lustre/osd-ldiskfs/osd_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/osd/osd_handler.c
  39  *
  40  * Top-level entry points into osd module
  41  *
  42  * Author: Nikita Danilov <nikita@clusterfs.com>
  43  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  44  */
  45
  46 #ifndef EXPORT_SYMTAB
  47 # define EXPORT_SYMTAB
  48 #endif
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52
  53 /* LUSTRE_VERSION_CODE */
  54 #include <lustre_ver.h>
  55 /* prerequisite for linux/xattr.h */
  56 #include <linux/types.h>
  57 /* prerequisite for linux/xattr.h */
  58 #include <linux/fs.h>
  59 /* XATTR_{REPLACE,CREATE} */
  60 #include <linux/xattr.h>
  61 /* simple_mkdir() */
  62 #include <lvfs.h>
  63
  64 /*
  65  * struct OBD_{ALLOC,FREE}*()
  66  * OBD_FAIL_CHECK
  67  */
  68 #include <obd_support.h>
  69 /* struct ptlrpc_thread */
  70 #include <lustre_net.h>
  71
  72 /* fid_is_local() */
  73 #include <lustre_fid.h>
  74
  75 #include "osd_internal.h"
  76 #include "osd_igif.h"
  77
  78 /* llo_* api support */
  79 #include <md_object.h>
  80
  81 #ifdef HAVE_LDISKFS_PDO
  82 int ldiskfs_pdo = 1;
  83 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  84                 "ldiskfs with parallel directory operations");
  85 #else
  86 int ldiskfs_pdo = 0;
  87 #endif
  88
  89 static const char dot[] = ".";
  90 static const char dotdot[] = "..";
  91 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  92
  93 struct osd_directory {
  94         struct iam_container od_container;
  95         struct iam_descr     od_descr;
  96 };
  97
  98 struct osd_object {
  99         struct dt_object       oo_dt;
 100         /**
 101          * Inode for file system object represented by this osd_object. This
 102          * inode is pinned for the whole duration of lu_object life.
 103          *
 104          * Not modified concurrently (either setup early during object
 105          * creation, or assigned by osd_object_create() under write lock).
 106          */
 107         struct inode          *oo_inode;
 108         /**
 109          * to protect index ops.
 110          */
 111         struct htree_lock_head *oo_hl_head;
 112         cfs_rw_semaphore_t     oo_ext_idx_sem;
 113         cfs_rw_semaphore_t     oo_sem;
 114         struct osd_directory  *oo_dir;
 115         /** protects inode attributes. */
 116         cfs_spinlock_t         oo_guard;
 117         /**
 118          * Following two members are used to indicate the presence of dot and
 119          * dotdot in the given directory. This is required for interop mode
 120          * (b11826).
 121          */
 122         int                    oo_compat_dot_created;
 123         int                    oo_compat_dotdot_created;
 124
 125         const struct lu_env   *oo_owner;
 126 #ifdef CONFIG_LOCKDEP
 127         struct lockdep_map     oo_dep_map;
 128 #endif
 129 };
 130
 131 static const struct lu_object_operations      osd_lu_obj_ops;
 132 static const struct lu_device_operations      osd_lu_ops;
 133 static       struct lu_context_key            osd_key;
 134 static const struct dt_object_operations      osd_obj_ops;
 135 static const struct dt_object_operations      osd_obj_ea_ops;
 136 static const struct dt_body_operations        osd_body_ops;
 137 static const struct dt_body_operations        osd_body_ops_new;
 138 static const struct dt_index_operations       osd_index_iam_ops;
 139 static const struct dt_index_operations       osd_index_ea_ops;
 140
 141 #define OSD_TRACK_DECLARES
 142 #ifdef OSD_TRACK_DECLARES
 143 #define OSD_DECLARE_OP(oh, op)   {                               \
 144         LASSERT(oh->ot_handle == NULL);                          \
 145         ((oh)->ot_declare_ ##op)++; }
 146 #define OSD_EXEC_OP(handle, op)     {                            \
 147         struct osd_thandle *oh;                                  \
 148         oh = container_of0(handle, struct osd_thandle, ot_super);\
 149         if (((oh)->ot_declare_ ##op) > 0) {                      \
 150                 ((oh)->ot_declare_ ##op)--;                      \
 151         }                                                        \
 152         }
 153 #else
 154 #define OSD_DECLARE_OP(oh, op)
 155 #define OSD_EXEC_OP(oh, op)
 156 #endif
 157
 158 /* There are at most 10 uid/gids are affected in a transaction, and
 159  * that's rename case:
 160  * - 2 for source parent uid & gid;
 161  * - 2 for source child uid & gid ('..' entry update when the child
 162  *   is directory);
 163  * - 2 for target parent uid & gid;
 164  * - 2 for target child uid & gid (if the target child exists);
 165  * - 2 for root uid & gid (last_rcvd, llog, etc);
 166  *
 167  * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
 168  * the id type of each id in the ot_id_array.
 169  */
 170 #define OSD_MAX_UGID_CNT        10
 171
 172 struct osd_thandle {
 173         struct thandle          ot_super;
 174         handle_t               *ot_handle;
 175         struct journal_callback ot_jcb;
 176         cfs_list_t              ot_dcb_list;
 177         /* Link to the device, for debugging. */
 178         struct lu_ref_link     *ot_dev_link;
 179         unsigned short          ot_credits;
 180         unsigned short          ot_id_cnt;
 181         unsigned short          ot_id_type;
 182         uid_t                   ot_id_array[OSD_MAX_UGID_CNT];
 183
 184 #ifdef OSD_TRACK_DECLARES
 185         unsigned char           ot_declare_attr_set;
 186         unsigned char           ot_declare_punch;
 187         unsigned char           ot_declare_xattr_set;
 188         unsigned char           ot_declare_create;
 189         unsigned char           ot_declare_destroy;
 190         unsigned char           ot_declare_ref_add;
 191         unsigned char           ot_declare_ref_del;
 192         unsigned char           ot_declare_write;
 193         unsigned char           ot_declare_insert;
 194         unsigned char           ot_declare_delete;
 195 #endif
 196
 197 #if OSD_THANDLE_STATS
 198         /** time when this handle was allocated */
 199         cfs_time_t oth_alloced;
 200
 201         /** time when this thanle was started */
 202         cfs_time_t oth_started;
 203 #endif
 204 };
 205
 206 /**
 207  * Basic transaction credit op
 208  */
 209 enum dt_txn_op {
 210         DTO_INDEX_INSERT,
 211         DTO_INDEX_DELETE,
 212         DTO_INDEX_UPDATE,
 213         DTO_OBJECT_CREATE,
 214         DTO_OBJECT_DELETE,
 215         DTO_ATTR_SET_BASE,
 216         DTO_XATTR_SET,
 217         DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
 218         DTO_WRITE_BASE,
 219         DTO_WRITE_BLOCK,
 220         DTO_ATTR_SET_CHOWN,
 221
 222         DTO_NR
 223 };
 224
 225 /*
 226  * Helpers.
 227  */
 228 static int lu_device_is_osd(const struct lu_device *d)
 229 {
 230         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
 231 }
 232
 233 static struct osd_device *osd_dt_dev(const struct dt_device *d)
 234 {
 235         LASSERT(lu_device_is_osd(&d->dd_lu_dev));
 236         return container_of0(d, struct osd_device, od_dt_dev);
 237 }
 238
 239 static struct osd_device *osd_dev(const struct lu_device *d)
 240 {
 241         LASSERT(lu_device_is_osd(d));
 242         return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
 243 }
 244
 245 static struct osd_device *osd_obj2dev(const struct osd_object *o)
 246 {
 247         return osd_dev(o->oo_dt.do_lu.lo_dev);
 248 }
 249
 250 static struct super_block *osd_sb(const struct osd_device *dev)
 251 {
 252         return dev->od_mount->lmi_mnt->mnt_sb;
 253 }
 254
 255 static int osd_object_is_root(const struct osd_object *obj)
 256 {
 257         return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
 258 }
 259
 260 static struct osd_object *osd_obj(const struct lu_object *o)
 261 {
 262         LASSERT(lu_device_is_osd(o->lo_dev));
 263         return container_of0(o, struct osd_object, oo_dt.do_lu);
 264 }
 265
 266 static struct osd_object *osd_dt_obj(const struct dt_object *d)
 267 {
 268         return osd_obj(&d->do_lu);
 269 }
 270
 271 static struct lu_device *osd2lu_dev(struct osd_device *osd)
 272 {
 273         return &osd->od_dt_dev.dd_lu_dev;
 274 }
 275
 276 static journal_t *osd_journal(const struct osd_device *dev)
 277 {
 278         return LDISKFS_SB(osd_sb(dev))->s_journal;
 279 }
 280
 281 static int osd_has_index(const struct osd_object *obj)
 282 {
 283         return obj->oo_dt.do_index_ops != NULL;
 284 }
 285
 286 static int osd_object_invariant(const struct lu_object *l)
 287 {
 288         return osd_invariant(osd_obj(l));
 289 }
 290
 291 #ifdef HAVE_QUOTA_SUPPORT
 292 static inline void
 293 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
 294 {
 295         struct md_ucred    *uc = md_ucred(env);
 296         struct cred        *tc;
 297
 298         LASSERT(uc != NULL);
 299
 300         save->oc_uid = current_fsuid();
 301         save->oc_gid = current_fsgid();
 302         save->oc_cap = current_cap();
 303         if ((tc = prepare_creds())) {
 304                 tc->fsuid         = uc->mu_fsuid;
 305                 tc->fsgid         = uc->mu_fsgid;
 306                 commit_creds(tc);
 307         }
 308         /* XXX not suboptimal */
 309         cfs_curproc_cap_unpack(uc->mu_cap);
 310 }
 311
 312 static inline void
 313 osd_pop_ctxt(struct osd_ctxt *save)
 314 {
 315         struct cred *tc;
 316
 317         if ((tc = prepare_creds())) {
 318                 tc->fsuid         = save->oc_uid;
 319                 tc->fsgid         = save->oc_gid;
 320                 tc->cap_effective = save->oc_cap;
 321                 commit_creds(tc);
 322         }
 323 }
 324 #endif
 325
 326 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
 327 {
 328         return lu_context_key_get(&env->le_ctx, &osd_key);
 329 }
 330
 331 /*
 332  * Concurrency: doesn't matter
 333  */
 334 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 335 {
 336         return osd_oti_get(env)->oti_r_locks > 0;
 337 }
 338
 339 /*
 340  * Concurrency: doesn't matter
 341  */
 342 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 343 {
 344         struct osd_thread_info *oti = osd_oti_get(env);
 345         return oti->oti_w_locks > 0 && o->oo_owner == env;
 346 }
 347
 348 /*
 349  * Concurrency: doesn't access mutable data
 350  */
 351 static int osd_root_get(const struct lu_env *env,
 352                         struct dt_device *dev, struct lu_fid *f)
 353 {
 354         struct inode *inode;
 355
 356         inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
 357         LU_IGIF_BUILD(f, inode->i_ino, inode->i_generation);
 358         return 0;
 359 }
 360
 361 static inline int osd_qid_type(struct osd_thandle *oh, int i)
 362 {
 363         return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
 364 }
 365
 366 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
 367 {
 368         oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
 369 }
 370
 371 static void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
 372                             int type, uid_t id, struct inode *inode)
 373 {
 374 #ifdef CONFIG_QUOTA
 375         int i, allocated = 0;
 376         struct osd_object *obj;
 377
 378         LASSERT(dt != NULL);
 379         LASSERT(oh != NULL);
 380         LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
 381                  oh->ot_id_cnt);
 382
 383         /* id entry is allocated in the quota file */
 384         if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
 385                 allocated = 1;
 386
 387         for (i = 0; i < oh->ot_id_cnt; i++) {
 388                 if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
 389                         return;
 390         }
 391
 392         if (unlikely(i >= OSD_MAX_UGID_CNT)) {
 393                 CERROR("more than %d uid/gids for a transaction?\n", i);
 394                 return;
 395         }
 396
 397         oh->ot_id_array[i] = id;
 398         osd_qid_set_type(oh, i, type);
 399         oh->ot_id_cnt++;
 400         obj = osd_dt_obj(dt);
 401         oh->ot_credits += (allocated || id == 0) ?
 402                 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
 403 #endif
 404 }
 405
 406 /*
 407  * OSD object methods.
 408  */
 409
 410 /*
 411  * Concurrency: no concurrent access is possible that early in object
 412  * life-cycle.
 413  */
 414 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 415                                           const struct lu_object_header *hdr,
 416                                           struct lu_device *d)
 417 {
 418         struct osd_object *mo;
 419
 420         OBD_ALLOC_PTR(mo);
 421         if (mo != NULL) {
 422                 struct lu_object *l;
 423
 424                 l = &mo->oo_dt.do_lu;
 425                 dt_object_init(&mo->oo_dt, NULL, d);
 426                 if (osd_dev(d)->od_iop_mode)
 427                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 428                 else
 429                         mo->oo_dt.do_ops = &osd_obj_ops;
 430
 431                 l->lo_ops = &osd_lu_obj_ops;
 432                 cfs_init_rwsem(&mo->oo_sem);
 433                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 434                 cfs_spin_lock_init(&mo->oo_guard);
 435                 return l;
 436         } else {
 437                 return NULL;
 438         }
 439 }
 440
 441 /*
 442  * retrieve object from backend ext fs.
 443  **/
 444 static struct inode *osd_iget(struct osd_thread_info *info,
 445                               struct osd_device *dev,
 446                               const struct osd_inode_id *id)
 447 {
 448         struct inode *inode = NULL;
 449
 450 #ifdef HAVE_EXT4_LDISKFS
 451         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 452         if (IS_ERR(inode))
 453         /* Newer kernels return an error instead of a NULL pointer */
 454                 inode = NULL;
 455 #else
 456         inode = iget(osd_sb(dev), id->oii_ino);
 457 #endif
 458         if (inode == NULL) {
 459                 CERROR("no inode\n");
 460                 inode = ERR_PTR(-EACCES);
 461         } else if (id->oii_gen != OSD_OII_NOGEN &&
 462                    inode->i_generation != id->oii_gen) {
 463                 iput(inode);
 464                 inode = ERR_PTR(-ESTALE);
 465         } else if (inode->i_nlink == 0) {
 466                 /* due to parallel readdir and unlink,
 467                 * we can have dead inode here. */
 468                 CWARN("stale inode\n");
 469                 make_bad_inode(inode);
 470                 iput(inode);
 471                 inode = ERR_PTR(-ESTALE);
 472         } else if (is_bad_inode(inode)) {
 473                 CERROR("bad inode %lx\n",inode->i_ino);
 474                 iput(inode);
 475                 inode = ERR_PTR(-ENOENT);
 476         } else {
 477                 /* Do not update file c/mtime in ldiskfs.
 478                  * NB: we don't have any lock to protect this because we don't
 479                  * have reference on osd_object now, but contention with
 480                  * another lookup + attr_set can't happen in the tiny window
 481                  * between if (...) and set S_NOCMTIME. */
 482                 if (!(inode->i_flags & S_NOCMTIME))
 483                         inode->i_flags |= S_NOCMTIME;
 484         }
 485         return inode;
 486 }
 487
 488 static int osd_fid_lookup(const struct lu_env *env,
 489                           struct osd_object *obj, const struct lu_fid *fid)
 490 {
 491         struct osd_thread_info *info;
 492         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
 493         struct osd_device      *dev;
 494         struct osd_inode_id    *id;
 495         struct inode           *inode;
 496         int                     result;
 497
 498         LINVRNT(osd_invariant(obj));
 499         LASSERT(obj->oo_inode == NULL);
 500         LASSERTF(fid_is_sane(fid) || osd_fid_is_root(fid), DFID, PFID(fid));
 501         /*
 502          * This assertion checks that osd layer sees only local
 503          * fids. Unfortunately it is somewhat expensive (does a
 504          * cache-lookup). Disabling it for production/acceptance-testing.
 505          */
 506         LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
 507
 508         ENTRY;
 509
 510         info = osd_oti_get(env);
 511         dev  = osd_dev(ldev);
 512         id   = &info->oti_id;
 513
 514         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 515                 RETURN(-ENOENT);
 516
 517         result = osd_oi_lookup(info, osd_fid2oi(dev, fid), fid, id);
 518         if (result != 0) {
 519                 if (result == -ENOENT)
 520                         result = 0;
 521                 goto out;
 522         }
 523
 524         inode = osd_iget(info, dev, id);
 525         if (IS_ERR(inode)) {
 526                 /*
 527                  * If fid wasn't found in oi, inode-less object is
 528                  * created, for which lu_object_exists() returns
 529                  * false. This is used in a (frequent) case when
 530                  * objects are created as locking anchors or
 531                  * place holders for objects yet to be created.
 532                  */
 533                 result = PTR_ERR(inode);
 534                 goto out;
 535         }
 536
 537         obj->oo_inode = inode;
 538         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 539         if (dev->od_iop_mode) {
 540                 obj->oo_compat_dot_created = 1;
 541                 obj->oo_compat_dotdot_created = 1;
 542         }
 543
 544         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 545                 goto out;
 546
 547         LASSERT(obj->oo_hl_head == NULL);
 548         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 549         if (obj->oo_hl_head == NULL) {
 550                 obj->oo_inode = NULL;
 551                 iput(inode);
 552                 result = -ENOMEM;
 553         }
 554 out:
 555         LINVRNT(osd_invariant(obj));
 556
 557         RETURN(result);
 558 }
 559
 560 /*
 561  * Concurrency: shouldn't matter.
 562  */
 563 static void osd_object_init0(struct osd_object *obj)
 564 {
 565         LASSERT(obj->oo_inode != NULL);
 566         obj->oo_dt.do_body_ops = &osd_body_ops;
 567         obj->oo_dt.do_lu.lo_header->loh_attr |=
 568                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 569 }
 570
 571 /*
 572  * Concurrency: no concurrent access is possible that early in object
 573  * life-cycle.
 574  */
 575 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 576                            const struct lu_object_conf *unused)
 577 {
 578         struct osd_object *obj = osd_obj(l);
 579         int result;
 580
 581         LINVRNT(osd_invariant(obj));
 582
 583         result = osd_fid_lookup(env, obj, lu_object_fid(l));
 584         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 585         if (result == 0) {
 586                 if (obj->oo_inode != NULL)
 587                         osd_object_init0(obj);
 588         }
 589         LINVRNT(osd_invariant(obj));
 590         return result;
 591 }
 592
 593 /*
 594  * Concurrency: no concurrent access is possible that late in object
 595  * life-cycle.
 596  */
 597 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 598 {
 599         struct osd_object *obj = osd_obj(l);
 600
 601         LINVRNT(osd_invariant(obj));
 602
 603         dt_object_fini(&obj->oo_dt);
 604         if (obj->oo_hl_head != NULL)
 605                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 606         OBD_FREE_PTR(obj);
 607 }
 608
 609 /**
 610  * IAM Iterator
 611  */
 612 static struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
 613                                              const struct iam_container *bag)
 614 {
 615         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 616                                            osd_oti_get(env)->oti_it_ipd);
 617 }
 618
 619 static struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
 620                                               const struct iam_container *bag)
 621 {
 622         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 623                                            osd_oti_get(env)->oti_idx_ipd);
 624 }
 625
 626 static void osd_ipd_put(const struct lu_env *env,
 627                         const struct iam_container *bag,
 628                         struct iam_path_descr *ipd)
 629 {
 630         bag->ic_descr->id_ops->id_ipd_free(ipd);
 631 }
 632
 633 /*
 634  * Concurrency: no concurrent access is possible that late in object
 635  * life-cycle.
 636  */
 637 static void osd_index_fini(struct osd_object *o)
 638 {
 639         struct iam_container *bag;
 640
 641         if (o->oo_dir != NULL) {
 642                 bag = &o->oo_dir->od_container;
 643                 if (o->oo_inode != NULL) {
 644                         if (bag->ic_object == o->oo_inode)
 645                                 iam_container_fini(bag);
 646                 }
 647                 OBD_FREE_PTR(o->oo_dir);
 648                 o->oo_dir = NULL;
 649         }
 650 }
 651
 652 /*
 653  * Concurrency: no concurrent access is possible that late in object
 654  * life-cycle (for all existing callers, that is. New callers have to provide
 655  * their own locking.)
 656  */
 657 static int osd_inode_unlinked(const struct inode *inode)
 658 {
 659         return inode->i_nlink == 0;
 660 }
 661
 662 enum {
 663         OSD_TXN_OI_DELETE_CREDITS    = 20,
 664         OSD_TXN_INODE_DELETE_CREDITS = 20
 665 };
 666
 667 /*
 668  * Journal
 669  */
 670
 671 #if OSD_THANDLE_STATS
 672 /**
 673  * Set time when the handle is allocated
 674  */
 675 static void osd_th_alloced(struct osd_thandle *oth)
 676 {
 677         oth->oth_alloced = cfs_time_current();
 678 }
 679
 680 /**
 681  * Set time when the handle started
 682  */
 683 static void osd_th_started(struct osd_thandle *oth)
 684 {
 685         oth->oth_started = cfs_time_current();
 686 }
 687
 688 /**
 689  * Helper function to convert time interval to microseconds packed in
 690  * long int (default time units for the counter in "stats" initialized
 691  * by lu_time_init() )
 692  */
 693 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 694 {
 695         struct timeval val;
 696
 697         cfs_duration_usec(cfs_time_sub(end, start), &val);
 698         return val.tv_sec * 1000000 + val.tv_usec;
 699 }
 700
 701 /**
 702  * Check whether the we deal with this handle for too long.
 703  */
 704 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 705                                 cfs_time_t alloced, cfs_time_t started,
 706                                 cfs_time_t closed)
 707 {
 708         cfs_time_t now = cfs_time_current();
 709
 710         LASSERT(dev != NULL);
 711
 712         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 713                             interval_to_usec(alloced, started));
 714         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 715                             interval_to_usec(started, closed));
 716         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 717                             interval_to_usec(closed, now));
 718
 719         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 720                 CWARN("transaction handle %p was open for too long: "
 721                       "now "CFS_TIME_T" ,"
 722                       "alloced "CFS_TIME_T" ,"
 723                       "started "CFS_TIME_T" ,"
 724                       "closed "CFS_TIME_T"\n",
 725                       oth, now, alloced, started, closed);
 726                 libcfs_debug_dumpstack(NULL);
 727         }
 728 }
 729
 730 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 731 {                                                                       \
 732         cfs_time_t __closed = cfs_time_current();                       \
 733         cfs_time_t __alloced = oth->oth_alloced;                        \
 734         cfs_time_t __started = oth->oth_started;                        \
 735                                                                         \
 736         expr;                                                           \
 737         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 738 }
 739
 740 #else /* OSD_THANDLE_STATS */
 741
 742 #define osd_th_alloced(h)                  do {} while(0)
 743 #define osd_th_started(h)                  do {} while(0)
 744 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 745
 746 #endif /* OSD_THANDLE_STATS */
 747
 748 /*
 749  * Concurrency: doesn't access mutable data.
 750  */
 751 static int osd_param_is_sane(const struct osd_device *dev,
 752                              const struct thandle *th)
 753 {
 754         struct osd_thandle *oh;
 755         oh = container_of0(th, struct osd_thandle, ot_super);
 756         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 757 }
 758
 759 /*
 760  * Concurrency: shouldn't matter.
 761  */
 762 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 763 static void osd_trans_commit_cb(struct super_block *sb,
 764                                 struct journal_callback *jcb, int error)
 765 #else
 766 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 767 #endif
 768 {
 769         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 770         struct thandle     *th  = &oh->ot_super;
 771         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 772         struct dt_txn_commit_cb *dcb, *tmp;
 773
 774         LASSERT(oh->ot_handle == NULL);
 775
 776         if (error)
 777                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 778
 779         dt_txn_hook_commit(th);
 780
 781         /* call per-transaction callbacks if any */
 782         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage)
 783                 dcb->dcb_func(NULL, th, dcb, error);
 784
 785         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 786         lu_device_put(lud);
 787         th->th_dev = NULL;
 788
 789         lu_context_exit(&th->th_ctx);
 790         lu_context_fini(&th->th_ctx);
 791         OBD_FREE_PTR(oh);
 792 }
 793
 794 static struct thandle *osd_trans_create(const struct lu_env *env,
 795                                         struct dt_device *d)
 796 {
 797         struct osd_thread_info *oti = osd_oti_get(env);
 798         struct osd_thandle     *oh;
 799         struct thandle         *th;
 800         ENTRY;
 801
 802         th = ERR_PTR(-ENOMEM);
 803         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 804         if (oh != NULL) {
 805                 th = &oh->ot_super;
 806                 th->th_dev = d;
 807                 th->th_result = 0;
 808                 th->th_tags = LCT_TX_HANDLE;
 809                 oh->ot_credits = 0;
 810                 oti->oti_dev = osd_dt_dev(d);
 811                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 812                 osd_th_alloced(oh);
 813         }
 814         RETURN(th);
 815 }
 816
 817 /*
 818  * Concurrency: shouldn't matter.
 819  */
 820 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 821                     struct thandle *th)
 822 {
 823         struct osd_thread_info *oti = osd_oti_get(env);
 824         struct osd_device  *dev = osd_dt_dev(d);
 825         handle_t           *jh;
 826         struct osd_thandle *oh;
 827         int rc;
 828
 829         ENTRY;
 830
 831         LASSERT(current->journal_info == NULL);
 832
 833         oh = container_of0(th, struct osd_thandle, ot_super);
 834         LASSERT(oh != NULL);
 835         LASSERT(oh->ot_handle == NULL);
 836
 837         rc = dt_txn_hook_start(env, d, th);
 838         if (rc != 0)
 839                 GOTO(out, rc);
 840
 841         if (!osd_param_is_sane(dev, th)) {
 842                 CWARN("%s: too many transaction credits (%d > %d)\n",
 843                       d->dd_lu_dev.ld_obd->obd_name, oh->ot_credits,
 844                       osd_journal(dev)->j_max_transaction_buffers);
 845                 /* XXX Limit the credits to 'max_transaction_buffers', and
 846                  *     let the underlying filesystem to catch the error if
 847                  *     we really need so many credits.
 848                  *
 849                  *     This should be removed when we can calculate the
 850                  *     credits precisely. */
 851                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 852 #ifdef OSD_TRACK_DECLARES
 853                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 854                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 855                        oh->ot_declare_xattr_set);
 856                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 857                        oh->ot_declare_create, oh->ot_declare_ref_add,
 858                        oh->ot_declare_ref_del, oh->ot_declare_write);
 859                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 860                        oh->ot_declare_insert, oh->ot_declare_delete,
 861                        oh->ot_declare_destroy);
 862 #endif
 863         }
 864
 865         /*
 866          * XXX temporary stuff. Some abstraction layer should
 867          * be used.
 868          */
 869         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 870         osd_th_started(oh);
 871         if (!IS_ERR(jh)) {
 872                 oh->ot_handle = jh;
 873                 LASSERT(oti->oti_txns == 0);
 874                 lu_context_init(&th->th_ctx, th->th_tags);
 875                 lu_context_enter(&th->th_ctx);
 876
 877                 lu_device_get(&d->dd_lu_dev);
 878                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 879                                              "osd-tx", th);
 880
 881                 /*
 882                  * XXX: current rule is that we first start tx,
 883                  *      then lock object(s), but we can't use
 884                  *      this rule for data (due to locking specifics
 885                  *      in ldiskfs). also in long-term we'd like to
 886                  *      use usually-used (locks;tx) ordering. so,
 887                  *      UGLY thing is that we'll use one ordering for
 888                  *      data (ofd) and reverse ordering for metadata
 889                  *      (mdd). then at some point we'll fix the latter
 890                  */
 891                 if (lu_device_is_md(&d->dd_lu_dev)) {
 892                         LASSERT(oti->oti_r_locks == 0);
 893                         LASSERT(oti->oti_w_locks == 0);
 894                 }
 895
 896                 oti->oti_txns++;
 897                 rc = 0;
 898         } else {
 899                 rc = PTR_ERR(jh);
 900         }
 901 out:
 902         RETURN(rc);
 903 }
 904
 905 /*
 906  * Concurrency: shouldn't matter.
 907  */
 908 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 909 {
 910         int                     rc = 0;
 911         struct osd_thandle     *oh;
 912         struct osd_thread_info *oti = osd_oti_get(env);
 913
 914         ENTRY;
 915
 916         oh = container_of0(th, struct osd_thandle, ot_super);
 917
 918         if (oh->ot_handle != NULL) {
 919                 handle_t *hdl = oh->ot_handle;
 920
 921                 hdl->h_sync = th->th_sync;
 922
 923                 /*
 924                  * add commit callback
 925                  * notice we don't do this in osd_trans_start()
 926                  * as underlying transaction can change during truncate
 927                  */
 928                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 929                                          &oh->ot_jcb);
 930
 931                 LASSERT(oti->oti_txns == 1);
 932                 oti->oti_txns--;
 933                 /*
 934                  * XXX: current rule is that we first start tx,
 935                  *      then lock object(s), but we can't use
 936                  *      this rule for data (due to locking specifics
 937                  *      in ldiskfs). also in long-term we'd like to
 938                  *      use usually-used (locks;tx) ordering. so,
 939                  *      UGLY thing is that we'll use one ordering for
 940                  *      data (ofd) and reverse ordering for metadata
 941                  *      (mdd). then at some point we'll fix the latter
 942                  */
 943                 if (lu_device_is_md(&th->th_dev->dd_lu_dev)) {
 944                         LASSERT(oti->oti_r_locks == 0);
 945                         LASSERT(oti->oti_w_locks == 0);
 946                 }
 947                 rc = dt_txn_hook_stop(env, th);
 948                 if (rc != 0)
 949                         CERROR("Failure in transaction hook: %d\n", rc);
 950                 oh->ot_handle = NULL;
 951                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 952                                   rc = ldiskfs_journal_stop(hdl));
 953                 if (rc != 0)
 954                         CERROR("Failure to stop transaction: %d\n", rc);
 955         } else {
 956                 OBD_FREE_PTR(oh);
 957         }
 958
 959         RETURN(rc);
 960 }
 961
 962 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 963 {
 964         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 965                                                ot_super);
 966
 967         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 968
 969         return 0;
 970 }
 971
 972 /*
 973  * Called just before object is freed. Releases all resources except for
 974  * object itself (that is released by osd_object_free()).
 975  *
 976  * Concurrency: no concurrent access is possible that late in object
 977  * life-cycle.
 978  */
 979 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 980 {
 981         struct osd_object *obj   = osd_obj(l);
 982         struct inode      *inode = obj->oo_inode;
 983
 984         LINVRNT(osd_invariant(obj));
 985
 986         /*
 987          * If object is unlinked remove fid->ino mapping from object index.
 988          */
 989
 990         osd_index_fini(obj);
 991         if (inode != NULL) {
 992                 iput(inode);
 993                 obj->oo_inode = NULL;
 994         }
 995 }
 996
 997 /*
 998  * Concurrency: ->loo_object_release() is called under site spin-lock.
 999  */
1000 static void osd_object_release(const struct lu_env *env,
1001                                struct lu_object *l)
1002 {
1003 }
1004
1005 /*
1006  * Concurrency: shouldn't matter.
1007  */
1008 static int osd_object_print(const struct lu_env *env, void *cookie,
1009                             lu_printer_t p, const struct lu_object *l)
1010 {
1011         struct osd_object *o = osd_obj(l);
1012         struct iam_descr  *d;
1013
1014         if (o->oo_dir != NULL)
1015                 d = o->oo_dir->od_container.ic_descr;
1016         else
1017                 d = NULL;
1018         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1019                     o, o->oo_inode,
1020                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1021                     o->oo_inode ? o->oo_inode->i_generation : 0,
1022                     d ? d->id_ops->id_name : "plain");
1023 }
1024
1025 /*
1026  * Concurrency: shouldn't matter.
1027  */
1028 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1029                cfs_kstatfs_t *sfs)
1030 {
1031         struct osd_device *osd = osd_dt_dev(d);
1032         struct super_block *sb = osd_sb(osd);
1033         int result = 0;
1034
1035         cfs_spin_lock(&osd->od_osfs_lock);
1036         /* cache 1 second */
1037         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
1038                 result = ll_do_statfs(sb, &osd->od_kstatfs);
1039                 if (likely(result == 0)) /* N.B. statfs can't really fail */
1040                         osd->od_osfs_age = cfs_time_current_64();
1041         }
1042
1043         if (likely(result == 0))
1044                 *sfs = osd->od_kstatfs;
1045         cfs_spin_unlock(&osd->od_osfs_lock);
1046
1047         return result;
1048 }
1049
1050 /*
1051  * Concurrency: doesn't access mutable data.
1052  */
1053 static void osd_conf_get(const struct lu_env *env,
1054                          const struct dt_device *dev,
1055                          struct dt_device_param *param)
1056 {
1057         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1058
1059         /*
1060          * XXX should be taken from not-yet-existing fs abstraction layer.
1061          */
1062         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1063         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1064         param->ddp_block_shift  = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
1065         param->ddp_mntopts      = 0;
1066         if (test_opt(sb, XATTR_USER))
1067                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1068         if (test_opt(sb, POSIX_ACL))
1069                 param->ddp_mntopts |= MNTOPT_ACL;
1070
1071 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1072         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1073                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1074         else
1075 #endif
1076                 param->ddp_max_ea_size = sb->s_blocksize;
1077
1078 }
1079
1080 /**
1081  * Helper function to get and fill the buffer with input values.
1082  */
1083 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1084 {
1085         struct lu_buf *buf;
1086
1087         buf = &osd_oti_get(env)->oti_buf;
1088         buf->lb_buf = area;
1089         buf->lb_len = len;
1090         return buf;
1091 }
1092
1093 /*
1094  * Concurrency: shouldn't matter.
1095  */
1096 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1097 {
1098         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
1099         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1100 }
1101
1102 /**
1103  * Start commit for OSD device.
1104  *
1105  * An implementation of dt_commit_async method for OSD device.
1106  * Asychronously starts underlayng fs sync and thereby a transaction
1107  * commit.
1108  *
1109  * \param env environment
1110  * \param d dt device
1111  *
1112  * \see dt_device_operations
1113  */
1114 static int osd_commit_async(const struct lu_env *env,
1115                             struct dt_device *d)
1116 {
1117         struct super_block *s = osd_sb(osd_dt_dev(d));
1118         ENTRY;
1119
1120         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
1121         RETURN(s->s_op->sync_fs(s, 0));
1122 }
1123
1124 /*
1125  * Concurrency: shouldn't matter.
1126  */
1127
1128 static void osd_ro(const struct lu_env *env, struct dt_device *d)
1129 {
1130         struct super_block *sb = osd_sb(osd_dt_dev(d));
1131         ENTRY;
1132
1133         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
1134
1135         __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1136         EXIT;
1137 }
1138
1139 /*
1140  * Concurrency: serialization provided by callers.
1141  */
1142 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1143                               int mode, unsigned long timeout, __u32 alg,
1144                               struct lustre_capa_key *keys)
1145 {
1146         struct osd_device *dev = osd_dt_dev(d);
1147         ENTRY;
1148
1149         dev->od_fl_capa = mode;
1150         dev->od_capa_timeout = timeout;
1151         dev->od_capa_alg = alg;
1152         dev->od_capa_keys = keys;
1153         RETURN(0);
1154 }
1155
1156 /**
1157  * Concurrency: serialization provided by callers.
1158  */
1159 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1160                                struct dt_quota_ctxt *ctxt, void *data)
1161 {
1162         struct obd_device *obd = (void *)ctxt;
1163         struct vfsmount *mnt = (struct vfsmount *)data;
1164         ENTRY;
1165
1166         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1167         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1168         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1169         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1170         obd->obd_lvfs_ctxt.fs = get_ds();
1171
1172         EXIT;
1173 }
1174
1175 /**
1176  * Note: we do not count into QUOTA here.
1177  * If we mount with --data_journal we may need more.
1178  */
1179 static const int osd_dto_credits_noquota[DTO_NR] = {
1180         /**
1181          * Insert/Delete.
1182          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1183          * SINGLEDATA_TRANS_BLOCKS(8)
1184          * XXX Note: maybe iam need more, since iam have more level than
1185          *           EXT3 htree.
1186          */
1187         [DTO_INDEX_INSERT]  = 16,
1188         [DTO_INDEX_DELETE]  = 16,
1189         /**
1190          * Unused now
1191          */
1192         [DTO_INDEX_UPDATE]  = 16,
1193         /**
1194          * Create a object. The same as create object in EXT3.
1195          * DATA_TRANS_BLOCKS(14) +
1196          * INDEX_EXTRA_BLOCKS(8) +
1197          * 3(inode bits, groups, GDT)
1198          */
1199         [DTO_OBJECT_CREATE] = 25,
1200         /**
1201          * XXX: real credits to be fixed
1202          */
1203         [DTO_OBJECT_DELETE] = 25,
1204         /**
1205          * Attr set credits (inode)
1206          */
1207         [DTO_ATTR_SET_BASE] = 1,
1208         /**
1209          * Xattr set. The same as xattr of EXT3.
1210          * DATA_TRANS_BLOCKS(14)
1211          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1212          * are also counted in. Do not know why?
1213          */
1214         [DTO_XATTR_SET]     = 14,
1215         [DTO_LOG_REC]       = 14,
1216         /**
1217          * credits for inode change during write.
1218          */
1219         [DTO_WRITE_BASE]    = 3,
1220         /**
1221          * credits for single block write.
1222          */
1223         [DTO_WRITE_BLOCK]   = 14,
1224         /**
1225          * Attr set credits for chown.
1226          * This is extra credits for setattr, and it is null without quota
1227          */
1228         [DTO_ATTR_SET_CHOWN]= 0
1229 };
1230
1231 static const struct dt_device_operations osd_dt_ops = {
1232         .dt_root_get       = osd_root_get,
1233         .dt_statfs         = osd_statfs,
1234         .dt_trans_create   = osd_trans_create,
1235         .dt_trans_start    = osd_trans_start,
1236         .dt_trans_stop     = osd_trans_stop,
1237         .dt_trans_cb_add   = osd_trans_cb_add,
1238         .dt_conf_get       = osd_conf_get,
1239         .dt_sync           = osd_sync,
1240         .dt_ro             = osd_ro,
1241         .dt_commit_async   = osd_commit_async,
1242         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1243         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1244 };
1245
1246 static void osd_object_read_lock(const struct lu_env *env,
1247                                  struct dt_object *dt, unsigned role)
1248 {
1249         struct osd_object *obj = osd_dt_obj(dt);
1250         struct osd_thread_info *oti = osd_oti_get(env);
1251
1252         LINVRNT(osd_invariant(obj));
1253
1254         LASSERT(obj->oo_owner != env);
1255         cfs_down_read_nested(&obj->oo_sem, role);
1256
1257         LASSERT(obj->oo_owner == NULL);
1258         oti->oti_r_locks++;
1259 }
1260
1261 static void osd_object_write_lock(const struct lu_env *env,
1262                                   struct dt_object *dt, unsigned role)
1263 {
1264         struct osd_object *obj = osd_dt_obj(dt);
1265         struct osd_thread_info *oti = osd_oti_get(env);
1266
1267         LINVRNT(osd_invariant(obj));
1268
1269         LASSERT(obj->oo_owner != env);
1270         cfs_down_write_nested(&obj->oo_sem, role);
1271
1272         LASSERT(obj->oo_owner == NULL);
1273         obj->oo_owner = env;
1274         oti->oti_w_locks++;
1275 }
1276
1277 static void osd_object_read_unlock(const struct lu_env *env,
1278                                    struct dt_object *dt)
1279 {
1280         struct osd_object *obj = osd_dt_obj(dt);
1281         struct osd_thread_info *oti = osd_oti_get(env);
1282
1283         LINVRNT(osd_invariant(obj));
1284
1285         LASSERT(oti->oti_r_locks > 0);
1286         oti->oti_r_locks--;
1287         cfs_up_read(&obj->oo_sem);
1288 }
1289
1290 static void osd_object_write_unlock(const struct lu_env *env,
1291                                     struct dt_object *dt)
1292 {
1293         struct osd_object *obj = osd_dt_obj(dt);
1294         struct osd_thread_info *oti = osd_oti_get(env);
1295
1296         LINVRNT(osd_invariant(obj));
1297
1298         LASSERT(obj->oo_owner == env);
1299         LASSERT(oti->oti_w_locks > 0);
1300         oti->oti_w_locks--;
1301         obj->oo_owner = NULL;
1302         cfs_up_write(&obj->oo_sem);
1303 }
1304
1305 static int osd_object_write_locked(const struct lu_env *env,
1306                                    struct dt_object *dt)
1307 {
1308         struct osd_object *obj = osd_dt_obj(dt);
1309
1310         LINVRNT(osd_invariant(obj));
1311
1312         return obj->oo_owner == env;
1313 }
1314
1315 static int capa_is_sane(const struct lu_env *env,
1316                         struct osd_device *dev,
1317                         struct lustre_capa *capa,
1318                         struct lustre_capa_key *keys)
1319 {
1320         struct osd_thread_info *oti = osd_oti_get(env);
1321         struct lustre_capa *tcapa = &oti->oti_capa;
1322         struct obd_capa *oc;
1323         int i, rc = 0;
1324         ENTRY;
1325
1326         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1327         if (oc) {
1328                 if (capa_is_expired(oc)) {
1329                         DEBUG_CAPA(D_ERROR, capa, "expired");
1330                         rc = -ESTALE;
1331                 }
1332                 capa_put(oc);
1333                 RETURN(rc);
1334         }
1335
1336         if (capa_is_expired_sec(capa)) {
1337                 DEBUG_CAPA(D_ERROR, capa, "expired");
1338                 RETURN(-ESTALE);
1339         }
1340
1341         cfs_spin_lock(&capa_lock);
1342         for (i = 0; i < 2; i++) {
1343                 if (keys[i].lk_keyid == capa->lc_keyid) {
1344                         oti->oti_capa_key = keys[i];
1345                         break;
1346                 }
1347         }
1348         cfs_spin_unlock(&capa_lock);
1349
1350         if (i == 2) {
1351                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1352                 RETURN(-ESTALE);
1353         }
1354
1355         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1356         if (rc)
1357                 RETURN(rc);
1358
1359         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1360                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1361                 RETURN(-EACCES);
1362         }
1363
1364         oc = capa_add(dev->od_capa_hash, capa);
1365         capa_put(oc);
1366
1367         RETURN(0);
1368 }
1369
1370 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1371                            struct lustre_capa *capa, __u64 opc)
1372 {
1373         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1374         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1375         struct md_capainfo *ci;
1376         int rc;
1377
1378         if (!dev->od_fl_capa)
1379                 return 0;
1380
1381         if (capa == BYPASS_CAPA)
1382                 return 0;
1383
1384         ci = md_capainfo(env);
1385         if (unlikely(!ci))
1386                 return 0;
1387
1388         if (ci->mc_auth == LC_ID_NONE)
1389                 return 0;
1390
1391         if (!capa) {
1392                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1393                 return -EACCES;
1394         }
1395
1396         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1397                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1398                            PFID(fid));
1399                 return -EACCES;
1400         }
1401
1402         if (!capa_opc_supported(capa, opc)) {
1403                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1404                 return -EACCES;
1405         }
1406
1407         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1408                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1409                 return -EACCES;
1410         }
1411
1412         return 0;
1413 }
1414
1415 static struct timespec *osd_inode_time(const struct lu_env *env,
1416                                        struct inode *inode, __u64 seconds)
1417 {
1418         struct osd_thread_info *oti = osd_oti_get(env);
1419         struct timespec        *t   = &oti->oti_time;
1420
1421         t->tv_sec  = seconds;
1422         t->tv_nsec = 0;
1423         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1424         return t;
1425 }
1426
1427
1428 static void osd_inode_getattr(const struct lu_env *env,
1429                               struct inode *inode, struct lu_attr *attr)
1430 {
1431         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1432                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1433                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1434
1435         attr->la_atime      = LTIME_S(inode->i_atime);
1436         attr->la_mtime      = LTIME_S(inode->i_mtime);
1437         attr->la_ctime      = LTIME_S(inode->i_ctime);
1438         attr->la_mode       = inode->i_mode;
1439         attr->la_size       = i_size_read(inode);
1440         attr->la_blocks     = inode->i_blocks;
1441         attr->la_uid        = inode->i_uid;
1442         attr->la_gid        = inode->i_gid;
1443         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1444         attr->la_nlink      = inode->i_nlink;
1445         attr->la_rdev       = inode->i_rdev;
1446         attr->la_blksize    = ll_inode_blksize(inode);
1447         attr->la_blkbits    = inode->i_blkbits;
1448 }
1449
1450 static int osd_attr_get(const struct lu_env *env,
1451                         struct dt_object *dt,
1452                         struct lu_attr *attr,
1453                         struct lustre_capa *capa)
1454 {
1455         struct osd_object *obj = osd_dt_obj(dt);
1456
1457         LASSERT(dt_object_exists(dt));
1458         LINVRNT(osd_invariant(obj));
1459
1460         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1461                 return -EACCES;
1462
1463         cfs_spin_lock(&obj->oo_guard);
1464         osd_inode_getattr(env, obj->oo_inode, attr);
1465         cfs_spin_unlock(&obj->oo_guard);
1466         return 0;
1467 }
1468
1469 static int osd_declare_attr_set(const struct lu_env *env,
1470                                 struct dt_object *dt,
1471                                 const struct lu_attr *attr,
1472                                 struct thandle *handle)
1473 {
1474         struct osd_thandle *oh;
1475         struct osd_object *obj;
1476
1477         LASSERT(dt != NULL);
1478         LASSERT(handle != NULL);
1479
1480         obj = osd_dt_obj(dt);
1481         LASSERT(osd_invariant(obj));
1482
1483         oh = container_of0(handle, struct osd_thandle, ot_super);
1484         LASSERT(oh->ot_handle == NULL);
1485
1486         OSD_DECLARE_OP(oh, attr_set);
1487         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1488
1489         if (attr && attr->la_valid & LA_UID) {
1490                 if (obj->oo_inode)
1491                         osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
1492                                         obj->oo_inode);
1493                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1494         }
1495         if (attr && attr->la_valid & LA_GID) {
1496                 if (obj->oo_inode)
1497                         osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
1498                                         obj->oo_inode);
1499                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1500         }
1501
1502         return 0;
1503 }
1504
1505 static int osd_inode_setattr(const struct lu_env *env,
1506                              struct inode *inode, const struct lu_attr *attr)
1507 {
1508         __u64 bits;
1509
1510         bits = attr->la_valid;
1511
1512         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1513
1514 #ifdef HAVE_QUOTA_SUPPORT
1515         if ((bits & LA_UID && attr->la_uid != inode->i_uid) ||
1516             (bits & LA_GID && attr->la_gid != inode->i_gid)) {
1517                 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1518                 struct iattr iattr;
1519                 int rc;
1520
1521                 iattr.ia_valid = 0;
1522                 if (bits & LA_UID)
1523                         iattr.ia_valid |= ATTR_UID;
1524                 if (bits & LA_GID)
1525                         iattr.ia_valid |= ATTR_GID;
1526                 iattr.ia_uid = attr->la_uid;
1527                 iattr.ia_gid = attr->la_gid;
1528                 osd_push_ctxt(env, save);
1529                 rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1530                 osd_pop_ctxt(save);
1531                 if (rc != 0)
1532                         return rc;
1533         }
1534 #endif
1535
1536         if (bits & LA_ATIME)
1537                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1538         if (bits & LA_CTIME)
1539                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1540         if (bits & LA_MTIME)
1541                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1542         if (bits & LA_SIZE) {
1543                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1544                 i_size_write(inode, attr->la_size);
1545         }
1546
1547 #if 0
1548         /* OSD should not change "i_blocks" which is used by quota.
1549          * "i_blocks" should be changed by ldiskfs only. */
1550         if (bits & LA_BLOCKS)
1551                 inode->i_blocks = attr->la_blocks;
1552 #endif
1553         if (bits & LA_MODE)
1554                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1555                         (attr->la_mode & ~S_IFMT);
1556         if (bits & LA_UID)
1557                 inode->i_uid    = attr->la_uid;
1558         if (bits & LA_GID)
1559                 inode->i_gid    = attr->la_gid;
1560         if (bits & LA_NLINK)
1561                 inode->i_nlink  = attr->la_nlink;
1562         if (bits & LA_RDEV)
1563                 inode->i_rdev   = attr->la_rdev;
1564
1565         if (bits & LA_FLAGS) {
1566                 /* always keep S_NOCMTIME */
1567                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1568                                  S_NOCMTIME;
1569         }
1570         return 0;
1571 }
1572
1573 static int osd_attr_set(const struct lu_env *env,
1574                         struct dt_object *dt,
1575                         const struct lu_attr *attr,
1576                         struct thandle *handle,
1577                         struct lustre_capa *capa)
1578 {
1579         struct osd_object *obj = osd_dt_obj(dt);
1580         int rc;
1581
1582         LASSERT(handle != NULL);
1583         LASSERT(dt_object_exists(dt));
1584         LASSERT(osd_invariant(obj));
1585
1586         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1587                 return -EACCES;
1588
1589         OSD_EXEC_OP(handle, attr_set);
1590
1591         cfs_spin_lock(&obj->oo_guard);
1592         rc = osd_inode_setattr(env, obj->oo_inode, attr);
1593         cfs_spin_unlock(&obj->oo_guard);
1594
1595         if (!rc)
1596                 obj->oo_inode->i_sb->s_op->dirty_inode(obj->oo_inode);
1597         return rc;
1598 }
1599
1600 /*
1601  * Object creation.
1602  *
1603  * XXX temporary solution.
1604  */
1605 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1606                           struct lu_attr *attr, struct thandle *th)
1607 {
1608         return 0;
1609 }
1610
1611 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1612                            struct lu_attr *attr, struct thandle *th)
1613 {
1614         osd_object_init0(obj);
1615         if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1616                 unlock_new_inode(obj->oo_inode);
1617         return 0;
1618 }
1619
1620 static struct dentry * osd_child_dentry_get(const struct lu_env *env,
1621                                             struct osd_object *obj,
1622                                             const char *name,
1623                                             const int namelen)
1624 {
1625         struct osd_thread_info *info   = osd_oti_get(env);
1626         struct dentry *child_dentry = &info->oti_child_dentry;
1627         struct dentry *obj_dentry = &info->oti_obj_dentry;
1628
1629         obj_dentry->d_inode = obj->oo_inode;
1630         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
1631         obj_dentry->d_name.hash = 0;
1632
1633         child_dentry->d_name.hash = 0;
1634         child_dentry->d_parent = obj_dentry;
1635         child_dentry->d_name.name = name;
1636         child_dentry->d_name.len = namelen;
1637         return child_dentry;
1638 }
1639
1640
1641 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1642                       cfs_umode_t mode,
1643                       struct dt_allocation_hint *hint,
1644                       struct thandle *th)
1645 {
1646         int result;
1647         struct osd_device  *osd = osd_obj2dev(obj);
1648         struct osd_thandle *oth;
1649         struct dt_object   *parent;
1650         struct inode       *inode;
1651 #ifdef HAVE_QUOTA_SUPPORT
1652         struct osd_ctxt    *save = &info->oti_ctxt;
1653 #endif
1654
1655         LINVRNT(osd_invariant(obj));
1656         LASSERT(obj->oo_inode == NULL);
1657         LASSERT(obj->oo_hl_head == NULL);
1658
1659         if (S_ISDIR(mode) && ldiskfs_pdo) {
1660                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1661                 if (obj->oo_hl_head == NULL)
1662                         return -ENOMEM;
1663         }
1664
1665         oth = container_of(th, struct osd_thandle, ot_super);
1666         LASSERT(oth->ot_handle->h_transaction != NULL);
1667
1668         if (hint && hint->dah_parent)
1669                 parent = hint->dah_parent;
1670         else
1671                 parent = osd->od_obj_area;
1672
1673         LASSERT(parent != NULL);
1674         LASSERT(osd_dt_obj(parent)->oo_inode->i_op != NULL);
1675
1676 #ifdef HAVE_QUOTA_SUPPORT
1677         osd_push_ctxt(info->oti_env, save);
1678 #endif
1679         inode = ldiskfs_create_inode(oth->ot_handle,
1680                                      osd_dt_obj(parent)->oo_inode, mode);
1681 #ifdef HAVE_QUOTA_SUPPORT
1682         osd_pop_ctxt(save);
1683 #endif
1684         if (!IS_ERR(inode)) {
1685                 /* Do not update file c/mtime in ldiskfs.
1686                  * NB: don't need any lock because no contention at this
1687                  * early stage */
1688                 inode->i_flags |= S_NOCMTIME;
1689                 obj->oo_inode = inode;
1690                 result = 0;
1691         } else {
1692                 if (obj->oo_hl_head != NULL) {
1693                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1694                         obj->oo_hl_head = NULL;
1695                 }
1696                 result = PTR_ERR(inode);
1697         }
1698         LINVRNT(osd_invariant(obj));
1699         return result;
1700 }
1701
1702 enum {
1703         OSD_NAME_LEN = 255
1704 };
1705
1706 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1707                      struct lu_attr *attr,
1708                      struct dt_allocation_hint *hint,
1709                      struct dt_object_format *dof,
1710                      struct thandle *th)
1711 {
1712         int result;
1713         struct osd_thandle *oth;
1714         struct osd_device *osd = osd_obj2dev(obj);
1715         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1716
1717         LASSERT(S_ISDIR(attr->la_mode));
1718
1719         oth = container_of(th, struct osd_thandle, ot_super);
1720         LASSERT(oth->ot_handle->h_transaction != NULL);
1721         result = osd_mkfile(info, obj, mode, hint, th);
1722         if (result == 0 && osd->od_iop_mode == 0) {
1723                 LASSERT(obj->oo_inode != NULL);
1724                 /*
1725                  * XXX uh-oh... call low-level iam function directly.
1726                  */
1727
1728                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1729                                          sizeof (struct osd_fid_pack),
1730                                          oth->ot_handle);
1731         }
1732         return result;
1733 }
1734
1735 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1736                         struct lu_attr *attr,
1737                         struct dt_allocation_hint *hint,
1738                         struct dt_object_format *dof,
1739                         struct thandle *th)
1740 {
1741         int result;
1742         struct osd_thandle *oth;
1743         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1744
1745         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1746
1747         LASSERT(S_ISREG(attr->la_mode));
1748
1749         oth = container_of(th, struct osd_thandle, ot_super);
1750         LASSERT(oth->ot_handle->h_transaction != NULL);
1751
1752         result = osd_mkfile(info, obj, mode, hint, th);
1753         if (result == 0) {
1754                 LASSERT(obj->oo_inode != NULL);
1755                 if (feat->dif_flags & DT_IND_VARKEY)
1756                         result = iam_lvar_create(obj->oo_inode,
1757                                                  feat->dif_keysize_max,
1758                                                  feat->dif_ptrsize,
1759                                                  feat->dif_recsize_max,
1760                                                  oth->ot_handle);
1761                 else
1762                         result = iam_lfix_create(obj->oo_inode,
1763                                                  feat->dif_keysize_max,
1764                                                  feat->dif_ptrsize,
1765                                                  feat->dif_recsize_max,
1766                                                  oth->ot_handle);
1767
1768         }
1769         return result;
1770 }
1771
1772 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1773                      struct lu_attr *attr,
1774                      struct dt_allocation_hint *hint,
1775                      struct dt_object_format *dof,
1776                      struct thandle *th)
1777 {
1778         LASSERT(S_ISREG(attr->la_mode));
1779         return osd_mkfile(info, obj, (attr->la_mode &
1780                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1781 }
1782
1783 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1784                      struct lu_attr *attr,
1785                      struct dt_allocation_hint *hint,
1786                      struct dt_object_format *dof,
1787                      struct thandle *th)
1788 {
1789         LASSERT(S_ISLNK(attr->la_mode));
1790         return osd_mkfile(info, obj, (attr->la_mode &
1791                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1792 }
1793
1794 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1795                      struct lu_attr *attr,
1796                      struct dt_allocation_hint *hint,
1797                      struct dt_object_format *dof,
1798                      struct thandle *th)
1799 {
1800         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1801         int result;
1802
1803         LINVRNT(osd_invariant(obj));
1804         LASSERT(obj->oo_inode == NULL);
1805         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1806                 S_ISFIFO(mode) || S_ISSOCK(mode));
1807
1808         result = osd_mkfile(info, obj, mode, hint, th);
1809         if (result == 0) {
1810                 LASSERT(obj->oo_inode != NULL);
1811                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1812         }
1813         LINVRNT(osd_invariant(obj));
1814         return result;
1815 }
1816
1817 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1818                               struct lu_attr *,
1819                               struct dt_allocation_hint *hint,
1820                               struct dt_object_format *dof,
1821                               struct thandle *);
1822
1823 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1824 {
1825         osd_obj_type_f result;
1826
1827         switch (type) {
1828         case DFT_DIR:
1829                 result = osd_mkdir;
1830                 break;
1831         case DFT_REGULAR:
1832                 result = osd_mkreg;
1833                 break;
1834         case DFT_SYM:
1835                 result = osd_mksym;
1836                 break;
1837         case DFT_NODE:
1838                 result = osd_mknod;
1839                 break;
1840         case DFT_INDEX:
1841                 result = osd_mk_index;
1842                 break;
1843
1844         default:
1845                 LBUG();
1846                 break;
1847         }
1848         return result;
1849 }
1850
1851
1852 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1853                         struct dt_object *parent, cfs_umode_t child_mode)
1854 {
1855         LASSERT(ah);
1856
1857         memset(ah, 0, sizeof(*ah));
1858         ah->dah_parent = parent;
1859         ah->dah_mode = child_mode;
1860 }
1861
1862 /**
1863  * Helper function for osd_object_create()
1864  *
1865  * \retval 0, on success
1866  */
1867 static int __osd_object_create(struct osd_thread_info *info,
1868                                struct osd_object *obj, struct lu_attr *attr,
1869                                struct dt_allocation_hint *hint,
1870                                struct dt_object_format *dof,
1871                                struct thandle *th)
1872 {
1873
1874         int result;
1875
1876         result = osd_create_pre(info, obj, attr, th);
1877         if (result == 0) {
1878                 result = osd_create_type_f(dof->dof_type)(info, obj,
1879                                            attr, hint, dof, th);
1880                 if (result == 0)
1881                         result = osd_create_post(info, obj, attr, th);
1882         }
1883         return result;
1884 }
1885
1886 /**
1887  * Helper function for osd_object_create()
1888  *
1889  * \retval 0, on success
1890  */
1891 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1892                            const struct lu_fid *fid, struct thandle *th)
1893 {
1894         struct osd_thread_info *info = osd_oti_get(env);
1895         struct osd_inode_id    *id   = &info->oti_id;
1896         struct osd_device      *osd  = osd_obj2dev(obj);
1897         struct md_ucred        *uc   = md_ucred(env);
1898
1899         LASSERT(obj->oo_inode != NULL);
1900         LASSERT(uc != NULL);
1901
1902         id->oii_ino = obj->oo_inode->i_ino;
1903         id->oii_gen = obj->oo_inode->i_generation;
1904
1905         return osd_oi_insert(info, osd_fid2oi(osd, fid), fid, id, th,
1906                              uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
1907 }
1908
1909 static int osd_declare_object_create(const struct lu_env *env,
1910                                      struct dt_object *dt,
1911                                      struct lu_attr *attr,
1912                                      struct dt_allocation_hint *hint,
1913                                      struct dt_object_format *dof,
1914                                      struct thandle *handle)
1915 {
1916         struct osd_thandle *oh;
1917
1918         LASSERT(handle != NULL);
1919
1920         oh = container_of0(handle, struct osd_thandle, ot_super);
1921         LASSERT(oh->ot_handle == NULL);
1922
1923         OSD_DECLARE_OP(oh, create);
1924         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1925         /* XXX: So far, only normal fid needs be inserted into the oi,
1926          *      things could be changed later. Revise following code then. */
1927         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1928                 OSD_DECLARE_OP(oh, insert);
1929                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1930         }
1931         /* If this is directory, then we expect . and .. to be inserted as
1932          * well. The one directory block always needs to be created for the
1933          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1934          * block), there is no danger of needing a tree for the first block.
1935          */
1936         if (attr && S_ISDIR(attr->la_mode)) {
1937                 OSD_DECLARE_OP(oh, insert);
1938                 OSD_DECLARE_OP(oh, insert);
1939                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1940         }
1941
1942         if (attr) {
1943                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1944                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1945         }
1946         return 0;
1947 }
1948
1949 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1950                              struct lu_attr *attr,
1951                              struct dt_allocation_hint *hint,
1952                              struct dt_object_format *dof,
1953                              struct thandle *th)
1954 {
1955         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1956         struct osd_object      *obj    = osd_dt_obj(dt);
1957         struct osd_thread_info *info   = osd_oti_get(env);
1958         int result;
1959
1960         ENTRY;
1961
1962         LINVRNT(osd_invariant(obj));
1963         LASSERT(!dt_object_exists(dt));
1964         LASSERT(osd_write_locked(env, obj));
1965         LASSERT(th != NULL);
1966
1967         OSD_EXEC_OP(th, create);
1968
1969         result = __osd_object_create(info, obj, attr, hint, dof, th);
1970         if (result == 0)
1971                 result = __osd_oi_insert(env, obj, fid, th);
1972
1973         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1974         LASSERT(osd_invariant(obj));
1975         RETURN(result);
1976 }
1977
1978 /**
1979  * Called to destroy on-disk representation of the object
1980  *
1981  * Concurrency: must be locked
1982  */
1983 static int osd_declare_object_destroy(const struct lu_env *env,
1984                                       struct dt_object *dt,
1985                                       struct thandle *th)
1986 {
1987         struct osd_object      *obj = osd_dt_obj(dt);
1988         struct inode           *inode = obj->oo_inode;
1989         struct osd_thandle     *oh;
1990         ENTRY;
1991
1992         oh = container_of0(th, struct osd_thandle, ot_super);
1993         LASSERT(oh->ot_handle == NULL);
1994         LASSERT(inode);
1995
1996         OSD_DECLARE_OP(oh, destroy);
1997         OSD_DECLARE_OP(oh, delete);
1998         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
1999         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2000
2001         osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
2002         osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
2003
2004         RETURN(0);
2005 }
2006
2007 static int osd_object_destroy(const struct lu_env *env,
2008                               struct dt_object *dt,
2009                               struct thandle *th)
2010 {
2011         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2012         struct osd_object      *obj = osd_dt_obj(dt);
2013         struct inode           *inode = obj->oo_inode;
2014         struct osd_device      *osd = osd_obj2dev(obj);
2015         struct osd_thandle     *oh;
2016         int                     result;
2017         ENTRY;
2018
2019         oh = container_of0(th, struct osd_thandle, ot_super);
2020         LASSERT(oh->ot_handle);
2021         LASSERT(inode);
2022         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2023
2024         if (S_ISDIR(inode->i_mode)) {
2025                 LASSERT(osd_inode_unlinked(inode) ||
2026                         inode->i_nlink == 1);
2027                 cfs_spin_lock(&obj->oo_guard);
2028                 inode->i_nlink = 0;
2029                 cfs_spin_unlock(&obj->oo_guard);
2030                 inode->i_sb->s_op->dirty_inode(inode);
2031         } else {
2032                 LASSERT(osd_inode_unlinked(inode));
2033         }
2034
2035         OSD_EXEC_OP(th, destroy);
2036
2037         result = osd_oi_delete(osd_oti_get(env),
2038                                osd_fid2oi(osd, fid), fid, th);
2039
2040         /* XXX: add to ext3 orphan list */
2041         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2042
2043         /* not needed in the cache anymore */
2044         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2045
2046         RETURN(0);
2047 }
2048
2049 /**
2050  * Helper function for osd_xattr_set()
2051  */
2052 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2053                            const struct lu_buf *buf, const char *name, int fl)
2054 {
2055         struct osd_object      *obj      = osd_dt_obj(dt);
2056         struct inode           *inode    = obj->oo_inode;
2057         struct osd_thread_info *info     = osd_oti_get(env);
2058         struct dentry          *dentry   = &info->oti_child_dentry;
2059         int                     fs_flags = 0;
2060         int  rc;
2061
2062         LASSERT(dt_object_exists(dt));
2063         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2064         LASSERT(osd_write_locked(env, obj));
2065
2066         if (fl & LU_XATTR_REPLACE)
2067                 fs_flags |= XATTR_REPLACE;
2068
2069         if (fl & LU_XATTR_CREATE)
2070                 fs_flags |= XATTR_CREATE;
2071
2072         dentry->d_inode = inode;
2073         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2074                                    buf->lb_len, fs_flags);
2075         return rc;
2076 }
2077
2078 /**
2079  * Put the fid into lustre_mdt_attrs, and then place the structure
2080  * inode's ea. This fid should not be altered during the life time
2081  * of the inode.
2082  *
2083  * \retval +ve, on success
2084  * \retval -ve, on error
2085  *
2086  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2087  */
2088 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2089                           const struct lu_fid *fid)
2090 {
2091         struct osd_thread_info  *info      = osd_oti_get(env);
2092         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2093
2094         lustre_lma_init(mdt_attrs, fid);
2095         lustre_lma_swab(mdt_attrs);
2096         return __osd_xattr_set(env, dt,
2097                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2098                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2099
2100 }
2101
2102 /**
2103  * Helper function to form igif
2104  */
2105 static inline void osd_igif_get(const struct lu_env *env, struct inode  *inode,
2106                                 struct lu_fid *fid)
2107 {
2108         LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
2109 }
2110
2111 /**
2112  * Helper function to pack the fid, ldiskfs stores fid in packed format.
2113  */
2114 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
2115                   struct lu_fid *befider)
2116 {
2117         fid_cpu_to_be(befider, (struct lu_fid *)fid);
2118         memcpy(pack->fp_area, befider, sizeof(*befider));
2119         pack->fp_len =  sizeof(*befider) + 1;
2120 }
2121
2122 /**
2123  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2124  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2125  * To have compatilibility with 1.8 ldiskfs driver we need to have
2126  * magic number at start of fid data.
2127  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2128  * its inmemory API.
2129  */
2130 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2131                                   const struct dt_rec *fid)
2132 {
2133         param->edp_magic = LDISKFS_LUFID_MAGIC;
2134         param->edp_len =  sizeof(struct lu_fid) + 1;
2135
2136         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2137                       (struct lu_fid *)fid);
2138 }
2139
2140 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
2141 {
2142         int result;
2143
2144         result = 0;
2145         switch (pack->fp_len) {
2146         case sizeof *fid + 1:
2147                 memcpy(fid, pack->fp_area, sizeof *fid);
2148                 fid_be_to_cpu(fid, fid);
2149                 break;
2150         default:
2151                 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
2152                 result = -EIO;
2153         }
2154         return result;
2155 }
2156
2157 /**
2158  * Try to read the fid from inode ea into dt_rec, if return value
2159  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2160  *
2161  * \param fid object fid.
2162  *
2163  * \retval 0 on success
2164  */
2165 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2166                           __u32 ino, struct lu_fid *fid)
2167 {
2168         struct osd_thread_info  *info      = osd_oti_get(env);
2169         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2170         struct lu_device        *ldev   = obj->oo_dt.do_lu.lo_dev;
2171         struct dentry           *dentry = &info->oti_child_dentry;
2172         struct osd_inode_id     *id     = &info->oti_id;
2173         struct osd_device       *dev;
2174         struct inode            *inode;
2175         int                      rc;
2176
2177         ENTRY;
2178         dev  = osd_dev(ldev);
2179
2180         id->oii_ino = ino;
2181         id->oii_gen = OSD_OII_NOGEN;
2182
2183         inode = osd_iget(info, dev, id);
2184         if (IS_ERR(inode)) {
2185                 rc = PTR_ERR(inode);
2186                 GOTO(out,rc);
2187         }
2188         dentry->d_inode = inode;
2189
2190         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2191         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
2192                                    sizeof *mdt_attrs);
2193
2194         /* Check LMA compatibility */
2195         if (rc > 0 &&
2196             (mdt_attrs->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP))) {
2197                 CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
2198                       inode->i_ino, le32_to_cpu(mdt_attrs->lma_incompat) &
2199                       ~LMA_INCOMPAT_SUPP);
2200                 return -ENOSYS;
2201         }
2202
2203         if (rc > 0) {
2204                 lustre_lma_swab(mdt_attrs);
2205                 memcpy(fid, &mdt_attrs->lma_self_fid, sizeof(*fid));
2206                 rc = 0;
2207         } else if (rc == -ENODATA) {
2208                 osd_igif_get(env, inode, fid);
2209                 rc = 0;
2210         }
2211         iput(inode);
2212 out:
2213         RETURN(rc);
2214 }
2215
2216 /**
2217  * OSD layer object create function for interoperability mode (b11826).
2218  * This is mostly similar to osd_object_create(). Only difference being, fid is
2219  * inserted into inode ea here.
2220  *
2221  * \retval   0, on success
2222  * \retval -ve, on error
2223  */
2224 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2225                              struct lu_attr *attr,
2226                              struct dt_allocation_hint *hint,
2227                              struct dt_object_format *dof,
2228                              struct thandle *th)
2229 {
2230         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2231         struct osd_object      *obj    = osd_dt_obj(dt);
2232         struct osd_thread_info *info   = osd_oti_get(env);
2233         int result;
2234
2235         ENTRY;
2236
2237         LASSERT(osd_invariant(obj));
2238         LASSERT(!dt_object_exists(dt));
2239         LASSERT(osd_write_locked(env, obj));
2240         LASSERT(th != NULL);
2241
2242         OSD_EXEC_OP(th, create);
2243
2244         result = __osd_object_create(info, obj, attr, hint, dof, th);
2245
2246         /* objects under osd root shld have igif fid, so dont add fid EA */
2247         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2248                 result = osd_ea_fid_set(env, dt, fid);
2249
2250         if (result == 0)
2251                 result = __osd_oi_insert(env, obj, fid, th);
2252
2253         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2254         LINVRNT(osd_invariant(obj));
2255         RETURN(result);
2256 }
2257
2258 static int osd_declare_object_ref_add(const struct lu_env *env,
2259                                struct dt_object *dt,
2260                                struct thandle *handle)
2261 {
2262         struct osd_thandle *oh;
2263
2264         /* it's possible that object doesn't exist yet */
2265         LASSERT(handle != NULL);
2266
2267         oh = container_of0(handle, struct osd_thandle, ot_super);
2268         LASSERT(oh->ot_handle == NULL);
2269
2270         OSD_DECLARE_OP(oh, ref_add);
2271         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2272
2273         return 0;
2274 }
2275
2276 /*
2277  * Concurrency: @dt is write locked.
2278  */
2279 static int osd_object_ref_add(const struct lu_env *env,
2280                               struct dt_object *dt,
2281                               struct thandle *th)
2282 {
2283         struct osd_object *obj = osd_dt_obj(dt);
2284         struct inode *inode = obj->oo_inode;
2285
2286         LINVRNT(osd_invariant(obj));
2287         LASSERT(dt_object_exists(dt));
2288         LASSERT(osd_write_locked(env, obj));
2289         LASSERT(th != NULL);
2290
2291         OSD_EXEC_OP(th, ref_add);
2292
2293         /*
2294          * DIR_NLINK feature is set for compatibility reasons if:
2295          * 1) nlinks > LDISKFS_LINK_MAX, or
2296          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2297          *
2298          * It is easier to always set this flag (rather than check and set),
2299          * since it has less overhead, and the superblock will be dirtied
2300          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2301          * do not actually care whether this flag is set or not.
2302          */
2303         cfs_spin_lock(&obj->oo_guard);
2304         inode->i_nlink++;
2305         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2306                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2307                     inode->i_nlink == 2)
2308                         inode->i_nlink = 1;
2309         }
2310         LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
2311         cfs_spin_unlock(&obj->oo_guard);
2312         inode->i_sb->s_op->dirty_inode(inode);
2313         LINVRNT(osd_invariant(obj));
2314
2315         return 0;
2316 }
2317
2318 static int osd_declare_object_ref_del(const struct lu_env *env,
2319                                struct dt_object *dt,
2320                                struct thandle *handle)
2321 {
2322         struct osd_thandle *oh;
2323
2324         LASSERT(dt_object_exists(dt));
2325         LASSERT(handle != NULL);
2326
2327         oh = container_of0(handle, struct osd_thandle, ot_super);
2328         LASSERT(oh->ot_handle == NULL);
2329
2330         OSD_DECLARE_OP(oh, ref_del);
2331         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2332
2333         return 0;
2334 }
2335
2336 /*
2337  * Concurrency: @dt is write locked.
2338  */
2339 static int osd_object_ref_del(const struct lu_env *env,
2340                               struct dt_object *dt,
2341                               struct thandle *th)
2342 {
2343         struct osd_object *obj = osd_dt_obj(dt);
2344         struct inode *inode = obj->oo_inode;
2345
2346         LINVRNT(osd_invariant(obj));
2347         LASSERT(dt_object_exists(dt));
2348         LASSERT(osd_write_locked(env, obj));
2349         LASSERT(th != NULL);
2350
2351         OSD_EXEC_OP(th, ref_del);
2352
2353         cfs_spin_lock(&obj->oo_guard);
2354         LASSERT(inode->i_nlink > 0);
2355         inode->i_nlink--;
2356         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2357          * then the nlink count is 1. Don't let it be set to 0 or the directory
2358          * inode will be deleted incorrectly. */
2359         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2360                 inode->i_nlink++;
2361         cfs_spin_unlock(&obj->oo_guard);
2362         inode->i_sb->s_op->dirty_inode(inode);
2363         LINVRNT(osd_invariant(obj));
2364
2365         return 0;
2366 }
2367
2368 /*
2369  * Get the 64-bit version for an inode.
2370  */
2371 static int osd_object_version_get(const struct lu_env *env,
2372                                   struct dt_object *dt, dt_obj_version_t *ver)
2373 {
2374         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2375
2376         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2377                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2378         *ver = LDISKFS_I(inode)->i_fs_version;
2379         return 0;
2380 }
2381
2382 /*
2383  * Concurrency: @dt is read locked.
2384  */
2385 static int osd_xattr_get(const struct lu_env *env,
2386                          struct dt_object *dt,
2387                          struct lu_buf *buf,
2388                          const char *name,
2389                          struct lustre_capa *capa)
2390 {
2391         struct osd_object      *obj    = osd_dt_obj(dt);
2392         struct inode           *inode  = obj->oo_inode;
2393         struct osd_thread_info *info   = osd_oti_get(env);
2394         struct dentry          *dentry = &info->oti_obj_dentry;
2395
2396         /* version get is not real XATTR but uses xattr API */
2397         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2398                 /* for version we are just using xattr API but change inode
2399                  * field instead */
2400                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2401                 osd_object_version_get(env, dt, buf->lb_buf);
2402                 return sizeof(dt_obj_version_t);
2403         }
2404
2405         LASSERT(dt_object_exists(dt));
2406         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2407         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2408
2409         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2410                 return -EACCES;
2411
2412         dentry->d_inode = inode;
2413         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2414 }
2415
2416
2417 static int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
2418                                  const struct lu_buf *buf, const char *name,
2419                                  int fl, struct thandle *handle)
2420 {
2421         struct osd_thandle *oh;
2422
2423         LASSERT(handle != NULL);
2424
2425         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2426                 /* no credits for version */
2427                 return 0;
2428         }
2429
2430         oh = container_of0(handle, struct osd_thandle, ot_super);
2431         LASSERT(oh->ot_handle == NULL);
2432
2433         OSD_DECLARE_OP(oh, xattr_set);
2434         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2435
2436         return 0;
2437 }
2438
2439 /*
2440  * Set the 64-bit version for object
2441  */
2442 static void osd_object_version_set(const struct lu_env *env,
2443                                    struct dt_object *dt,
2444                                    dt_obj_version_t *new_version)
2445 {
2446         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2447
2448         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2449                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2450
2451         LDISKFS_I(inode)->i_fs_version = *new_version;
2452         /** Version is set after all inode operations are finished,
2453          *  so we should mark it dirty here */
2454         inode->i_sb->s_op->dirty_inode(inode);
2455 }
2456
2457 /*
2458  * Concurrency: @dt is write locked.
2459  */
2460 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2461                          const struct lu_buf *buf, const char *name, int fl,
2462                          struct thandle *handle, struct lustre_capa *capa)
2463 {
2464         LASSERT(handle != NULL);
2465
2466         /* version set is not real XATTR */
2467         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2468                 /* for version we are just using xattr API but change inode
2469                  * field instead */
2470                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2471                 osd_object_version_set(env, dt, buf->lb_buf);
2472                 return sizeof(dt_obj_version_t);
2473         }
2474
2475         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2476                 return -EACCES;
2477
2478         OSD_EXEC_OP(handle, xattr_set);
2479         return __osd_xattr_set(env, dt, buf, name, fl);
2480 }
2481
2482 /*
2483  * Concurrency: @dt is read locked.
2484  */
2485 static int osd_xattr_list(const struct lu_env *env,
2486                           struct dt_object *dt,
2487                           struct lu_buf *buf,
2488                           struct lustre_capa *capa)
2489 {
2490         struct osd_object      *obj    = osd_dt_obj(dt);
2491         struct inode           *inode  = obj->oo_inode;
2492         struct osd_thread_info *info   = osd_oti_get(env);
2493         struct dentry          *dentry = &info->oti_obj_dentry;
2494
2495         LASSERT(dt_object_exists(dt));
2496         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2497         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2498
2499         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2500                 return -EACCES;
2501
2502         dentry->d_inode = inode;
2503         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2504 }
2505
2506 static int osd_declare_xattr_del(const struct lu_env *env,
2507                                  struct dt_object *dt,
2508                                 const char *name,
2509                                 struct thandle *handle)
2510 {
2511         struct osd_thandle *oh;
2512
2513         LASSERT(dt_object_exists(dt));
2514         LASSERT(handle != NULL);
2515
2516         oh = container_of0(handle, struct osd_thandle, ot_super);
2517         LASSERT(oh->ot_handle == NULL);
2518
2519         OSD_DECLARE_OP(oh, xattr_set);
2520         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2521
2522         return 0;
2523 }
2524
2525 /*
2526  * Concurrency: @dt is write locked.
2527  */
2528 static int osd_xattr_del(const struct lu_env *env,
2529                          struct dt_object *dt,
2530                          const char *name,
2531                          struct thandle *handle,
2532                          struct lustre_capa *capa)
2533 {
2534         struct osd_object      *obj    = osd_dt_obj(dt);
2535         struct inode           *inode  = obj->oo_inode;
2536         struct osd_thread_info *info   = osd_oti_get(env);
2537         struct dentry          *dentry = &info->oti_obj_dentry;
2538         int                     rc;
2539
2540         LASSERT(dt_object_exists(dt));
2541         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2542         LASSERT(osd_write_locked(env, obj));
2543         LASSERT(handle != NULL);
2544
2545         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2546                 return -EACCES;
2547
2548         OSD_EXEC_OP(handle, xattr_set);
2549
2550         dentry->d_inode = inode;
2551         rc = inode->i_op->removexattr(dentry, name);
2552         return rc;
2553 }
2554
2555 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2556                                      struct dt_object *dt,
2557                                      struct lustre_capa *old,
2558                                      __u64 opc)
2559 {
2560         struct osd_thread_info *info = osd_oti_get(env);
2561         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2562         struct osd_object *obj = osd_dt_obj(dt);
2563         struct osd_device *dev = osd_obj2dev(obj);
2564         struct lustre_capa_key *key = &info->oti_capa_key;
2565         struct lustre_capa *capa = &info->oti_capa;
2566         struct obd_capa *oc;
2567         struct md_capainfo *ci;
2568         int rc;
2569         ENTRY;
2570
2571         if (!dev->od_fl_capa)
2572                 RETURN(ERR_PTR(-ENOENT));
2573
2574         LASSERT(dt_object_exists(dt));
2575         LINVRNT(osd_invariant(obj));
2576
2577         /* renewal sanity check */
2578         if (old && osd_object_auth(env, dt, old, opc))
2579                 RETURN(ERR_PTR(-EACCES));
2580
2581         ci = md_capainfo(env);
2582         if (unlikely(!ci))
2583                 RETURN(ERR_PTR(-ENOENT));
2584
2585         switch (ci->mc_auth) {
2586         case LC_ID_NONE:
2587                 RETURN(NULL);
2588         case LC_ID_PLAIN:
2589                 capa->lc_uid = obj->oo_inode->i_uid;
2590                 capa->lc_gid = obj->oo_inode->i_gid;
2591                 capa->lc_flags = LC_ID_PLAIN;
2592                 break;
2593         case LC_ID_CONVERT: {
2594                 __u32 d[4], s[4];
2595
2596                 s[0] = obj->oo_inode->i_uid;
2597                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2598                 s[2] = obj->oo_inode->i_gid;
2599                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2600                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2601                 if (unlikely(rc))
2602                         RETURN(ERR_PTR(rc));
2603
2604                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2605                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2606                 capa->lc_flags = LC_ID_CONVERT;
2607                 break;
2608         }
2609         default:
2610                 RETURN(ERR_PTR(-EINVAL));
2611         }
2612
2613         capa->lc_fid = *fid;
2614         capa->lc_opc = opc;
2615         capa->lc_flags |= dev->od_capa_alg << 24;
2616         capa->lc_timeout = dev->od_capa_timeout;
2617         capa->lc_expiry = 0;
2618
2619         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2620         if (oc) {
2621                 LASSERT(!capa_is_expired(oc));
2622                 RETURN(oc);
2623         }
2624
2625         cfs_spin_lock(&capa_lock);
2626         *key = dev->od_capa_keys[1];
2627         cfs_spin_unlock(&capa_lock);
2628
2629         capa->lc_keyid = key->lk_keyid;
2630         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2631
2632         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2633         if (rc) {
2634                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2635                 RETURN(ERR_PTR(rc));
2636         }
2637
2638         oc = capa_add(dev->od_capa_hash, capa);
2639         RETURN(oc);
2640 }
2641
2642 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2643 {
2644         int rc;
2645         struct osd_object      *obj    = osd_dt_obj(dt);
2646         struct inode           *inode  = obj->oo_inode;
2647         struct osd_thread_info *info   = osd_oti_get(env);
2648         struct dentry          *dentry = &info->oti_obj_dentry;
2649         struct file            *file   = &info->oti_file;
2650         ENTRY;
2651
2652         dentry->d_inode = inode;
2653         file->f_dentry = dentry;
2654         file->f_mapping = inode->i_mapping;
2655         file->f_op = inode->i_fop;
2656         LOCK_INODE_MUTEX(inode);
2657         rc = file->f_op->fsync(file, dentry, 0);
2658         UNLOCK_INODE_MUTEX(inode);
2659         RETURN(rc);
2660 }
2661
2662 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2663                         void **data)
2664 {
2665         struct osd_object *obj = osd_dt_obj(dt);
2666         ENTRY;
2667
2668         *data = (void *)obj->oo_inode;
2669         RETURN(0);
2670 }
2671
2672 /*
2673  * Index operations.
2674  */
2675
2676 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2677                            const struct dt_index_features *feat)
2678 {
2679         struct iam_descr *descr;
2680
2681         if (osd_object_is_root(o))
2682                 return feat == &dt_directory_features;
2683
2684         LASSERT(o->oo_dir != NULL);
2685
2686         descr = o->oo_dir->od_container.ic_descr;
2687         if (feat == &dt_directory_features) {
2688                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2689                         return 1;
2690                 else
2691                         return 0;
2692         } else {
2693                 return
2694                         feat->dif_keysize_min <= descr->id_key_size &&
2695                         descr->id_key_size <= feat->dif_keysize_max &&
2696                         feat->dif_recsize_min <= descr->id_rec_size &&
2697                         descr->id_rec_size <= feat->dif_recsize_max &&
2698                         !(feat->dif_flags & (DT_IND_VARKEY |
2699                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2700                         ergo(feat->dif_flags & DT_IND_UPDATE,
2701                              1 /* XXX check that object (and file system) is
2702                                 * writable */);
2703         }
2704 }
2705
2706 static int osd_iam_container_init(const struct lu_env *env,
2707                                   struct osd_object *obj,
2708                                   struct osd_directory *dir)
2709 {
2710         struct iam_container *bag = &dir->od_container;
2711         int result;
2712
2713         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2714         if (result != 0)
2715                 return result;
2716
2717         result = iam_container_setup(bag);
2718         if (result != 0)
2719                 goto out;
2720
2721         if (osd_obj2dev(obj)->od_iop_mode) {
2722                 u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag);
2723
2724                 bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode,
2725                                                 ptr, 0, &result);
2726         }
2727
2728  out:
2729         if (result == 0)
2730                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2731         else
2732                 iam_container_fini(bag);
2733
2734         return result;
2735 }
2736
2737
2738 /*
2739  * Concurrency: no external locking is necessary.
2740  */
2741 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2742                          const struct dt_index_features *feat)
2743 {
2744         int result;
2745         int ea_dir = 0;
2746         struct osd_object *obj = osd_dt_obj(dt);
2747         struct osd_device *osd = osd_obj2dev(obj);
2748
2749         LINVRNT(osd_invariant(obj));
2750         LASSERT(dt_object_exists(dt));
2751
2752         if (osd_object_is_root(obj)) {
2753                 dt->do_index_ops = &osd_index_ea_ops;
2754                 result = 0;
2755         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2756                 dt->do_index_ops = &osd_index_ea_ops;
2757                 if (S_ISDIR(obj->oo_inode->i_mode))
2758                         result = 0;
2759                 else
2760                         result = -ENOTDIR;
2761                 ea_dir = 1;
2762         } else if (!osd_has_index(obj)) {
2763                 struct osd_directory *dir;
2764
2765                 OBD_ALLOC_PTR(dir);
2766                 if (dir != NULL) {
2767
2768                         cfs_spin_lock(&obj->oo_guard);
2769                         if (obj->oo_dir == NULL)
2770                                 obj->oo_dir = dir;
2771                         else
2772                                 /*
2773                                  * Concurrent thread allocated container data.
2774                                  */
2775                                 OBD_FREE_PTR(dir);
2776                         cfs_spin_unlock(&obj->oo_guard);
2777                         /*
2778                          * Now, that we have container data, serialize its
2779                          * initialization.
2780                          */
2781                         cfs_down_write(&obj->oo_ext_idx_sem);
2782                         /*
2783                          * recheck under lock.
2784                          */
2785                         if (!osd_has_index(obj))
2786                                 result = osd_iam_container_init(env, obj, dir);
2787                         else
2788                                 result = 0;
2789                         cfs_up_write(&obj->oo_ext_idx_sem);
2790                 } else {
2791                         result = -ENOMEM;
2792                 }
2793         } else {
2794                 result = 0;
2795         }
2796
2797         if (result == 0 && ea_dir == 0) {
2798                 if (!osd_iam_index_probe(env, obj, feat))
2799                         result = -ENOTDIR;
2800         }
2801         LINVRNT(osd_invariant(obj));
2802
2803         return result;
2804 }
2805
2806 static const struct dt_object_operations osd_obj_ops = {
2807         .do_read_lock         = osd_object_read_lock,
2808         .do_write_lock        = osd_object_write_lock,
2809         .do_read_unlock       = osd_object_read_unlock,
2810         .do_write_unlock      = osd_object_write_unlock,
2811         .do_write_locked      = osd_object_write_locked,
2812         .do_attr_get          = osd_attr_get,
2813         .do_declare_attr_set  = osd_declare_attr_set,
2814         .do_attr_set          = osd_attr_set,
2815         .do_ah_init           = osd_ah_init,
2816         .do_declare_create    = osd_declare_object_create,
2817         .do_create            = osd_object_create,
2818         .do_declare_destroy   = osd_declare_object_destroy,
2819         .do_destroy           = osd_object_destroy,
2820         .do_index_try         = osd_index_try,
2821         .do_declare_ref_add   = osd_declare_object_ref_add,
2822         .do_ref_add           = osd_object_ref_add,
2823         .do_declare_ref_del   = osd_declare_object_ref_del,
2824         .do_ref_del           = osd_object_ref_del,
2825         .do_xattr_get         = osd_xattr_get,
2826         .do_declare_xattr_set = osd_declare_xattr_set,
2827         .do_xattr_set         = osd_xattr_set,
2828         .do_declare_xattr_del = osd_declare_xattr_del,
2829         .do_xattr_del         = osd_xattr_del,
2830         .do_xattr_list        = osd_xattr_list,
2831         .do_capa_get          = osd_capa_get,
2832         .do_object_sync       = osd_object_sync,
2833         .do_data_get          = osd_data_get,
2834 };
2835
2836 /**
2837  * dt_object_operations for interoperability mode
2838  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2839  */
2840 static const struct dt_object_operations osd_obj_ea_ops = {
2841         .do_read_lock         = osd_object_read_lock,
2842         .do_write_lock        = osd_object_write_lock,
2843         .do_read_unlock       = osd_object_read_unlock,
2844         .do_write_unlock      = osd_object_write_unlock,
2845         .do_write_locked      = osd_object_write_locked,
2846         .do_attr_get          = osd_attr_get,
2847         .do_declare_attr_set  = osd_declare_attr_set,
2848         .do_attr_set          = osd_attr_set,
2849         .do_ah_init           = osd_ah_init,
2850         .do_declare_create    = osd_declare_object_create,
2851         .do_create            = osd_object_ea_create,
2852         .do_declare_destroy   = osd_declare_object_destroy,
2853         .do_destroy           = osd_object_destroy,
2854         .do_index_try         = osd_index_try,
2855         .do_declare_ref_add   = osd_declare_object_ref_add,
2856         .do_ref_add           = osd_object_ref_add,
2857         .do_declare_ref_del   = osd_declare_object_ref_del,
2858         .do_ref_del           = osd_object_ref_del,
2859         .do_xattr_get         = osd_xattr_get,
2860         .do_declare_xattr_set = osd_declare_xattr_set,
2861         .do_xattr_set         = osd_xattr_set,
2862         .do_declare_xattr_del = osd_declare_xattr_del,
2863         .do_xattr_del         = osd_xattr_del,
2864         .do_xattr_list        = osd_xattr_list,
2865         .do_capa_get          = osd_capa_get,
2866         .do_object_sync       = osd_object_sync,
2867         .do_data_get          = osd_data_get,
2868 };
2869
2870 /*
2871  * Body operations.
2872  */
2873
2874 /*
2875  * XXX: Another layering violation for now.
2876  *
2877  * We don't want to use ->f_op->read methods, because generic file write
2878  *
2879  *         - serializes on ->i_sem, and
2880  *
2881  *         - does a lot of extra work like balance_dirty_pages(),
2882  *
2883  * which doesn't work for globally shared files like /last-received.
2884  */
2885 static int osd_ldiskfs_readlink(struct inode *inode, char *buffer, int buflen)
2886 {
2887         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2888
2889         memcpy(buffer, (char*)ei->i_data, buflen);
2890
2891         return  buflen;
2892 }
2893
2894 static int osd_ldiskfs_read(struct inode *inode, void *buf, int size,
2895                             loff_t *offs)
2896 {
2897         struct buffer_head *bh;
2898         unsigned long block;
2899         int osize = size;
2900         int blocksize;
2901         int csize;
2902         int boffs;
2903         int err;
2904
2905         /* prevent reading after eof */
2906         spin_lock(&inode->i_lock);
2907         if (i_size_read(inode) < *offs + size) {
2908                 size = i_size_read(inode) - *offs;
2909                 spin_unlock(&inode->i_lock);
2910                 if (size < 0) {
2911                         CDEBUG(D_EXT2, "size %llu is too short to read @%llu\n",
2912                                i_size_read(inode), *offs);
2913                         return -EBADR;
2914                 } else if (size == 0) {
2915                         return 0;
2916                 }
2917         } else {
2918                 spin_unlock(&inode->i_lock);
2919         }
2920
2921         blocksize = 1 << inode->i_blkbits;
2922
2923         while (size > 0) {
2924                 block = *offs >> inode->i_blkbits;
2925                 boffs = *offs & (blocksize - 1);
2926                 csize = min(blocksize - boffs, size);
2927                 bh = ldiskfs_bread(NULL, inode, block, 0, &err);
2928                 if (!bh) {
2929                         CERROR("can't read block: %d\n", err);
2930                         return err;
2931                 }
2932
2933                 memcpy(buf, bh->b_data + boffs, csize);
2934                 brelse(bh);
2935
2936                 *offs += csize;
2937                 buf += csize;
2938                 size -= csize;
2939         }
2940         return osize;
2941 }
2942
2943 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
2944                         struct lu_buf *buf, loff_t *pos,
2945                         struct lustre_capa *capa)
2946 {
2947         struct osd_object      *obj    = osd_dt_obj(dt);
2948         struct inode           *inode  = obj->oo_inode;
2949         int rc;
2950
2951         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2952                 RETURN(-EACCES);
2953
2954         /* Read small symlink from inode body as we need to maintain correct
2955          * on-disk symlinks for ldiskfs.
2956          */
2957         if (S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
2958             (buf->lb_len <= sizeof (LDISKFS_I(inode)->i_data)))
2959                 rc = osd_ldiskfs_readlink(inode, buf->lb_buf, buf->lb_len);
2960         else
2961                 rc = osd_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
2962
2963         return rc;
2964 }
2965
2966 static int osd_ldiskfs_writelink(struct inode *inode, char *buffer, int buflen)
2967 {
2968
2969         memcpy((char*)&LDISKFS_I(inode)->i_data, (char *)buffer,
2970                buflen);
2971         LDISKFS_I(inode)->i_disksize = buflen;
2972         i_size_write(inode, buflen);
2973         inode->i_sb->s_op->dirty_inode(inode);
2974
2975         return 0;
2976 }
2977
2978 static int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
2979                                     loff_t *offs, handle_t *handle)
2980 {
2981         struct buffer_head *bh = NULL;
2982         loff_t offset = *offs;
2983         loff_t new_size = i_size_read(inode);
2984         unsigned long block;
2985         int blocksize = 1 << inode->i_blkbits;
2986         int err = 0;
2987         int size;
2988         int boffs;
2989         int dirty_inode = 0;
2990
2991         while (bufsize > 0) {
2992                 if (bh != NULL)
2993                         brelse(bh);
2994
2995                 block = offset >> inode->i_blkbits;
2996                 boffs = offset & (blocksize - 1);
2997                 size = min(blocksize - boffs, bufsize);
2998                 bh = ldiskfs_bread(handle, inode, block, 1, &err);
2999                 if (!bh) {
3000                         CERROR("can't read/create block: %d\n", err);
3001                         break;
3002                 }
3003
3004                 err = ldiskfs_journal_get_write_access(handle, bh);
3005                 if (err) {
3006                         CERROR("journal_get_write_access() returned error %d\n",
3007                                err);
3008                         break;
3009                 }
3010                 LASSERTF(boffs + size <= bh->b_size,
3011                          "boffs %d size %d bh->b_size %lu",
3012                          boffs, size, (unsigned long)bh->b_size);
3013                 memcpy(bh->b_data + boffs, buf, size);
3014                 err = ldiskfs_journal_dirty_metadata(handle, bh);
3015                 if (err)
3016                         break;
3017
3018                 if (offset + size > new_size)
3019                         new_size = offset + size;
3020                 offset += size;
3021                 bufsize -= size;
3022                 buf += size;
3023         }
3024         if (bh)
3025                 brelse(bh);
3026
3027         /* correct in-core and on-disk sizes */
3028         if (new_size > i_size_read(inode)) {
3029                 spin_lock(&inode->i_lock);
3030                 if (new_size > i_size_read(inode))
3031                         i_size_write(inode, new_size);
3032                 if (i_size_read(inode) > LDISKFS_I(inode)->i_disksize) {
3033                         LDISKFS_I(inode)->i_disksize = i_size_read(inode);
3034                         dirty_inode = 1;
3035                 }
3036                 spin_unlock(&inode->i_lock);
3037                 if (dirty_inode)
3038                         inode->i_sb->s_op->dirty_inode(inode);
3039         }
3040
3041         if (err == 0)
3042                 *offs = offset;
3043         return err;
3044 }
3045
3046 static ssize_t osd_declare_write(const struct lu_env *env, struct dt_object *dt,
3047                                  const loff_t size, loff_t pos,
3048                                  struct thandle *handle)
3049 {
3050         struct osd_thandle *oh;
3051         int credits;
3052
3053         LASSERT(handle != NULL);
3054
3055         oh = container_of0(handle, struct osd_thandle, ot_super);
3056         LASSERT(oh->ot_handle == NULL);
3057
3058         /* XXX: size == 0 or INT_MAX indicating a catalog header update or
3059          *      llog write, see comment in mdd_declare_llog_record().
3060          *
3061          *      This hack should be removed in 2.3
3062          */
3063         if (size == DECLARE_LLOG_REWRITE)
3064                 credits = 2;
3065         else if (size == DECLARE_LLOG_WRITE)
3066                 credits = 6;
3067         else
3068                 credits = osd_dto_credits_noquota[DTO_WRITE_BLOCK];
3069
3070         OSD_DECLARE_OP(oh, write);
3071         oh->ot_credits += credits;
3072
3073         if (osd_dt_obj(dt)->oo_inode == NULL)
3074                 return 0;
3075
3076         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3077                         osd_dt_obj(dt)->oo_inode);
3078         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3079                         osd_dt_obj(dt)->oo_inode);
3080         return 0;
3081 }
3082
3083 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
3084                          const struct lu_buf *buf, loff_t *pos,
3085                          struct thandle *handle, struct lustre_capa *capa,
3086                          int ignore_quota)
3087 {
3088         struct osd_object  *obj   = osd_dt_obj(dt);
3089         struct inode       *inode = obj->oo_inode;
3090         struct osd_thandle *oh;
3091         ssize_t            result = 0;
3092 #ifdef HAVE_QUOTA_SUPPORT
3093         cfs_cap_t           save = cfs_curproc_cap_pack();
3094 #endif
3095
3096         LASSERT(handle != NULL);
3097
3098         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
3099                 RETURN(-EACCES);
3100
3101         oh = container_of(handle, struct osd_thandle, ot_super);
3102         LASSERT(oh->ot_handle->h_transaction != NULL);
3103 #ifdef HAVE_QUOTA_SUPPORT
3104         if (ignore_quota)
3105                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3106         else
3107                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3108 #endif
3109         /* Write small symlink to inode body as we need to maintain correct
3110          * on-disk symlinks for ldiskfs.
3111          */
3112         if(S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
3113            (buf->lb_len < sizeof (LDISKFS_I(inode)->i_data)))
3114                 result = osd_ldiskfs_writelink(inode, buf->lb_buf, buf->lb_len);
3115         else
3116                 result = osd_ldiskfs_write_record(inode, buf->lb_buf,
3117                                                   buf->lb_len, pos,
3118                                                   oh->ot_handle);
3119 #ifdef HAVE_QUOTA_SUPPORT
3120         cfs_curproc_cap_unpack(save);
3121 #endif
3122         if (result == 0)
3123                 result = buf->lb_len;
3124         return result;
3125 }
3126
3127 /*
3128  * in some cases we may need declare methods for objects being created
3129  * e.g., when we create symlink
3130  */
3131 static const struct dt_body_operations osd_body_ops_new = {
3132         .dbo_declare_write = osd_declare_write,
3133 };
3134
3135 static const struct dt_body_operations osd_body_ops = {
3136         .dbo_read          = osd_read,
3137         .dbo_declare_write = osd_declare_write,
3138         .dbo_write         = osd_write
3139 };
3140
3141 static int osd_index_declare_iam_delete(const struct lu_env *env,
3142                                         struct dt_object *dt,
3143                                         const struct dt_key *key,
3144                                         struct thandle *handle)
3145 {
3146         struct osd_thandle    *oh;
3147
3148         oh = container_of0(handle, struct osd_thandle, ot_super);
3149         LASSERT(oh->ot_handle == NULL);
3150
3151         OSD_DECLARE_OP(oh, delete);
3152         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3153
3154         return 0;
3155 }
3156
3157 /**
3158  *      delete a (key, value) pair from index \a dt specified by \a key
3159  *
3160  *      \param  dt      osd index object
3161  *      \param  key     key for index
3162  *      \param  rec     record reference
3163  *      \param  handle  transaction handler
3164  *
3165  *      \retval  0  success
3166  *      \retval -ve   failure
3167  */
3168
3169 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3170                                 const struct dt_key *key, struct thandle *handle,
3171                                 struct lustre_capa *capa)
3172 {
3173         struct osd_object     *obj = osd_dt_obj(dt);
3174         struct osd_thandle    *oh;
3175         struct iam_path_descr *ipd;
3176         struct iam_container  *bag = &obj->oo_dir->od_container;
3177         int rc;
3178
3179         ENTRY;
3180
3181         LINVRNT(osd_invariant(obj));
3182         LASSERT(dt_object_exists(dt));
3183         LASSERT(bag->ic_object == obj->oo_inode);
3184         LASSERT(handle != NULL);
3185
3186         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3187                 RETURN(-EACCES);
3188
3189         OSD_EXEC_OP(handle, delete);
3190
3191         ipd = osd_idx_ipd_get(env, bag);
3192         if (unlikely(ipd == NULL))
3193                 RETURN(-ENOMEM);
3194
3195         oh = container_of0(handle, struct osd_thandle, ot_super);
3196         LASSERT(oh->ot_handle != NULL);
3197         LASSERT(oh->ot_handle->h_transaction != NULL);
3198
3199         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3200         osd_ipd_put(env, bag, ipd);
3201         LINVRNT(osd_invariant(obj));
3202         RETURN(rc);
3203 }
3204
3205 static int osd_index_declare_ea_delete(const struct lu_env *env,
3206                                        struct dt_object *dt,
3207                                        const struct dt_key *key,
3208                                        struct thandle *handle)
3209 {
3210         struct osd_thandle *oh;
3211
3212         LASSERT(dt_object_exists(dt));
3213         LASSERT(handle != NULL);
3214
3215         oh = container_of0(handle, struct osd_thandle, ot_super);
3216         LASSERT(oh->ot_handle == NULL);
3217
3218         OSD_DECLARE_OP(oh, delete);
3219         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3220
3221         LASSERT(osd_dt_obj(dt)->oo_inode);
3222         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3223                         osd_dt_obj(dt)->oo_inode);
3224         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3225                         osd_dt_obj(dt)->oo_inode);
3226
3227         return 0;
3228 }
3229
3230 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3231                                           struct dt_rec *fid)
3232 {
3233         struct osd_fid_pack *rec;
3234         int rc = -ENODATA;
3235
3236         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3237                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3238                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3239         }
3240         RETURN(rc);
3241 }
3242
3243 /**
3244  * Index delete function for interoperability mode (b11826).
3245  * It will remove the directory entry added by osd_index_ea_insert().
3246  * This entry is needed to maintain name->fid mapping.
3247  *
3248  * \param key,  key i.e. file entry to be deleted
3249  *
3250  * \retval   0, on success
3251  * \retval -ve, on error
3252  */
3253 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3254                                const struct dt_key *key, struct thandle *handle,
3255                                struct lustre_capa *capa)
3256 {
3257         struct osd_object          *obj    = osd_dt_obj(dt);
3258         struct inode               *dir    = obj->oo_inode;
3259         struct dentry              *dentry;
3260         struct osd_thandle         *oh;
3261         struct ldiskfs_dir_entry_2 *de;
3262         struct buffer_head         *bh;
3263         struct htree_lock          *hlock = NULL;
3264
3265         int rc;
3266
3267         ENTRY;
3268
3269         LINVRNT(osd_invariant(obj));
3270         LASSERT(dt_object_exists(dt));
3271         LASSERT(handle != NULL);
3272
3273         OSD_EXEC_OP(handle, delete);
3274
3275         oh = container_of(handle, struct osd_thandle, ot_super);
3276         LASSERT(oh->ot_handle != NULL);
3277         LASSERT(oh->ot_handle->h_transaction != NULL);
3278
3279         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3280                 RETURN(-EACCES);
3281
3282         dentry = osd_child_dentry_get(env, obj,
3283                                       (char *)key, strlen((char *)key));
3284
3285         if (obj->oo_hl_head != NULL) {
3286                 hlock = osd_oti_get(env)->oti_hlock;
3287                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3288                                    dir, LDISKFS_HLOCK_DEL);
3289         } else {
3290                 cfs_down_write(&obj->oo_ext_idx_sem);
3291         }
3292
3293         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3294         if (bh) {
3295                 rc = ldiskfs_delete_entry(oh->ot_handle,
3296                                           dir, de, bh);
3297                 brelse(bh);
3298         } else {
3299                 rc = -ENOENT;
3300         }
3301         if (hlock != NULL)
3302                 ldiskfs_htree_unlock(hlock);
3303         else
3304                 cfs_up_write(&obj->oo_ext_idx_sem);
3305
3306         LASSERT(osd_invariant(obj));
3307         RETURN(rc);
3308 }
3309
3310 /**
3311  *      Lookup index for \a key and copy record to \a rec.
3312  *
3313  *      \param  dt      osd index object
3314  *      \param  key     key for index
3315  *      \param  rec     record reference
3316  *
3317  *      \retval  +ve  success : exact mach
3318  *      \retval  0    return record with key not greater than \a key
3319  *      \retval -ve   failure
3320  */
3321 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3322                                 struct dt_rec *rec, const struct dt_key *key,
3323                                 struct lustre_capa *capa)
3324 {
3325         struct osd_object     *obj = osd_dt_obj(dt);
3326         struct iam_path_descr *ipd;
3327         struct iam_container  *bag = &obj->oo_dir->od_container;
3328         struct osd_thread_info *oti = osd_oti_get(env);
3329         struct iam_iterator    *it = &oti->oti_idx_it;
3330         struct iam_rec *iam_rec;
3331         int rc;
3332         ENTRY;
3333
3334         LASSERT(osd_invariant(obj));
3335         LASSERT(dt_object_exists(dt));
3336         LASSERT(bag->ic_object == obj->oo_inode);
3337
3338         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3339                 RETURN(-EACCES);
3340
3341         ipd = osd_idx_ipd_get(env, bag);
3342         if (IS_ERR(ipd))
3343                 RETURN(-ENOMEM);
3344
3345         /* got ipd now we can start iterator. */
3346         iam_it_init(it, bag, 0, ipd);
3347
3348         rc = iam_it_get(it, (struct iam_key *)key);
3349         if (rc >= 0) {
3350                 if (S_ISDIR(obj->oo_inode->i_mode))
3351                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3352                 else
3353                         iam_rec = (struct iam_rec *) rec;
3354
3355                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3356                 if (S_ISDIR(obj->oo_inode->i_mode))
3357                         osd_fid_unpack((struct lu_fid *) rec,
3358                                        (struct osd_fid_pack *)iam_rec);
3359         }
3360         iam_it_put(it);
3361         iam_it_fini(it);
3362         osd_ipd_put(env, bag, ipd);
3363
3364         LINVRNT(osd_invariant(obj));
3365
3366         RETURN(rc);
3367 }
3368
3369 static int osd_index_declare_iam_insert(const struct lu_env *env,
3370                                         struct dt_object *dt,
3371                                         const struct dt_rec *rec,
3372                                         const struct dt_key *key,
3373                                         struct thandle *handle)
3374 {
3375         struct osd_thandle *oh;
3376
3377         LASSERT(dt_object_exists(dt));
3378         LASSERT(handle != NULL);
3379
3380         oh = container_of0(handle, struct osd_thandle, ot_super);
3381         LASSERT(oh->ot_handle == NULL);
3382
3383         OSD_DECLARE_OP(oh, insert);
3384         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3385
3386         return 0;
3387 }
3388
3389 /**
3390  *      Inserts (key, value) pair in \a dt index object.
3391  *
3392  *      \param  dt      osd index object
3393  *      \param  key     key for index
3394  *      \param  rec     record reference
3395  *      \param  th      transaction handler
3396  *
3397  *      \retval  0  success
3398  *      \retval -ve failure
3399  */
3400 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3401                                 const struct dt_rec *rec, const struct dt_key *key,
3402                                 struct thandle *th, struct lustre_capa *capa,
3403                                 int ignore_quota)
3404 {
3405         struct osd_object     *obj = osd_dt_obj(dt);
3406         struct iam_path_descr *ipd;
3407         struct osd_thandle    *oh;
3408         struct iam_container  *bag = &obj->oo_dir->od_container;
3409 #ifdef HAVE_QUOTA_SUPPORT
3410         cfs_cap_t              save = cfs_curproc_cap_pack();
3411 #endif
3412         struct osd_thread_info *oti = osd_oti_get(env);
3413         struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_ldp;
3414         int rc;
3415
3416         ENTRY;
3417
3418         LINVRNT(osd_invariant(obj));
3419         LASSERT(dt_object_exists(dt));
3420         LASSERT(bag->ic_object == obj->oo_inode);
3421         LASSERT(th != NULL);
3422
3423         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3424                 return -EACCES;
3425
3426         OSD_EXEC_OP(th, insert);
3427
3428         ipd = osd_idx_ipd_get(env, bag);
3429         if (unlikely(ipd == NULL))
3430                 RETURN(-ENOMEM);
3431
3432         oh = container_of0(th, struct osd_thandle, ot_super);
3433         LASSERT(oh->ot_handle != NULL);
3434         LASSERT(oh->ot_handle->h_transaction != NULL);
3435 #ifdef HAVE_QUOTA_SUPPORT
3436         if (ignore_quota)
3437                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3438         else
3439                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3440 #endif
3441         if (S_ISDIR(obj->oo_inode->i_mode))
3442                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3443         else
3444                 iam_rec = (struct iam_rec *) rec;
3445         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3446                         iam_rec, ipd);
3447 #ifdef HAVE_QUOTA_SUPPORT
3448         cfs_curproc_cap_unpack(save);
3449 #endif
3450         osd_ipd_put(env, bag, ipd);
3451         LINVRNT(osd_invariant(obj));
3452         RETURN(rc);
3453 }
3454
3455 /**
3456  * Calls ldiskfs_add_entry() to add directory entry
3457  * into the directory. This is required for
3458  * interoperability mode (b11826)
3459  *
3460  * \retval   0, on success
3461  * \retval -ve, on error
3462  */
3463 static int __osd_ea_add_rec(struct osd_thread_info *info,
3464                             struct osd_object *pobj,
3465                             struct inode  *cinode,
3466                             const char *name,
3467                             const struct dt_rec *fid,
3468                             struct htree_lock *hlock,
3469                             struct thandle *th)
3470 {
3471         struct ldiskfs_dentry_param *ldp;
3472         struct dentry      *child;
3473         struct osd_thandle *oth;
3474         int rc;
3475
3476         oth = container_of(th, struct osd_thandle, ot_super);
3477         LASSERT(oth->ot_handle != NULL);
3478         LASSERT(oth->ot_handle->h_transaction != NULL);
3479
3480         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3481
3482         if (fid_is_igif((struct lu_fid *)fid) ||
3483             fid_is_norm((struct lu_fid *)fid)) {
3484                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3485                 osd_get_ldiskfs_dirent_param(ldp, fid);
3486                 child->d_fsdata = (void*) ldp;
3487         } else
3488                 child->d_fsdata = NULL;
3489         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3490
3491         RETURN(rc);
3492 }
3493
3494 /**
3495  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3496  * into the directory.Also sets flags into osd object to
3497  * indicate dot and dotdot are created. This is required for
3498  * interoperability mode (b11826)
3499  *
3500  * \param dir   directory for dot and dotdot fixup.
3501  * \param obj   child object for linking
3502  *
3503  * \retval   0, on success
3504  * \retval -ve, on error
3505  */
3506 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3507                               struct osd_object *dir,
3508                               struct inode  *parent_dir, const char *name,
3509                               const struct dt_rec *dot_fid,
3510                               const struct dt_rec *dot_dot_fid,
3511                               struct thandle *th)
3512 {
3513         struct inode            *inode  = dir->oo_inode;
3514         struct ldiskfs_dentry_param *dot_ldp;
3515         struct ldiskfs_dentry_param *dot_dot_ldp;
3516         struct osd_thandle      *oth;
3517         int result = 0;
3518
3519         oth = container_of(th, struct osd_thandle, ot_super);
3520         LASSERT(oth->ot_handle->h_transaction != NULL);
3521         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3522
3523         if (strcmp(name, dot) == 0) {
3524                 if (dir->oo_compat_dot_created) {
3525                         result = -EEXIST;
3526                 } else {
3527                         LASSERT(inode == parent_dir);
3528                         dir->oo_compat_dot_created = 1;
3529                         result = 0;
3530                 }
3531         } else if(strcmp(name, dotdot) == 0) {
3532                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3533                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3534
3535                 if (!dir->oo_compat_dot_created)
3536                         return -EINVAL;
3537                 if (fid_seq((struct lu_fid *)dot_fid) >= FID_SEQ_NORMAL) {
3538                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3539                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3540                 } else {
3541                         dot_ldp = NULL;
3542                         dot_dot_ldp = NULL;
3543                 }
3544                 /* in case of rename, dotdot is already created */
3545                 if (dir->oo_compat_dotdot_created) {
3546                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3547                                                 dot_dot_fid, NULL, th);
3548                 }
3549
3550                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3551                                                 inode, dot_ldp, dot_dot_ldp);
3552                 if (result == 0)
3553                        dir->oo_compat_dotdot_created = 1;
3554         }
3555
3556         return result;
3557 }
3558
3559
3560 /**
3561  * It will call the appropriate osd_add* function and return the
3562  * value, return by respective functions.
3563  */
3564 static int osd_ea_add_rec(const struct lu_env *env,
3565                           struct osd_object *pobj,
3566                           struct inode *cinode,
3567                           const char *name,
3568                           const struct dt_rec *fid,
3569                           struct thandle *th)
3570 {
3571         struct osd_thread_info    *info   = osd_oti_get(env);
3572         struct htree_lock         *hlock;
3573         int rc;
3574
3575         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3576
3577         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3578                                                    name[2] =='\0'))) {
3579                 if (hlock != NULL) {
3580                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3581                                            pobj->oo_inode, 0);
3582                 } else {
3583                         cfs_down_write(&pobj->oo_ext_idx_sem);
3584                 }
3585                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3586                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3587                                         fid, th);
3588         } else {
3589                 if (hlock != NULL) {
3590                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3591                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3592                 } else {
3593                         cfs_down_write(&pobj->oo_ext_idx_sem);
3594                 }
3595
3596                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3597                                       hlock, th);
3598         }
3599         if (hlock != NULL)
3600                 ldiskfs_htree_unlock(hlock);
3601         else
3602                 cfs_up_write(&pobj->oo_ext_idx_sem);
3603
3604         return rc;
3605 }
3606
3607 /**
3608  * Calls ->lookup() to find dentry. From dentry get inode and
3609  * read inode's ea to get fid. This is required for  interoperability
3610  * mode (b11826)
3611  *
3612  * \retval   0, on success
3613  * \retval -ve, on error
3614  */
3615 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3616                              struct dt_rec *rec, const struct dt_key *key)
3617 {
3618         struct inode               *dir    = obj->oo_inode;
3619         struct dentry              *dentry;
3620         struct ldiskfs_dir_entry_2 *de;
3621         struct buffer_head         *bh;
3622         struct lu_fid              *fid = (struct lu_fid *) rec;
3623         struct htree_lock          *hlock = NULL;
3624         int ino;
3625         int rc;
3626
3627         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3628
3629         dentry = osd_child_dentry_get(env, obj,
3630                                       (char *)key, strlen((char *)key));
3631
3632         if (obj->oo_hl_head != NULL) {
3633                 hlock = osd_oti_get(env)->oti_hlock;
3634                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3635                                    dir, LDISKFS_HLOCK_LOOKUP);
3636         } else {
3637                 cfs_down_read(&obj->oo_ext_idx_sem);
3638         }
3639
3640         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3641         if (bh) {
3642                 ino = le32_to_cpu(de->inode);
3643                 rc = osd_get_fid_from_dentry(de, rec);
3644
3645                 /* done with de, release bh */
3646                 brelse(bh);
3647                 if (rc != 0)
3648                         rc = osd_ea_fid_get(env, obj, ino, fid);
3649         } else {
3650                 rc = -ENOENT;
3651         }
3652
3653         if (hlock != NULL)
3654                 ldiskfs_htree_unlock(hlock);
3655         else
3656                 cfs_up_read(&obj->oo_ext_idx_sem);
3657         RETURN (rc);
3658 }
3659
3660 /**
3661  * Find the osd object for given fid.
3662  *
3663  * \param fid need to find the osd object having this fid
3664  *
3665  * \retval osd_object on success
3666  * \retval        -ve on error
3667  */
3668 struct osd_object *osd_object_find(const struct lu_env *env,
3669                                    struct dt_object *dt,
3670                                    const struct lu_fid *fid)
3671 {
3672         struct lu_device         *ludev = dt->do_lu.lo_dev;
3673         struct osd_object        *child = NULL;
3674         struct lu_object         *luch;
3675         struct lu_object         *lo;
3676
3677         luch = lu_object_find(env, ludev, fid, NULL);
3678         if (!IS_ERR(luch)) {
3679                 if (lu_object_exists(luch)) {
3680                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3681                         if (lo != NULL)
3682                                 child = osd_obj(lo);
3683                         else
3684                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3685                                                 "lu_object can't be located"
3686                                                 ""DFID"\n", PFID(fid));
3687
3688                         if (child == NULL) {
3689                                 lu_object_put(env, luch);
3690                                 CERROR("Unable to get osd_object\n");
3691                                 child = ERR_PTR(-ENOENT);
3692                         }
3693                 } else {
3694                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3695                                         "lu_object does not exists "DFID"\n",
3696                                         PFID(fid));
3697                         child = ERR_PTR(-ENOENT);
3698                 }
3699         } else
3700                 child = (void *)luch;
3701
3702         return child;
3703 }
3704
3705 /**
3706  * Put the osd object once done with it.
3707  *
3708  * \param obj osd object that needs to be put
3709  */
3710 static inline void osd_object_put(const struct lu_env *env,
3711                                   struct osd_object *obj)
3712 {
3713         lu_object_put(env, &obj->oo_dt.do_lu);
3714 }
3715
3716 static int osd_index_declare_ea_insert(const struct lu_env *env,
3717                                        struct dt_object *dt,
3718                                        const struct dt_rec *rec,
3719                                        const struct dt_key *key,
3720                                        struct thandle *handle)
3721 {
3722         struct osd_thandle *oh;
3723
3724         LASSERT(dt_object_exists(dt));
3725         LASSERT(handle != NULL);
3726
3727         oh = container_of0(handle, struct osd_thandle, ot_super);
3728         LASSERT(oh->ot_handle == NULL);
3729
3730         OSD_DECLARE_OP(oh, insert);
3731         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3732
3733         LASSERT(osd_dt_obj(dt)->oo_inode);
3734         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3735                         osd_dt_obj(dt)->oo_inode);
3736         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3737                         osd_dt_obj(dt)->oo_inode);
3738
3739         return 0;
3740 }
3741
3742 /**
3743  * Index add function for interoperability mode (b11826).
3744  * It will add the directory entry.This entry is needed to
3745  * maintain name->fid mapping.
3746  *
3747  * \param key it is key i.e. file entry to be inserted
3748  * \param rec it is value of given key i.e. fid
3749  *
3750  * \retval   0, on success
3751  * \retval -ve, on error
3752  */
3753 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3754                                const struct dt_rec *rec,
3755                                const struct dt_key *key, struct thandle *th,
3756                                struct lustre_capa *capa, int ignore_quota)
3757 {
3758         struct osd_object        *obj   = osd_dt_obj(dt);
3759         struct lu_fid            *fid   = (struct lu_fid *) rec;
3760         const char               *name  = (const char *)key;
3761         struct osd_object        *child;
3762 #ifdef HAVE_QUOTA_SUPPORT
3763         cfs_cap_t                 save  = cfs_curproc_cap_pack();
3764 #endif
3765         int rc;
3766
3767         ENTRY;
3768
3769         LASSERT(osd_invariant(obj));
3770         LASSERT(dt_object_exists(dt));
3771         LASSERT(th != NULL);
3772
3773         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3774                 RETURN(-EACCES);
3775
3776         child = osd_object_find(env, dt, fid);
3777         if (!IS_ERR(child)) {
3778 #ifdef HAVE_QUOTA_SUPPORT
3779                 if (ignore_quota)
3780                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3781                 else
3782                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3783 #endif
3784                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3785 #ifdef HAVE_QUOTA_SUPPORT
3786                 cfs_curproc_cap_unpack(save);
3787 #endif
3788                 osd_object_put(env, child);
3789         } else {
3790                 rc = PTR_ERR(child);
3791         }
3792
3793         LASSERT(osd_invariant(obj));
3794         RETURN(rc);
3795 }
3796
3797 /**
3798  *  Initialize osd Iterator for given osd index object.
3799  *
3800  *  \param  dt      osd index object
3801  */
3802
3803 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3804                                      struct dt_object *dt,
3805                                      __u32 unused,
3806                                      struct lustre_capa *capa)
3807 {
3808         struct osd_it_iam         *it;
3809         struct osd_thread_info *oti = osd_oti_get(env);
3810         struct osd_object     *obj = osd_dt_obj(dt);
3811         struct lu_object      *lo  = &dt->do_lu;
3812         struct iam_path_descr *ipd;
3813         struct iam_container  *bag = &obj->oo_dir->od_container;
3814
3815         LASSERT(lu_object_exists(lo));
3816
3817         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3818                 return ERR_PTR(-EACCES);
3819
3820         it = &oti->oti_it;
3821         ipd = osd_it_ipd_get(env, bag);
3822         if (likely(ipd != NULL)) {
3823                 it->oi_obj = obj;
3824                 it->oi_ipd = ipd;
3825                 lu_object_get(lo);
3826                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3827                 return (struct dt_it *)it;
3828         }
3829         return ERR_PTR(-ENOMEM);
3830 }
3831
3832 /**
3833  * free given Iterator.
3834  */
3835
3836 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3837 {
3838         struct osd_it_iam     *it = (struct osd_it_iam *)di;
3839         struct osd_object *obj = it->oi_obj;
3840
3841         iam_it_fini(&it->oi_it);
3842         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3843         lu_object_put(env, &obj->oo_dt.do_lu);
3844 }
3845
3846 /**
3847  *  Move Iterator to record specified by \a key
3848  *
3849  *  \param  di      osd iterator
3850  *  \param  key     key for index
3851  *
3852  *  \retval +ve  di points to record with least key not larger than key
3853  *  \retval  0   di points to exact matched key
3854  *  \retval -ve  failure
3855  */
3856
3857 static int osd_it_iam_get(const struct lu_env *env,
3858                       struct dt_it *di, const struct dt_key *key)
3859 {
3860         struct osd_it_iam *it = (struct osd_it_iam *)di;
3861
3862         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3863 }
3864
3865 /**
3866  *  Release Iterator
3867  *
3868  *  \param  di      osd iterator
3869  */
3870
3871 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3872 {
3873         struct osd_it_iam *it = (struct osd_it_iam *)di;
3874
3875         iam_it_put(&it->oi_it);
3876 }
3877
3878 /**
3879  *  Move iterator by one record
3880  *
3881  *  \param  di      osd iterator
3882  *
3883  *  \retval +1   end of container reached
3884  *  \retval  0   success
3885  *  \retval -ve  failure
3886  */
3887
3888 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3889 {
3890         struct osd_it_iam *it = (struct osd_it_iam *)di;
3891
3892         return iam_it_next(&it->oi_it);
3893 }
3894
3895 /**
3896  * Return pointer to the key under iterator.
3897  */
3898
3899 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3900                                  const struct dt_it *di)
3901 {
3902         struct osd_it_iam *it = (struct osd_it_iam *)di;
3903
3904         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3905 }
3906
3907 /**
3908  * Return size of key under iterator (in bytes)
3909  */
3910
3911 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3912 {
3913         struct osd_it_iam *it = (struct osd_it_iam *)di;
3914
3915         return iam_it_key_size(&it->oi_it);
3916 }
3917
3918 static inline void osd_it_append_attrs(struct lu_dirent*ent,
3919                                        __u32 attr,
3920                                        int len,
3921                                        __u16 type)
3922 {
3923         struct luda_type        *lt;
3924         const unsigned           align = sizeof(struct luda_type) - 1;
3925
3926         /* check if file type is required */
3927         if (attr & LUDA_TYPE) {
3928                         len = (len + align) & ~align;
3929
3930                         lt = (void *) ent->lde_name + len;
3931                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3932                         ent->lde_attrs |= LUDA_TYPE;
3933         }
3934
3935         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3936 }
3937
3938 /**
3939  * build lu direct from backend fs dirent.
3940  */
3941
3942 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3943                                       struct lu_fid *fid,
3944                                       __u64 offset,
3945                                       char *name,
3946                                       __u16 namelen,
3947                                       __u16 type,
3948                                       __u32 attr)
3949 {
3950         fid_cpu_to_le(&ent->lde_fid, fid);
3951         ent->lde_attrs = LUDA_FID;
3952
3953         ent->lde_hash = cpu_to_le64(offset);
3954         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3955
3956         strncpy(ent->lde_name, name, namelen);
3957         ent->lde_namelen = cpu_to_le16(namelen);
3958
3959         /* append lustre attributes */
3960         osd_it_append_attrs(ent, attr, namelen, type);
3961 }
3962
3963 /**
3964  * Return pointer to the record under iterator.
3965  */
3966 static int osd_it_iam_rec(const struct lu_env *env,
3967                           const struct dt_it *di,
3968                           struct dt_rec *dtrec,
3969                           __u32 attr)
3970 {
3971         struct osd_it_iam *it        = (struct osd_it_iam *)di;
3972         struct osd_thread_info *info = osd_oti_get(env);
3973         struct lu_fid     *fid       = &info->oti_fid;
3974         const struct osd_fid_pack *rec;
3975         struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3976         char *name;
3977         int namelen;
3978         __u64 hash;
3979         int rc;
3980
3981         name = (char *)iam_it_key_get(&it->oi_it);
3982         if (IS_ERR(name))
3983                 RETURN(PTR_ERR(name));
3984
3985         namelen = iam_it_key_size(&it->oi_it);
3986
3987         rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
3988         if (IS_ERR(rec))
3989                 RETURN(PTR_ERR(rec));
3990
3991         rc = osd_fid_unpack(fid, rec);
3992         if (rc)
3993                 RETURN(rc);
3994
3995         hash = iam_it_store(&it->oi_it);
3996
3997         /* IAM does not store object type in IAM index (dir) */
3998         osd_it_pack_dirent(lde, fid, hash, name, namelen,
3999                            0, LUDA_FID);
4000
4001         return 0;
4002 }
4003
4004 /**
4005  * Returns cookie for current Iterator position.
4006  */
4007 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4008 {
4009         struct osd_it_iam *it = (struct osd_it_iam *)di;
4010
4011         return iam_it_store(&it->oi_it);
4012 }
4013
4014 /**
4015  * Restore iterator from cookie.
4016  *
4017  * \param  di      osd iterator
4018  * \param  hash    Iterator location cookie
4019  *
4020  * \retval +ve  di points to record with least key not larger than key.
4021  * \retval  0   di points to exact matched key
4022  * \retval -ve  failure
4023  */
4024
4025 static int osd_it_iam_load(const struct lu_env *env,
4026                        const struct dt_it *di, __u64 hash)
4027 {
4028         struct osd_it_iam *it = (struct osd_it_iam *)di;
4029
4030         return iam_it_load(&it->oi_it, hash);
4031 }
4032
4033 static const struct dt_index_operations osd_index_iam_ops = {
4034         .dio_lookup         = osd_index_iam_lookup,
4035         .dio_declare_insert = osd_index_declare_iam_insert,
4036         .dio_insert         = osd_index_iam_insert,
4037         .dio_declare_delete = osd_index_declare_iam_delete,
4038         .dio_delete         = osd_index_iam_delete,
4039         .dio_it     = {
4040                 .init     = osd_it_iam_init,
4041                 .fini     = osd_it_iam_fini,
4042                 .get      = osd_it_iam_get,
4043                 .put      = osd_it_iam_put,
4044                 .next     = osd_it_iam_next,
4045                 .key      = osd_it_iam_key,
4046                 .key_size = osd_it_iam_key_size,
4047                 .rec      = osd_it_iam_rec,
4048                 .store    = osd_it_iam_store,
4049                 .load     = osd_it_iam_load
4050         }
4051 };
4052
4053 /**
4054  * Creates or initializes iterator context.
4055  *
4056  * \retval struct osd_it_ea, iterator structure on success
4057  *
4058  */
4059 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4060                                     struct dt_object *dt,
4061                                     __u32 attr,
4062                                     struct lustre_capa *capa)
4063 {
4064         struct osd_object       *obj  = osd_dt_obj(dt);
4065         struct osd_thread_info  *info = osd_oti_get(env);
4066         struct osd_it_ea        *it   = &info->oti_it_ea;
4067         struct lu_object        *lo   = &dt->do_lu;
4068         struct dentry           *obj_dentry = &info->oti_it_dentry;
4069         ENTRY;
4070         LASSERT(lu_object_exists(lo));
4071
4072         obj_dentry->d_inode = obj->oo_inode;
4073         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4074         obj_dentry->d_name.hash = 0;
4075
4076         it->oie_rd_dirent       = 0;
4077         it->oie_it_dirent       = 0;
4078         it->oie_dirent          = NULL;
4079         it->oie_buf             = info->oti_it_ea_buf;
4080         it->oie_obj             = obj;
4081         it->oie_file.f_pos      = 0;
4082         it->oie_file.f_dentry   = obj_dentry;
4083         if (attr & LUDA_64BITHASH)
4084                 it->oie_file.f_flags = O_64BITHASH;
4085         else
4086                 it->oie_file.f_flags = O_32BITHASH;
4087         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4088         it->oie_file.f_op         = obj->oo_inode->i_fop;
4089         it->oie_file.private_data = NULL;
4090         lu_object_get(lo);
4091         RETURN((struct dt_it *) it);
4092 }
4093
4094 /**
4095  * Destroy or finishes iterator context.
4096  *
4097  * \param di iterator structure to be destroyed
4098  */
4099 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4100 {
4101         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4102         struct osd_object    *obj  = it->oie_obj;
4103         struct inode       *inode  = obj->oo_inode;
4104
4105         ENTRY;
4106         it->oie_file.f_op->release(inode, &it->oie_file);
4107         lu_object_put(env, &obj->oo_dt.do_lu);
4108         EXIT;
4109 }
4110
4111 /**
4112  * It position the iterator at given key, so that next lookup continues from
4113  * that key Or it is similar to dio_it->load() but based on a key,
4114  * rather than file position.
4115  *
4116  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4117  * to the beginning.
4118  *
4119  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4120  */
4121 static int osd_it_ea_get(const struct lu_env *env,
4122                          struct dt_it *di, const struct dt_key *key)
4123 {
4124         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4125
4126         ENTRY;
4127         LASSERT(((const char *)key)[0] == '\0');
4128         it->oie_file.f_pos      = 0;
4129         it->oie_rd_dirent       = 0;
4130         it->oie_it_dirent       = 0;
4131         it->oie_dirent          = NULL;
4132
4133         RETURN(+1);
4134 }
4135
4136 /**
4137  * Does nothing
4138  */
4139 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4140 {
4141 }
4142
4143 /**
4144  * It is called internally by ->readdir(). It fills the
4145  * iterator's in-memory data structure with required
4146  * information i.e. name, namelen, rec_size etc.
4147  *
4148  * \param buf in which information to be filled in.
4149  * \param name name of the file in given dir
4150  *
4151  * \retval 0 on success
4152  * \retval 1 on buffer full
4153  */
4154 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4155                                loff_t offset, __u64 ino,
4156                                unsigned d_type)
4157 {
4158         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4159         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4160         struct lu_fid           *fid  = &ent->oied_fid;
4161         struct osd_fid_pack     *rec;
4162         ENTRY;
4163
4164         /* this should never happen */
4165         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4166                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4167                 RETURN(-EIO);
4168         }
4169
4170         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4171             OSD_IT_EA_BUFSIZE)
4172                 RETURN(1);
4173
4174         if (d_type & LDISKFS_DIRENT_LUFID) {
4175                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4176
4177                 if (osd_fid_unpack(fid, rec) != 0)
4178                         fid_zero(fid);
4179
4180                 d_type &= ~LDISKFS_DIRENT_LUFID;
4181         } else {
4182                 fid_zero(fid);
4183         }
4184
4185         ent->oied_ino     = ino;
4186         ent->oied_off     = offset;
4187         ent->oied_namelen = namelen;
4188         ent->oied_type    = d_type;
4189
4190         memcpy(ent->oied_name, name, namelen);
4191
4192         it->oie_rd_dirent++;
4193         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4194         RETURN(0);
4195 }
4196
4197 /**
4198  * Calls ->readdir() to load a directory entry at a time
4199  * and stored it in iterator's in-memory data structure.
4200  *
4201  * \param di iterator's in memory structure
4202  *
4203  * \retval   0 on success
4204  * \retval -ve on error
4205  */
4206 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4207                                const struct dt_it *di)
4208 {
4209         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4210         struct osd_object  *obj   = it->oie_obj;
4211         struct inode       *inode = obj->oo_inode;
4212         struct htree_lock  *hlock = NULL;
4213         int                 result = 0;
4214
4215         ENTRY;
4216         it->oie_dirent = it->oie_buf;
4217         it->oie_rd_dirent = 0;
4218
4219         if (obj->oo_hl_head != NULL) {
4220                 hlock = osd_oti_get(env)->oti_hlock;
4221                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4222                                    inode, LDISKFS_HLOCK_READDIR);
4223         } else {
4224                 cfs_down_read(&obj->oo_ext_idx_sem);
4225         }
4226
4227         result = inode->i_fop->readdir(&it->oie_file, it,
4228                                        (filldir_t) osd_ldiskfs_filldir);
4229
4230         if (hlock != NULL)
4231                 ldiskfs_htree_unlock(hlock);
4232         else
4233                 cfs_up_read(&obj->oo_ext_idx_sem);
4234
4235         if (it->oie_rd_dirent == 0) {
4236                 result = -EIO;
4237         } else {
4238                 it->oie_dirent = it->oie_buf;
4239                 it->oie_it_dirent = 1;
4240         }
4241
4242         RETURN(result);
4243 }
4244
4245 /**
4246  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4247  * to load a directory entry at a time and stored it in
4248  * iterator's in-memory data structure.
4249  *
4250  * \param di iterator's in memory structure
4251  *
4252  * \retval +ve iterator reached to end
4253  * \retval   0 iterator not reached to end
4254  * \retval -ve on error
4255  */
4256 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4257 {
4258         struct osd_it_ea *it = (struct osd_it_ea *)di;
4259         int rc;
4260
4261         ENTRY;
4262
4263         if (it->oie_it_dirent < it->oie_rd_dirent) {
4264                 it->oie_dirent =
4265                         (void *) it->oie_dirent +
4266                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4267                                        it->oie_dirent->oied_namelen);
4268                 it->oie_it_dirent++;
4269                 RETURN(0);
4270         } else {
4271                 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
4272                         rc = +1;
4273                 else
4274                         rc = osd_ldiskfs_it_fill(env, di);
4275         }
4276
4277         RETURN(rc);
4278 }
4279
4280 /**
4281  * Returns the key at current position from iterator's in memory structure.
4282  *
4283  * \param di iterator's in memory structure
4284  *
4285  * \retval key i.e. struct dt_key on success
4286  */
4287 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4288                                     const struct dt_it *di)
4289 {
4290         struct osd_it_ea *it = (struct osd_it_ea *)di;
4291         ENTRY;
4292         RETURN((struct dt_key *)it->oie_dirent->oied_name);
4293 }
4294
4295 /**
4296  * Returns the key's size at current position from iterator's in memory structure.
4297  *
4298  * \param di iterator's in memory structure
4299  *
4300  * \retval key_size i.e. struct dt_key on success
4301  */
4302 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4303 {
4304         struct osd_it_ea *it = (struct osd_it_ea *)di;
4305         ENTRY;
4306         RETURN(it->oie_dirent->oied_namelen);
4307 }
4308
4309
4310 /**
4311  * Returns the value (i.e. fid/igif) at current position from iterator's
4312  * in memory structure.
4313  *
4314  * \param di struct osd_it_ea, iterator's in memory structure
4315  * \param attr attr requested for dirent.
4316  * \param lde lustre dirent
4317  *
4318  * \retval   0 no error and \param lde has correct lustre dirent.
4319  * \retval -ve on error
4320  */
4321 static inline int osd_it_ea_rec(const struct lu_env *env,
4322                                 const struct dt_it *di,
4323                                 struct dt_rec *dtrec,
4324                                 __u32 attr)
4325 {
4326         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4327         struct osd_object       *obj    = it->oie_obj;
4328         struct lu_fid           *fid    = &it->oie_dirent->oied_fid;
4329         struct lu_dirent        *lde    = (struct lu_dirent *)dtrec;
4330         int    rc = 0;
4331
4332         ENTRY;
4333
4334         if (!fid_is_sane(fid))
4335                 rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
4336
4337         if (rc == 0)
4338                 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4339                                    it->oie_dirent->oied_name,
4340                                    it->oie_dirent->oied_namelen,
4341                                    it->oie_dirent->oied_type,
4342                                    attr);
4343         RETURN(rc);
4344 }
4345
4346 /**
4347  * Returns a cookie for current position of the iterator head, so that
4348  * user can use this cookie to load/start the iterator next time.
4349  *
4350  * \param di iterator's in memory structure
4351  *
4352  * \retval cookie for current position, on success
4353  */
4354 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4355 {
4356         struct osd_it_ea *it = (struct osd_it_ea *)di;
4357         ENTRY;
4358         RETURN(it->oie_dirent->oied_off);
4359 }
4360
4361 /**
4362  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4363  * to load a directory entry at a time and stored it i inn,
4364  * in iterator's in-memory data structure.
4365  *
4366  * \param di struct osd_it_ea, iterator's in memory structure
4367  *
4368  * \retval +ve on success
4369  * \retval -ve on error
4370  */
4371 static int osd_it_ea_load(const struct lu_env *env,
4372                           const struct dt_it *di, __u64 hash)
4373 {
4374         struct osd_it_ea *it = (struct osd_it_ea *)di;
4375         int rc;
4376
4377         ENTRY;
4378         it->oie_file.f_pos = hash;
4379
4380         rc =  osd_ldiskfs_it_fill(env, di);
4381         if (rc == 0)
4382                 rc = +1;
4383
4384         RETURN(rc);
4385 }
4386
4387 /**
4388  * Index lookup function for interoperability mode (b11826).
4389  *
4390  * \param key,  key i.e. file name to be searched
4391  *
4392  * \retval +ve, on success
4393  * \retval -ve, on error
4394  */
4395 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4396                                struct dt_rec *rec, const struct dt_key *key,
4397                                struct lustre_capa *capa)
4398 {
4399         struct osd_object *obj = osd_dt_obj(dt);
4400         int rc = 0;
4401
4402         ENTRY;
4403
4404         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4405         LINVRNT(osd_invariant(obj));
4406
4407         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4408                 return -EACCES;
4409
4410         rc = osd_ea_lookup_rec(env, obj, rec, key);
4411
4412         if (rc == 0)
4413                 rc = +1;
4414         RETURN(rc);
4415 }
4416
4417 /**
4418  * Index and Iterator operations for interoperability
4419  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4420  */
4421 static const struct dt_index_operations osd_index_ea_ops = {
4422         .dio_lookup         = osd_index_ea_lookup,
4423         .dio_declare_insert = osd_index_declare_ea_insert,
4424         .dio_insert         = osd_index_ea_insert,
4425         .dio_declare_delete = osd_index_declare_ea_delete,
4426         .dio_delete         = osd_index_ea_delete,
4427         .dio_it     = {
4428                 .init     = osd_it_ea_init,
4429                 .fini     = osd_it_ea_fini,
4430                 .get      = osd_it_ea_get,
4431                 .put      = osd_it_ea_put,
4432                 .next     = osd_it_ea_next,
4433                 .key      = osd_it_ea_key,
4434                 .key_size = osd_it_ea_key_size,
4435                 .rec      = osd_it_ea_rec,
4436                 .store    = osd_it_ea_store,
4437                 .load     = osd_it_ea_load
4438         }
4439 };
4440
4441 static void *osd_key_init(const struct lu_context *ctx,
4442                           struct lu_context_key *key)
4443 {
4444         struct osd_thread_info *info;
4445
4446         OBD_ALLOC_PTR(info);
4447         if (info == NULL)
4448                 return ERR_PTR(-ENOMEM);
4449
4450         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4451         if (info->oti_it_ea_buf == NULL)
4452                 goto out_free_info;
4453
4454         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4455
4456         info->oti_hlock = ldiskfs_htree_lock_alloc();
4457         if (info->oti_hlock == NULL)
4458                 goto out_free_ea;
4459
4460         return info;
4461
4462  out_free_ea:
4463         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4464  out_free_info:
4465         OBD_FREE_PTR(info);
4466         return ERR_PTR(-ENOMEM);
4467 }
4468
4469 static void osd_key_fini(const struct lu_context *ctx,
4470                          struct lu_context_key *key, void* data)
4471 {
4472         struct osd_thread_info *info = data;
4473
4474         if (info->oti_hlock != NULL)
4475                 ldiskfs_htree_lock_free(info->oti_hlock);
4476         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4477         OBD_FREE_PTR(info);
4478 }
4479
4480 static void osd_key_exit(const struct lu_context *ctx,
4481                          struct lu_context_key *key, void *data)
4482 {
4483         struct osd_thread_info *info = data;
4484
4485         LASSERT(info->oti_r_locks == 0);
4486         LASSERT(info->oti_w_locks == 0);
4487         LASSERT(info->oti_txns    == 0);
4488 }
4489
4490 /* type constructor/destructor: osd_type_init, osd_type_fini */
4491 LU_TYPE_INIT_FINI(osd, &osd_key);
4492
4493 static struct lu_context_key osd_key = {
4494         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
4495         .lct_init = osd_key_init,
4496         .lct_fini = osd_key_fini,
4497         .lct_exit = osd_key_exit
4498 };
4499
4500
4501 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4502                            const char *name, struct lu_device *next)
4503 {
4504         return osd_procfs_init(osd_dev(d), name);
4505 }
4506
4507 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4508 {
4509         struct osd_thread_info *info = osd_oti_get(env);
4510         ENTRY;
4511         if (o->od_obj_area != NULL) {
4512                 lu_object_put(env, &o->od_obj_area->do_lu);
4513                 o->od_obj_area = NULL;
4514         }
4515         if (o->od_oi_table != NULL)
4516                 osd_oi_fini(info, &o->od_oi_table, o->od_oi_count);
4517
4518         RETURN(0);
4519 }
4520
4521 static int osd_mount(const struct lu_env *env,
4522                      struct osd_device *o, struct lustre_cfg *cfg)
4523 {
4524         struct lustre_mount_info *lmi;
4525         const char               *dev  = lustre_cfg_string(cfg, 0);
4526         struct lustre_disk_data  *ldd;
4527         struct lustre_sb_info    *lsi;
4528
4529         ENTRY;
4530         if (o->od_mount != NULL) {
4531                 CERROR("Already mounted (%s)\n", dev);
4532                 RETURN(-EEXIST);
4533         }
4534
4535         /* get mount */
4536         lmi = server_get_mount(dev);
4537         if (lmi == NULL) {
4538                 CERROR("Cannot get mount info for %s!\n", dev);
4539                 RETURN(-EFAULT);
4540         }
4541
4542         LASSERT(lmi != NULL);
4543         /* save lustre_mount_info in dt_device */
4544         o->od_mount = lmi;
4545
4546         lsi = s2lsi(lmi->lmi_sb);
4547         ldd = lsi->lsi_ldd;
4548
4549         if (ldd->ldd_flags & LDD_F_IAM_DIR) {
4550                 o->od_iop_mode = 0;
4551                 LCONSOLE_WARN("OSD: IAM mode enabled\n");
4552         } else
4553                 o->od_iop_mode = 1;
4554
4555         o->od_obj_area = NULL;
4556         RETURN(0);
4557 }
4558
4559 static struct lu_device *osd_device_fini(const struct lu_env *env,
4560                                          struct lu_device *d)
4561 {
4562         int rc;
4563         ENTRY;
4564
4565         shrink_dcache_sb(osd_sb(osd_dev(d)));
4566         osd_sync(env, lu2dt_dev(d));
4567
4568         rc = osd_procfs_fini(osd_dev(d));
4569         if (rc) {
4570                 CERROR("proc fini error %d \n", rc);
4571                 RETURN (ERR_PTR(rc));
4572         }
4573
4574         if (osd_dev(d)->od_mount)
4575                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
4576                                  osd_dev(d)->od_mount->lmi_mnt);
4577         osd_dev(d)->od_mount = NULL;
4578
4579         RETURN(NULL);
4580 }
4581
4582 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4583                                           struct lu_device_type *t,
4584                                           struct lustre_cfg *cfg)
4585 {
4586         struct lu_device  *l;
4587         struct osd_device *o;
4588
4589         OBD_ALLOC_PTR(o);
4590         if (o != NULL) {
4591                 int result;
4592
4593                 result = dt_device_init(&o->od_dt_dev, t);
4594                 if (result == 0) {
4595                         l = osd2lu_dev(o);
4596                         l->ld_ops = &osd_lu_ops;
4597                         o->od_dt_dev.dd_ops = &osd_dt_ops;
4598                         cfs_spin_lock_init(&o->od_osfs_lock);
4599                         o->od_osfs_age = cfs_time_shift_64(-1000);
4600                         o->od_capa_hash = init_capa_hash();
4601                         if (o->od_capa_hash == NULL) {
4602                                 dt_device_fini(&o->od_dt_dev);
4603                                 l = ERR_PTR(-ENOMEM);
4604                         }
4605                 } else
4606                         l = ERR_PTR(result);
4607
4608                 if (IS_ERR(l))
4609                         OBD_FREE_PTR(o);
4610         } else
4611                 l = ERR_PTR(-ENOMEM);
4612         return l;
4613 }
4614
4615 static struct lu_device *osd_device_free(const struct lu_env *env,
4616                                          struct lu_device *d)
4617 {
4618         struct osd_device *o = osd_dev(d);
4619         ENTRY;
4620
4621         cleanup_capa_hash(o->od_capa_hash);
4622         dt_device_fini(&o->od_dt_dev);
4623         OBD_FREE_PTR(o);
4624         RETURN(NULL);
4625 }
4626
4627 static int osd_process_config(const struct lu_env *env,
4628                               struct lu_device *d, struct lustre_cfg *cfg)
4629 {
4630         struct osd_device *o = osd_dev(d);
4631         int err;
4632         ENTRY;
4633
4634         switch(cfg->lcfg_command) {
4635         case LCFG_SETUP:
4636                 err = osd_mount(env, o, cfg);
4637                 break;
4638         case LCFG_CLEANUP:
4639                 err = osd_shutdown(env, o);
4640                 break;
4641         default:
4642                 err = -ENOSYS;
4643         }
4644
4645         RETURN(err);
4646 }
4647
4648 static int osd_recovery_complete(const struct lu_env *env,
4649                                  struct lu_device *d)
4650 {
4651         RETURN(0);
4652 }
4653
4654 static int osd_prepare(const struct lu_env *env,
4655                        struct lu_device *pdev,
4656                        struct lu_device *dev)
4657 {
4658         struct osd_device *osd = osd_dev(dev);
4659         struct lustre_sb_info *lsi;
4660         struct lustre_disk_data *ldd;
4661         struct lustre_mount_info  *lmi;
4662         struct osd_thread_info *oti = osd_oti_get(env);
4663         struct dt_object *d;
4664         int result;
4665
4666         ENTRY;
4667         /* 1. initialize oi before any file create or file open */
4668         result = osd_oi_init(oti, &osd->od_oi_table,
4669                              &osd->od_dt_dev, lu2md_dev(pdev));
4670         if (result < 0)
4671                 RETURN(result);
4672
4673         LASSERT(result > 0);
4674         osd->od_oi_count = result;
4675
4676         lmi = osd->od_mount;
4677         lsi = s2lsi(lmi->lmi_sb);
4678         ldd = lsi->lsi_ldd;
4679
4680         /* 2. setup local objects */
4681         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4682         if (result)
4683                 goto out;
4684
4685         /* 3. open remote object dir */
4686         d = dt_store_open(env, lu2dt_dev(dev), "",
4687                           remote_obj_dir, &oti->oti_fid);
4688         if (!IS_ERR(d)) {
4689                 osd->od_obj_area = d;
4690                 result = 0;
4691         } else {
4692                 result = PTR_ERR(d);
4693                 osd->od_obj_area = NULL;
4694         }
4695
4696 out:
4697         RETURN(result);
4698 }
4699
4700 static const struct lu_object_operations osd_lu_obj_ops = {
4701         .loo_object_init      = osd_object_init,
4702         .loo_object_delete    = osd_object_delete,
4703         .loo_object_release   = osd_object_release,
4704         .loo_object_free      = osd_object_free,
4705         .loo_object_print     = osd_object_print,
4706         .loo_object_invariant = osd_object_invariant
4707 };
4708
4709 static const struct lu_device_operations osd_lu_ops = {
4710         .ldo_object_alloc      = osd_object_alloc,
4711         .ldo_process_config    = osd_process_config,
4712         .ldo_recovery_complete = osd_recovery_complete,
4713         .ldo_prepare           = osd_prepare,
4714 };
4715
4716 static const struct lu_device_type_operations osd_device_type_ops = {
4717         .ldto_init = osd_type_init,
4718         .ldto_fini = osd_type_fini,
4719
4720         .ldto_start = osd_type_start,
4721         .ldto_stop  = osd_type_stop,
4722
4723         .ldto_device_alloc = osd_device_alloc,
4724         .ldto_device_free  = osd_device_free,
4725
4726         .ldto_device_init    = osd_device_init,
4727         .ldto_device_fini    = osd_device_fini
4728 };
4729
4730 static struct lu_device_type osd_device_type = {
4731         .ldt_tags     = LU_DEVICE_DT,
4732         .ldt_name     = LUSTRE_OSD_NAME,
4733         .ldt_ops      = &osd_device_type_ops,
4734         .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4735 };
4736
4737 /*
4738  * lprocfs legacy support.
4739  */
4740 static struct obd_ops osd_obd_device_ops = {
4741         .o_owner = THIS_MODULE
4742 };
4743
4744 static struct lu_local_obj_desc llod_osd_rem_obj_dir = {
4745         .llod_name      = remote_obj_dir,
4746         .llod_oid       = OSD_REM_OBJ_DIR_OID,
4747         .llod_is_index  = 1,
4748         .llod_feat      = &dt_directory_features,
4749 };
4750
4751 static int __init osd_mod_init(void)
4752 {
4753         struct lprocfs_static_vars lvars;
4754
4755         osd_oi_mod_init();
4756         llo_local_obj_register(&llod_osd_rem_obj_dir);
4757         lprocfs_osd_init_vars(&lvars);
4758         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4759                                    LUSTRE_OSD_NAME, &osd_device_type);
4760 }
4761
4762 static void __exit osd_mod_exit(void)
4763 {
4764         llo_local_obj_unregister(&llod_osd_rem_obj_dir);
4765         class_unregister_type(LUSTRE_OSD_NAME);
4766 }
4767
4768 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4769 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4770 MODULE_LICENSE("GPL");
4771
4772 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);