fs abstraction layer.
 871          */
 872         param->ddp_max_name_len = LDISKFS_NAME_LEN;
 873         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
 874         param->ddp_block_shift  = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
 875         param->ddp_mntopts      = 0;
 876         if (test_opt(sb, XATTR_USER))
 877                 param->ddp_mntopts |= MNTOPT_USERXATTR;
 878         if (test_opt(sb, POSIX_ACL))
 879                 param->ddp_mntopts |= MNTOPT_ACL;
 880
 881 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
 882         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
 883                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
 884         else
 885 #endif
 886                 param->ddp_max_ea_size = sb->s_blocksize;
 887
 888 }
 889
 890 /**
 891  * Helper function to get and fill the buffer with input values.
 892  */
 893 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
 894 {
 895         struct lu_buf *buf;
 896
 897         buf = &osd_oti_get(env)->oti_buf;
 898         buf->lb_buf = area;
 899         buf->lb_len = len;
 900         return buf;
 901 }
 902
 903 /*
 904  * Concurrency: shouldn't matter.
 905  */
 906 static int osd_sync(const struct lu_env *env, struct dt_device *d)
 907 {
 908         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
 909         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
 910 }
 911
 912 /**
 913  * Start commit for OSD device.
 914  *
 915  * An implementation of dt_commit_async method for OSD device.
 916  * Asychronously starts underlayng fs sync and thereby a transaction
 917  * commit.
 918  *
 919  * \param env environment
 920  * \param d dt device
 921  *
 922  * \see dt_device_operations
 923  */
 924 static int osd_commit_async(const struct lu_env *env,
 925                             struct dt_device *d)
 926 {
 927         struct super_block *s = osd_sb(osd_dt_dev(d));
 928         ENTRY;
 929
 930         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
 931         RETURN(s->s_op->sync_fs(s, 0));
 932 }
 933
 934 /*
 935  * Concurrency: shouldn't matter.
 936  */
 937
 938 static int osd_ro(const struct lu_env *env, struct dt_device *d)
 939 {
 940         struct super_block *sb = osd_sb(osd_dt_dev(d));
 941         int rc;
 942         ENTRY;
 943
 944         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
 945
 946         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
 947         RETURN(rc);
 948 }
 949
 950 /*
 951  * Concurrency: serialization provided by callers.
 952  */
 953 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
 954                               int mode, unsigned long timeout, __u32 alg,
 955                               struct lustre_capa_key *keys)
 956 {
 957         struct osd_device *dev = osd_dt_dev(d);
 958         ENTRY;
 959
 960         dev->od_fl_capa = mode;
 961         dev->od_capa_timeout = timeout;
 962         dev->od_capa_alg = alg;
 963         dev->od_capa_keys = keys;
 964         RETURN(0);
 965 }
 966
 967 /**
 968  * Concurrency: serialization provided by callers.
 969  */
 970 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
 971                                struct dt_quota_ctxt *ctxt, void *data)
 972 {
 973         struct obd_device *obd = (void *)ctxt;
 974         struct vfsmount *mnt = (struct vfsmount *)data;
 975         ENTRY;
 976
 977         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
 978         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
 979         obd->obd_lvfs_ctxt.pwdmnt = mnt;
 980         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
 981         obd->obd_lvfs_ctxt.fs = get_ds();
 982
 983         EXIT;
 984 }
 985
 986 /**
 987  * Note: we do not count into QUOTA here.
 988  * If we mount with --data_journal we may need more.
 989  */
 990 const int osd_dto_credits_noquota[DTO_NR] = {
 991         /**
 992          * Insert/Delete.
 993          * INDEX_EXTRA_TRANS_BLOCKS(8) +
 994          * SINGLEDATA_TRANS_BLOCKS(8)
 995          * XXX Note: maybe iam need more, since iam have more level than
 996          *           EXT3 htree.
 997          */
 998         [DTO_INDEX_INSERT]  = 16,
 999         [DTO_INDEX_DELETE]  = 16,
1000         /**
1001          * Unused now
1002          */
1003         [DTO_INDEX_UPDATE]  = 16,
1004         /**
1005          * Create a object. The same as create object in EXT3.
1006          * DATA_TRANS_BLOCKS(14) +
1007          * INDEX_EXTRA_BLOCKS(8) +
1008          * 3(inode bits, groups, GDT)
1009          */
1010         [DTO_OBJECT_CREATE] = 25,
1011         /**
1012          * XXX: real credits to be fixed
1013          */
1014         [DTO_OBJECT_DELETE] = 25,
1015         /**
1016          * Attr set credits (inode)
1017          */
1018         [DTO_ATTR_SET_BASE] = 1,
1019         /**
1020          * Xattr set. The same as xattr of EXT3.
1021          * DATA_TRANS_BLOCKS(14)
1022          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1023          * are also counted in. Do not know why?
1024          */
1025         [DTO_XATTR_SET]     = 14,
1026         [DTO_LOG_REC]       = 14,
1027         /**
1028          * credits for inode change during write.
1029          */
1030         [DTO_WRITE_BASE]    = 3,
1031         /**
1032          * credits for single block write.
1033          */
1034         [DTO_WRITE_BLOCK]   = 14,
1035         /**
1036          * Attr set credits for chown.
1037          * This is extra credits for setattr, and it is null without quota
1038          */
1039         [DTO_ATTR_SET_CHOWN]= 0
1040 };
1041
1042 static const struct dt_device_operations osd_dt_ops = {
1043         .dt_root_get       = osd_root_get,
1044         .dt_statfs         = osd_statfs,
1045         .dt_trans_create   = osd_trans_create,
1046         .dt_trans_start    = osd_trans_start,
1047         .dt_trans_stop     = osd_trans_stop,
1048         .dt_trans_cb_add   = osd_trans_cb_add,
1049         .dt_conf_get       = osd_conf_get,
1050         .dt_sync           = osd_sync,
1051         .dt_ro             = osd_ro,
1052         .dt_commit_async   = osd_commit_async,
1053         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1054         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1055 };
1056
1057 static void osd_object_read_lock(const struct lu_env *env,
1058                                  struct dt_object *dt, unsigned role)
1059 {
1060         struct osd_object *obj = osd_dt_obj(dt);
1061         struct osd_thread_info *oti = osd_oti_get(env);
1062
1063         LINVRNT(osd_invariant(obj));
1064
1065         LASSERT(obj->oo_owner != env);
1066         cfs_down_read_nested(&obj->oo_sem, role);
1067
1068         LASSERT(obj->oo_owner == NULL);
1069         oti->oti_r_locks++;
1070 }
1071
1072 static void osd_object_write_lock(const struct lu_env *env,
1073                                   struct dt_object *dt, unsigned role)
1074 {
1075         struct osd_object *obj = osd_dt_obj(dt);
1076         struct osd_thread_info *oti = osd_oti_get(env);
1077
1078         LINVRNT(osd_invariant(obj));
1079
1080         LASSERT(obj->oo_owner != env);
1081         cfs_down_write_nested(&obj->oo_sem, role);
1082
1083         LASSERT(obj->oo_owner == NULL);
1084         obj->oo_owner = env;
1085         oti->oti_w_locks++;
1086 }
1087
1088 static void osd_object_read_unlock(const struct lu_env *env,
1089                                    struct dt_object *dt)
1090 {
1091         struct osd_object *obj = osd_dt_obj(dt);
1092         struct osd_thread_info *oti = osd_oti_get(env);
1093
1094         LINVRNT(osd_invariant(obj));
1095
1096         LASSERT(oti->oti_r_locks > 0);
1097         oti->oti_r_locks--;
1098         cfs_up_read(&obj->oo_sem);
1099 }
1100
1101 static void osd_object_write_unlock(const struct lu_env *env,
1102                                     struct dt_object *dt)
1103 {
1104         struct osd_object *obj = osd_dt_obj(dt);
1105         struct osd_thread_info *oti = osd_oti_get(env);
1106
1107         LINVRNT(osd_invariant(obj));
1108
1109         LASSERT(obj->oo_owner == env);
1110         LASSERT(oti->oti_w_locks > 0);
1111         oti->oti_w_locks--;
1112         obj->oo_owner = NULL;
1113         cfs_up_write(&obj->oo_sem);
1114 }
1115
1116 static int osd_object_write_locked(const struct lu_env *env,
1117                                    struct dt_object *dt)
1118 {
1119         struct osd_object *obj = osd_dt_obj(dt);
1120
1121         LINVRNT(osd_invariant(obj));
1122
1123         return obj->oo_owner == env;
1124 }
1125
1126 static int capa_is_sane(const struct lu_env *env,
1127                         struct osd_device *dev,
1128                         struct lustre_capa *capa,
1129                         struct lustre_capa_key *keys)
1130 {
1131         struct osd_thread_info *oti = osd_oti_get(env);
1132         struct lustre_capa *tcapa = &oti->oti_capa;
1133         struct obd_capa *oc;
1134         int i, rc = 0;
1135         ENTRY;
1136
1137         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1138         if (oc) {
1139                 if (capa_is_expired(oc)) {
1140                         DEBUG_CAPA(D_ERROR, capa, "expired");
1141                         rc = -ESTALE;
1142                 }
1143                 capa_put(oc);
1144                 RETURN(rc);
1145         }
1146
1147         if (capa_is_expired_sec(capa)) {
1148                 DEBUG_CAPA(D_ERROR, capa, "expired");
1149                 RETURN(-ESTALE);
1150         }
1151
1152         cfs_spin_lock(&capa_lock);
1153         for (i = 0; i < 2; i++) {
1154                 if (keys[i].lk_keyid == capa->lc_keyid) {
1155                         oti->oti_capa_key = keys[i];
1156                         break;
1157                 }
1158         }
1159         cfs_spin_unlock(&capa_lock);
1160
1161         if (i == 2) {
1162                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1163                 RETURN(-ESTALE);
1164         }
1165
1166         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1167         if (rc)
1168                 RETURN(rc);
1169
1170         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1171                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1172                 RETURN(-EACCES);
1173         }
1174
1175         oc = capa_add(dev->od_capa_hash, capa);
1176         capa_put(oc);
1177
1178         RETURN(0);
1179 }
1180
1181 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1182                     struct lustre_capa *capa, __u64 opc)
1183 {
1184         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1185         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1186         struct md_capainfo *ci;
1187         int rc;
1188
1189         if (!dev->od_fl_capa)
1190                 return 0;
1191
1192         if (capa == BYPASS_CAPA)
1193                 return 0;
1194
1195         ci = md_capainfo(env);
1196         if (unlikely(!ci))
1197                 return 0;
1198
1199         if (ci->mc_auth == LC_ID_NONE)
1200                 return 0;
1201
1202         if (!capa) {
1203                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1204                 return -EACCES;
1205         }
1206
1207         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1208                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1209                            PFID(fid));
1210                 return -EACCES;
1211         }
1212
1213         if (!capa_opc_supported(capa, opc)) {
1214                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1215                 return -EACCES;
1216         }
1217
1218         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1219                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1220                 return -EACCES;
1221         }
1222
1223         return 0;
1224 }
1225
1226 static struct timespec *osd_inode_time(const struct lu_env *env,
1227                                        struct inode *inode, __u64 seconds)
1228 {
1229         struct osd_thread_info *oti = osd_oti_get(env);
1230         struct timespec        *t   = &oti->oti_time;
1231
1232         t->tv_sec  = seconds;
1233         t->tv_nsec = 0;
1234         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1235         return t;
1236 }
1237
1238
1239 static void osd_inode_getattr(const struct lu_env *env,
1240                               struct inode *inode, struct lu_attr *attr)
1241 {
1242         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1243                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1244                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1245
1246         attr->la_atime      = LTIME_S(inode->i_atime);
1247         attr->la_mtime      = LTIME_S(inode->i_mtime);
1248         attr->la_ctime      = LTIME_S(inode->i_ctime);
1249         attr->la_mode       = inode->i_mode;
1250         attr->la_size       = i_size_read(inode);
1251         attr->la_blocks     = inode->i_blocks;
1252         attr->la_uid        = inode->i_uid;
1253         attr->la_gid        = inode->i_gid;
1254         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1255         attr->la_nlink      = inode->i_nlink;
1256         attr->la_rdev       = inode->i_rdev;
1257         attr->la_blksize    = ll_inode_blksize(inode);
1258         attr->la_blkbits    = inode->i_blkbits;
1259 }
1260
1261 static int osd_attr_get(const struct lu_env *env,
1262                         struct dt_object *dt,
1263                         struct lu_attr *attr,
1264                         struct lustre_capa *capa)
1265 {
1266         struct osd_object *obj = osd_dt_obj(dt);
1267
1268         LASSERT(dt_object_exists(dt));
1269         LINVRNT(osd_invariant(obj));
1270
1271         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1272                 return -EACCES;
1273
1274         cfs_spin_lock(&obj->oo_guard);
1275         osd_inode_getattr(env, obj->oo_inode, attr);
1276         cfs_spin_unlock(&obj->oo_guard);
1277         return 0;
1278 }
1279
1280 static int osd_declare_attr_set(const struct lu_env *env,
1281                                 struct dt_object *dt,
1282                                 const struct lu_attr *attr,
1283                                 struct thandle *handle)
1284 {
1285         struct osd_thandle *oh;
1286         struct osd_object *obj;
1287
1288         LASSERT(dt != NULL);
1289         LASSERT(handle != NULL);
1290
1291         obj = osd_dt_obj(dt);
1292         LASSERT(osd_invariant(obj));
1293
1294         oh = container_of0(handle, struct osd_thandle, ot_super);
1295         LASSERT(oh->ot_handle == NULL);
1296
1297         OSD_DECLARE_OP(oh, attr_set);
1298         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1299
1300         if (attr && attr->la_valid & LA_UID) {
1301                 if (obj->oo_inode)
1302                         osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
1303                                         obj->oo_inode);
1304                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1305         }
1306         if (attr && attr->la_valid & LA_GID) {
1307                 if (obj->oo_inode)
1308                         osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
1309                                         obj->oo_inode);
1310                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1311         }
1312
1313         return 0;
1314 }
1315
1316 static int osd_inode_setattr(const struct lu_env *env,
1317                              struct inode *inode, const struct lu_attr *attr)
1318 {
1319         __u64 bits;
1320
1321         bits = attr->la_valid;
1322
1323         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1324
1325         if (bits & LA_ATIME)
1326                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1327         if (bits & LA_CTIME)
1328                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1329         if (bits & LA_MTIME)
1330                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1331         if (bits & LA_SIZE) {
1332                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1333                 i_size_write(inode, attr->la_size);
1334         }
1335
1336 #if 0
1337         /* OSD should not change "i_blocks" which is used by quota.
1338          * "i_blocks" should be changed by ldiskfs only. */
1339         if (bits & LA_BLOCKS)
1340                 inode->i_blocks = attr->la_blocks;
1341 #endif
1342         if (bits & LA_MODE)
1343                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1344                         (attr->la_mode & ~S_IFMT);
1345         if (bits & LA_UID)
1346                 inode->i_uid    = attr->la_uid;
1347         if (bits & LA_GID)
1348                 inode->i_gid    = attr->la_gid;
1349         if (bits & LA_NLINK)
1350                 inode->i_nlink  = attr->la_nlink;
1351         if (bits & LA_RDEV)
1352                 inode->i_rdev   = attr->la_rdev;
1353
1354         if (bits & LA_FLAGS) {
1355                 /* always keep S_NOCMTIME */
1356                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1357                                  S_NOCMTIME;
1358         }
1359         return 0;
1360 }
1361
1362 static int osd_attr_set(const struct lu_env *env,
1363                         struct dt_object *dt,
1364                         const struct lu_attr *attr,
1365                         struct thandle *handle,
1366                         struct lustre_capa *capa)
1367 {
1368         struct osd_object *obj = osd_dt_obj(dt);
1369         struct inode      *inode;
1370         int rc;
1371
1372         LASSERT(handle != NULL);
1373         LASSERT(dt_object_exists(dt));
1374         LASSERT(osd_invariant(obj));
1375
1376         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1377                 return -EACCES;
1378
1379         OSD_EXEC_OP(handle, attr_set);
1380
1381         inode = obj->oo_inode;
1382 #ifdef HAVE_QUOTA_SUPPORT
1383         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1384             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1385                 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1386                 struct iattr iattr;
1387                 int rc;
1388
1389                 iattr.ia_valid = 0;
1390                 if (attr->la_valid & LA_UID)
1391                         iattr.ia_valid |= ATTR_UID;
1392                 if (attr->la_valid & LA_GID)
1393                         iattr.ia_valid |= ATTR_GID;
1394                 iattr.ia_uid = attr->la_uid;
1395                 iattr.ia_gid = attr->la_gid;
1396                 osd_push_ctxt(env, save);
1397                 rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1398                 osd_pop_ctxt(save);
1399                 if (rc != 0)
1400                         return rc;
1401         }
1402 #endif
1403
1404         cfs_spin_lock(&obj->oo_guard);
1405         rc = osd_inode_setattr(env, inode, attr);
1406         cfs_spin_unlock(&obj->oo_guard);
1407
1408         if (!rc)
1409                 inode->i_sb->s_op->dirty_inode(inode);
1410         return rc;
1411 }
1412
1413 /*
1414  * Object creation.
1415  *
1416  * XXX temporary solution.
1417  */
1418 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1419                           struct lu_attr *attr, struct thandle *th)
1420 {
1421         return 0;
1422 }
1423
1424 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1425                            struct lu_attr *attr, struct thandle *th)
1426 {
1427         osd_object_init0(obj);
1428         if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1429                 unlock_new_inode(obj->oo_inode);
1430         return 0;
1431 }
1432
1433 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1434                                     struct osd_object *obj,
1435                                     const char *name, const int namelen)
1436 {
1437         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1438 }
1439
1440 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1441                       cfs_umode_t mode,
1442                       struct dt_allocation_hint *hint,
1443                       struct thandle *th)
1444 {
1445         int result;
1446         struct osd_device  *osd = osd_obj2dev(obj);
1447         struct osd_thandle *oth;
1448         struct dt_object   *parent = NULL;
1449         struct inode       *inode;
1450 #ifdef HAVE_QUOTA_SUPPORT
1451         struct osd_ctxt    *save = &info->oti_ctxt;
1452 #endif
1453
1454         LINVRNT(osd_invariant(obj));
1455         LASSERT(obj->oo_inode == NULL);
1456         LASSERT(obj->oo_hl_head == NULL);
1457
1458         if (S_ISDIR(mode) && ldiskfs_pdo) {
1459                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1460                 if (obj->oo_hl_head == NULL)
1461                         return -ENOMEM;
1462         }
1463
1464         oth = container_of(th, struct osd_thandle, ot_super);
1465         LASSERT(oth->ot_handle->h_transaction != NULL);
1466
1467         if (hint && hint->dah_parent)
1468                 parent = hint->dah_parent;
1469
1470 #ifdef HAVE_QUOTA_SUPPORT
1471         osd_push_ctxt(info->oti_env, save);
1472 #endif
1473         inode = ldiskfs_create_inode(oth->ot_handle,
1474                                      parent ? osd_dt_obj(parent)->oo_inode :
1475                                               osd_sb(osd)->s_root->d_inode,
1476                                      mode);
1477 #ifdef HAVE_QUOTA_SUPPORT
1478         osd_pop_ctxt(save);
1479 #endif
1480         if (!IS_ERR(inode)) {
1481                 /* Do not update file c/mtime in ldiskfs.
1482                  * NB: don't need any lock because no contention at this
1483                  * early stage */
1484                 inode->i_flags |= S_NOCMTIME;
1485                 obj->oo_inode = inode;
1486                 result = 0;
1487         } else {
1488                 if (obj->oo_hl_head != NULL) {
1489                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1490                         obj->oo_hl_head = NULL;
1491                 }
1492                 result = PTR_ERR(inode);
1493         }
1494         LINVRNT(osd_invariant(obj));
1495         return result;
1496 }
1497
1498 enum {
1499         OSD_NAME_LEN = 255
1500 };
1501
1502 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1503                      struct lu_attr *attr,
1504                      struct dt_allocation_hint *hint,
1505                      struct dt_object_format *dof,
1506                      struct thandle *th)
1507 {
1508         int result;
1509         struct osd_thandle *oth;
1510         struct osd_device *osd = osd_obj2dev(obj);
1511         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1512
1513         LASSERT(S_ISDIR(attr->la_mode));
1514
1515         oth = container_of(th, struct osd_thandle, ot_super);
1516         LASSERT(oth->ot_handle->h_transaction != NULL);
1517         result = osd_mkfile(info, obj, mode, hint, th);
1518         if (result == 0 && osd->od_iop_mode == 0) {
1519                 LASSERT(obj->oo_inode != NULL);
1520                 /*
1521                  * XXX uh-oh... call low-level iam function directly.
1522                  */
1523
1524                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1525                                          sizeof (struct osd_fid_pack),
1526                                          oth->ot_handle);
1527         }
1528         return result;
1529 }
1530
1531 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1532                         struct lu_attr *attr,
1533                         struct dt_allocation_hint *hint,
1534                         struct dt_object_format *dof,
1535                         struct thandle *th)
1536 {
1537         int result;
1538         struct osd_thandle *oth;
1539         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1540
1541         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1542
1543         LASSERT(S_ISREG(attr->la_mode));
1544
1545         oth = container_of(th, struct osd_thandle, ot_super);
1546         LASSERT(oth->ot_handle->h_transaction != NULL);
1547
1548         result = osd_mkfile(info, obj, mode, hint, th);
1549         if (result == 0) {
1550                 LASSERT(obj->oo_inode != NULL);
1551                 if (feat->dif_flags & DT_IND_VARKEY)
1552                         result = iam_lvar_create(obj->oo_inode,
1553                                                  feat->dif_keysize_max,
1554                                                  feat->dif_ptrsize,
1555                                                  feat->dif_recsize_max,
1556                                                  oth->ot_handle);
1557                 else
1558                         result = iam_lfix_create(obj->oo_inode,
1559                                                  feat->dif_keysize_max,
1560                                                  feat->dif_ptrsize,
1561                                                  feat->dif_recsize_max,
1562                                                  oth->ot_handle);
1563
1564         }
1565         return result;
1566 }
1567
1568 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1569                      struct lu_attr *attr,
1570                      struct dt_allocation_hint *hint,
1571                      struct dt_object_format *dof,
1572                      struct thandle *th)
1573 {
1574         LASSERT(S_ISREG(attr->la_mode));
1575         return osd_mkfile(info, obj, (attr->la_mode &
1576                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1577 }
1578
1579 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1580                      struct lu_attr *attr,
1581                      struct dt_allocation_hint *hint,
1582                      struct dt_object_format *dof,
1583                      struct thandle *th)
1584 {
1585         LASSERT(S_ISLNK(attr->la_mode));
1586         return osd_mkfile(info, obj, (attr->la_mode &
1587                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1588 }
1589
1590 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1591                      struct lu_attr *attr,
1592                      struct dt_allocation_hint *hint,
1593                      struct dt_object_format *dof,
1594                      struct thandle *th)
1595 {
1596         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1597         int result;
1598
1599         LINVRNT(osd_invariant(obj));
1600         LASSERT(obj->oo_inode == NULL);
1601         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1602                 S_ISFIFO(mode) || S_ISSOCK(mode));
1603
1604         result = osd_mkfile(info, obj, mode, hint, th);
1605         if (result == 0) {
1606                 LASSERT(obj->oo_inode != NULL);
1607                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1608         }
1609         LINVRNT(osd_invariant(obj));
1610         return result;
1611 }
1612
1613 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1614                               struct lu_attr *,
1615                               struct dt_allocation_hint *hint,
1616                               struct dt_object_format *dof,
1617                               struct thandle *);
1618
1619 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1620 {
1621         osd_obj_type_f result;
1622
1623         switch (type) {
1624         case DFT_DIR:
1625                 result = osd_mkdir;
1626                 break;
1627         case DFT_REGULAR:
1628                 result = osd_mkreg;
1629                 break;
1630         case DFT_SYM:
1631                 result = osd_mksym;
1632                 break;
1633         case DFT_NODE:
1634                 result = osd_mknod;
1635                 break;
1636         case DFT_INDEX:
1637                 result = osd_mk_index;
1638                 break;
1639
1640         default:
1641                 LBUG();
1642                 break;
1643         }
1644         return result;
1645 }
1646
1647
1648 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1649                         struct dt_object *parent, cfs_umode_t child_mode)
1650 {
1651         LASSERT(ah);
1652
1653         memset(ah, 0, sizeof(*ah));
1654         ah->dah_parent = parent;
1655         ah->dah_mode = child_mode;
1656 }
1657
1658 /**
1659  * Helper function for osd_object_create()
1660  *
1661  * \retval 0, on success
1662  */
1663 static int __osd_object_create(struct osd_thread_info *info,
1664                                struct osd_object *obj, struct lu_attr *attr,
1665                                struct dt_allocation_hint *hint,
1666                                struct dt_object_format *dof,
1667                                struct thandle *th)
1668 {
1669
1670         int result;
1671
1672         result = osd_create_pre(info, obj, attr, th);
1673         if (result == 0) {
1674                 result = osd_create_type_f(dof->dof_type)(info, obj,
1675                                            attr, hint, dof, th);
1676                 if (result == 0)
1677                         result = osd_create_post(info, obj, attr, th);
1678         }
1679         return result;
1680 }
1681
1682 /**
1683  * Helper function for osd_object_create()
1684  *
1685  * \retval 0, on success
1686  */
1687 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1688                            const struct lu_fid *fid, struct thandle *th)
1689 {
1690         struct osd_thread_info *info = osd_oti_get(env);
1691         struct osd_inode_id    *id   = &info->oti_id;
1692         struct osd_device      *osd  = osd_obj2dev(obj);
1693         struct md_ucred        *uc   = md_ucred(env);
1694
1695         LASSERT(obj->oo_inode != NULL);
1696         LASSERT(uc != NULL);
1697
1698         id->oii_ino = obj->oo_inode->i_ino;
1699         id->oii_gen = obj->oo_inode->i_generation;
1700
1701         return osd_oi_insert(info, osd, fid, id, th,
1702                              uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
1703 }
1704
1705 static int osd_declare_object_create(const struct lu_env *env,
1706                                      struct dt_object *dt,
1707                                      struct lu_attr *attr,
1708                                      struct dt_allocation_hint *hint,
1709                                      struct dt_object_format *dof,
1710                                      struct thandle *handle)
1711 {
1712         struct osd_thandle *oh;
1713
1714         LASSERT(handle != NULL);
1715
1716         oh = container_of0(handle, struct osd_thandle, ot_super);
1717         LASSERT(oh->ot_handle == NULL);
1718
1719         OSD_DECLARE_OP(oh, create);
1720         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1721         /* XXX: So far, only normal fid needs be inserted into the oi,
1722          *      things could be changed later. Revise following code then. */
1723         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1724                 OSD_DECLARE_OP(oh, insert);
1725                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1726         }
1727         /* If this is directory, then we expect . and .. to be inserted as
1728          * well. The one directory block always needs to be created for the
1729          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1730          * block), there is no danger of needing a tree for the first block.
1731          */
1732         if (attr && S_ISDIR(attr->la_mode)) {
1733                 OSD_DECLARE_OP(oh, insert);
1734                 OSD_DECLARE_OP(oh, insert);
1735                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1736         }
1737
1738         if (attr) {
1739                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1740                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1741         }
1742         return 0;
1743 }
1744
1745 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1746                              struct lu_attr *attr,
1747                              struct dt_allocation_hint *hint,
1748                              struct dt_object_format *dof,
1749                              struct thandle *th)
1750 {
1751         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1752         struct osd_object      *obj    = osd_dt_obj(dt);
1753         struct osd_thread_info *info   = osd_oti_get(env);
1754         int result;
1755
1756         ENTRY;
1757
1758         LINVRNT(osd_invariant(obj));
1759         LASSERT(!dt_object_exists(dt));
1760         LASSERT(osd_write_locked(env, obj));
1761         LASSERT(th != NULL);
1762
1763         OSD_EXEC_OP(th, create);
1764
1765         result = __osd_object_create(info, obj, attr, hint, dof, th);
1766         if (result == 0)
1767                 result = __osd_oi_insert(env, obj, fid, th);
1768
1769         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1770         LASSERT(osd_invariant(obj));
1771         RETURN(result);
1772 }
1773
1774 /**
1775  * Called to destroy on-disk representation of the object
1776  *
1777  * Concurrency: must be locked
1778  */
1779 static int osd_declare_object_destroy(const struct lu_env *env,
1780                                       struct dt_object *dt,
1781                                       struct thandle *th)
1782 {
1783         struct osd_object  *obj = osd_dt_obj(dt);
1784         struct inode       *inode = obj->oo_inode;
1785         struct osd_thandle *oh;
1786
1787         ENTRY;
1788
1789         oh = container_of0(th, struct osd_thandle, ot_super);
1790         LASSERT(oh->ot_handle == NULL);
1791         LASSERT(inode);
1792
1793         OSD_DECLARE_OP(oh, destroy);
1794         OSD_DECLARE_OP(oh, delete);
1795         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
1796         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
1797
1798         osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
1799         osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
1800
1801         RETURN(0);
1802 }
1803
1804 static int osd_object_destroy(const struct lu_env *env,
1805                               struct dt_object *dt,
1806                               struct thandle *th)
1807 {
1808         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
1809         struct osd_object      *obj = osd_dt_obj(dt);
1810         struct inode           *inode = obj->oo_inode;
1811         struct osd_device      *osd = osd_obj2dev(obj);
1812         struct osd_thandle     *oh;
1813         int                     result;
1814         ENTRY;
1815
1816         oh = container_of0(th, struct osd_thandle, ot_super);
1817         LASSERT(oh->ot_handle);
1818         LASSERT(inode);
1819         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1820
1821         if (S_ISDIR(inode->i_mode)) {
1822                 LASSERT(osd_inode_unlinked(inode) ||
1823                         inode->i_nlink == 1);
1824                 cfs_spin_lock(&obj->oo_guard);
1825                 inode->i_nlink = 0;
1826                 cfs_spin_unlock(&obj->oo_guard);
1827                 inode->i_sb->s_op->dirty_inode(inode);
1828         } else {
1829                 LASSERT(osd_inode_unlinked(inode));
1830         }
1831
1832         OSD_EXEC_OP(th, destroy);
1833
1834         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
1835
1836         /* XXX: add to ext3 orphan list */
1837         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
1838
1839         /* not needed in the cache anymore */
1840         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
1841
1842         RETURN(0);
1843 }
1844
1845 /**
1846  * Helper function for osd_xattr_set()
1847  */
1848 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1849                            const struct lu_buf *buf, const char *name, int fl)
1850 {
1851         struct osd_object      *obj      = osd_dt_obj(dt);
1852         struct inode           *inode    = obj->oo_inode;
1853         struct osd_thread_info *info     = osd_oti_get(env);
1854         struct dentry          *dentry   = &info->oti_child_dentry;
1855         int                     fs_flags = 0;
1856         int                     rc;
1857
1858         LASSERT(dt_object_exists(dt));
1859         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1860         LASSERT(osd_write_locked(env, obj));
1861
1862         if (fl & LU_XATTR_REPLACE)
1863                 fs_flags |= XATTR_REPLACE;
1864
1865         if (fl & LU_XATTR_CREATE)
1866                 fs_flags |= XATTR_CREATE;
1867
1868         dentry->d_inode = inode;
1869         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
1870                                    buf->lb_len, fs_flags);
1871         return rc;
1872 }
1873
1874 /**
1875  * Put the fid into lustre_mdt_attrs, and then place the structure
1876  * inode's ea. This fid should not be altered during the life time
1877  * of the inode.
1878  *
1879  * \retval +ve, on success
1880  * \retval -ve, on error
1881  *
1882  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
1883  */
1884 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
1885                           const struct lu_fid *fid)
1886 {
1887         struct osd_thread_info  *info      = osd_oti_get(env);
1888         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1889
1890         lustre_lma_init(mdt_attrs, fid);
1891         lustre_lma_swab(mdt_attrs);
1892         return __osd_xattr_set(env, dt,
1893                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
1894                                XATTR_NAME_LMA, LU_XATTR_CREATE);
1895
1896 }
1897
1898 /**
1899  * Helper function to form igif
1900  */
1901 static inline void osd_igif_get(const struct lu_env *env, struct inode  *inode,
1902                                 struct lu_fid *fid)
1903 {
1904         LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
1905 }
1906
1907 /**
1908  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
1909  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
1910  * To have compatilibility with 1.8 ldiskfs driver we need to have
1911  * magic number at start of fid data.
1912  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
1913  * its inmemory API.
1914  */
1915 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
1916                                   const struct dt_rec *fid)
1917 {
1918         param->edp_magic = LDISKFS_LUFID_MAGIC;
1919         param->edp_len =  sizeof(struct lu_fid) + 1;
1920
1921         fid_cpu_to_be((struct lu_fid *)param->edp_data,
1922                       (struct lu_fid *)fid);
1923 }
1924
1925 /**
1926  * Try to read the fid from inode ea into dt_rec, if return value
1927  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
1928  *
1929  * \param fid object fid.
1930  *
1931  * \retval 0 on success
1932  */
1933 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
1934                           __u32 ino, struct lu_fid *fid)
1935 {
1936         struct osd_thread_info  *info      = osd_oti_get(env);
1937         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1938         struct lu_device        *ldev   = obj->oo_dt.do_lu.lo_dev;
1939         struct dentry           *dentry = &info->oti_child_dentry;
1940         struct osd_inode_id     *id     = &info->oti_id;
1941         struct osd_device       *dev;
1942         struct inode            *inode;
1943         int                      rc;
1944
1945         ENTRY;
1946         dev  = osd_dev(ldev);
1947
1948         id->oii_ino = ino;
1949         id->oii_gen = OSD_OII_NOGEN;
1950
1951         inode = osd_iget(info, dev, id);
1952         if (IS_ERR(inode)) {
1953                 rc = PTR_ERR(inode);
1954                 GOTO(out,rc);
1955         }
1956         dentry->d_inode = inode;
1957
1958         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1959         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
1960                                    sizeof *mdt_attrs);
1961
1962         /* Check LMA compatibility */
1963         if (rc > 0 &&
1964             (mdt_attrs->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP))) {
1965                 CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
1966                       inode->i_ino, le32_to_cpu(mdt_attrs->lma_incompat) &
1967                       ~LMA_INCOMPAT_SUPP);
1968                 return -ENOSYS;
1969         }
1970
1971         if (rc > 0) {
1972                 lustre_lma_swab(mdt_attrs);
1973                 memcpy(fid, &mdt_attrs->lma_self_fid, sizeof(*fid));
1974                 rc = 0;
1975         } else if (rc == -ENODATA) {
1976                 osd_igif_get(env, inode, fid);
1977                 rc = 0;
1978         }
1979         iput(inode);
1980 out:
1981         RETURN(rc);
1982 }
1983
1984 /**
1985  * OSD layer object create function for interoperability mode (b11826).
1986  * This is mostly similar to osd_object_create(). Only difference being, fid is
1987  * inserted into inode ea here.
1988  *
1989  * \retval   0, on success
1990  * \retval -ve, on error
1991  */
1992 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
1993                                 struct lu_attr *attr,
1994                                 struct dt_allocation_hint *hint,
1995                                 struct dt_object_format *dof,
1996                                 struct thandle *th)
1997 {
1998         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1999         struct osd_object      *obj    = osd_dt_obj(dt);
2000         struct osd_thread_info *info   = osd_oti_get(env);
2001         int                     result;
2002
2003         ENTRY;
2004
2005         LASSERT(osd_invariant(obj));
2006         LASSERT(!dt_object_exists(dt));
2007         LASSERT(osd_write_locked(env, obj));
2008         LASSERT(th != NULL);
2009
2010         OSD_EXEC_OP(th, create);
2011
2012         result = __osd_object_create(info, obj, attr, hint, dof, th);
2013         /* objects under osd root shld have igif fid, so dont add fid EA */
2014         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2015                 result = osd_ea_fid_set(env, dt, fid);
2016
2017         if (result == 0)
2018                 result = __osd_oi_insert(env, obj, fid, th);
2019
2020         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2021         LINVRNT(osd_invariant(obj));
2022         RETURN(result);
2023 }
2024
2025 static int osd_declare_object_ref_add(const struct lu_env *env,
2026                                       struct dt_object *dt,
2027                                       struct thandle *handle)
2028 {
2029         struct osd_thandle *oh;
2030
2031         /* it's possible that object doesn't exist yet */
2032         LASSERT(handle != NULL);
2033
2034         oh = container_of0(handle, struct osd_thandle, ot_super);
2035         LASSERT(oh->ot_handle == NULL);
2036
2037         OSD_DECLARE_OP(oh, ref_add);
2038         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2039
2040         return 0;
2041 }
2042
2043 /*
2044  * Concurrency: @dt is write locked.
2045  */
2046 static int osd_object_ref_add(const struct lu_env *env,
2047                               struct dt_object *dt, struct thandle *th)
2048 {
2049         struct osd_object *obj = osd_dt_obj(dt);
2050         struct inode      *inode = obj->oo_inode;
2051
2052         LINVRNT(osd_invariant(obj));
2053         LASSERT(dt_object_exists(dt));
2054         LASSERT(osd_write_locked(env, obj));
2055         LASSERT(th != NULL);
2056
2057         OSD_EXEC_OP(th, ref_add);
2058
2059         /*
2060          * DIR_NLINK feature is set for compatibility reasons if:
2061          * 1) nlinks > LDISKFS_LINK_MAX, or
2062          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2063          *
2064          * It is easier to always set this flag (rather than check and set),
2065          * since it has less overhead, and the superblock will be dirtied
2066          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2067          * do not actually care whether this flag is set or not.
2068          */
2069         cfs_spin_lock(&obj->oo_guard);
2070         inode->i_nlink++;
2071         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2072                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2073                     inode->i_nlink == 2)
2074                         inode->i_nlink = 1;
2075         }
2076         LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
2077         cfs_spin_unlock(&obj->oo_guard);
2078         inode->i_sb->s_op->dirty_inode(inode);
2079         LINVRNT(osd_invariant(obj));
2080
2081         return 0;
2082 }
2083
2084 static int osd_declare_object_ref_del(const struct lu_env *env,
2085                                       struct dt_object *dt,
2086                                       struct thandle *handle)
2087 {
2088         struct osd_thandle *oh;
2089
2090         LASSERT(dt_object_exists(dt));
2091         LASSERT(handle != NULL);
2092
2093         oh = container_of0(handle, struct osd_thandle, ot_super);
2094         LASSERT(oh->ot_handle == NULL);
2095
2096         OSD_DECLARE_OP(oh, ref_del);
2097         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2098
2099         return 0;
2100 }
2101
2102 /*
2103  * Concurrency: @dt is write locked.
2104  */
2105 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2106                               struct thandle *th)
2107 {
2108         struct osd_object *obj = osd_dt_obj(dt);
2109         struct inode      *inode = obj->oo_inode;
2110
2111         LINVRNT(osd_invariant(obj));
2112         LASSERT(dt_object_exists(dt));
2113         LASSERT(osd_write_locked(env, obj));
2114         LASSERT(th != NULL);
2115
2116         OSD_EXEC_OP(th, ref_del);
2117
2118         cfs_spin_lock(&obj->oo_guard);
2119         LASSERT(inode->i_nlink > 0);
2120         inode->i_nlink--;
2121         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2122          * then the nlink count is 1. Don't let it be set to 0 or the directory
2123          * inode will be deleted incorrectly. */
2124         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2125                 inode->i_nlink++;
2126         cfs_spin_unlock(&obj->oo_guard);
2127         inode->i_sb->s_op->dirty_inode(inode);
2128         LINVRNT(osd_invariant(obj));
2129
2130         return 0;
2131 }
2132
2133 /*
2134  * Get the 64-bit version for an inode.
2135  */
2136 static int osd_object_version_get(const struct lu_env *env,
2137                                   struct dt_object *dt, dt_obj_version_t *ver)
2138 {
2139         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2140
2141         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2142                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2143         *ver = LDISKFS_I(inode)->i_fs_version;
2144         return 0;
2145 }
2146
2147 /*
2148  * Concurrency: @dt is read locked.
2149  */
2150 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2151                          struct lu_buf *buf, const char *name,
2152                          struct lustre_capa *capa)
2153 {
2154         struct osd_object      *obj    = osd_dt_obj(dt);
2155         struct inode           *inode  = obj->oo_inode;
2156         struct osd_thread_info *info   = osd_oti_get(env);
2157         struct dentry          *dentry = &info->oti_obj_dentry;
2158
2159         /* version get is not real XATTR but uses xattr API */
2160         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2161                 /* for version we are just using xattr API but change inode
2162                  * field instead */
2163                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2164                 osd_object_version_get(env, dt, buf->lb_buf);
2165                 return sizeof(dt_obj_version_t);
2166         }
2167
2168         LASSERT(dt_object_exists(dt));
2169         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2170         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2171
2172         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2173                 return -EACCES;
2174
2175         dentry->d_inode = inode;
2176         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2177 }
2178
2179
2180 static int osd_declare_xattr_set(const struct lu_env *env,
2181                                  struct dt_object *dt,
2182                                  const struct lu_buf *buf, const char *name,
2183                                  int fl, struct thandle *handle)
2184 {
2185         struct osd_thandle *oh;
2186
2187         LASSERT(handle != NULL);
2188
2189         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2190                 /* no credits for version */
2191                 return 0;
2192         }
2193
2194         oh = container_of0(handle, struct osd_thandle, ot_super);
2195         LASSERT(oh->ot_handle == NULL);
2196
2197         OSD_DECLARE_OP(oh, xattr_set);
2198         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2199
2200         return 0;
2201 }
2202
2203 /*
2204  * Set the 64-bit version for object
2205  */
2206 static void osd_object_version_set(const struct lu_env *env,
2207                                    struct dt_object *dt,
2208                                    dt_obj_version_t *new_version)
2209 {
2210         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2211
2212         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2213                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2214
2215         LDISKFS_I(inode)->i_fs_version = *new_version;
2216         /** Version is set after all inode operations are finished,
2217          *  so we should mark it dirty here */
2218         inode->i_sb->s_op->dirty_inode(inode);
2219 }
2220
2221 /*
2222  * Concurrency: @dt is write locked.
2223  */
2224 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2225                          const struct lu_buf *buf, const char *name, int fl,
2226                          struct thandle *handle, struct lustre_capa *capa)
2227 {
2228         LASSERT(handle != NULL);
2229
2230         /* version set is not real XATTR */
2231         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2232                 /* for version we are just using xattr API but change inode
2233                  * field instead */
2234                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2235                 osd_object_version_set(env, dt, buf->lb_buf);
2236                 return sizeof(dt_obj_version_t);
2237         }
2238
2239         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2240                 return -EACCES;
2241
2242         OSD_EXEC_OP(handle, xattr_set);
2243         return __osd_xattr_set(env, dt, buf, name, fl);
2244 }
2245
2246 /*
2247  * Concurrency: @dt is read locked.
2248  */
2249 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2250                           struct lu_buf *buf, struct lustre_capa *capa)
2251 {
2252         struct osd_object      *obj    = osd_dt_obj(dt);
2253         struct inode           *inode  = obj->oo_inode;
2254         struct osd_thread_info *info   = osd_oti_get(env);
2255         struct dentry          *dentry = &info->oti_obj_dentry;
2256
2257         LASSERT(dt_object_exists(dt));
2258         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2259         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2260
2261         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2262                 return -EACCES;
2263
2264         dentry->d_inode = inode;
2265         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2266 }
2267
2268 static int osd_declare_xattr_del(const struct lu_env *env,
2269                                  struct dt_object *dt, const char *name,
2270                                  struct thandle *handle)
2271 {
2272         struct osd_thandle *oh;
2273
2274         LASSERT(dt_object_exists(dt));
2275         LASSERT(handle != NULL);
2276
2277         oh = container_of0(handle, struct osd_thandle, ot_super);
2278         LASSERT(oh->ot_handle == NULL);
2279
2280         OSD_DECLARE_OP(oh, xattr_set);
2281         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2282
2283         return 0;
2284 }
2285
2286 /*
2287  * Concurrency: @dt is write locked.
2288  */
2289 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2290                          const char *name, struct thandle *handle,
2291                          struct lustre_capa *capa)
2292 {
2293         struct osd_object      *obj    = osd_dt_obj(dt);
2294         struct inode           *inode  = obj->oo_inode;
2295         struct osd_thread_info *info   = osd_oti_get(env);
2296         struct dentry          *dentry = &info->oti_obj_dentry;
2297         int                     rc;
2298
2299         LASSERT(dt_object_exists(dt));
2300         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2301         LASSERT(osd_write_locked(env, obj));
2302         LASSERT(handle != NULL);
2303
2304         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2305                 return -EACCES;
2306
2307         OSD_EXEC_OP(handle, xattr_set);
2308
2309         dentry->d_inode = inode;
2310         rc = inode->i_op->removexattr(dentry, name);
2311         return rc;
2312 }
2313
2314 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2315                                      struct dt_object *dt,
2316                                      struct lustre_capa *old,
2317                                      __u64 opc)
2318 {
2319         struct osd_thread_info *info = osd_oti_get(env);
2320         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2321         struct osd_object *obj = osd_dt_obj(dt);
2322         struct osd_device *dev = osd_obj2dev(obj);
2323         struct lustre_capa_key *key = &info->oti_capa_key;
2324         struct lustre_capa *capa = &info->oti_capa;
2325         struct obd_capa *oc;
2326         struct md_capainfo *ci;
2327         int rc;
2328         ENTRY;
2329
2330         if (!dev->od_fl_capa)
2331                 RETURN(ERR_PTR(-ENOENT));
2332
2333         LASSERT(dt_object_exists(dt));
2334         LINVRNT(osd_invariant(obj));
2335
2336         /* renewal sanity check */
2337         if (old && osd_object_auth(env, dt, old, opc))
2338                 RETURN(ERR_PTR(-EACCES));
2339
2340         ci = md_capainfo(env);
2341         if (unlikely(!ci))
2342                 RETURN(ERR_PTR(-ENOENT));
2343
2344         switch (ci->mc_auth) {
2345         case LC_ID_NONE:
2346                 RETURN(NULL);
2347         case LC_ID_PLAIN:
2348                 capa->lc_uid = obj->oo_inode->i_uid;
2349                 capa->lc_gid = obj->oo_inode->i_gid;
2350                 capa->lc_flags = LC_ID_PLAIN;
2351                 break;
2352         case LC_ID_CONVERT: {
2353                 __u32 d[4], s[4];
2354
2355                 s[0] = obj->oo_inode->i_uid;
2356                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2357                 s[2] = obj->oo_inode->i_gid;
2358                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2359                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2360                 if (unlikely(rc))
2361                         RETURN(ERR_PTR(rc));
2362
2363                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2364                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2365                 capa->lc_flags = LC_ID_CONVERT;
2366                 break;
2367         }
2368         default:
2369                 RETURN(ERR_PTR(-EINVAL));
2370         }
2371
2372         capa->lc_fid = *fid;
2373         capa->lc_opc = opc;
2374         capa->lc_flags |= dev->od_capa_alg << 24;
2375         capa->lc_timeout = dev->od_capa_timeout;
2376         capa->lc_expiry = 0;
2377
2378         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2379         if (oc) {
2380                 LASSERT(!capa_is_expired(oc));
2381                 RETURN(oc);
2382         }
2383
2384         cfs_spin_lock(&capa_lock);
2385         *key = dev->od_capa_keys[1];
2386         cfs_spin_unlock(&capa_lock);
2387
2388         capa->lc_keyid = key->lk_keyid;
2389         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2390
2391         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2392         if (rc) {
2393                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2394                 RETURN(ERR_PTR(rc));
2395         }
2396
2397         oc = capa_add(dev->od_capa_hash, capa);
2398         RETURN(oc);
2399 }
2400
2401 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2402 {
2403         struct osd_object      *obj    = osd_dt_obj(dt);
2404         struct inode           *inode  = obj->oo_inode;
2405         struct osd_thread_info *info   = osd_oti_get(env);
2406         struct dentry          *dentry = &info->oti_obj_dentry;
2407         struct file            *file   = &info->oti_file;
2408         int                     rc;
2409
2410         ENTRY;
2411
2412         dentry->d_inode = inode;
2413         file->f_dentry = dentry;
2414         file->f_mapping = inode->i_mapping;
2415         file->f_op = inode->i_fop;
2416         LOCK_INODE_MUTEX(inode);
2417         rc = file->f_op->fsync(file, dentry, 0);
2418         UNLOCK_INODE_MUTEX(inode);
2419         RETURN(rc);
2420 }
2421
2422 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2423                         void **data)
2424 {
2425         struct osd_object *obj = osd_dt_obj(dt);
2426         ENTRY;
2427
2428         *data = (void *)obj->oo_inode;
2429         RETURN(0);
2430 }
2431
2432 /*
2433  * Index operations.
2434  */
2435
2436 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2437                            const struct dt_index_features *feat)
2438 {
2439         struct iam_descr *descr;
2440
2441         if (osd_object_is_root(o))
2442                 return feat == &dt_directory_features;
2443
2444         LASSERT(o->oo_dir != NULL);
2445
2446         descr = o->oo_dir->od_container.ic_descr;
2447         if (feat == &dt_directory_features) {
2448                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2449                         return 1;
2450                 else
2451                         return 0;
2452         } else {
2453                 return
2454                         feat->dif_keysize_min <= descr->id_key_size &&
2455                         descr->id_key_size <= feat->dif_keysize_max &&
2456                         feat->dif_recsize_min <= descr->id_rec_size &&
2457                         descr->id_rec_size <= feat->dif_recsize_max &&
2458                         !(feat->dif_flags & (DT_IND_VARKEY |
2459                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2460                         ergo(feat->dif_flags & DT_IND_UPDATE,
2461                              1 /* XXX check that object (and file system) is
2462                                 * writable */);
2463         }
2464 }
2465
2466 static int osd_iam_container_init(const struct lu_env *env,
2467                                   struct osd_object *obj,
2468                                   struct osd_directory *dir)
2469 {
2470         struct iam_container *bag = &dir->od_container;
2471         int result;
2472
2473         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2474         if (result != 0)
2475                 return result;
2476
2477         result = iam_container_setup(bag);
2478         if (result != 0)
2479                 goto out;
2480
2481         if (osd_obj2dev(obj)->od_iop_mode) {
2482                 u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag);
2483
2484                 bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode,
2485                                                 ptr, 0, &result);
2486         }
2487
2488  out:
2489         if (result == 0)
2490                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2491         else
2492                 iam_container_fini(bag);
2493
2494         return result;
2495 }
2496
2497
2498 /*
2499  * Concurrency: no external locking is necessary.
2500  */
2501 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2502                          const struct dt_index_features *feat)
2503 {
2504         int result;
2505         int ea_dir = 0;
2506         struct osd_object *obj = osd_dt_obj(dt);
2507         struct osd_device *osd = osd_obj2dev(obj);
2508
2509         LINVRNT(osd_invariant(obj));
2510         LASSERT(dt_object_exists(dt));
2511
2512         if (osd_object_is_root(obj)) {
2513                 dt->do_index_ops = &osd_index_ea_ops;
2514                 result = 0;
2515         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2516                 dt->do_index_ops = &osd_index_ea_ops;
2517                 if (S_ISDIR(obj->oo_inode->i_mode))
2518                         result = 0;
2519                 else
2520                         result = -ENOTDIR;
2521                 ea_dir = 1;
2522         } else if (!osd_has_index(obj)) {
2523                 struct osd_directory *dir;
2524
2525                 OBD_ALLOC_PTR(dir);
2526                 if (dir != NULL) {
2527
2528                         cfs_spin_lock(&obj->oo_guard);
2529                         if (obj->oo_dir == NULL)
2530                                 obj->oo_dir = dir;
2531                         else
2532                                 /*
2533                                  * Concurrent thread allocated container data.
2534                                  */
2535                                 OBD_FREE_PTR(dir);
2536                         cfs_spin_unlock(&obj->oo_guard);
2537                         /*
2538                          * Now, that we have container data, serialize its
2539                          * initialization.
2540                          */
2541                         cfs_down_write(&obj->oo_ext_idx_sem);
2542                         /*
2543                          * recheck under lock.
2544                          */
2545                         if (!osd_has_index(obj))
2546                                 result = osd_iam_container_init(env, obj, dir);
2547                         else
2548                                 result = 0;
2549                         cfs_up_write(&obj->oo_ext_idx_sem);
2550                 } else {
2551                         result = -ENOMEM;
2552                 }
2553         } else {
2554                 result = 0;
2555         }
2556
2557         if (result == 0 && ea_dir == 0) {
2558                 if (!osd_iam_index_probe(env, obj, feat))
2559                         result = -ENOTDIR;
2560         }
2561         LINVRNT(osd_invariant(obj));
2562
2563         return result;
2564 }
2565
2566 static const struct dt_object_operations osd_obj_ops = {
2567         .do_read_lock         = osd_object_read_lock,
2568         .do_write_lock        = osd_object_write_lock,
2569         .do_read_unlock       = osd_object_read_unlock,
2570         .do_write_unlock      = osd_object_write_unlock,
2571         .do_write_locked      = osd_object_write_locked,
2572         .do_attr_get          = osd_attr_get,
2573         .do_declare_attr_set  = osd_declare_attr_set,
2574         .do_attr_set          = osd_attr_set,
2575         .do_ah_init           = osd_ah_init,
2576         .do_declare_create    = osd_declare_object_create,
2577         .do_create            = osd_object_create,
2578         .do_declare_destroy   = osd_declare_object_destroy,
2579         .do_destroy           = osd_object_destroy,
2580         .do_index_try         = osd_index_try,
2581         .do_declare_ref_add   = osd_declare_object_ref_add,
2582         .do_ref_add           = osd_object_ref_add,
2583         .do_declare_ref_del   = osd_declare_object_ref_del,
2584         .do_ref_del           = osd_object_ref_del,
2585         .do_xattr_get         = osd_xattr_get,
2586         .do_declare_xattr_set = osd_declare_xattr_set,
2587         .do_xattr_set         = osd_xattr_set,
2588         .do_declare_xattr_del = osd_declare_xattr_del,
2589         .do_xattr_del         = osd_xattr_del,
2590         .do_xattr_list        = osd_xattr_list,
2591         .do_capa_get          = osd_capa_get,
2592         .do_object_sync       = osd_object_sync,
2593         .do_data_get          = osd_data_get,
2594 };
2595
2596 /**
2597  * dt_object_operations for interoperability mode
2598  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2599  */
2600 static const struct dt_object_operations osd_obj_ea_ops = {
2601         .do_read_lock         = osd_object_read_lock,
2602         .do_write_lock        = osd_object_write_lock,
2603         .do_read_unlock       = osd_object_read_unlock,
2604         .do_write_unlock      = osd_object_write_unlock,
2605         .do_write_locked      = osd_object_write_locked,
2606         .do_attr_get          = osd_attr_get,
2607         .do_declare_attr_set  = osd_declare_attr_set,
2608         .do_attr_set          = osd_attr_set,
2609         .do_ah_init           = osd_ah_init,
2610         .do_declare_create    = osd_declare_object_create,
2611         .do_create            = osd_object_ea_create,
2612         .do_declare_destroy   = osd_declare_object_destroy,
2613         .do_destroy           = osd_object_destroy,
2614         .do_index_try         = osd_index_try,
2615         .do_declare_ref_add   = osd_declare_object_ref_add,
2616         .do_ref_add           = osd_object_ref_add,
2617         .do_declare_ref_del   = osd_declare_object_ref_del,
2618         .do_ref_del           = osd_object_ref_del,
2619         .do_xattr_get         = osd_xattr_get,
2620         .do_declare_xattr_set = osd_declare_xattr_set,
2621         .do_xattr_set         = osd_xattr_set,
2622         .do_declare_xattr_del = osd_declare_xattr_del,
2623         .do_xattr_del         = osd_xattr_del,
2624         .do_xattr_list        = osd_xattr_list,
2625         .do_capa_get          = osd_capa_get,
2626         .do_object_sync       = osd_object_sync,
2627         .do_data_get          = osd_data_get,
2628 };
2629
2630 static int osd_index_declare_iam_delete(const struct lu_env *env,
2631                                         struct dt_object *dt,
2632                                         const struct dt_key *key,
2633                                         struct thandle *handle)
2634 {
2635         struct osd_thandle    *oh;
2636
2637         oh = container_of0(handle, struct osd_thandle, ot_super);
2638         LASSERT(oh->ot_handle == NULL);
2639
2640         OSD_DECLARE_OP(oh, delete);
2641         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2642
2643         return 0;
2644 }
2645
2646 /**
2647  *      delete a (key, value) pair from index \a dt specified by \a key
2648  *
2649  *      \param  dt      osd index object
2650  *      \param  key     key for index
2651  *      \param  rec     record reference
2652  *      \param  handle  transaction handler
2653  *
2654  *      \retval  0  success
2655  *      \retval -ve   failure
2656  */
2657
2658 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2659                                 const struct dt_key *key,
2660                                 struct thandle *handle,
2661                                 struct lustre_capa *capa)
2662 {
2663         struct osd_object     *obj = osd_dt_obj(dt);
2664         struct osd_thandle    *oh;
2665         struct iam_path_descr *ipd;
2666         struct iam_container  *bag = &obj->oo_dir->od_container;
2667         int                    rc;
2668
2669         ENTRY;
2670
2671         LINVRNT(osd_invariant(obj));
2672         LASSERT(dt_object_exists(dt));
2673         LASSERT(bag->ic_object == obj->oo_inode);
2674         LASSERT(handle != NULL);
2675
2676         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2677                 RETURN(-EACCES);
2678
2679         OSD_EXEC_OP(handle, delete);
2680
2681         ipd = osd_idx_ipd_get(env, bag);
2682         if (unlikely(ipd == NULL))
2683                 RETURN(-ENOMEM);
2684
2685         oh = container_of0(handle, struct osd_thandle, ot_super);
2686         LASSERT(oh->ot_handle != NULL);
2687         LASSERT(oh->ot_handle->h_transaction != NULL);
2688
2689         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2690         osd_ipd_put(env, bag, ipd);
2691         LINVRNT(osd_invariant(obj));
2692         RETURN(rc);
2693 }
2694
2695 static int osd_index_declare_ea_delete(const struct lu_env *env,
2696                                        struct dt_object *dt,
2697                                        const struct dt_key *key,
2698                                        struct thandle *handle)
2699 {
2700         struct osd_thandle *oh;
2701
2702         LASSERT(dt_object_exists(dt));
2703         LASSERT(handle != NULL);
2704
2705         oh = container_of0(handle, struct osd_thandle, ot_super);
2706         LASSERT(oh->ot_handle == NULL);
2707
2708         OSD_DECLARE_OP(oh, delete);
2709         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2710
2711         LASSERT(osd_dt_obj(dt)->oo_inode);
2712         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
2713                         osd_dt_obj(dt)->oo_inode);
2714         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
2715                         osd_dt_obj(dt)->oo_inode);
2716
2717         return 0;
2718 }
2719
2720 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2721                                           struct dt_rec *fid)
2722 {
2723         struct osd_fid_pack *rec;
2724         int                  rc = -ENODATA;
2725
2726         if (de->file_type & LDISKFS_DIRENT_LUFID) {
2727                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2728                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2729         }
2730         RETURN(rc);
2731 }
2732
2733 /**
2734  * Index delete function for interoperability mode (b11826).
2735  * It will remove the directory entry added by osd_index_ea_insert().
2736  * This entry is needed to maintain name->fid mapping.
2737  *
2738  * \param key,  key i.e. file entry to be deleted
2739  *
2740  * \retval   0, on success
2741  * \retval -ve, on error
2742  */
2743 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2744                                const struct dt_key *key,
2745                                struct thandle *handle,
2746                                struct lustre_capa *capa)
2747 {
2748         struct osd_object          *obj    = osd_dt_obj(dt);
2749         struct inode               *dir    = obj->oo_inode;
2750         struct dentry              *dentry;
2751         struct osd_thandle         *oh;
2752         struct ldiskfs_dir_entry_2 *de;
2753         struct buffer_head         *bh;
2754         struct htree_lock          *hlock = NULL;
2755         int                         rc;
2756
2757         ENTRY;
2758
2759         LINVRNT(osd_invariant(obj));
2760         LASSERT(dt_object_exists(dt));
2761         LASSERT(handle != NULL);
2762
2763         OSD_EXEC_OP(handle, delete);
2764
2765         oh = container_of(handle, struct osd_thandle, ot_super);
2766         LASSERT(oh->ot_handle != NULL);
2767         LASSERT(oh->ot_handle->h_transaction != NULL);
2768
2769         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2770                 RETURN(-EACCES);
2771
2772         dentry = osd_child_dentry_get(env, obj,
2773                                       (char *)key, strlen((char *)key));
2774
2775         if (obj->oo_hl_head != NULL) {
2776                 hlock = osd_oti_get(env)->oti_hlock;
2777                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
2778                                    dir, LDISKFS_HLOCK_DEL);
2779         } else {
2780                 cfs_down_write(&obj->oo_ext_idx_sem);
2781         }
2782
2783         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
2784         if (bh) {
2785                 rc = ldiskfs_delete_entry(oh->ot_handle,
2786                                           dir, de, bh);
2787                 brelse(bh);
2788         } else {
2789                 rc = -ENOENT;
2790         }
2791         if (hlock != NULL)
2792                 ldiskfs_htree_unlock(hlock);
2793         else
2794                 cfs_up_write(&obj->oo_ext_idx_sem);
2795
2796         LASSERT(osd_invariant(obj));
2797         RETURN(rc);
2798 }
2799
2800 /**
2801  *      Lookup index for \a key and copy record to \a rec.
2802  *
2803  *      \param  dt      osd index object
2804  *      \param  key     key for index
2805  *      \param  rec     record reference
2806  *
2807  *      \retval  +ve  success : exact mach
2808  *      \retval  0    return record with key not greater than \a key
2809  *      \retval -ve   failure
2810  */
2811 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
2812                                 struct dt_rec *rec, const struct dt_key *key,
2813                                 struct lustre_capa *capa)
2814 {
2815         struct osd_object      *obj = osd_dt_obj(dt);
2816         struct iam_path_descr  *ipd;
2817         struct iam_container   *bag = &obj->oo_dir->od_container;
2818         struct osd_thread_info *oti = osd_oti_get(env);
2819         struct iam_iterator    *it = &oti->oti_idx_it;
2820         struct iam_rec         *iam_rec;
2821         int                     rc;
2822
2823         ENTRY;
2824
2825         LASSERT(osd_invariant(obj));
2826         LASSERT(dt_object_exists(dt));
2827         LASSERT(bag->ic_object == obj->oo_inode);
2828
2829         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2830                 RETURN(-EACCES);
2831
2832         ipd = osd_idx_ipd_get(env, bag);
2833         if (IS_ERR(ipd))
2834                 RETURN(-ENOMEM);
2835
2836         /* got ipd now we can start iterator. */
2837         iam_it_init(it, bag, 0, ipd);
2838
2839         rc = iam_it_get(it, (struct iam_key *)key);
2840         if (rc >= 0) {
2841                 if (S_ISDIR(obj->oo_inode->i_mode))
2842                         iam_rec = (struct iam_rec *)oti->oti_ldp;
2843                 else
2844                         iam_rec = (struct iam_rec *) rec;
2845
2846                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
2847                 if (S_ISDIR(obj->oo_inode->i_mode))
2848                         osd_fid_unpack((struct lu_fid *) rec,
2849                                        (struct osd_fid_pack *)iam_rec);
2850         }
2851         iam_it_put(it);
2852         iam_it_fini(it);
2853         osd_ipd_put(env, bag, ipd);
2854
2855         LINVRNT(osd_invariant(obj));
2856
2857         RETURN(rc);
2858 }
2859
2860 static int osd_index_declare_iam_insert(const struct lu_env *env,
2861                                         struct dt_object *dt,
2862                                         const struct dt_rec *rec,
2863                                         const struct dt_key *key,
2864                                         struct thandle *handle)
2865 {
2866         struct osd_thandle *oh;
2867
2868         LASSERT(dt_object_exists(dt));
2869         LASSERT(handle != NULL);
2870
2871         oh = container_of0(handle, struct osd_thandle, ot_super);
2872         LASSERT(oh->ot_handle == NULL);
2873
2874         OSD_DECLARE_OP(oh, insert);
2875         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
2876
2877         return 0;
2878 }
2879
2880 /**
2881  *      Inserts (key, value) pair in \a dt index object.
2882  *
2883  *      \param  dt      osd index object
2884  *      \param  key     key for index
2885  *      \param  rec     record reference
2886  *      \param  th      transaction handler
2887  *
2888  *      \retval  0  success
2889  *      \retval -ve failure
2890  */
2891 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
2892                                 const struct dt_rec *rec,
2893                                 const struct dt_key *key, struct thandle *th,
2894                                 struct lustre_capa *capa, int ignore_quota)
2895 {
2896         struct osd_object     *obj = osd_dt_obj(dt);
2897         struct iam_path_descr *ipd;
2898         struct osd_thandle    *oh;
2899         struct iam_container  *bag = &obj->oo_dir->od_container;
2900 #ifdef HAVE_QUOTA_SUPPORT
2901         cfs_cap_t              save = cfs_curproc_cap_pack();
2902 #endif
2903         struct osd_thread_info *oti = osd_oti_get(env);
2904         struct iam_rec         *iam_rec = (struct iam_rec *)oti->oti_ldp;
2905         int                     rc;
2906
2907         ENTRY;
2908
2909         LINVRNT(osd_invariant(obj));
2910         LASSERT(dt_object_exists(dt));
2911         LASSERT(bag->ic_object == obj->oo_inode);
2912         LASSERT(th != NULL);
2913
2914         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2915                 return -EACCES;
2916
2917         OSD_EXEC_OP(th, insert);
2918
2919         ipd = osd_idx_ipd_get(env, bag);
2920         if (unlikely(ipd == NULL))
2921                 RETURN(-ENOMEM);
2922
2923         oh = container_of0(th, struct osd_thandle, ot_super);
2924         LASSERT(oh->ot_handle != NULL);
2925         LASSERT(oh->ot_handle->h_transaction != NULL);
2926 #ifdef HAVE_QUOTA_SUPPORT
2927         if (ignore_quota)
2928                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
2929         else
2930                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
2931 #endif
2932         if (S_ISDIR(obj->oo_inode->i_mode))
2933                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
2934         else
2935                 iam_rec = (struct iam_rec *) rec;
2936         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
2937                         iam_rec, ipd);
2938 #ifdef HAVE_QUOTA_SUPPORT
2939         cfs_curproc_cap_unpack(save);
2940 #endif
2941         osd_ipd_put(env, bag, ipd);
2942         LINVRNT(osd_invariant(obj));
2943         RETURN(rc);
2944 }
2945
2946 /**
2947  * Calls ldiskfs_add_entry() to add directory entry
2948  * into the directory. This is required for
2949  * interoperability mode (b11826)
2950  *
2951  * \retval   0, on success
2952  * \retval -ve, on error
2953  */
2954 static int __osd_ea_add_rec(struct osd_thread_info *info,
2955                             struct osd_object *pobj, struct inode  *cinode,
2956                             const char *name, const struct dt_rec *fid,
2957                             struct htree_lock *hlock, struct thandle *th)
2958 {
2959         struct ldiskfs_dentry_param *ldp;
2960         struct dentry               *child;
2961         struct osd_thandle          *oth;
2962         int                          rc;
2963
2964         oth = container_of(th, struct osd_thandle, ot_super);
2965         LASSERT(oth->ot_handle != NULL);
2966         LASSERT(oth->ot_handle->h_transaction != NULL);
2967
2968         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
2969
2970         /* XXX: remove fid_is_igif() check here.
2971          * IGIF check is just to handle insertion of .. when it is 'ROOT',
2972          * it is IGIF now but needs FID in dir entry as well for readdir
2973          * to work.
2974          * LU-838 should fix that and remove fid_is_igif() check */
2975         if (fid_is_igif((struct lu_fid *)fid) ||
2976             fid_is_norm((struct lu_fid *)fid)) {
2977                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2978                 osd_get_ldiskfs_dirent_param(ldp, fid);
2979                 child->d_fsdata = (void *)ldp;
2980         } else {
2981                 child->d_fsdata = NULL;
2982         }
2983         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
2984
2985         RETURN(rc);
2986 }
2987
2988 /**
2989  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
2990  * into the directory.Also sets flags into osd object to
2991  * indicate dot and dotdot are created. This is required for
2992  * interoperability mode (b11826)
2993  *
2994  * \param dir   directory for dot and dotdot fixup.
2995  * \param obj   child object for linking
2996  *
2997  * \retval   0, on success
2998  * \retval -ve, on error
2999  */
3000 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3001                               struct osd_object *dir,
3002                               struct inode  *parent_dir, const char *name,
3003                               const struct dt_rec *dot_fid,
3004                               const struct dt_rec *dot_dot_fid,
3005                               struct thandle *th)
3006 {
3007         struct inode                *inode = dir->oo_inode;
3008         struct ldiskfs_dentry_param *dot_ldp;
3009         struct ldiskfs_dentry_param *dot_dot_ldp;
3010         struct osd_thandle          *oth;
3011         int result = 0;
3012
3013         oth = container_of(th, struct osd_thandle, ot_super);
3014         LASSERT(oth->ot_handle->h_transaction != NULL);
3015         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3016
3017         if (strcmp(name, dot) == 0) {
3018                 if (dir->oo_compat_dot_created) {
3019                         result = -EEXIST;
3020                 } else {
3021                         LASSERT(inode == parent_dir);
3022                         dir->oo_compat_dot_created = 1;
3023                         result = 0;
3024                 }
3025         } else if(strcmp(name, dotdot) == 0) {
3026                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3027                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3028
3029                 if (!dir->oo_compat_dot_created)
3030                         return -EINVAL;
3031                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3032                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3033                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3034                 } else {
3035                         dot_ldp = NULL;
3036                         dot_dot_ldp = NULL;
3037                 }
3038                 /* in case of rename, dotdot is already created */
3039                 if (dir->oo_compat_dotdot_created) {
3040                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3041                                                 dot_dot_fid, NULL, th);
3042                 }
3043
3044                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3045                                                 inode, dot_ldp, dot_dot_ldp);
3046                 if (result == 0)
3047                        dir->oo_compat_dotdot_created = 1;
3048         }
3049
3050         return result;
3051 }
3052
3053
3054 /**
3055  * It will call the appropriate osd_add* function and return the
3056  * value, return by respective functions.
3057  */
3058 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3059                           struct inode *cinode, const char *name,
3060                           const struct dt_rec *fid, struct thandle *th)
3061 {
3062         struct osd_thread_info *info   = osd_oti_get(env);
3063         struct htree_lock      *hlock;
3064         int                     rc;
3065
3066         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3067
3068         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3069                                                    name[2] =='\0'))) {
3070                 if (hlock != NULL) {
3071                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3072                                            pobj->oo_inode, 0);
3073                 } else {
3074                         cfs_down_write(&pobj->oo_ext_idx_sem);
3075                 }
3076                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3077                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3078                                         fid, th);
3079         } else {
3080                 if (hlock != NULL) {
3081                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3082                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3083                 } else {
3084                         cfs_down_write(&pobj->oo_ext_idx_sem);
3085                 }
3086
3087                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3088                                       hlock, th);
3089         }
3090         if (hlock != NULL)
3091                 ldiskfs_htree_unlock(hlock);
3092         else
3093                 cfs_up_write(&pobj->oo_ext_idx_sem);
3094
3095         return rc;
3096 }
3097
3098 /**
3099  * Calls ->lookup() to find dentry. From dentry get inode and
3100  * read inode's ea to get fid. This is required for  interoperability
3101  * mode (b11826)
3102  *
3103  * \retval   0, on success
3104  * \retval -ve, on error
3105  */
3106 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3107                              struct dt_rec *rec, const struct dt_key *key)
3108 {
3109         struct inode               *dir    = obj->oo_inode;
3110         struct dentry              *dentry;
3111         struct ldiskfs_dir_entry_2 *de;
3112         struct buffer_head         *bh;
3113         struct lu_fid              *fid = (struct lu_fid *) rec;
3114         struct htree_lock          *hlock = NULL;
3115         int                         ino;
3116         int                         rc;
3117
3118         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3119
3120         dentry = osd_child_dentry_get(env, obj,
3121                                       (char *)key, strlen((char *)key));
3122
3123         if (obj->oo_hl_head != NULL) {
3124                 hlock = osd_oti_get(env)->oti_hlock;
3125                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3126                                    dir, LDISKFS_HLOCK_LOOKUP);
3127         } else {
3128                 cfs_down_read(&obj->oo_ext_idx_sem);
3129         }
3130
3131         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3132         if (bh) {
3133                 ino = le32_to_cpu(de->inode);
3134                 rc = osd_get_fid_from_dentry(de, rec);
3135
3136                 /* done with de, release bh */
3137                 brelse(bh);
3138                 if (rc != 0)
3139                         rc = osd_ea_fid_get(env, obj, ino, fid);
3140         } else {
3141                 rc = -ENOENT;
3142         }
3143
3144         if (hlock != NULL)
3145                 ldiskfs_htree_unlock(hlock);
3146         else
3147                 cfs_up_read(&obj->oo_ext_idx_sem);
3148         RETURN (rc);
3149 }
3150
3151 /**
3152  * Find the osd object for given fid.
3153  *
3154  * \param fid need to find the osd object having this fid
3155  *
3156  * \retval osd_object on success
3157  * \retval        -ve on error
3158  */
3159 struct osd_object *osd_object_find(const struct lu_env *env,
3160                                    struct dt_object *dt,
3161                                    const struct lu_fid *fid)
3162 {
3163         struct lu_device  *ludev = dt->do_lu.lo_dev;
3164         struct osd_object *child = NULL;
3165         struct lu_object  *luch;
3166         struct lu_object  *lo;
3167
3168         luch = lu_object_find(env, ludev, fid, NULL);
3169         if (!IS_ERR(luch)) {
3170                 if (lu_object_exists(luch)) {
3171                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3172                         if (lo != NULL)
3173                                 child = osd_obj(lo);
3174                         else
3175                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3176                                                 "lu_object can't be located"
3177                                                 ""DFID"\n", PFID(fid));
3178
3179                         if (child == NULL) {
3180                                 lu_object_put(env, luch);
3181                                 CERROR("Unable to get osd_object\n");
3182                                 child = ERR_PTR(-ENOENT);
3183                         }
3184                 } else {
3185                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3186                                         "lu_object does not exists "DFID"\n",
3187                                         PFID(fid));
3188                         child = ERR_PTR(-ENOENT);
3189                 }
3190         } else
3191                 child = (void *)luch;
3192
3193         return child;
3194 }
3195
3196 /**
3197  * Put the osd object once done with it.
3198  *
3199  * \param obj osd object that needs to be put
3200  */
3201 static inline void osd_object_put(const struct lu_env *env,
3202                                   struct osd_object *obj)
3203 {
3204         lu_object_put(env, &obj->oo_dt.do_lu);
3205 }
3206
3207 static int osd_index_declare_ea_insert(const struct lu_env *env,
3208                                        struct dt_object *dt,
3209                                        const struct dt_rec *rec,
3210                                        const struct dt_key *key,
3211                                        struct thandle *handle)
3212 {
3213         struct osd_thandle *oh;
3214
3215         LASSERT(dt_object_exists(dt));
3216         LASSERT(handle != NULL);
3217
3218         oh = container_of0(handle, struct osd_thandle, ot_super);
3219         LASSERT(oh->ot_handle == NULL);
3220
3221         OSD_DECLARE_OP(oh, insert);
3222         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3223
3224         LASSERT(osd_dt_obj(dt)->oo_inode);
3225         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3226                         osd_dt_obj(dt)->oo_inode);
3227         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3228                         osd_dt_obj(dt)->oo_inode);
3229
3230         return 0;
3231 }
3232
3233 /**
3234  * Index add function for interoperability mode (b11826).
3235  * It will add the directory entry.This entry is needed to
3236  * maintain name->fid mapping.
3237  *
3238  * \param key it is key i.e. file entry to be inserted
3239  * \param rec it is value of given key i.e. fid
3240  *
3241  * \retval   0, on success
3242  * \retval -ve, on error
3243  */
3244 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3245                                const struct dt_rec *rec,
3246                                const struct dt_key *key, struct thandle *th,
3247                                struct lustre_capa *capa, int ignore_quota)
3248 {
3249         struct osd_object *obj   = osd_dt_obj(dt);
3250         struct lu_fid     *fid   = (struct lu_fid *) rec;
3251         const char        *name  = (const char *)key;
3252         struct osd_object *child;
3253 #ifdef HAVE_QUOTA_SUPPORT
3254         cfs_cap_t          save  = cfs_curproc_cap_pack();
3255 #endif
3256         int                rc;
3257
3258         ENTRY;
3259
3260         LASSERT(osd_invariant(obj));
3261         LASSERT(dt_object_exists(dt));
3262         LASSERT(th != NULL);
3263
3264         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3265                 RETURN(-EACCES);
3266
3267         child = osd_object_find(env, dt, fid);
3268         if (!IS_ERR(child)) {
3269 #ifdef HAVE_QUOTA_SUPPORT
3270                 if (ignore_quota)
3271                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3272                 else
3273                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3274 #endif
3275                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3276 #ifdef HAVE_QUOTA_SUPPORT
3277                 cfs_curproc_cap_unpack(save);
3278 #endif
3279                 osd_object_put(env, child);
3280         } else {
3281                 rc = PTR_ERR(child);
3282         }
3283
3284         LASSERT(osd_invariant(obj));
3285         RETURN(rc);
3286 }
3287
3288 /**
3289  *  Initialize osd Iterator for given osd index object.
3290  *
3291  *  \param  dt      osd index object
3292  */
3293
3294 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3295                                      struct dt_object *dt,
3296                                      __u32 unused,
3297                                      struct lustre_capa *capa)
3298 {
3299         struct osd_it_iam      *it;
3300         struct osd_thread_info *oti = osd_oti_get(env);
3301         struct osd_object      *obj = osd_dt_obj(dt);
3302         struct lu_object       *lo  = &dt->do_lu;
3303         struct iam_path_descr  *ipd;
3304         struct iam_container   *bag = &obj->oo_dir->od_container;
3305
3306         LASSERT(lu_object_exists(lo));
3307
3308         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3309                 return ERR_PTR(-EACCES);
3310
3311         it = &oti->oti_it;
3312         ipd = osd_it_ipd_get(env, bag);
3313         if (likely(ipd != NULL)) {
3314                 it->oi_obj = obj;
3315                 it->oi_ipd = ipd;
3316                 lu_object_get(lo);
3317                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3318                 return (struct dt_it *)it;
3319         }
3320         return ERR_PTR(-ENOMEM);
3321 }
3322
3323 /**
3324  * free given Iterator.
3325  */
3326
3327 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3328 {
3329         struct osd_it_iam *it = (struct osd_it_iam *)di;
3330         struct osd_object *obj = it->oi_obj;
3331
3332         iam_it_fini(&it->oi_it);
3333         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3334         lu_object_put(env, &obj->oo_dt.do_lu);
3335 }
3336
3337 /**
3338  *  Move Iterator to record specified by \a key
3339  *
3340  *  \param  di      osd iterator
3341  *  \param  key     key for index
3342  *
3343  *  \retval +ve  di points to record with least key not larger than key
3344  *  \retval  0   di points to exact matched key
3345  *  \retval -ve  failure
3346  */
3347
3348 static int osd_it_iam_get(const struct lu_env *env,
3349                           struct dt_it *di, const struct dt_key *key)
3350 {
3351         struct osd_it_iam *it = (struct osd_it_iam *)di;
3352
3353         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3354 }
3355
3356 /**
3357  *  Release Iterator
3358  *
3359  *  \param  di      osd iterator
3360  */
3361
3362 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3363 {
3364         struct osd_it_iam *it = (struct osd_it_iam *)di;
3365
3366         iam_it_put(&it->oi_it);
3367 }
3368
3369 /**
3370  *  Move iterator by one record
3371  *
3372  *  \param  di      osd iterator
3373  *
3374  *  \retval +1   end of container reached
3375  *  \retval  0   success
3376  *  \retval -ve  failure
3377  */
3378
3379 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3380 {
3381         struct osd_it_iam *it = (struct osd_it_iam *)di;
3382
3383         return iam_it_next(&it->oi_it);
3384 }
3385
3386 /**
3387  * Return pointer to the key under iterator.
3388  */
3389
3390 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3391                                  const struct dt_it *di)
3392 {
3393         struct osd_it_iam *it = (struct osd_it_iam *)di;
3394
3395         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3396 }
3397
3398 /**
3399  * Return size of key under iterator (in bytes)
3400  */
3401
3402 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3403 {
3404         struct osd_it_iam *it = (struct osd_it_iam *)di;
3405
3406         return iam_it_key_size(&it->oi_it);
3407 }
3408
3409 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3410                                        int len, __u16 type)
3411 {
3412         struct luda_type *lt;
3413         const unsigned    align = sizeof(struct luda_type) - 1;
3414
3415         /* check if file type is required */
3416         if (attr & LUDA_TYPE) {
3417                         len = (len + align) & ~align;
3418
3419                         lt = (void *) ent->lde_name + len;
3420                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3421                         ent->lde_attrs |= LUDA_TYPE;
3422         }
3423
3424         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3425 }
3426
3427 /**
3428  * build lu direct from backend fs dirent.
3429  */
3430
3431 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3432                                       struct lu_fid *fid, __u64 offset,
3433                                       char *name, __u16 namelen,
3434                                       __u16 type, __u32 attr)
3435 {
3436         fid_cpu_to_le(&ent->lde_fid, fid);
3437         ent->lde_attrs = LUDA_FID;
3438
3439         ent->lde_hash = cpu_to_le64(offset);
3440         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3441
3442         strncpy(ent->lde_name, name, namelen);
3443         ent->lde_namelen = cpu_to_le16(namelen);
3444
3445         /* append lustre attributes */
3446         osd_it_append_attrs(ent, attr, namelen, type);
3447 }
3448
3449 /**
3450  * Return pointer to the record under iterator.
3451  */
3452 static int osd_it_iam_rec(const struct lu_env *env,
3453                           const struct dt_it *di,
3454                           struct dt_rec *dtrec, __u32 attr)
3455 {
3456         struct osd_it_iam *it        = (struct osd_it_iam *)di;
3457         struct osd_thread_info *info = osd_oti_get(env);
3458         struct lu_fid     *fid       = &info->oti_fid;
3459         const struct osd_fid_pack *rec;
3460         struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3461         char *name;
3462         int namelen;
3463         __u64 hash;
3464         int rc;
3465
3466         name = (char *)iam_it_key_get(&it->oi_it);
3467         if (IS_ERR(name))
3468                 RETURN(PTR_ERR(name));
3469
3470         namelen = iam_it_key_size(&it->oi_it);
3471
3472         rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
3473         if (IS_ERR(rec))
3474                 RETURN(PTR_ERR(rec));
3475
3476         rc = osd_fid_unpack(fid, rec);
3477         if (rc)
3478                 RETURN(rc);
3479
3480         hash = iam_it_store(&it->oi_it);
3481
3482         /* IAM does not store object type in IAM index (dir) */
3483         osd_it_pack_dirent(lde, fid, hash, name, namelen,
3484                            0, LUDA_FID);
3485
3486         return 0;
3487 }
3488
3489 /**
3490  * Returns cookie for current Iterator position.
3491  */
3492 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3493 {
3494         struct osd_it_iam *it = (struct osd_it_iam *)di;
3495
3496         return iam_it_store(&it->oi_it);
3497 }
3498
3499 /**
3500  * Restore iterator from cookie.
3501  *
3502  * \param  di      osd iterator
3503  * \param  hash    Iterator location cookie
3504  *
3505  * \retval +ve  di points to record with least key not larger than key.
3506  * \retval  0   di points to exact matched key
3507  * \retval -ve  failure
3508  */
3509
3510 static int osd_it_iam_load(const struct lu_env *env,
3511                            const struct dt_it *di, __u64 hash)
3512 {
3513         struct osd_it_iam *it = (struct osd_it_iam *)di;
3514
3515         return iam_it_load(&it->oi_it, hash);
3516 }
3517
3518 static const struct dt_index_operations osd_index_iam_ops = {
3519         .dio_lookup         = osd_index_iam_lookup,
3520         .dio_declare_insert = osd_index_declare_iam_insert,
3521         .dio_insert         = osd_index_iam_insert,
3522         .dio_declare_delete = osd_index_declare_iam_delete,
3523         .dio_delete         = osd_index_iam_delete,
3524         .dio_it     = {
3525                 .init     = osd_it_iam_init,
3526                 .fini     = osd_it_iam_fini,
3527                 .get      = osd_it_iam_get,
3528                 .put      = osd_it_iam_put,
3529                 .next     = osd_it_iam_next,
3530                 .key      = osd_it_iam_key,
3531                 .key_size = osd_it_iam_key_size,
3532                 .rec      = osd_it_iam_rec,
3533                 .store    = osd_it_iam_store,
3534                 .load     = osd_it_iam_load
3535         }
3536 };
3537
3538 /**
3539  * Creates or initializes iterator context.
3540  *
3541  * \retval struct osd_it_ea, iterator structure on success
3542  *
3543  */
3544 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3545                                     struct dt_object *dt,
3546                                     __u32 attr,
3547                                     struct lustre_capa *capa)
3548 {
3549         struct osd_object       *obj  = osd_dt_obj(dt);
3550         struct osd_thread_info  *info = osd_oti_get(env);
3551         struct osd_it_ea        *it   = &info->oti_it_ea;
3552         struct lu_object        *lo   = &dt->do_lu;
3553         struct dentry           *obj_dentry = &info->oti_it_dentry;
3554         ENTRY;
3555         LASSERT(lu_object_exists(lo));
3556
3557         obj_dentry->d_inode = obj->oo_inode;
3558         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3559         obj_dentry->d_name.hash = 0;
3560
3561         it->oie_rd_dirent       = 0;
3562         it->oie_it_dirent       = 0;
3563         it->oie_dirent          = NULL;
3564         it->oie_buf             = info->oti_it_ea_buf;
3565         it->oie_obj             = obj;
3566         it->oie_file.f_pos      = 0;
3567         it->oie_file.f_dentry   = obj_dentry;
3568         if (attr & LUDA_64BITHASH)
3569                 it->oie_file.f_flags = O_64BITHASH;
3570         else
3571                 it->oie_file.f_flags = O_32BITHASH;
3572         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
3573         it->oie_file.f_op         = obj->oo_inode->i_fop;
3574         it->oie_file.private_data = NULL;
3575         lu_object_get(lo);
3576         RETURN((struct dt_it *) it);
3577 }
3578
3579 /**
3580  * Destroy or finishes iterator context.
3581  *
3582  * \param di iterator structure to be destroyed
3583  */
3584 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3585 {
3586         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3587         struct osd_object    *obj  = it->oie_obj;
3588         struct inode       *inode  = obj->oo_inode;
3589
3590         ENTRY;
3591         it->oie_file.f_op->release(inode, &it->oie_file);
3592         lu_object_put(env, &obj->oo_dt.do_lu);
3593         EXIT;
3594 }
3595
3596 /**
3597  * It position the iterator at given key, so that next lookup continues from
3598  * that key Or it is similar to dio_it->load() but based on a key,
3599  * rather than file position.
3600  *
3601  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3602  * to the beginning.
3603  *
3604  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3605  */
3606 static int osd_it_ea_get(const struct lu_env *env,
3607                          struct dt_it *di, const struct dt_key *key)
3608 {
3609         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3610
3611         ENTRY;
3612         LASSERT(((const char *)key)[0] == '\0');
3613         it->oie_file.f_pos      = 0;
3614         it->oie_rd_dirent       = 0;
3615         it->oie_it_dirent       = 0;
3616         it->oie_dirent          = NULL;
3617
3618         RETURN(+1);
3619 }
3620
3621 /**
3622  * Does nothing
3623  */
3624 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3625 {
3626 }
3627
3628 /**
3629  * It is called internally by ->readdir(). It fills the
3630  * iterator's in-memory data structure with required
3631  * information i.e. name, namelen, rec_size etc.
3632  *
3633  * \param buf in which information to be filled in.
3634  * \param name name of the file in given dir
3635  *
3636  * \retval 0 on success
3637  * \retval 1 on buffer full
3638  */
3639 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3640                                loff_t offset, __u64 ino,
3641                                unsigned d_type)
3642 {
3643         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
3644         struct osd_it_ea_dirent *ent  = it->oie_dirent;
3645         struct lu_fid           *fid  = &ent->oied_fid;
3646         struct osd_fid_pack     *rec;
3647         ENTRY;
3648
3649         /* this should never happen */
3650         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3651                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3652                 RETURN(-EIO);
3653         }
3654
3655         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3656             OSD_IT_EA_BUFSIZE)
3657                 RETURN(1);
3658
3659         if (d_type & LDISKFS_DIRENT_LUFID) {
3660                 rec = (struct osd_fid_pack*) (name + namelen + 1);
3661
3662                 if (osd_fid_unpack(fid, rec) != 0)
3663                         fid_zero(fid);
3664
3665                 d_type &= ~LDISKFS_DIRENT_LUFID;
3666         } else {
3667                 fid_zero(fid);
3668         }
3669
3670         ent->oied_ino     = ino;
3671         ent->oied_off     = offset;
3672         ent->oied_namelen = namelen;
3673         ent->oied_type    = d_type;
3674
3675         memcpy(ent->oied_name, name, namelen);
3676
3677         it->oie_rd_dirent++;
3678         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
3679         RETURN(0);
3680 }
3681
3682 /**
3683  * Calls ->readdir() to load a directory entry at a time
3684  * and stored it in iterator's in-memory data structure.
3685  *
3686  * \param di iterator's in memory structure
3687  *
3688  * \retval   0 on success
3689  * \retval -ve on error
3690  */
3691 static int osd_ldiskfs_it_fill(const struct lu_env *env,
3692                                const struct dt_it *di)
3693 {
3694         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
3695         struct osd_object  *obj   = it->oie_obj;
3696         struct inode       *inode = obj->oo_inode;
3697         struct htree_lock  *hlock = NULL;
3698         int                 result = 0;
3699
3700         ENTRY;
3701         it->oie_dirent = it->oie_buf;
3702         it->oie_rd_dirent = 0;
3703
3704         if (obj->oo_hl_head != NULL) {
3705                 hlock = osd_oti_get(env)->oti_hlock;
3706                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3707                                    inode, LDISKFS_HLOCK_READDIR);
3708         } else {
3709                 cfs_down_read(&obj->oo_ext_idx_sem);
3710         }
3711
3712         result = inode->i_fop->readdir(&it->oie_file, it,
3713                                        (filldir_t) osd_ldiskfs_filldir);
3714
3715         if (hlock != NULL)
3716                 ldiskfs_htree_unlock(hlock);
3717         else
3718                 cfs_up_read(&obj->oo_ext_idx_sem);
3719
3720         if (it->oie_rd_dirent == 0) {
3721                 result = -EIO;
3722         } else {
3723                 it->oie_dirent = it->oie_buf;
3724                 it->oie_it_dirent = 1;
3725         }
3726
3727         RETURN(result);
3728 }
3729
3730 /**
3731  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3732  * to load a directory entry at a time and stored it in
3733  * iterator's in-memory data structure.
3734  *
3735  * \param di iterator's in memory structure
3736  *
3737  * \retval +ve iterator reached to end
3738  * \retval   0 iterator not reached to end
3739  * \retval -ve on error
3740  */
3741 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
3742 {
3743         struct osd_it_ea *it = (struct osd_it_ea *)di;
3744         int rc;
3745
3746         ENTRY;
3747
3748         if (it->oie_it_dirent < it->oie_rd_dirent) {
3749                 it->oie_dirent =
3750                         (void *) it->oie_dirent +
3751                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
3752                                        it->oie_dirent->oied_namelen);
3753                 it->oie_it_dirent++;
3754                 RETURN(0);
3755         } else {
3756                 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
3757                         rc = +1;
3758                 else
3759                         rc = osd_ldiskfs_it_fill(env, di);
3760         }
3761
3762         RETURN(rc);
3763 }
3764
3765 /**
3766  * Returns the key at current position from iterator's in memory structure.
3767  *
3768  * \param di iterator's in memory structure
3769  *
3770  * \retval key i.e. struct dt_key on success
3771  */
3772 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
3773                                     const struct dt_it *di)
3774 {
3775         struct osd_it_ea *it = (struct osd_it_ea *)di;
3776
3777         return (struct dt_key *)it->oie_dirent->oied_name;
3778 }
3779
3780 /**
3781  * Returns the key's size at current position from iterator's in memory structure.
3782  *
3783  * \param di iterator's in memory structure
3784  *
3785  * \retval key_size i.e. struct dt_key on success
3786  */
3787 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
3788 {
3789         struct osd_it_ea *it = (struct osd_it_ea *)di;
3790
3791         return it->oie_dirent->oied_namelen;
3792 }
3793
3794
3795 /**
3796  * Returns the value (i.e. fid/igif) at current position from iterator's
3797  * in memory structure.
3798  *
3799  * \param di struct osd_it_ea, iterator's in memory structure
3800  * \param attr attr requested for dirent.
3801  * \param lde lustre dirent
3802  *
3803  * \retval   0 no error and \param lde has correct lustre dirent.
3804  * \retval -ve on error
3805  */
3806 static inline int osd_it_ea_rec(const struct lu_env *env,
3807                                 const struct dt_it *di,
3808                                 struct dt_rec *dtrec, __u32 attr)
3809 {
3810         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
3811         struct osd_object       *obj    = it->oie_obj;
3812         struct lu_fid           *fid    = &it->oie_dirent->oied_fid;
3813         struct lu_dirent        *lde    = (struct lu_dirent *)dtrec;
3814         int    rc = 0;
3815
3816         ENTRY;
3817
3818         if (!fid_is_sane(fid))
3819                 rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
3820
3821         if (rc == 0)
3822                 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
3823                                    it->oie_dirent->oied_name,
3824                                    it->oie_dirent->oied_namelen,
3825                                    it->oie_dirent->oied_type,
3826                                    attr);
3827         RETURN(rc);
3828 }
3829
3830 /**
3831  * Returns a cookie for current position of the iterator head, so that
3832  * user can use this cookie to load/start the iterator next time.
3833  *
3834  * \param di iterator's in memory structure
3835  *
3836  * \retval cookie for current position, on success
3837  */
3838 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
3839 {
3840         struct osd_it_ea *it = (struct osd_it_ea *)di;
3841
3842         return it->oie_dirent->oied_off;
3843 }
3844
3845 /**
3846  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3847  * to load a directory entry at a time and stored it i inn,
3848  * in iterator's in-memory data structure.
3849  *
3850  * \param di struct osd_it_ea, iterator's in memory structure
3851  *
3852  * \retval +ve on success
3853  * \retval -ve on error
3854  */
3855 static int osd_it_ea_load(const struct lu_env *env,
3856                           const struct dt_it *di, __u64 hash)
3857 {
3858         struct osd_it_ea *it = (struct osd_it_ea *)di;
3859         int rc;
3860
3861         ENTRY;
3862         it->oie_file.f_pos = hash;
3863
3864         rc =  osd_ldiskfs_it_fill(env, di);
3865         if (rc == 0)
3866                 rc = +1;
3867
3868         RETURN(rc);
3869 }
3870
3871 /**
3872  * Index lookup function for interoperability mode (b11826).
3873  *
3874  * \param key,  key i.e. file name to be searched
3875  *
3876  * \retval +ve, on success
3877  * \retval -ve, on error
3878  */
3879 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
3880                                struct dt_rec *rec, const struct dt_key *key,
3881                                struct lustre_capa *capa)
3882 {
3883         struct osd_object *obj = osd_dt_obj(dt);
3884         int rc = 0;
3885
3886         ENTRY;
3887
3888         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
3889         LINVRNT(osd_invariant(obj));
3890
3891         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3892                 return -EACCES;
3893
3894         rc = osd_ea_lookup_rec(env, obj, rec, key);
3895
3896         if (rc == 0)
3897                 rc = +1;
3898         RETURN(rc);
3899 }
3900
3901 /**
3902  * Index and Iterator operations for interoperability
3903  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3904  */
3905 static const struct dt_index_operations osd_index_ea_ops = {
3906         .dio_lookup         = osd_index_ea_lookup,
3907         .dio_declare_insert = osd_index_declare_ea_insert,
3908         .dio_insert         = osd_index_ea_insert,
3909         .dio_declare_delete = osd_index_declare_ea_delete,
3910         .dio_delete         = osd_index_ea_delete,
3911         .dio_it     = {
3912                 .init     = osd_it_ea_init,
3913                 .fini     = osd_it_ea_fini,
3914                 .get      = osd_it_ea_get,
3915                 .put      = osd_it_ea_put,
3916                 .next     = osd_it_ea_next,
3917                 .key      = osd_it_ea_key,
3918                 .key_size = osd_it_ea_key_size,
3919                 .rec      = osd_it_ea_rec,
3920                 .store    = osd_it_ea_store,
3921                 .load     = osd_it_ea_load
3922         }
3923 };
3924
3925 static void *osd_key_init(const struct lu_context *ctx,
3926                           struct lu_context_key *key)
3927 {
3928         struct osd_thread_info *info;
3929
3930         OBD_ALLOC_PTR(info);
3931         if (info == NULL)
3932                 return ERR_PTR(-ENOMEM);
3933
3934         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3935         if (info->oti_it_ea_buf == NULL)
3936                 goto out_free_info;
3937
3938         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
3939
3940         info->oti_hlock = ldiskfs_htree_lock_alloc();
3941         if (info->oti_hlock == NULL)
3942                 goto out_free_ea;
3943
3944         return info;
3945
3946  out_free_ea:
3947         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3948  out_free_info:
3949         OBD_FREE_PTR(info);
3950         return ERR_PTR(-ENOMEM);
3951 }
3952
3953 static void osd_key_fini(const struct lu_context *ctx,
3954                          struct lu_context_key *key, void* data)
3955 {
3956         struct osd_thread_info *info = data;
3957
3958         if (info->oti_hlock != NULL)
3959                 ldiskfs_htree_lock_free(info->oti_hlock);
3960         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3961         OBD_FREE_PTR(info);
3962 }
3963
3964 static void osd_key_exit(const struct lu_context *ctx,
3965                          struct lu_context_key *key, void *data)
3966 {
3967         struct osd_thread_info *info = data;
3968
3969         LASSERT(info->oti_r_locks == 0);
3970         LASSERT(info->oti_w_locks == 0);
3971         LASSERT(info->oti_txns    == 0);
3972 }
3973
3974 /* type constructor/destructor: osd_type_init, osd_type_fini */
3975 LU_TYPE_INIT_FINI(osd, &osd_key);
3976
3977 struct lu_context_key osd_key = {
3978         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
3979         .lct_init = osd_key_init,
3980         .lct_fini = osd_key_fini,
3981         .lct_exit = osd_key_exit
3982 };
3983
3984
3985 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
3986                            const char *name, struct lu_device *next)
3987 {
3988         return osd_procfs_init(osd_dev(d), name);
3989 }
3990
3991 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
3992 {
3993         struct osd_thread_info *info = osd_oti_get(env);
3994
3995         ENTRY;
3996
3997         if (o->od_oi_table != NULL)
3998                 osd_oi_fini(info, o);
3999
4000         if (o->od_fsops) {
4001                 fsfilt_put_ops(o->od_fsops);
4002                 o->od_fsops = NULL;
4003         }
4004
4005         RETURN(0);
4006 }
4007
4008 static int osd_mount(const struct lu_env *env,
4009                      struct osd_device *o, struct lustre_cfg *cfg)
4010 {
4011         struct lustre_mount_info *lmi;
4012         const char               *dev  = lustre_cfg_string(cfg, 0);
4013         struct lustre_disk_data  *ldd;
4014         struct lustre_sb_info    *lsi;
4015         int                       rc = 0;
4016
4017         ENTRY;
4018
4019         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4020         if (o->od_fsops == NULL) {
4021                 CERROR("Can't find fsfilt_ldiskfs\n");
4022                 RETURN(-ENOTSUPP);
4023         }
4024
4025         if (o->od_mount != NULL) {
4026                 CERROR("Already mounted (%s)\n", dev);
4027                 RETURN(-EEXIST);
4028         }
4029
4030         /* get mount */
4031         lmi = server_get_mount(dev);
4032         if (lmi == NULL) {
4033                 CERROR("Cannot get mount info for %s!\n", dev);
4034                 RETURN(-EFAULT);
4035         }
4036
4037         LASSERT(lmi != NULL);
4038         /* save lustre_mount_info in dt_device */
4039         o->od_mount = lmi;
4040         o->od_mnt = lmi->lmi_mnt;
4041
4042         lsi = s2lsi(lmi->lmi_sb);
4043         ldd = lsi->lsi_ldd;
4044
4045         if (ldd->ldd_flags & LDD_F_IAM_DIR) {
4046                 o->od_iop_mode = 0;
4047                 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", dev);
4048         } else
4049                 o->od_iop_mode = 1;
4050
4051         if (ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
4052                 rc = osd_compat_init(o);
4053                 if (rc)
4054                         CERROR("%s: can't initialize compats: %d\n", dev, rc);
4055         }
4056
4057         RETURN(rc);
4058 }
4059
4060 static struct lu_device *osd_device_fini(const struct lu_env *env,
4061                                          struct lu_device *d)
4062 {
4063         int rc;
4064         ENTRY;
4065
4066         osd_compat_fini(osd_dev(d));
4067
4068         shrink_dcache_sb(osd_sb(osd_dev(d)));
4069         osd_sync(env, lu2dt_dev(d));
4070
4071         rc = osd_procfs_fini(osd_dev(d));
4072         if (rc) {
4073                 CERROR("proc fini error %d \n", rc);
4074                 RETURN (ERR_PTR(rc));
4075         }
4076
4077         if (osd_dev(d)->od_mount)
4078                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
4079                                  osd_dev(d)->od_mount->lmi_mnt);
4080         osd_dev(d)->od_mount = NULL;
4081
4082         RETURN(NULL);
4083 }
4084
4085 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4086                                           struct lu_device_type *t,
4087                                           struct lustre_cfg *cfg)
4088 {
4089         struct lu_device  *l;
4090         struct osd_device *o;
4091
4092         OBD_ALLOC_PTR(o);
4093         if (o != NULL) {
4094                 int result;
4095
4096                 result = dt_device_init(&o->od_dt_dev, t);
4097                 if (result == 0) {
4098                         l = osd2lu_dev(o);
4099                         l->ld_ops = &osd_lu_ops;
4100                         o->od_dt_dev.dd_ops = &osd_dt_ops;
4101                         cfs_spin_lock_init(&o->od_osfs_lock);
4102                         o->od_osfs_age = cfs_time_shift_64(-1000);
4103                         o->od_capa_hash = init_capa_hash();
4104                         if (o->od_capa_hash == NULL) {
4105                                 dt_device_fini(&o->od_dt_dev);
4106                                 l = ERR_PTR(-ENOMEM);
4107                         }
4108                 } else
4109                         l = ERR_PTR(result);
4110
4111                 if (IS_ERR(l))
4112                         OBD_FREE_PTR(o);
4113         } else
4114                 l = ERR_PTR(-ENOMEM);
4115         return l;
4116 }
4117
4118 static struct lu_device *osd_device_free(const struct lu_env *env,
4119                                          struct lu_device *d)
4120 {
4121         struct osd_device *o = osd_dev(d);
4122         ENTRY;
4123
4124         cleanup_capa_hash(o->od_capa_hash);
4125         dt_device_fini(&o->od_dt_dev);
4126         OBD_FREE_PTR(o);
4127         RETURN(NULL);
4128 }
4129
4130 static int osd_process_config(const struct lu_env *env,
4131                               struct lu_device *d, struct lustre_cfg *cfg)
4132 {
4133         struct osd_device *o = osd_dev(d);
4134         int err;
4135         ENTRY;
4136
4137         switch(cfg->lcfg_command) {
4138         case LCFG_SETUP:
4139                 err = osd_mount(env, o, cfg);
4140                 break;
4141         case LCFG_CLEANUP:
4142                 err = osd_shutdown(env, o);
4143                 break;
4144         default:
4145                 err = -ENOSYS;
4146         }
4147
4148         RETURN(err);
4149 }
4150
4151 static int osd_recovery_complete(const struct lu_env *env,
4152                                  struct lu_device *d)
4153 {
4154         RETURN(0);
4155 }
4156
4157 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4158                        struct lu_device *dev)
4159 {
4160         struct osd_device      *osd = osd_dev(dev);
4161         struct osd_thread_info *oti = osd_oti_get(env);
4162         int                     result;
4163
4164         ENTRY;
4165
4166         /* 1. initialize oi before any file create or file open */
4167         result = osd_oi_init(oti, osd);
4168         if (result < 0)
4169                 RETURN(result);
4170
4171         if (!lu_device_is_md(pdev))
4172                 RETURN(0);
4173
4174         /* 2. setup local objects */
4175         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4176         RETURN(result);
4177 }
4178
4179 static const struct lu_object_operations osd_lu_obj_ops = {
4180         .loo_object_init      = osd_object_init,
4181         .loo_object_delete    = osd_object_delete,
4182         .loo_object_release   = osd_object_release,
4183         .loo_object_free      = osd_object_free,
4184         .loo_object_print     = osd_object_print,
4185         .loo_object_invariant = osd_object_invariant
4186 };
4187
4188 const struct lu_device_operations osd_lu_ops = {
4189         .ldo_object_alloc      = osd_object_alloc,
4190         .ldo_process_config    = osd_process_config,
4191         .ldo_recovery_complete = osd_recovery_complete,
4192         .ldo_prepare           = osd_prepare,
4193 };
4194
4195 static const struct lu_device_type_operations osd_device_type_ops = {
4196         .ldto_init = osd_type_init,
4197         .ldto_fini = osd_type_fini,
4198
4199         .ldto_start = osd_type_start,
4200         .ldto_stop  = osd_type_stop,
4201
4202         .ldto_device_alloc = osd_device_alloc,
4203         .ldto_device_free  = osd_device_free,
4204
4205         .ldto_device_init    = osd_device_init,
4206         .ldto_device_fini    = osd_device_fini
4207 };
4208
4209 static struct lu_device_type osd_device_type = {
4210         .ldt_tags     = LU_DEVICE_DT,
4211         .ldt_name     = LUSTRE_OSD_NAME,
4212         .ldt_ops      = &osd_device_type_ops,
4213         .ldt_ctx_tags = LCT_LOCAL,
4214 };
4215
4216 /*
4217  * lprocfs legacy support.
4218  */
4219 static struct obd_ops osd_obd_device_ops = {
4220         .o_owner = THIS_MODULE
4221 };
4222
4223 static int __init osd_mod_init(void)
4224 {
4225         struct lprocfs_static_vars lvars;
4226
4227         osd_oi_mod_init();
4228         lprocfs_osd_init_vars(&lvars);
4229         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4230                                    LUSTRE_OSD_NAME, &osd_device_type);
4231 }
4232
4233 static void __exit osd_mod_exit(void)
4234 {
4235         class_unregister_type(LUSTRE_OSD_NAME);
4236 }
4237
4238 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4239 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4240 MODULE_LICENSE("GPL");
4241
4242 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);