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 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * Copyright (c) 2011 Whamcloud, Inc.
  37  */
  38 /*
  39  * This file is part of Lustre, http://www.lustre.org/
  40  * Lustre is a trademark of Sun Microsystems, Inc.
  41  *
  42  * lustre/osd/osd_handler.c
  43  *
  44  * Top-level entry points into osd module
  45  *
  46  * Author: Nikita Danilov <nikita@clusterfs.com>
  47  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  48  */
  49
  50 #ifndef EXPORT_SYMTAB
  51 # define EXPORT_SYMTAB
  52 #endif
  53 #define DEBUG_SUBSYSTEM S_MDS
  54
  55 #include <linux/module.h>
  56
  57 /* LUSTRE_VERSION_CODE */
  58 #include <lustre_ver.h>
  59 /* prerequisite for linux/xattr.h */
  60 #include <linux/types.h>
  61 /* prerequisite for linux/xattr.h */
  62 #include <linux/fs.h>
  63 /* XATTR_{REPLACE,CREATE} */
  64 #include <linux/xattr.h>
  65 /* simple_mkdir() */
  66 #include <lvfs.h>
  67
  68 /*
  69  * struct OBD_{ALLOC,FREE}*()
  70  * OBD_FAIL_CHECK
  71  */
  72 #include <obd_support.h>
  73 /* struct ptlrpc_thread */
  74 #include <lustre_net.h>
  75
  76 /* fid_is_local() */
  77 #include <lustre_fid.h>
  78
  79 #include "osd_internal.h"
  80 #include "osd_igif.h"
  81
  82 /* llo_* api support */
  83 #include <md_object.h>
  84
  85 #ifdef HAVE_LDISKFS_PDO
  86 int ldiskfs_pdo = 1;
  87 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  88                 "ldiskfs with parallel directory operations");
  89 #else
  90 int ldiskfs_pdo = 0;
  91 #endif
  92
  93 static const char dot[] = ".";
  94 static const char dotdot[] = "..";
  95 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  96
  97 struct osd_directory {
  98         struct iam_container od_container;
  99         struct iam_descr     od_descr;
 100 };
 101
 102 struct osd_object {
 103         struct dt_object       oo_dt;
 104         /**
 105          * Inode for file system object represented by this osd_object. This
 106          * inode is pinned for the whole duration of lu_object life.
 107          *
 108          * Not modified concurrently (either setup early during object
 109          * creation, or assigned by osd_object_create() under write lock).
 110          */
 111         struct inode          *oo_inode;
 112         /**
 113          * to protect index ops.
 114          */
 115         struct htree_lock_head *oo_hl_head;
 116         cfs_rw_semaphore_t     oo_ext_idx_sem;
 117         cfs_rw_semaphore_t     oo_sem;
 118         struct osd_directory  *oo_dir;
 119         /** protects inode attributes. */
 120         cfs_spinlock_t         oo_guard;
 121         /**
 122          * Following two members are used to indicate the presence of dot and
 123          * dotdot in the given directory. This is required for interop mode
 124          * (b11826).
 125          */
 126         int                    oo_compat_dot_created;
 127         int                    oo_compat_dotdot_created;
 128
 129         const struct lu_env   *oo_owner;
 130 #ifdef CONFIG_LOCKDEP
 131         struct lockdep_map     oo_dep_map;
 132 #endif
 133 };
 134
 135 static const struct lu_object_operations      osd_lu_obj_ops;
 136 static const struct lu_device_operations      osd_lu_ops;
 137 static       struct lu_context_key            osd_key;
 138 static const struct dt_object_operations      osd_obj_ops;
 139 static const struct dt_object_operations      osd_obj_ea_ops;
 140 static const struct dt_body_operations        osd_body_ops;
 141 static const struct dt_body_operations        osd_body_ops_new;
 142 static const struct dt_index_operations       osd_index_iam_ops;
 143 static const struct dt_index_operations       osd_index_ea_ops;
 144
 145 #define OSD_TRACK_DECLARES
 146 #ifdef OSD_TRACK_DECLARES
 147 #define OSD_DECLARE_OP(oh, op)   {                               \
 148         LASSERT(oh->ot_handle == NULL);                          \
 149         ((oh)->ot_declare_ ##op)++; }
 150 #define OSD_EXEC_OP(handle, op)     {                            \
 151         struct osd_thandle *oh;                                  \
 152         oh = container_of0(handle, struct osd_thandle, ot_super);\
 153         if (((oh)->ot_declare_ ##op) > 0) {                      \
 154                 ((oh)->ot_declare_ ##op)--;                      \
 155         }                                                        \
 156         }
 157 #else
 158 #define OSD_DECLARE_OP(oh, op)
 159 #define OSD_EXEC_OP(oh, op)
 160 #endif
 161
 162 /* There are at most 10 uid/gids are affected in a transaction, and
 163  * that's rename case:
 164  * - 2 for source parent uid & gid;
 165  * - 2 for source child uid & gid ('..' entry update when the child
 166  *   is directory);
 167  * - 2 for target parent uid & gid;
 168  * - 2 for target child uid & gid (if the target child exists);
 169  * - 2 for root uid & gid (last_rcvd, llog, etc);
 170  *
 171  * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
 172  * the id type of each id in the ot_id_array.
 173  */
 174 #define OSD_MAX_UGID_CNT        10
 175
 176 struct osd_thandle {
 177         struct thandle          ot_super;
 178         handle_t               *ot_handle;
 179         struct journal_callback ot_jcb;
 180         cfs_list_t              ot_dcb_list;
 181         /* Link to the device, for debugging. */
 182         struct lu_ref_link     *ot_dev_link;
 183         unsigned short          ot_credits;
 184         unsigned short          ot_id_cnt;
 185         unsigned short          ot_id_type;
 186         uid_t                   ot_id_array[OSD_MAX_UGID_CNT];
 187
 188 #ifdef OSD_TRACK_DECLARES
 189         unsigned char           ot_declare_attr_set;
 190         unsigned char           ot_declare_punch;
 191         unsigned char           ot_declare_xattr_set;
 192         unsigned char           ot_declare_create;
 193         unsigned char           ot_declare_destroy;
 194         unsigned char           ot_declare_ref_add;
 195         unsigned char           ot_declare_ref_del;
 196         unsigned char           ot_declare_write;
 197         unsigned char           ot_declare_insert;
 198         unsigned char           ot_declare_delete;
 199 #endif
 200
 201 #if OSD_THANDLE_STATS
 202         /** time when this handle was allocated */
 203         cfs_time_t oth_alloced;
 204
 205         /** time when this thanle was started */
 206         cfs_time_t oth_started;
 207 #endif
 208 };
 209
 210 /**
 211  * Basic transaction credit op
 212  */
 213 enum dt_txn_op {
 214         DTO_INDEX_INSERT,
 215         DTO_INDEX_DELETE,
 216         DTO_INDEX_UPDATE,
 217         DTO_OBJECT_CREATE,
 218         DTO_OBJECT_DELETE,
 219         DTO_ATTR_SET_BASE,
 220         DTO_XATTR_SET,
 221         DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
 222         DTO_WRITE_BASE,
 223         DTO_WRITE_BLOCK,
 224         DTO_ATTR_SET_CHOWN,
 225
 226         DTO_NR
 227 };
 228
 229 /*
 230  * Helpers.
 231  */
 232 static int lu_device_is_osd(const struct lu_device *d)
 233 {
 234         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
 235 }
 236
 237 static struct osd_device *osd_dt_dev(const struct dt_device *d)
 238 {
 239         LASSERT(lu_device_is_osd(&d->dd_lu_dev));
 240         return container_of0(d, struct osd_device, od_dt_dev);
 241 }
 242
 243 static struct osd_device *osd_dev(const struct lu_device *d)
 244 {
 245         LASSERT(lu_device_is_osd(d));
 246         return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
 247 }
 248
 249 static struct osd_device *osd_obj2dev(const struct osd_object *o)
 250 {
 251         return osd_dev(o->oo_dt.do_lu.lo_dev);
 252 }
 253
 254 static struct super_block *osd_sb(const struct osd_device *dev)
 255 {
 256         return dev->od_mount->lmi_mnt->mnt_sb;
 257 }
 258
 259 static int osd_object_is_root(const struct osd_object *obj)
 260 {
 261         return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
 262 }
 263
 264 static struct osd_object *osd_obj(const struct lu_object *o)
 265 {
 266         LASSERT(lu_device_is_osd(o->lo_dev));
 267         return container_of0(o, struct osd_object, oo_dt.do_lu);
 268 }
 269
 270 static struct osd_object *osd_dt_obj(const struct dt_object *d)
 271 {
 272         return osd_obj(&d->do_lu);
 273 }
 274
 275 static struct lu_device *osd2lu_dev(struct osd_device *osd)
 276 {
 277         return &osd->od_dt_dev.dd_lu_dev;
 278 }
 279
 280 static journal_t *osd_journal(const struct osd_device *dev)
 281 {
 282         return LDISKFS_SB(osd_sb(dev))->s_journal;
 283 }
 284
 285 static int osd_has_index(const struct osd_object *obj)
 286 {
 287         return obj->oo_dt.do_index_ops != NULL;
 288 }
 289
 290 static int osd_object_invariant(const struct lu_object *l)
 291 {
 292         return osd_invariant(osd_obj(l));
 293 }
 294
 295 #ifdef HAVE_QUOTA_SUPPORT
 296 static inline void
 297 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
 298 {
 299         struct md_ucred    *uc = md_ucred(env);
 300         struct cred        *tc;
 301
 302         LASSERT(uc != NULL);
 303
 304         save->oc_uid = current_fsuid();
 305         save->oc_gid = current_fsgid();
 306         save->oc_cap = current_cap();
 307         if ((tc = prepare_creds())) {
 308                 tc->fsuid         = uc->mu_fsuid;
 309                 tc->fsgid         = uc->mu_fsgid;
 310                 commit_creds(tc);
 311         }
 312         /* XXX not suboptimal */
 313         cfs_curproc_cap_unpack(uc->mu_cap);
 314 }
 315
 316 static inline void
 317 osd_pop_ctxt(struct osd_ctxt *save)
 318 {
 319         struct cred *tc;
 320
 321         if ((tc = prepare_creds())) {
 322                 tc->fsuid         = save->oc_uid;
 323                 tc->fsgid         = save->oc_gid;
 324                 tc->cap_effective = save->oc_cap;
 325                 commit_creds(tc);
 326         }
 327 }
 328 #endif
 329
 330 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
 331 {
 332         return lu_context_key_get(&env->le_ctx, &osd_key);
 333 }
 334
 335 /*
 336  * Concurrency: doesn't matter
 337  */
 338 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 339 {
 340         return osd_oti_get(env)->oti_r_locks > 0;
 341 }
 342
 343 /*
 344  * Concurrency: doesn't matter
 345  */
 346 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 347 {
 348         struct osd_thread_info *oti = osd_oti_get(env);
 349         return oti->oti_w_locks > 0 && o->oo_owner == env;
 350 }
 351
 352 /*
 353  * Concurrency: doesn't access mutable data
 354  */
 355 static int osd_root_get(const struct lu_env *env,
 356                         struct dt_device *dev, struct lu_fid *f)
 357 {
 358         struct inode *inode;
 359
 360         inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
 361         LU_IGIF_BUILD(f, inode->i_ino, inode->i_generation);
 362         return 0;
 363 }
 364
 365 static inline int osd_qid_type(struct osd_thandle *oh, int i)
 366 {
 367         return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
 368 }
 369
 370 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
 371 {
 372         oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
 373 }
 374
 375 static void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
 376                             int type, uid_t id, struct inode *inode)
 377 {
 378 #ifdef CONFIG_QUOTA
 379         int i, allocated = 0;
 380         struct osd_object *obj;
 381
 382         LASSERT(dt != NULL);
 383         LASSERT(oh != NULL);
 384         LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
 385                  oh->ot_id_cnt);
 386
 387         /* id entry is allocated in the quota file */
 388         if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
 389                 allocated = 1;
 390
 391         for (i = 0; i < oh->ot_id_cnt; i++) {
 392                 if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
 393                         return;
 394         }
 395
 396         if (unlikely(i >= OSD_MAX_UGID_CNT)) {
 397                 CERROR("more than %d uid/gids for a transaction?\n", i);
 398                 return;
 399         }
 400
 401         oh->ot_id_array[i] = id;
 402         osd_qid_set_type(oh, i, type);
 403         oh->ot_id_cnt++;
 404         obj = osd_dt_obj(dt);
 405         oh->ot_credits += (allocated || id == 0) ?
 406                 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
 407 #endif
 408 }
 409
 410 /*
 411  * OSD object methods.
 412  */
 413
 414 /*
 415  * Concurrency: no concurrent access is possible that early in object
 416  * life-cycle.
 417  */
 418 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 419                                           const struct lu_object_header *hdr,
 420                                           struct lu_device *d)
 421 {
 422         struct osd_object *mo;
 423
 424         OBD_ALLOC_PTR(mo);
 425         if (mo != NULL) {
 426                 struct lu_object *l;
 427
 428                 l = &mo->oo_dt.do_lu;
 429                 dt_object_init(&mo->oo_dt, NULL, d);
 430                 if (osd_dev(d)->od_iop_mode)
 431                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 432                 else
 433                         mo->oo_dt.do_ops = &osd_obj_ops;
 434
 435                 l->lo_ops = &osd_lu_obj_ops;
 436                 cfs_init_rwsem(&mo->oo_sem);
 437                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 438                 cfs_spin_lock_init(&mo->oo_guard);
 439                 return l;
 440         } else {
 441                 return NULL;
 442         }
 443 }
 444
 445 /*
 446  * retrieve object from backend ext fs.
 447  **/
 448 static struct inode *osd_iget(struct osd_thread_info *info,
 449                               struct osd_device *dev,
 450                               const struct osd_inode_id *id)
 451 {
 452         struct inode *inode = NULL;
 453
 454 #ifdef HAVE_EXT4_LDISKFS
 455         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 456         if (IS_ERR(inode))
 457         /* Newer kernels return an error instead of a NULL pointer */
 458                 inode = NULL;
 459 #else
 460         inode = iget(osd_sb(dev), id->oii_ino);
 461 #endif
 462         if (inode == NULL) {
 463                 CERROR("no inode\n");
 464                 inode = ERR_PTR(-EACCES);
 465         } else if (id->oii_gen != OSD_OII_NOGEN &&
 466                    inode->i_generation != id->oii_gen) {
 467                 iput(inode);
 468                 inode = ERR_PTR(-ESTALE);
 469         } else if (inode->i_nlink == 0) {
 470                 /* due to parallel readdir and unlink,
 471                 * we can have dead inode here. */
 472                 CWARN("stale inode\n");
 473                 make_bad_inode(inode);
 474                 iput(inode);
 475                 inode = ERR_PTR(-ESTALE);
 476         } else if (is_bad_inode(inode)) {
 477                 CERROR("bad inode %lx\n",inode->i_ino);
 478                 iput(inode);
 479                 inode = ERR_PTR(-ENOENT);
 480         } else {
 481                 /* Do not update file c/mtime in ldiskfs.
 482                  * NB: we don't have any lock to protect this because we don't
 483                  * have reference on osd_object now, but contention with
 484                  * another lookup + attr_set can't happen in the tiny window
 485                  * between if (...) and set S_NOCMTIME. */
 486                 if (!(inode->i_flags & S_NOCMTIME))
 487                         inode->i_flags |= S_NOCMTIME;
 488         }
 489         return inode;
 490 }
 491
 492 static int osd_fid_lookup(const struct lu_env *env,
 493                           struct osd_object *obj, const struct lu_fid *fid)
 494 {
 495         struct osd_thread_info *info;
 496         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
 497         struct osd_device      *dev;
 498         struct osd_inode_id    *id;
 499         struct inode           *inode;
 500         int                     result;
 501
 502         LINVRNT(osd_invariant(obj));
 503         LASSERT(obj->oo_inode == NULL);
 504         LASSERTF(fid_is_sane(fid) || osd_fid_is_root(fid), DFID, PFID(fid));
 505         /*
 506          * This assertion checks that osd layer sees only local
 507          * fids. Unfortunately it is somewhat expensive (does a
 508          * cache-lookup). Disabling it for production/acceptance-testing.
 509          */
 510         LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
 511
 512         ENTRY;
 513
 514         info = osd_oti_get(env);
 515         dev  = osd_dev(ldev);
 516         id   = &info->oti_id;
 517
 518         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 519                 RETURN(-ENOENT);
 520
 521         result = osd_oi_lookup(info, osd_fid2oi(dev, fid), fid, id);
 522         if (result != 0) {
 523                 if (result == -ENOENT)
 524                         result = 0;
 525                 goto out;
 526         }
 527
 528         inode = osd_iget(info, dev, id);
 529         if (IS_ERR(inode)) {
 530                 /*
 531                  * If fid wasn't found in oi, inode-less object is
 532                  * created, for which lu_object_exists() returns
 533                  * false. This is used in a (frequent) case when
 534                  * objects are created as locking anchors or
 535                  * place holders for objects yet to be created.
 536                  */
 537                 result = PTR_ERR(inode);
 538                 goto out;
 539         }
 540
 541         obj->oo_inode = inode;
 542         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 543         if (dev->od_iop_mode) {
 544                 obj->oo_compat_dot_created = 1;
 545                 obj->oo_compat_dotdot_created = 1;
 546         }
 547
 548         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 549                 goto out;
 550
 551         LASSERT(obj->oo_hl_head == NULL);
 552         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 553         if (obj->oo_hl_head == NULL) {
 554                 obj->oo_inode = NULL;
 555                 iput(inode);
 556                 result = -ENOMEM;
 557         }
 558 out:
 559         LINVRNT(osd_invariant(obj));
 560
 561         RETURN(result);
 562 }
 563
 564 /*
 565  * Concurrency: shouldn't matter.
 566  */
 567 static void osd_object_init0(struct osd_object *obj)
 568 {
 569         LASSERT(obj->oo_inode != NULL);
 570         obj->oo_dt.do_body_ops = &osd_body_ops;
 571         obj->oo_dt.do_lu.lo_header->loh_attr |=
 572                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 573 }
 574
 575 /*
 576  * Concurrency: no concurrent access is possible that early in object
 577  * life-cycle.
 578  */
 579 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 580                            const struct lu_object_conf *unused)
 581 {
 582         struct osd_object *obj = osd_obj(l);
 583         int result;
 584
 585         LINVRNT(osd_invariant(obj));
 586
 587         result = osd_fid_lookup(env, obj, lu_object_fid(l));
 588         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 589         if (result == 0) {
 590                 if (obj->oo_inode != NULL)
 591                         osd_object_init0(obj);
 592         }
 593         LINVRNT(osd_invariant(obj));
 594         return result;
 595 }
 596
 597 /*
 598  * Concurrency: no concurrent access is possible that late in object
 599  * life-cycle.
 600  */
 601 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 602 {
 603         struct osd_object *obj = osd_obj(l);
 604
 605         LINVRNT(osd_invariant(obj));
 606
 607         dt_object_fini(&obj->oo_dt);
 608         if (obj->oo_hl_head != NULL)
 609                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 610         OBD_FREE_PTR(obj);
 611 }
 612
 613 /**
 614  * IAM Iterator
 615  */
 616 static struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
 617                                              const struct iam_container *bag)
 618 {
 619         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 620                                            osd_oti_get(env)->oti_it_ipd);
 621 }
 622
 623 static struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
 624                                               const struct iam_container *bag)
 625 {
 626         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 627                                            osd_oti_get(env)->oti_idx_ipd);
 628 }
 629
 630 static void osd_ipd_put(const struct lu_env *env,
 631                         const struct iam_container *bag,
 632                         struct iam_path_descr *ipd)
 633 {
 634         bag->ic_descr->id_ops->id_ipd_free(ipd);
 635 }
 636
 637 /*
 638  * Concurrency: no concurrent access is possible that late in object
 639  * life-cycle.
 640  */
 641 static void osd_index_fini(struct osd_object *o)
 642 {
 643         struct iam_container *bag;
 644
 645         if (o->oo_dir != NULL) {
 646                 bag = &o->oo_dir->od_container;
 647                 if (o->oo_inode != NULL) {
 648                         if (bag->ic_object == o->oo_inode)
 649                                 iam_container_fini(bag);
 650                 }
 651                 OBD_FREE_PTR(o->oo_dir);
 652                 o->oo_dir = NULL;
 653         }
 654 }
 655
 656 /*
 657  * Concurrency: no concurrent access is possible that late in object
 658  * life-cycle (for all existing callers, that is. New callers have to provide
 659  * their own locking.)
 660  */
 661 static int osd_inode_unlinked(const struct inode *inode)
 662 {
 663         return inode->i_nlink == 0;
 664 }
 665
 666 enum {
 667         OSD_TXN_OI_DELETE_CREDITS    = 20,
 668         OSD_TXN_INODE_DELETE_CREDITS = 20
 669 };
 670
 671 /*
 672  * Journal
 673  */
 674
 675 #if OSD_THANDLE_STATS
 676 /**
 677  * Set time when the handle is allocated
 678  */
 679 static void osd_th_alloced(struct osd_thandle *oth)
 680 {
 681         oth->oth_alloced = cfs_time_current();
 682 }
 683
 684 /**
 685  * Set time when the handle started
 686  */
 687 static void osd_th_started(struct osd_thandle *oth)
 688 {
 689         oth->oth_started = cfs_time_current();
 690 }
 691
 692 /**
 693  * Helper function to convert time interval to microseconds packed in
 694  * long int (default time units for the counter in "stats" initialized
 695  * by lu_time_init() )
 696  */
 697 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 698 {
 699         struct timeval val;
 700
 701         cfs_duration_usec(cfs_time_sub(end, start), &val);
 702         return val.tv_sec * 1000000 + val.tv_usec;
 703 }
 704
 705 /**
 706  * Check whether the we deal with this handle for too long.
 707  */
 708 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 709                                 cfs_time_t alloced, cfs_time_t started,
 710                                 cfs_time_t closed)
 711 {
 712         cfs_time_t now = cfs_time_current();
 713
 714         LASSERT(dev != NULL);
 715
 716         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 717                             interval_to_usec(alloced, started));
 718         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 719                             interval_to_usec(started, closed));
 720         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 721                             interval_to_usec(closed, now));
 722
 723         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 724                 CWARN("transaction handle %p was open for too long: "
 725                       "now "CFS_TIME_T" ,"
 726                       "alloced "CFS_TIME_T" ,"
 727                       "started "CFS_TIME_T" ,"
 728                       "closed "CFS_TIME_T"\n",
 729                       oth, now, alloced, started, closed);
 730                 libcfs_debug_dumpstack(NULL);
 731         }
 732 }
 733
 734 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 735 {                                                                       \
 736         cfs_time_t __closed = cfs_time_current();                       \
 737         cfs_time_t __alloced = oth->oth_alloced;                        \
 738         cfs_time_t __started = oth->oth_started;                        \
 739                                                                         \
 740         expr;                                                           \
 741         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 742 }
 743
 744 #else /* OSD_THANDLE_STATS */
 745
 746 #define osd_th_alloced(h)                  do {} while(0)
 747 #define osd_th_started(h)                  do {} while(0)
 748 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 749
 750 #endif /* OSD_THANDLE_STATS */
 751
 752 /*
 753  * Concurrency: doesn't access mutable data.
 754  */
 755 static int osd_param_is_sane(const struct osd_device *dev,
 756                              const struct thandle *th)
 757 {
 758         struct osd_thandle *oh;
 759         oh = container_of0(th, struct osd_thandle, ot_super);
 760         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 761 }
 762
 763 /*
 764  * Concurrency: shouldn't matter.
 765  */
 766 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 767 static void osd_trans_commit_cb(struct super_block *sb,
 768                                 struct journal_callback *jcb, int error)
 769 #else
 770 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 771 #endif
 772 {
 773         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 774         struct thandle     *th  = &oh->ot_super;
 775         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 776         struct dt_txn_commit_cb *dcb, *tmp;
 777
 778         LASSERT(oh->ot_handle == NULL);
 779
 780         if (error)
 781                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 782
 783         dt_txn_hook_commit(th);
 784
 785         /* call per-transaction callbacks if any */
 786         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage)
 787                 dcb->dcb_func(NULL, th, dcb, error);
 788
 789         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 790         lu_device_put(lud);
 791         th->th_dev = NULL;
 792
 793         lu_context_exit(&th->th_ctx);
 794         lu_context_fini(&th->th_ctx);
 795         OBD_FREE_PTR(oh);
 796 }
 797
 798 static struct thandle *osd_trans_create(const struct lu_env *env,
 799                                         struct dt_device *d)
 800 {
 801         struct osd_thread_info *oti = osd_oti_get(env);
 802         struct osd_thandle     *oh;
 803         struct thandle         *th;
 804         ENTRY;
 805
 806         th = ERR_PTR(-ENOMEM);
 807         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 808         if (oh != NULL) {
 809                 th = &oh->ot_super;
 810                 th->th_dev = d;
 811                 th->th_result = 0;
 812                 th->th_tags = LCT_TX_HANDLE;
 813                 oh->ot_credits = 0;
 814                 oti->oti_dev = osd_dt_dev(d);
 815                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 816                 osd_th_alloced(oh);
 817         }
 818         RETURN(th);
 819 }
 820
 821 /*
 822  * Concurrency: shouldn't matter.
 823  */
 824 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 825                     struct thandle *th)
 826 {
 827         struct osd_thread_info *oti = osd_oti_get(env);
 828         struct osd_device  *dev = osd_dt_dev(d);
 829         handle_t           *jh;
 830         struct osd_thandle *oh;
 831         int rc;
 832
 833         ENTRY;
 834
 835         LASSERT(current->journal_info == NULL);
 836
 837         oh = container_of0(th, struct osd_thandle, ot_super);
 838         LASSERT(oh != NULL);
 839         LASSERT(oh->ot_handle == NULL);
 840
 841         rc = dt_txn_hook_start(env, d, th);
 842         if (rc != 0)
 843                 GOTO(out, rc);
 844
 845         if (!osd_param_is_sane(dev, th)) {
 846                 CWARN("%s: too many transaction credits (%d > %d)\n",
 847                       d->dd_lu_dev.ld_obd->obd_name, oh->ot_credits,
 848                       osd_journal(dev)->j_max_transaction_buffers);
 849                 /* XXX Limit the credits to 'max_transaction_buffers', and
 850                  *     let the underlying filesystem to catch the error if
 851                  *     we really need so many credits.
 852                  *
 853                  *     This should be removed when we can calculate the
 854                  *     credits precisely. */
 855                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 856 #ifdef OSD_TRACK_DECLARES
 857                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 858                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 859                        oh->ot_declare_xattr_set);
 860                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 861                        oh->ot_declare_create, oh->ot_declare_ref_add,
 862                        oh->ot_declare_ref_del, oh->ot_declare_write);
 863                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 864                        oh->ot_declare_insert, oh->ot_declare_delete,
 865                        oh->ot_declare_destroy);
 866 #endif
 867         }
 868
 869         /*
 870          * XXX temporary stuff. Some abstraction layer should
 871          * be used.
 872          */
 873         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 874         osd_th_started(oh);
 875         if (!IS_ERR(jh)) {
 876                 oh->ot_handle = jh;
 877                 LASSERT(oti->oti_txns == 0);
 878                 lu_context_init(&th->th_ctx, th->th_tags);
 879                 lu_context_enter(&th->th_ctx);
 880
 881                 lu_device_get(&d->dd_lu_dev);
 882                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 883                                              "osd-tx", th);
 884
 885                 /*
 886                  * XXX: current rule is that we first start tx,
 887                  *      then lock object(s), but we can't use
 888                  *      this rule for data (due to locking specifics
 889                  *      in ldiskfs). also in long-term we'd like to
 890                  *      use usually-used (locks;tx) ordering. so,
 891                  *      UGLY thing is that we'll use one ordering for
 892                  *      data (ofd) and reverse ordering for metadata
 893                  *      (mdd). then at some point we'll fix the latter
 894                  */
 895                 if (lu_device_is_md(&d->dd_lu_dev)) {
 896                         LASSERT(oti->oti_r_locks == 0);
 897                         LASSERT(oti->oti_w_locks == 0);
 898                 }
 899
 900                 oti->oti_txns++;
 901                 rc = 0;
 902         } else {
 903                 rc = PTR_ERR(jh);
 904         }
 905 out:
 906         RETURN(rc);
 907 }
 908
 909 /*
 910  * Concurrency: shouldn't matter.
 911  */
 912 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 913 {
 914         int                     rc = 0;
 915         struct osd_thandle     *oh;
 916         struct osd_thread_info *oti = osd_oti_get(env);
 917
 918         ENTRY;
 919
 920         oh = container_of0(th, struct osd_thandle, ot_super);
 921
 922         if (oh->ot_handle != NULL) {
 923                 handle_t *hdl = oh->ot_handle;
 924
 925                 hdl->h_sync = th->th_sync;
 926
 927                 /*
 928                  * add commit callback
 929                  * notice we don't do this in osd_trans_start()
 930                  * as underlying transaction can change during truncate
 931                  */
 932                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 933                                          &oh->ot_jcb);
 934
 935                 LASSERT(oti->oti_txns == 1);
 936                 oti->oti_txns--;
 937                 /*
 938                  * XXX: current rule is that we first start tx,
 939                  *      then lock object(s), but we can't use
 940                  *      this rule for data (due to locking specifics
 941                  *      in ldiskfs). also in long-term we'd like to
 942                  *      use usually-used (locks;tx) ordering. so,
 943                  *      UGLY thing is that we'll use one ordering for
 944                  *      data (ofd) and reverse ordering for metadata
 945                  *      (mdd). then at some point we'll fix the latter
 946                  */
 947                 if (lu_device_is_md(&th->th_dev->dd_lu_dev)) {
 948                         LASSERT(oti->oti_r_locks == 0);
 949                         LASSERT(oti->oti_w_locks == 0);
 950                 }
 951                 rc = dt_txn_hook_stop(env, th);
 952                 if (rc != 0)
 953                         CERROR("Failure in transaction hook: %d\n", rc);
 954                 oh->ot_handle = NULL;
 955                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 956                                   rc = ldiskfs_journal_stop(hdl));
 957                 if (rc != 0)
 958                         CERROR("Failure to stop transaction: %d\n", rc);
 959         } else {
 960                 OBD_FREE_PTR(oh);
 961         }
 962
 963         RETURN(rc);
 964 }
 965
 966 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 967 {
 968         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 969                                                ot_super);
 970
 971         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 972
 973         return 0;
 974 }
 975
 976 /*
 977  * Called just before object is freed. Releases all resources except for
 978  * object itself (that is released by osd_object_free()).
 979  *
 980  * Concurrency: no concurrent access is possible that late in object
 981  * life-cycle.
 982  */
 983 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 984 {
 985         struct osd_object *obj   = osd_obj(l);
 986         struct inode      *inode = obj->oo_inode;
 987
 988         LINVRNT(osd_invariant(obj));
 989
 990         /*
 991          * If object is unlinked remove fid->ino mapping from object index.
 992          */
 993
 994         osd_index_fini(obj);
 995         if (inode != NULL) {
 996                 iput(inode);
 997                 obj->oo_inode = NULL;
 998         }
 999 }
1000
1001 /*
1002  * Concurrency: ->loo_object_release() is called under site spin-lock.
1003  */
1004 static void osd_object_release(const struct lu_env *env,
1005                                struct lu_object *l)
1006 {
1007 }
1008
1009 /*
1010  * Concurrency: shouldn't matter.
1011  */
1012 static int osd_object_print(const struct lu_env *env, void *cookie,
1013                             lu_printer_t p, const struct lu_object *l)
1014 {
1015         struct osd_object *o = osd_obj(l);
1016         struct iam_descr  *d;
1017
1018         if (o->oo_dir != NULL)
1019                 d = o->oo_dir->od_container.ic_descr;
1020         else
1021                 d = NULL;
1022         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1023                     o, o->oo_inode,
1024                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1025                     o->oo_inode ? o->oo_inode->i_generation : 0,
1026                     d ? d->id_ops->id_name : "plain");
1027 }
1028
1029 /*
1030  * Concurrency: shouldn't matter.
1031  */
1032 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1033                cfs_kstatfs_t *sfs)
1034 {
1035         struct osd_device *osd = osd_dt_dev(d);
1036         struct super_block *sb = osd_sb(osd);
1037         int result = 0;
1038
1039         cfs_spin_lock(&osd->od_osfs_lock);
1040         /* cache 1 second */
1041         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
1042                 result = ll_do_statfs(sb, &osd->od_kstatfs);
1043                 if (likely(result == 0)) /* N.B. statfs can't really fail */
1044                         osd->od_osfs_age = cfs_time_current_64();
1045         }
1046
1047         if (likely(result == 0))
1048                 *sfs = osd->od_kstatfs;
1049         cfs_spin_unlock(&osd->od_osfs_lock);
1050
1051         return result;
1052 }
1053
1054 /*
1055  * Concurrency: doesn't access mutable data.
1056  */
1057 static void osd_conf_get(const struct lu_env *env,
1058                          const struct dt_device *dev,
1059                          struct dt_device_param *param)
1060 {
1061         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1062
1063         /*
1064          * XXX should be taken from not-yet-existing fs abstraction layer.
1065          */
1066         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1067         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1068         param->ddp_block_shift  = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
1069         param->ddp_mntopts      = 0;
1070         if (test_opt(sb, XATTR_USER))
1071                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1072         if (test_opt(sb, POSIX_ACL))
1073                 param->ddp_mntopts |= MNTOPT_ACL;
1074
1075 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1076         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1077                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1078         else
1079 #endif
1080                 param->ddp_max_ea_size = sb->s_blocksize;
1081
1082 }
1083
1084 /**
1085  * Helper function to get and fill the buffer with input values.
1086  */
1087 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1088 {
1089         struct lu_buf *buf;
1090
1091         buf = &osd_oti_get(env)->oti_buf;
1092         buf->lb_buf = area;
1093         buf->lb_len = len;
1094         return buf;
1095 }
1096
1097 /*
1098  * Concurrency: shouldn't matter.
1099  */
1100 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1101 {
1102         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
1103         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1104 }
1105
1106 /**
1107  * Start commit for OSD device.
1108  *
1109  * An implementation of dt_commit_async method for OSD device.
1110  * Asychronously starts underlayng fs sync and thereby a transaction
1111  * commit.
1112  *
1113  * \param env environment
1114  * \param d dt device
1115  *
1116  * \see dt_device_operations
1117  */
1118 static int osd_commit_async(const struct lu_env *env,
1119                             struct dt_device *d)
1120 {
1121         struct super_block *s = osd_sb(osd_dt_dev(d));
1122         ENTRY;
1123
1124         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
1125         RETURN(s->s_op->sync_fs(s, 0));
1126 }
1127
1128 /*
1129  * Concurrency: shouldn't matter.
1130  */
1131
1132 static void osd_ro(const struct lu_env *env, struct dt_device *d)
1133 {
1134         struct super_block *sb = osd_sb(osd_dt_dev(d));
1135         ENTRY;
1136
1137         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
1138
1139         __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1140         EXIT;
1141 }
1142
1143 /*
1144  * Concurrency: serialization provided by callers.
1145  */
1146 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1147                               int mode, unsigned long timeout, __u32 alg,
1148                               struct lustre_capa_key *keys)
1149 {
1150         struct osd_device *dev = osd_dt_dev(d);
1151         ENTRY;
1152
1153         dev->od_fl_capa = mode;
1154         dev->od_capa_timeout = timeout;
1155         dev->od_capa_alg = alg;
1156         dev->od_capa_keys = keys;
1157         RETURN(0);
1158 }
1159
1160 /**
1161  * Concurrency: serialization provided by callers.
1162  */
1163 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1164                                struct dt_quota_ctxt *ctxt, void *data)
1165 {
1166         struct obd_device *obd = (void *)ctxt;
1167         struct vfsmount *mnt = (struct vfsmount *)data;
1168         ENTRY;
1169
1170         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1171         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1172         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1173         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1174         obd->obd_lvfs_ctxt.fs = get_ds();
1175
1176         EXIT;
1177 }
1178
1179 /**
1180  * Note: we do not count into QUOTA here.
1181  * If we mount with --data_journal we may need more.
1182  */
1183 static const int osd_dto_credits_noquota[DTO_NR] = {
1184         /**
1185          * Insert/Delete.
1186          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1187          * SINGLEDATA_TRANS_BLOCKS(8)
1188          * XXX Note: maybe iam need more, since iam have more level than
1189          *           EXT3 htree.
1190          */
1191         [DTO_INDEX_INSERT]  = 16,
1192         [DTO_INDEX_DELETE]  = 16,
1193         /**
1194          * Unused now
1195          */
1196         [DTO_INDEX_UPDATE]  = 16,
1197         /**
1198          * Create a object. The same as create object in EXT3.
1199          * DATA_TRANS_BLOCKS(14) +
1200          * INDEX_EXTRA_BLOCKS(8) +
1201          * 3(inode bits, groups, GDT)
1202          */
1203         [DTO_OBJECT_CREATE] = 25,
1204         /**
1205          * XXX: real credits to be fixed
1206          */
1207         [DTO_OBJECT_DELETE] = 25,
1208         /**
1209          * Attr set credits (inode)
1210          */
1211         [DTO_ATTR_SET_BASE] = 1,
1212         /**
1213          * Xattr set. The same as xattr of EXT3.
1214          * DATA_TRANS_BLOCKS(14)
1215          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1216          * are also counted in. Do not know why?
1217          */
1218         [DTO_XATTR_SET]     = 14,
1219         [DTO_LOG_REC]       = 14,
1220         /**
1221          * credits for inode change during write.
1222          */
1223         [DTO_WRITE_BASE]    = 3,
1224         /**
1225          * credits for single block write.
1226          */
1227         [DTO_WRITE_BLOCK]   = 14,
1228         /**
1229          * Attr set credits for chown.
1230          * This is extra credits for setattr, and it is null without quota
1231          */
1232         [DTO_ATTR_SET_CHOWN]= 0
1233 };
1234
1235 static const struct dt_device_operations osd_dt_ops = {
1236         .dt_root_get       = osd_root_get,
1237         .dt_statfs         = osd_statfs,
1238         .dt_trans_create   = osd_trans_create,
1239         .dt_trans_start    = osd_trans_start,
1240         .dt_trans_stop     = osd_trans_stop,
1241         .dt_trans_cb_add   = osd_trans_cb_add,
1242         .dt_conf_get       = osd_conf_get,
1243         .dt_sync           = osd_sync,
1244         .dt_ro             = osd_ro,
1245         .dt_commit_async   = osd_commit_async,
1246         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1247         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1248 };
1249
1250 static void osd_object_read_lock(const struct lu_env *env,
1251                                  struct dt_object *dt, unsigned role)
1252 {
1253         struct osd_object *obj = osd_dt_obj(dt);
1254         struct osd_thread_info *oti = osd_oti_get(env);
1255
1256         LINVRNT(osd_invariant(obj));
1257
1258         LASSERT(obj->oo_owner != env);
1259         cfs_down_read_nested(&obj->oo_sem, role);
1260
1261         LASSERT(obj->oo_owner == NULL);
1262         oti->oti_r_locks++;
1263 }
1264
1265 static void osd_object_write_lock(const struct lu_env *env,
1266                                   struct dt_object *dt, unsigned role)
1267 {
1268         struct osd_object *obj = osd_dt_obj(dt);
1269         struct osd_thread_info *oti = osd_oti_get(env);
1270
1271         LINVRNT(osd_invariant(obj));
1272
1273         LASSERT(obj->oo_owner != env);
1274         cfs_down_write_nested(&obj->oo_sem, role);
1275
1276         LASSERT(obj->oo_owner == NULL);
1277         obj->oo_owner = env;
1278         oti->oti_w_locks++;
1279 }
1280
1281 static void osd_object_read_unlock(const struct lu_env *env,
1282                                    struct dt_object *dt)
1283 {
1284         struct osd_object *obj = osd_dt_obj(dt);
1285         struct osd_thread_info *oti = osd_oti_get(env);
1286
1287         LINVRNT(osd_invariant(obj));
1288
1289         LASSERT(oti->oti_r_locks > 0);
1290         oti->oti_r_locks--;
1291         cfs_up_read(&obj->oo_sem);
1292 }
1293
1294 static void osd_object_write_unlock(const struct lu_env *env,
1295                                     struct dt_object *dt)
1296 {
1297         struct osd_object *obj = osd_dt_obj(dt);
1298         struct osd_thread_info *oti = osd_oti_get(env);
1299
1300         LINVRNT(osd_invariant(obj));
1301
1302         LASSERT(obj->oo_owner == env);
1303         LASSERT(oti->oti_w_locks > 0);
1304         oti->oti_w_locks--;
1305         obj->oo_owner = NULL;
1306         cfs_up_write(&obj->oo_sem);
1307 }
1308
1309 static int osd_object_write_locked(const struct lu_env *env,
1310                                    struct dt_object *dt)
1311 {
1312         struct osd_object *obj = osd_dt_obj(dt);
1313
1314         LINVRNT(osd_invariant(obj));
1315
1316         return obj->oo_owner == env;
1317 }
1318
1319 static int capa_is_sane(const struct lu_env *env,
1320                         struct osd_device *dev,
1321                         struct lustre_capa *capa,
1322                         struct lustre_capa_key *keys)
1323 {
1324         struct osd_thread_info *oti = osd_oti_get(env);
1325         struct lustre_capa *tcapa = &oti->oti_capa;
1326         struct obd_capa *oc;
1327         int i, rc = 0;
1328         ENTRY;
1329
1330         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1331         if (oc) {
1332                 if (capa_is_expired(oc)) {
1333                         DEBUG_CAPA(D_ERROR, capa, "expired");
1334                         rc = -ESTALE;
1335                 }
1336                 capa_put(oc);
1337                 RETURN(rc);
1338         }
1339
1340         if (capa_is_expired_sec(capa)) {
1341                 DEBUG_CAPA(D_ERROR, capa, "expired");
1342                 RETURN(-ESTALE);
1343         }
1344
1345         cfs_spin_lock(&capa_lock);
1346         for (i = 0; i < 2; i++) {
1347                 if (keys[i].lk_keyid == capa->lc_keyid) {
1348                         oti->oti_capa_key = keys[i];
1349                         break;
1350                 }
1351         }
1352         cfs_spin_unlock(&capa_lock);
1353
1354         if (i == 2) {
1355                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1356                 RETURN(-ESTALE);
1357         }
1358
1359         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1360         if (rc)
1361                 RETURN(rc);
1362
1363         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1364                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1365                 RETURN(-EACCES);
1366         }
1367
1368         oc = capa_add(dev->od_capa_hash, capa);
1369         capa_put(oc);
1370
1371         RETURN(0);
1372 }
1373
1374 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1375                            struct lustre_capa *capa, __u64 opc)
1376 {
1377         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1378         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1379         struct md_capainfo *ci;
1380         int rc;
1381
1382         if (!dev->od_fl_capa)
1383                 return 0;
1384
1385         if (capa == BYPASS_CAPA)
1386                 return 0;
1387
1388         ci = md_capainfo(env);
1389         if (unlikely(!ci))
1390                 return 0;
1391
1392         if (ci->mc_auth == LC_ID_NONE)
1393                 return 0;
1394
1395         if (!capa) {
1396                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1397                 return -EACCES;
1398         }
1399
1400         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1401                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1402                            PFID(fid));
1403                 return -EACCES;
1404         }
1405
1406         if (!capa_opc_supported(capa, opc)) {
1407                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1408                 return -EACCES;
1409         }
1410
1411         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1412                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1413                 return -EACCES;
1414         }
1415
1416         return 0;
1417 }
1418
1419 static struct timespec *osd_inode_time(const struct lu_env *env,
1420                                        struct inode *inode, __u64 seconds)
1421 {
1422         struct osd_thread_info *oti = osd_oti_get(env);
1423         struct timespec        *t   = &oti->oti_time;
1424
1425         t->tv_sec  = seconds;
1426         t->tv_nsec = 0;
1427         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1428         return t;
1429 }
1430
1431
1432 static void osd_inode_getattr(const struct lu_env *env,
1433                               struct inode *inode, struct lu_attr *attr)
1434 {
1435         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1436                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1437                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1438
1439         attr->la_atime      = LTIME_S(inode->i_atime);
1440         attr->la_mtime      = LTIME_S(inode->i_mtime);
1441         attr->la_ctime      = LTIME_S(inode->i_ctime);
1442         attr->la_mode       = inode->i_mode;
1443         attr->la_size       = i_size_read(inode);
1444         attr->la_blocks     = inode->i_blocks;
1445         attr->la_uid        = inode->i_uid;
1446         attr->la_gid        = inode->i_gid;
1447         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1448         attr->la_nlink      = inode->i_nlink;
1449         attr->la_rdev       = inode->i_rdev;
1450         attr->la_blksize    = ll_inode_blksize(inode);
1451         attr->la_blkbits    = inode->i_blkbits;
1452 }
1453
1454 static int osd_attr_get(const struct lu_env *env,
1455                         struct dt_object *dt,
1456                         struct lu_attr *attr,
1457                         struct lustre_capa *capa)
1458 {
1459         struct osd_object *obj = osd_dt_obj(dt);
1460
1461         LASSERT(dt_object_exists(dt));
1462         LINVRNT(osd_invariant(obj));
1463
1464         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1465                 return -EACCES;
1466
1467         cfs_spin_lock(&obj->oo_guard);
1468         osd_inode_getattr(env, obj->oo_inode, attr);
1469         cfs_spin_unlock(&obj->oo_guard);
1470         return 0;
1471 }
1472
1473 static int osd_declare_attr_set(const struct lu_env *env,
1474                                 struct dt_object *dt,
1475                                 const struct lu_attr *attr,
1476                                 struct thandle *handle)
1477 {
1478         struct osd_thandle *oh;
1479         struct osd_object *obj;
1480
1481         LASSERT(dt != NULL);
1482         LASSERT(handle != NULL);
1483
1484         obj = osd_dt_obj(dt);
1485         LASSERT(osd_invariant(obj));
1486
1487         oh = container_of0(handle, struct osd_thandle, ot_super);
1488         LASSERT(oh->ot_handle == NULL);
1489
1490         OSD_DECLARE_OP(oh, attr_set);
1491         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1492
1493         if (attr && attr->la_valid & LA_UID) {
1494                 if (obj->oo_inode)
1495                         osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
1496                                         obj->oo_inode);
1497                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1498         }
1499         if (attr && attr->la_valid & LA_GID) {
1500                 if (obj->oo_inode)
1501                         osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
1502                                         obj->oo_inode);
1503                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1504         }
1505
1506         return 0;
1507 }
1508
1509 static int osd_inode_setattr(const struct lu_env *env,
1510                              struct inode *inode, const struct lu_attr *attr)
1511 {
1512         __u64 bits;
1513
1514         bits = attr->la_valid;
1515
1516         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1517
1518 #ifdef HAVE_QUOTA_SUPPORT
1519         if ((bits & LA_UID && attr->la_uid != inode->i_uid) ||
1520             (bits & LA_GID && attr->la_gid != inode->i_gid)) {
1521                 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1522                 struct iattr iattr;
1523                 int rc;
1524
1525                 iattr.ia_valid = 0;
1526                 if (bits & LA_UID)
1527                         iattr.ia_valid |= ATTR_UID;
1528                 if (bits & LA_GID)
1529                         iattr.ia_valid |= ATTR_GID;
1530                 iattr.ia_uid = attr->la_uid;
1531                 iattr.ia_gid = attr->la_gid;
1532                 osd_push_ctxt(env, save);
1533                 rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1534                 osd_pop_ctxt(save);
1535                 if (rc != 0)
1536                         return rc;
1537         }
1538 #endif
1539
1540         if (bits & LA_ATIME)
1541                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1542         if (bits & LA_CTIME)
1543                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1544         if (bits & LA_MTIME)
1545                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1546         if (bits & LA_SIZE) {
1547                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1548                 i_size_write(inode, attr->la_size);
1549         }
1550
1551 #if 0
1552         /* OSD should not change "i_blocks" which is used by quota.
1553          * "i_blocks" should be changed by ldiskfs only. */
1554         if (bits & LA_BLOCKS)
1555                 inode->i_blocks = attr->la_blocks;
1556 #endif
1557         if (bits & LA_MODE)
1558                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1559                         (attr->la_mode & ~S_IFMT);
1560         if (bits & LA_UID)
1561                 inode->i_uid    = attr->la_uid;
1562         if (bits & LA_GID)
1563                 inode->i_gid    = attr->la_gid;
1564         if (bits & LA_NLINK)
1565                 inode->i_nlink  = attr->la_nlink;
1566         if (bits & LA_RDEV)
1567                 inode->i_rdev   = attr->la_rdev;
1568
1569         if (bits & LA_FLAGS) {
1570                 /* always keep S_NOCMTIME */
1571                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1572                                  S_NOCMTIME;
1573         }
1574         return 0;
1575 }
1576
1577 static int osd_attr_set(const struct lu_env *env,
1578                         struct dt_object *dt,
1579                         const struct lu_attr *attr,
1580                         struct thandle *handle,
1581                         struct lustre_capa *capa)
1582 {
1583         struct osd_object *obj = osd_dt_obj(dt);
1584         int rc;
1585
1586         LASSERT(handle != NULL);
1587         LASSERT(dt_object_exists(dt));
1588         LASSERT(osd_invariant(obj));
1589
1590         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1591                 return -EACCES;
1592
1593         OSD_EXEC_OP(handle, attr_set);
1594
1595         cfs_spin_lock(&obj->oo_guard);
1596         rc = osd_inode_setattr(env, obj->oo_inode, attr);
1597         cfs_spin_unlock(&obj->oo_guard);
1598
1599         if (!rc)
1600                 obj->oo_inode->i_sb->s_op->dirty_inode(obj->oo_inode);
1601         return rc;
1602 }
1603
1604 /*
1605  * Object creation.
1606  *
1607  * XXX temporary solution.
1608  */
1609 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1610                           struct lu_attr *attr, struct thandle *th)
1611 {
1612         return 0;
1613 }
1614
1615 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1616                            struct lu_attr *attr, struct thandle *th)
1617 {
1618         osd_object_init0(obj);
1619         if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1620                 unlock_new_inode(obj->oo_inode);
1621         return 0;
1622 }
1623
1624 static struct dentry * osd_child_dentry_get(const struct lu_env *env,
1625                                             struct osd_object *obj,
1626                                             const char *name,
1627                                             const int namelen)
1628 {
1629         struct osd_thread_info *info   = osd_oti_get(env);
1630         struct dentry *child_dentry = &info->oti_child_dentry;
1631         struct dentry *obj_dentry = &info->oti_obj_dentry;
1632
1633         obj_dentry->d_inode = obj->oo_inode;
1634         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
1635         obj_dentry->d_name.hash = 0;
1636
1637         child_dentry->d_name.hash = 0;
1638         child_dentry->d_parent = obj_dentry;
1639         child_dentry->d_name.name = name;
1640         child_dentry->d_name.len = namelen;
1641         return child_dentry;
1642 }
1643
1644
1645 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1646                       cfs_umode_t mode,
1647                       struct dt_allocation_hint *hint,
1648                       struct thandle *th)
1649 {
1650         int result;
1651         struct osd_device  *osd = osd_obj2dev(obj);
1652         struct osd_thandle *oth;
1653         struct dt_object   *parent;
1654         struct inode       *inode;
1655 #ifdef HAVE_QUOTA_SUPPORT
1656         struct osd_ctxt    *save = &info->oti_ctxt;
1657 #endif
1658
1659         LINVRNT(osd_invariant(obj));
1660         LASSERT(obj->oo_inode == NULL);
1661         LASSERT(obj->oo_hl_head == NULL);
1662
1663         if (S_ISDIR(mode) && ldiskfs_pdo) {
1664                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1665                 if (obj->oo_hl_head == NULL)
1666                         return -ENOMEM;
1667         }
1668
1669         oth = container_of(th, struct osd_thandle, ot_super);
1670         LASSERT(oth->ot_handle->h_transaction != NULL);
1671
1672         if (hint && hint->dah_parent)
1673                 parent = hint->dah_parent;
1674         else
1675                 parent = osd->od_obj_area;
1676
1677         LASSERT(parent != NULL);
1678         LASSERT(osd_dt_obj(parent)->oo_inode->i_op != NULL);
1679
1680 #ifdef HAVE_QUOTA_SUPPORT
1681         osd_push_ctxt(info->oti_env, save);
1682 #endif
1683         inode = ldiskfs_create_inode(oth->ot_handle,
1684                                      osd_dt_obj(parent)->oo_inode, mode);
1685 #ifdef HAVE_QUOTA_SUPPORT
1686         osd_pop_ctxt(save);
1687 #endif
1688         if (!IS_ERR(inode)) {
1689                 /* Do not update file c/mtime in ldiskfs.
1690                  * NB: don't need any lock because no contention at this
1691                  * early stage */
1692                 inode->i_flags |= S_NOCMTIME;
1693                 obj->oo_inode = inode;
1694                 result = 0;
1695         } else {
1696                 if (obj->oo_hl_head != NULL) {
1697                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1698                         obj->oo_hl_head = NULL;
1699                 }
1700                 result = PTR_ERR(inode);
1701         }
1702         LINVRNT(osd_invariant(obj));
1703         return result;
1704 }
1705
1706 enum {
1707         OSD_NAME_LEN = 255
1708 };
1709
1710 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1711                      struct lu_attr *attr,
1712                      struct dt_allocation_hint *hint,
1713                      struct dt_object_format *dof,
1714                      struct thandle *th)
1715 {
1716         int result;
1717         struct osd_thandle *oth;
1718         struct osd_device *osd = osd_obj2dev(obj);
1719         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1720
1721         LASSERT(S_ISDIR(attr->la_mode));
1722
1723         oth = container_of(th, struct osd_thandle, ot_super);
1724         LASSERT(oth->ot_handle->h_transaction != NULL);
1725         result = osd_mkfile(info, obj, mode, hint, th);
1726         if (result == 0 && osd->od_iop_mode == 0) {
1727                 LASSERT(obj->oo_inode != NULL);
1728                 /*
1729                  * XXX uh-oh... call low-level iam function directly.
1730                  */
1731
1732                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1733                                          sizeof (struct osd_fid_pack),
1734                                          oth->ot_handle);
1735         }
1736         return result;
1737 }
1738
1739 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1740                         struct lu_attr *attr,
1741                         struct dt_allocation_hint *hint,
1742                         struct dt_object_format *dof,
1743                         struct thandle *th)
1744 {
1745         int result;
1746         struct osd_thandle *oth;
1747         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1748
1749         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1750
1751         LASSERT(S_ISREG(attr->la_mode));
1752
1753         oth = container_of(th, struct osd_thandle, ot_super);
1754         LASSERT(oth->ot_handle->h_transaction != NULL);
1755
1756         result = osd_mkfile(info, obj, mode, hint, th);
1757         if (result == 0) {
1758                 LASSERT(obj->oo_inode != NULL);
1759                 if (feat->dif_flags & DT_IND_VARKEY)
1760                         result = iam_lvar_create(obj->oo_inode,
1761                                                  feat->dif_keysize_max,
1762                                                  feat->dif_ptrsize,
1763                                                  feat->dif_recsize_max,
1764                                                  oth->ot_handle);
1765                 else
1766                         result = iam_lfix_create(obj->oo_inode,
1767                                                  feat->dif_keysize_max,
1768                                                  feat->dif_ptrsize,
1769                                                  feat->dif_recsize_max,
1770                                                  oth->ot_handle);
1771
1772         }
1773         return result;
1774 }
1775
1776 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1777                      struct lu_attr *attr,
1778                      struct dt_allocation_hint *hint,
1779                      struct dt_object_format *dof,
1780                      struct thandle *th)
1781 {
1782         LASSERT(S_ISREG(attr->la_mode));
1783         return osd_mkfile(info, obj, (attr->la_mode &
1784                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1785 }
1786
1787 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1788                      struct lu_attr *attr,
1789                      struct dt_allocation_hint *hint,
1790                      struct dt_object_format *dof,
1791                      struct thandle *th)
1792 {
1793         LASSERT(S_ISLNK(attr->la_mode));
1794         return osd_mkfile(info, obj, (attr->la_mode &
1795                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1796 }
1797
1798 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1799                      struct lu_attr *attr,
1800                      struct dt_allocation_hint *hint,
1801                      struct dt_object_format *dof,
1802                      struct thandle *th)
1803 {
1804         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1805         int result;
1806
1807         LINVRNT(osd_invariant(obj));
1808         LASSERT(obj->oo_inode == NULL);
1809         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1810                 S_ISFIFO(mode) || S_ISSOCK(mode));
1811
1812         result = osd_mkfile(info, obj, mode, hint, th);
1813         if (result == 0) {
1814                 LASSERT(obj->oo_inode != NULL);
1815                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1816         }
1817         LINVRNT(osd_invariant(obj));
1818         return result;
1819 }
1820
1821 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1822                               struct lu_attr *,
1823                               struct dt_allocation_hint *hint,
1824                               struct dt_object_format *dof,
1825                               struct thandle *);
1826
1827 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1828 {
1829         osd_obj_type_f result;
1830
1831         switch (type) {
1832         case DFT_DIR:
1833                 result = osd_mkdir;
1834                 break;
1835         case DFT_REGULAR:
1836                 result = osd_mkreg;
1837                 break;
1838         case DFT_SYM:
1839                 result = osd_mksym;
1840                 break;
1841         case DFT_NODE:
1842                 result = osd_mknod;
1843                 break;
1844         case DFT_INDEX:
1845                 result = osd_mk_index;
1846                 break;
1847
1848         default:
1849                 LBUG();
1850                 break;
1851         }
1852         return result;
1853 }
1854
1855
1856 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1857                         struct dt_object *parent, cfs_umode_t child_mode)
1858 {
1859         LASSERT(ah);
1860
1861         memset(ah, 0, sizeof(*ah));
1862         ah->dah_parent = parent;
1863         ah->dah_mode = child_mode;
1864 }
1865
1866 /**
1867  * Helper function for osd_object_create()
1868  *
1869  * \retval 0, on success
1870  */
1871 static int __osd_object_create(struct osd_thread_info *info,
1872                                struct osd_object *obj, struct lu_attr *attr,
1873                                struct dt_allocation_hint *hint,
1874                                struct dt_object_format *dof,
1875                                struct thandle *th)
1876 {
1877
1878         int result;
1879
1880         result = osd_create_pre(info, obj, attr, th);
1881         if (result == 0) {
1882                 result = osd_create_type_f(dof->dof_type)(info, obj,
1883                                            attr, hint, dof, th);
1884                 if (result == 0)
1885                         result = osd_create_post(info, obj, attr, th);
1886         }
1887         return result;
1888 }
1889
1890 /**
1891  * Helper function for osd_object_create()
1892  *
1893  * \retval 0, on success
1894  */
1895 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1896                            const struct lu_fid *fid, struct thandle *th)
1897 {
1898         struct osd_thread_info *info = osd_oti_get(env);
1899         struct osd_inode_id    *id   = &info->oti_id;
1900         struct osd_device      *osd  = osd_obj2dev(obj);
1901         struct md_ucred        *uc   = md_ucred(env);
1902
1903         LASSERT(obj->oo_inode != NULL);
1904         LASSERT(uc != NULL);
1905
1906         id->oii_ino = obj->oo_inode->i_ino;
1907         id->oii_gen = obj->oo_inode->i_generation;
1908
1909         return osd_oi_insert(info, osd_fid2oi(osd, fid), fid, id, th,
1910                              uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
1911 }
1912
1913 static int osd_declare_object_create(const struct lu_env *env,
1914                                      struct dt_object *dt,
1915                                      struct lu_attr *attr,
1916                                      struct dt_allocation_hint *hint,
1917                                      struct dt_object_format *dof,
1918                                      struct thandle *handle)
1919 {
1920         struct osd_thandle *oh;
1921
1922         LASSERT(handle != NULL);
1923
1924         oh = container_of0(handle, struct osd_thandle, ot_super);
1925         LASSERT(oh->ot_handle == NULL);
1926
1927         OSD_DECLARE_OP(oh, create);
1928         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1929         /* XXX: So far, only normal fid needs be inserted into the oi,
1930          *      things could be changed later. Revise following code then. */
1931         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1932                 OSD_DECLARE_OP(oh, insert);
1933                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1934         }
1935         /* If this is directory, then we expect . and .. to be inserted as
1936          * well. The one directory block always needs to be created for the
1937          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1938          * block), there is no danger of needing a tree for the first block.
1939          */
1940         if (attr && S_ISDIR(attr->la_mode)) {
1941                 OSD_DECLARE_OP(oh, insert);
1942                 OSD_DECLARE_OP(oh, insert);
1943                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1944         }
1945
1946         if (attr) {
1947                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1948                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1949         }
1950         return 0;
1951 }
1952
1953 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1954                              struct lu_attr *attr,
1955                              struct dt_allocation_hint *hint,
1956                              struct dt_object_format *dof,
1957                              struct thandle *th)
1958 {
1959         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1960         struct osd_object      *obj    = osd_dt_obj(dt);
1961         struct osd_thread_info *info   = osd_oti_get(env);
1962         int result;
1963
1964         ENTRY;
1965
1966         LINVRNT(osd_invariant(obj));
1967         LASSERT(!dt_object_exists(dt));
1968         LASSERT(osd_write_locked(env, obj));
1969         LASSERT(th != NULL);
1970
1971         OSD_EXEC_OP(th, create);
1972
1973         result = __osd_object_create(info, obj, attr, hint, dof, th);
1974         if (result == 0)
1975                 result = __osd_oi_insert(env, obj, fid, th);
1976
1977         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1978         LASSERT(osd_invariant(obj));
1979         RETURN(result);
1980 }
1981
1982 /**
1983  * Called to destroy on-disk representation of the object
1984  *
1985  * Concurrency: must be locked
1986  */
1987 static int osd_declare_object_destroy(const struct lu_env *env,
1988                                       struct dt_object *dt,
1989                                       struct thandle *th)
1990 {
1991         struct osd_object      *obj = osd_dt_obj(dt);
1992         struct inode           *inode = obj->oo_inode;
1993         struct osd_thandle     *oh;
1994         ENTRY;
1995
1996         oh = container_of0(th, struct osd_thandle, ot_super);
1997         LASSERT(oh->ot_handle == NULL);
1998         LASSERT(inode);
1999
2000         OSD_DECLARE_OP(oh, destroy);
2001         OSD_DECLARE_OP(oh, delete);
2002         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
2003         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2004
2005         osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
2006         osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
2007
2008         RETURN(0);
2009 }
2010
2011 static int osd_object_destroy(const struct lu_env *env,
2012                               struct dt_object *dt,
2013                               struct thandle *th)
2014 {
2015         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2016         struct osd_object      *obj = osd_dt_obj(dt);
2017         struct inode           *inode = obj->oo_inode;
2018         struct osd_device      *osd = osd_obj2dev(obj);
2019         struct osd_thandle     *oh;
2020         int                     result;
2021         ENTRY;
2022
2023         oh = container_of0(th, struct osd_thandle, ot_super);
2024         LASSERT(oh->ot_handle);
2025         LASSERT(inode);
2026         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2027
2028         if (S_ISDIR(inode->i_mode)) {
2029                 LASSERT(osd_inode_unlinked(inode) ||
2030                         inode->i_nlink == 1);
2031                 cfs_spin_lock(&obj->oo_guard);
2032                 inode->i_nlink = 0;
2033                 cfs_spin_unlock(&obj->oo_guard);
2034                 inode->i_sb->s_op->dirty_inode(inode);
2035         } else {
2036                 LASSERT(osd_inode_unlinked(inode));
2037         }
2038
2039         OSD_EXEC_OP(th, destroy);
2040
2041         result = osd_oi_delete(osd_oti_get(env),
2042                                osd_fid2oi(osd, fid), fid, th);
2043
2044         /* XXX: add to ext3 orphan list */
2045         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2046
2047         /* not needed in the cache anymore */
2048         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2049
2050         RETURN(0);
2051 }
2052
2053 /**
2054  * Helper function for osd_xattr_set()
2055  */
2056 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2057                            const struct lu_buf *buf, const char *name, int fl)
2058 {
2059         struct osd_object      *obj      = osd_dt_obj(dt);
2060         struct inode           *inode    = obj->oo_inode;
2061         struct osd_thread_info *info     = osd_oti_get(env);
2062         struct dentry          *dentry   = &info->oti_child_dentry;
2063         int                     fs_flags = 0;
2064         int  rc;
2065
2066         LASSERT(dt_object_exists(dt));
2067         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2068         LASSERT(osd_write_locked(env, obj));
2069
2070         if (fl & LU_XATTR_REPLACE)
2071                 fs_flags |= XATTR_REPLACE;
2072
2073         if (fl & LU_XATTR_CREATE)
2074                 fs_flags |= XATTR_CREATE;
2075
2076         dentry->d_inode = inode;
2077         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2078                                    buf->lb_len, fs_flags);
2079         return rc;
2080 }
2081
2082 /**
2083  * Put the fid into lustre_mdt_attrs, and then place the structure
2084  * inode's ea. This fid should not be altered during the life time
2085  * of the inode.
2086  *
2087  * \retval +ve, on success
2088  * \retval -ve, on error
2089  *
2090  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2091  */
2092 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2093                           const struct lu_fid *fid)
2094 {
2095         struct osd_thread_info  *info      = osd_oti_get(env);
2096         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2097
2098         lustre_lma_init(mdt_attrs, fid);
2099         lustre_lma_swab(mdt_attrs);
2100         return __osd_xattr_set(env, dt,
2101                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2102                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2103
2104 }
2105
2106 /**
2107  * Helper function to form igif
2108  */
2109 static inline void osd_igif_get(const struct lu_env *env, struct inode  *inode,
2110                                 struct lu_fid *fid)
2111 {
2112         LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
2113 }
2114
2115 /**
2116  * Helper function to pack the fid, ldiskfs stores fid in packed format.
2117  */
2118 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
2119                   struct lu_fid *befider)
2120 {
2121         fid_cpu_to_be(befider, (struct lu_fid *)fid);
2122         memcpy(pack->fp_area, befider, sizeof(*befider));
2123         pack->fp_len =  sizeof(*befider) + 1;
2124 }
2125
2126 /**
2127  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2128  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2129  * To have compatilibility with 1.8 ldiskfs driver we need to have
2130  * magic number at start of fid data.
2131  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2132  * its inmemory API.
2133  */
2134 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2135                                   const struct dt_rec *fid)
2136 {
2137         param->edp_magic = LDISKFS_LUFID_MAGIC;
2138         param->edp_len =  sizeof(struct lu_fid) + 1;
2139
2140         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2141                       (struct lu_fid *)fid);
2142 }
2143
2144 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
2145 {
2146         int result;
2147
2148         result = 0;
2149         switch (pack->fp_len) {
2150         case sizeof *fid + 1:
2151                 memcpy(fid, pack->fp_area, sizeof *fid);
2152                 fid_be_to_cpu(fid, fid);
2153                 break;
2154         default:
2155                 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
2156                 result = -EIO;
2157         }
2158         return result;
2159 }
2160
2161 /**
2162  * Try to read the fid from inode ea into dt_rec, if return value
2163  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2164  *
2165  * \param fid object fid.
2166  *
2167  * \retval 0 on success
2168  */
2169 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2170                           __u32 ino, struct lu_fid *fid)
2171 {
2172         struct osd_thread_info  *info      = osd_oti_get(env);
2173         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2174         struct lu_device        *ldev   = obj->oo_dt.do_lu.lo_dev;
2175         struct dentry           *dentry = &info->oti_child_dentry;
2176         struct osd_inode_id     *id     = &info->oti_id;
2177         struct osd_device       *dev;
2178         struct inode            *inode;
2179         int                      rc;
2180
2181         ENTRY;
2182         dev  = osd_dev(ldev);
2183
2184         id->oii_ino = ino;
2185         id->oii_gen = OSD_OII_NOGEN;
2186
2187         inode = osd_iget(info, dev, id);
2188         if (IS_ERR(inode)) {
2189                 rc = PTR_ERR(inode);
2190                 GOTO(out,rc);
2191         }
2192         dentry->d_inode = inode;
2193
2194         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2195         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
2196                                    sizeof *mdt_attrs);
2197
2198         /* Check LMA compatibility */
2199         if (rc > 0 &&
2200             (mdt_attrs->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP))) {
2201                 CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
2202                       inode->i_ino, le32_to_cpu(mdt_attrs->lma_incompat) &
2203                       ~LMA_INCOMPAT_SUPP);
2204                 return -ENOSYS;
2205         }
2206
2207         if (rc > 0) {
2208                 lustre_lma_swab(mdt_attrs);
2209                 memcpy(fid, &mdt_attrs->lma_self_fid, sizeof(*fid));
2210                 rc = 0;
2211         } else if (rc == -ENODATA) {
2212                 osd_igif_get(env, inode, fid);
2213                 rc = 0;
2214         }
2215         iput(inode);
2216 out:
2217         RETURN(rc);
2218 }
2219
2220 /**
2221  * OSD layer object create function for interoperability mode (b11826).
2222  * This is mostly similar to osd_object_create(). Only difference being, fid is
2223  * inserted into inode ea here.
2224  *
2225  * \retval   0, on success
2226  * \retval -ve, on error
2227  */
2228 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2229                              struct lu_attr *attr,
2230                              struct dt_allocation_hint *hint,
2231                              struct dt_object_format *dof,
2232                              struct thandle *th)
2233 {
2234         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2235         struct osd_object      *obj    = osd_dt_obj(dt);
2236         struct osd_thread_info *info   = osd_oti_get(env);
2237         int result;
2238
2239         ENTRY;
2240
2241         LASSERT(osd_invariant(obj));
2242         LASSERT(!dt_object_exists(dt));
2243         LASSERT(osd_write_locked(env, obj));
2244         LASSERT(th != NULL);
2245
2246         OSD_EXEC_OP(th, create);
2247
2248         result = __osd_object_create(info, obj, attr, hint, dof, th);
2249
2250         /* objects under osd root shld have igif fid, so dont add fid EA */
2251         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2252                 result = osd_ea_fid_set(env, dt, fid);
2253
2254         if (result == 0)
2255                 result = __osd_oi_insert(env, obj, fid, th);
2256
2257         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2258         LINVRNT(osd_invariant(obj));
2259         RETURN(result);
2260 }
2261
2262 static int osd_declare_object_ref_add(const struct lu_env *env,
2263                                struct dt_object *dt,
2264                                struct thandle *handle)
2265 {
2266         struct osd_thandle *oh;
2267
2268         /* it's possible that object doesn't exist yet */
2269         LASSERT(handle != NULL);
2270
2271         oh = container_of0(handle, struct osd_thandle, ot_super);
2272         LASSERT(oh->ot_handle == NULL);
2273
2274         OSD_DECLARE_OP(oh, ref_add);
2275         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2276
2277         return 0;
2278 }
2279
2280 /*
2281  * Concurrency: @dt is write locked.
2282  */
2283 static int osd_object_ref_add(const struct lu_env *env,
2284                               struct dt_object *dt,
2285                               struct thandle *th)
2286 {
2287         struct osd_object *obj = osd_dt_obj(dt);
2288         struct inode *inode = obj->oo_inode;
2289
2290         LINVRNT(osd_invariant(obj));
2291         LASSERT(dt_object_exists(dt));
2292         LASSERT(osd_write_locked(env, obj));
2293         LASSERT(th != NULL);
2294
2295         OSD_EXEC_OP(th, ref_add);
2296
2297         /*
2298          * DIR_NLINK feature is set for compatibility reasons if:
2299          * 1) nlinks > LDISKFS_LINK_MAX, or
2300          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2301          *
2302          * It is easier to always set this flag (rather than check and set),
2303          * since it has less overhead, and the superblock will be dirtied
2304          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2305          * do not actually care whether this flag is set or not.
2306          */
2307         cfs_spin_lock(&obj->oo_guard);
2308         inode->i_nlink++;
2309         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2310                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2311                     inode->i_nlink == 2)
2312                         inode->i_nlink = 1;
2313         }
2314         LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
2315         cfs_spin_unlock(&obj->oo_guard);
2316         inode->i_sb->s_op->dirty_inode(inode);
2317         LINVRNT(osd_invariant(obj));
2318
2319         return 0;
2320 }
2321
2322 static int osd_declare_object_ref_del(const struct lu_env *env,
2323                                struct dt_object *dt,
2324                                struct thandle *handle)
2325 {
2326         struct osd_thandle *oh;
2327
2328         LASSERT(dt_object_exists(dt));
2329         LASSERT(handle != NULL);
2330
2331         oh = container_of0(handle, struct osd_thandle, ot_super);
2332         LASSERT(oh->ot_handle == NULL);
2333
2334         OSD_DECLARE_OP(oh, ref_del);
2335         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2336
2337         return 0;
2338 }
2339
2340 /*
2341  * Concurrency: @dt is write locked.
2342  */
2343 static int osd_object_ref_del(const struct lu_env *env,
2344                               struct dt_object *dt,
2345                               struct thandle *th)
2346 {
2347         struct osd_object *obj = osd_dt_obj(dt);
2348         struct inode *inode = obj->oo_inode;
2349
2350         LINVRNT(osd_invariant(obj));
2351         LASSERT(dt_object_exists(dt));
2352         LASSERT(osd_write_locked(env, obj));
2353         LASSERT(th != NULL);
2354
2355         OSD_EXEC_OP(th, ref_del);
2356
2357         cfs_spin_lock(&obj->oo_guard);
2358         LASSERT(inode->i_nlink > 0);
2359         inode->i_nlink--;
2360         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2361          * then the nlink count is 1. Don't let it be set to 0 or the directory
2362          * inode will be deleted incorrectly. */
2363         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2364                 inode->i_nlink++;
2365         cfs_spin_unlock(&obj->oo_guard);
2366         inode->i_sb->s_op->dirty_inode(inode);
2367         LINVRNT(osd_invariant(obj));
2368
2369         return 0;
2370 }
2371
2372 /*
2373  * Get the 64-bit version for an inode.
2374  */
2375 static int osd_object_version_get(const struct lu_env *env,
2376                                   struct dt_object *dt, dt_obj_version_t *ver)
2377 {
2378         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2379
2380         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2381                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2382         *ver = LDISKFS_I(inode)->i_fs_version;
2383         return 0;
2384 }
2385
2386 /*
2387  * Concurrency: @dt is read locked.
2388  */
2389 static int osd_xattr_get(const struct lu_env *env,
2390                          struct dt_object *dt,
2391                          struct lu_buf *buf,
2392                          const char *name,
2393                          struct lustre_capa *capa)
2394 {
2395         struct osd_object      *obj    = osd_dt_obj(dt);
2396         struct inode           *inode  = obj->oo_inode;
2397         struct osd_thread_info *info   = osd_oti_get(env);
2398         struct dentry          *dentry = &info->oti_obj_dentry;
2399
2400         /* version get is not real XATTR but uses xattr API */
2401         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2402                 /* for version we are just using xattr API but change inode
2403                  * field instead */
2404                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2405                 osd_object_version_get(env, dt, buf->lb_buf);
2406                 return sizeof(dt_obj_version_t);
2407         }
2408
2409         LASSERT(dt_object_exists(dt));
2410         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2411         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2412
2413         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2414                 return -EACCES;
2415
2416         dentry->d_inode = inode;
2417         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2418 }
2419
2420
2421 static int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
2422                                  const struct lu_buf *buf, const char *name,
2423                                  int fl, struct thandle *handle)
2424 {
2425         struct osd_thandle *oh;
2426
2427         LASSERT(handle != NULL);
2428
2429         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2430                 /* no credits for version */
2431                 return 0;
2432         }
2433
2434         oh = container_of0(handle, struct osd_thandle, ot_super);
2435         LASSERT(oh->ot_handle == NULL);
2436
2437         OSD_DECLARE_OP(oh, xattr_set);
2438         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2439
2440         return 0;
2441 }
2442
2443 /*
2444  * Set the 64-bit version for object
2445  */
2446 static void osd_object_version_set(const struct lu_env *env,
2447                                    struct dt_object *dt,
2448                                    dt_obj_version_t *new_version)
2449 {
2450         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2451
2452         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2453                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2454
2455         LDISKFS_I(inode)->i_fs_version = *new_version;
2456         /** Version is set after all inode operations are finished,
2457          *  so we should mark it dirty here */
2458         inode->i_sb->s_op->dirty_inode(inode);
2459 }
2460
2461 /*
2462  * Concurrency: @dt is write locked.
2463  */
2464 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2465                          const struct lu_buf *buf, const char *name, int fl,
2466                          struct thandle *handle, struct lustre_capa *capa)
2467 {
2468         LASSERT(handle != NULL);
2469
2470         /* version set is not real XATTR */
2471         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2472                 /* for version we are just using xattr API but change inode
2473                  * field instead */
2474                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2475                 osd_object_version_set(env, dt, buf->lb_buf);
2476                 return sizeof(dt_obj_version_t);
2477         }
2478
2479         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2480                 return -EACCES;
2481
2482         OSD_EXEC_OP(handle, xattr_set);
2483         return __osd_xattr_set(env, dt, buf, name, fl);
2484 }
2485
2486 /*
2487  * Concurrency: @dt is read locked.
2488  */
2489 static int osd_xattr_list(const struct lu_env *env,
2490                           struct dt_object *dt,
2491                           struct lu_buf *buf,
2492                           struct lustre_capa *capa)
2493 {
2494         struct osd_object      *obj    = osd_dt_obj(dt);
2495         struct inode           *inode  = obj->oo_inode;
2496         struct osd_thread_info *info   = osd_oti_get(env);
2497         struct dentry          *dentry = &info->oti_obj_dentry;
2498
2499         LASSERT(dt_object_exists(dt));
2500         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2501         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2502
2503         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2504                 return -EACCES;
2505
2506         dentry->d_inode = inode;
2507         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2508 }
2509
2510 static int osd_declare_xattr_del(const struct lu_env *env,
2511                                  struct dt_object *dt,
2512                                 const char *name,
2513                                 struct thandle *handle)
2514 {
2515         struct osd_thandle *oh;
2516
2517         LASSERT(dt_object_exists(dt));
2518         LASSERT(handle != NULL);
2519
2520         oh = container_of0(handle, struct osd_thandle, ot_super);
2521         LASSERT(oh->ot_handle == NULL);
2522
2523         OSD_DECLARE_OP(oh, xattr_set);
2524         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2525
2526         return 0;
2527 }
2528
2529 /*
2530  * Concurrency: @dt is write locked.
2531  */
2532 static int osd_xattr_del(const struct lu_env *env,
2533                          struct dt_object *dt,
2534                          const char *name,
2535                          struct thandle *handle,
2536                          struct lustre_capa *capa)
2537 {
2538         struct osd_object      *obj    = osd_dt_obj(dt);
2539         struct inode           *inode  = obj->oo_inode;
2540         struct osd_thread_info *info   = osd_oti_get(env);
2541         struct dentry          *dentry = &info->oti_obj_dentry;
2542         int                     rc;
2543
2544         LASSERT(dt_object_exists(dt));
2545         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2546         LASSERT(osd_write_locked(env, obj));
2547         LASSERT(handle != NULL);
2548
2549         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2550                 return -EACCES;
2551
2552         OSD_EXEC_OP(handle, xattr_set);
2553
2554         dentry->d_inode = inode;
2555         rc = inode->i_op->removexattr(dentry, name);
2556         return rc;
2557 }
2558
2559 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2560                                      struct dt_object *dt,
2561                                      struct lustre_capa *old,
2562                                      __u64 opc)
2563 {
2564         struct osd_thread_info *info = osd_oti_get(env);
2565         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2566         struct osd_object *obj = osd_dt_obj(dt);
2567         struct osd_device *dev = osd_obj2dev(obj);
2568         struct lustre_capa_key *key = &info->oti_capa_key;
2569         struct lustre_capa *capa = &info->oti_capa;
2570         struct obd_capa *oc;
2571         struct md_capainfo *ci;
2572         int rc;
2573         ENTRY;
2574
2575         if (!dev->od_fl_capa)
2576                 RETURN(ERR_PTR(-ENOENT));
2577
2578         LASSERT(dt_object_exists(dt));
2579         LINVRNT(osd_invariant(obj));
2580
2581         /* renewal sanity check */
2582         if (old && osd_object_auth(env, dt, old, opc))
2583                 RETURN(ERR_PTR(-EACCES));
2584
2585         ci = md_capainfo(env);
2586         if (unlikely(!ci))
2587                 RETURN(ERR_PTR(-ENOENT));
2588
2589         switch (ci->mc_auth) {
2590         case LC_ID_NONE:
2591                 RETURN(NULL);
2592         case LC_ID_PLAIN:
2593                 capa->lc_uid = obj->oo_inode->i_uid;
2594                 capa->lc_gid = obj->oo_inode->i_gid;
2595                 capa->lc_flags = LC_ID_PLAIN;
2596                 break;
2597         case LC_ID_CONVERT: {
2598                 __u32 d[4], s[4];
2599
2600                 s[0] = obj->oo_inode->i_uid;
2601                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2602                 s[2] = obj->oo_inode->i_gid;
2603                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2604                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2605                 if (unlikely(rc))
2606                         RETURN(ERR_PTR(rc));
2607
2608                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2609                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2610                 capa->lc_flags = LC_ID_CONVERT;
2611                 break;
2612         }
2613         default:
2614                 RETURN(ERR_PTR(-EINVAL));
2615         }
2616
2617         capa->lc_fid = *fid;
2618         capa->lc_opc = opc;
2619         capa->lc_flags |= dev->od_capa_alg << 24;
2620         capa->lc_timeout = dev->od_capa_timeout;
2621         capa->lc_expiry = 0;
2622
2623         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2624         if (oc) {
2625                 LASSERT(!capa_is_expired(oc));
2626                 RETURN(oc);
2627         }
2628
2629         cfs_spin_lock(&capa_lock);
2630         *key = dev->od_capa_keys[1];
2631         cfs_spin_unlock(&capa_lock);
2632
2633         capa->lc_keyid = key->lk_keyid;
2634         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2635
2636         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2637         if (rc) {
2638                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2639                 RETURN(ERR_PTR(rc));
2640         }
2641
2642         oc = capa_add(dev->od_capa_hash, capa);
2643         RETURN(oc);
2644 }
2645
2646 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2647 {
2648         int rc;
2649         struct osd_object      *obj    = osd_dt_obj(dt);
2650         struct inode           *inode  = obj->oo_inode;
2651         struct osd_thread_info *info   = osd_oti_get(env);
2652         struct dentry          *dentry = &info->oti_obj_dentry;
2653         struct file            *file   = &info->oti_file;
2654         ENTRY;
2655
2656         dentry->d_inode = inode;
2657         file->f_dentry = dentry;
2658         file->f_mapping = inode->i_mapping;
2659         file->f_op = inode->i_fop;
2660         LOCK_INODE_MUTEX(inode);
2661         rc = file->f_op->fsync(file, dentry, 0);
2662         UNLOCK_INODE_MUTEX(inode);
2663         RETURN(rc);
2664 }
2665
2666 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2667                         void **data)
2668 {
2669         struct osd_object *obj = osd_dt_obj(dt);
2670         ENTRY;
2671
2672         *data = (void *)obj->oo_inode;
2673         RETURN(0);
2674 }
2675
2676 /*
2677  * Index operations.
2678  */
2679
2680 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2681                            const struct dt_index_features *feat)
2682 {
2683         struct iam_descr *descr;
2684
2685         if (osd_object_is_root(o))
2686                 return feat == &dt_directory_features;
2687
2688         LASSERT(o->oo_dir != NULL);
2689
2690         descr = o->oo_dir->od_container.ic_descr;
2691         if (feat == &dt_directory_features) {
2692                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2693                         return 1;
2694                 else
2695                         return 0;
2696         } else {
2697                 return
2698                         feat->dif_keysize_min <= descr->id_key_size &&
2699                         descr->id_key_size <= feat->dif_keysize_max &&
2700                         feat->dif_recsize_min <= descr->id_rec_size &&
2701                         descr->id_rec_size <= feat->dif_recsize_max &&
2702                         !(feat->dif_flags & (DT_IND_VARKEY |
2703                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2704                         ergo(feat->dif_flags & DT_IND_UPDATE,
2705                              1 /* XXX check that object (and file system) is
2706                                 * writable */);
2707         }
2708 }
2709
2710 static int osd_iam_container_init(const struct lu_env *env,
2711                                   struct osd_object *obj,
2712                                   struct osd_directory *dir)
2713 {
2714         struct iam_container *bag = &dir->od_container;
2715         int result;
2716
2717         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2718         if (result != 0)
2719                 return result;
2720
2721         result = iam_container_setup(bag);
2722         if (result != 0)
2723                 goto out;
2724
2725         if (osd_obj2dev(obj)->od_iop_mode) {
2726                 u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag);
2727
2728                 bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode,
2729                                                 ptr, 0, &result);
2730         }
2731
2732  out:
2733         if (result == 0)
2734                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2735         else
2736                 iam_container_fini(bag);
2737
2738         return result;
2739 }
2740
2741
2742 /*
2743  * Concurrency: no external locking is necessary.
2744  */
2745 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2746                          const struct dt_index_features *feat)
2747 {
2748         int result;
2749         int ea_dir = 0;
2750         struct osd_object *obj = osd_dt_obj(dt);
2751         struct osd_device *osd = osd_obj2dev(obj);
2752
2753         LINVRNT(osd_invariant(obj));
2754         LASSERT(dt_object_exists(dt));
2755
2756         if (osd_object_is_root(obj)) {
2757                 dt->do_index_ops = &osd_index_ea_ops;
2758                 result = 0;
2759         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2760                 dt->do_index_ops = &osd_index_ea_ops;
2761                 if (S_ISDIR(obj->oo_inode->i_mode))
2762                         result = 0;
2763                 else
2764                         result = -ENOTDIR;
2765                 ea_dir = 1;
2766         } else if (!osd_has_index(obj)) {
2767                 struct osd_directory *dir;
2768
2769                 OBD_ALLOC_PTR(dir);
2770                 if (dir != NULL) {
2771
2772                         cfs_spin_lock(&obj->oo_guard);
2773                         if (obj->oo_dir == NULL)
2774                                 obj->oo_dir = dir;
2775                         else
2776                                 /*
2777                                  * Concurrent thread allocated container data.
2778                                  */
2779                                 OBD_FREE_PTR(dir);
2780                         cfs_spin_unlock(&obj->oo_guard);
2781                         /*
2782                          * Now, that we have container data, serialize its
2783                          * initialization.
2784                          */
2785                         cfs_down_write(&obj->oo_ext_idx_sem);
2786                         /*
2787                          * recheck under lock.
2788                          */
2789                         if (!osd_has_index(obj))
2790                                 result = osd_iam_container_init(env, obj, dir);
2791                         else
2792                                 result = 0;
2793                         cfs_up_write(&obj->oo_ext_idx_sem);
2794                 } else {
2795                         result = -ENOMEM;
2796                 }
2797         } else {
2798                 result = 0;
2799         }
2800
2801         if (result == 0 && ea_dir == 0) {
2802                 if (!osd_iam_index_probe(env, obj, feat))
2803                         result = -ENOTDIR;
2804         }
2805         LINVRNT(osd_invariant(obj));
2806
2807         return result;
2808 }
2809
2810 static const struct dt_object_operations osd_obj_ops = {
2811         .do_read_lock         = osd_object_read_lock,
2812         .do_write_lock        = osd_object_write_lock,
2813         .do_read_unlock       = osd_object_read_unlock,
2814         .do_write_unlock      = osd_object_write_unlock,
2815         .do_write_locked      = osd_object_write_locked,
2816         .do_attr_get          = osd_attr_get,
2817         .do_declare_attr_set  = osd_declare_attr_set,
2818         .do_attr_set          = osd_attr_set,
2819         .do_ah_init           = osd_ah_init,
2820         .do_declare_create    = osd_declare_object_create,
2821         .do_create            = osd_object_create,
2822         .do_declare_destroy   = osd_declare_object_destroy,
2823         .do_destroy           = osd_object_destroy,
2824         .do_index_try         = osd_index_try,
2825         .do_declare_ref_add   = osd_declare_object_ref_add,
2826         .do_ref_add           = osd_object_ref_add,
2827         .do_declare_ref_del   = osd_declare_object_ref_del,
2828         .do_ref_del           = osd_object_ref_del,
2829         .do_xattr_get         = osd_xattr_get,
2830         .do_declare_xattr_set = osd_declare_xattr_set,
2831         .do_xattr_set         = osd_xattr_set,
2832         .do_declare_xattr_del = osd_declare_xattr_del,
2833         .do_xattr_del         = osd_xattr_del,
2834         .do_xattr_list        = osd_xattr_list,
2835         .do_capa_get          = osd_capa_get,
2836         .do_object_sync       = osd_object_sync,
2837         .do_data_get          = osd_data_get,
2838 };
2839
2840 /**
2841  * dt_object_operations for interoperability mode
2842  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2843  */
2844 static const struct dt_object_operations osd_obj_ea_ops = {
2845         .do_read_lock         = osd_object_read_lock,
2846         .do_write_lock        = osd_object_write_lock,
2847         .do_read_unlock       = osd_object_read_unlock,
2848         .do_write_unlock      = osd_object_write_unlock,
2849         .do_write_locked      = osd_object_write_locked,
2850         .do_attr_get          = osd_attr_get,
2851         .do_declare_attr_set  = osd_declare_attr_set,
2852         .do_attr_set          = osd_attr_set,
2853         .do_ah_init           = osd_ah_init,
2854         .do_declare_create    = osd_declare_object_create,
2855         .do_create            = osd_object_ea_create,
2856         .do_declare_destroy   = osd_declare_object_destroy,
2857         .do_destroy           = osd_object_destroy,
2858         .do_index_try         = osd_index_try,
2859         .do_declare_ref_add   = osd_declare_object_ref_add,
2860         .do_ref_add           = osd_object_ref_add,
2861         .do_declare_ref_del   = osd_declare_object_ref_del,
2862         .do_ref_del           = osd_object_ref_del,
2863         .do_xattr_get         = osd_xattr_get,
2864         .do_declare_xattr_set = osd_declare_xattr_set,
2865         .do_xattr_set         = osd_xattr_set,
2866         .do_declare_xattr_del = osd_declare_xattr_del,
2867         .do_xattr_del         = osd_xattr_del,
2868         .do_xattr_list        = osd_xattr_list,
2869         .do_capa_get          = osd_capa_get,
2870         .do_object_sync       = osd_object_sync,
2871         .do_data_get          = osd_data_get,
2872 };
2873
2874 /*
2875  * Body operations.
2876  */
2877
2878 /*
2879  * XXX: Another layering violation for now.
2880  *
2881  * We don't want to use ->f_op->read methods, because generic file write
2882  *
2883  *         - serializes on ->i_sem, and
2884  *
2885  *         - does a lot of extra work like balance_dirty_pages(),
2886  *
2887  * which doesn't work for globally shared files like /last-received.
2888  */
2889 static int osd_ldiskfs_readlink(struct inode *inode, char *buffer, int buflen)
2890 {
2891         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2892
2893         memcpy(buffer, (char*)ei->i_data, buflen);
2894
2895         return  buflen;
2896 }
2897
2898 static int osd_ldiskfs_read(struct inode *inode, void *buf, int size,
2899                             loff_t *offs)
2900 {
2901         struct buffer_head *bh;
2902         unsigned long block;
2903         int osize = size;
2904         int blocksize;
2905         int csize;
2906         int boffs;
2907         int err;
2908
2909         /* prevent reading after eof */
2910         spin_lock(&inode->i_lock);
2911         if (i_size_read(inode) < *offs + size) {
2912                 size = i_size_read(inode) - *offs;
2913                 spin_unlock(&inode->i_lock);
2914                 if (size < 0) {
2915                         CDEBUG(D_EXT2, "size %llu is too short to read @%llu\n",
2916                                i_size_read(inode), *offs);
2917                         return -EBADR;
2918                 } else if (size == 0) {
2919                         return 0;
2920                 }
2921         } else {
2922                 spin_unlock(&inode->i_lock);
2923         }
2924
2925         blocksize = 1 << inode->i_blkbits;
2926
2927         while (size > 0) {
2928                 block = *offs >> inode->i_blkbits;
2929                 boffs = *offs & (blocksize - 1);
2930                 csize = min(blocksize - boffs, size);
2931                 bh = ldiskfs_bread(NULL, inode, block, 0, &err);
2932                 if (!bh) {
2933                         CERROR("can't read block: %d\n", err);
2934                         return err;
2935                 }
2936
2937                 memcpy(buf, bh->b_data + boffs, csize);
2938                 brelse(bh);
2939
2940                 *offs += csize;
2941                 buf += csize;
2942                 size -= csize;
2943         }
2944         return osize;
2945 }
2946
2947 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
2948                         struct lu_buf *buf, loff_t *pos,
2949                         struct lustre_capa *capa)
2950 {
2951         struct osd_object      *obj    = osd_dt_obj(dt);
2952         struct inode           *inode  = obj->oo_inode;
2953         int rc;
2954
2955         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2956                 RETURN(-EACCES);
2957
2958         /* Read small symlink from inode body as we need to maintain correct
2959          * on-disk symlinks for ldiskfs.
2960          */
2961         if (S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
2962             (buf->lb_len <= sizeof (LDISKFS_I(inode)->i_data)))
2963                 rc = osd_ldiskfs_readlink(inode, buf->lb_buf, buf->lb_len);
2964         else
2965                 rc = osd_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
2966
2967         return rc;
2968 }
2969
2970 static int osd_ldiskfs_writelink(struct inode *inode, char *buffer, int buflen)
2971 {
2972
2973         memcpy((char*)&LDISKFS_I(inode)->i_data, (char *)buffer,
2974                buflen);
2975         LDISKFS_I(inode)->i_disksize = buflen;
2976         i_size_write(inode, buflen);
2977         inode->i_sb->s_op->dirty_inode(inode);
2978
2979         return 0;
2980 }
2981
2982 static int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
2983                                     loff_t *offs, handle_t *handle)
2984 {
2985         struct buffer_head *bh = NULL;
2986         loff_t offset = *offs;
2987         loff_t new_size = i_size_read(inode);
2988         unsigned long block;
2989         int blocksize = 1 << inode->i_blkbits;
2990         int err = 0;
2991         int size;
2992         int boffs;
2993         int dirty_inode = 0;
2994
2995         while (bufsize > 0) {
2996                 if (bh != NULL)
2997                         brelse(bh);
2998
2999                 block = offset >> inode->i_blkbits;
3000                 boffs = offset & (blocksize - 1);
3001                 size = min(blocksize - boffs, bufsize);
3002                 bh = ldiskfs_bread(handle, inode, block, 1, &err);
3003                 if (!bh) {
3004                         CERROR("can't read/create block: %d\n", err);
3005                         break;
3006                 }
3007
3008                 err = ldiskfs_journal_get_write_access(handle, bh);
3009                 if (err) {
3010                         CERROR("journal_get_write_access() returned error %d\n",
3011                                err);
3012                         break;
3013                 }
3014                 LASSERTF(boffs + size <= bh->b_size,
3015                          "boffs %d size %d bh->b_size %lu",
3016                          boffs, size, (unsigned long)bh->b_size);
3017                 memcpy(bh->b_data + boffs, buf, size);
3018                 err = ldiskfs_journal_dirty_metadata(handle, bh);
3019                 if (err)
3020                         break;
3021
3022                 if (offset + size > new_size)
3023                         new_size = offset + size;
3024                 offset += size;
3025                 bufsize -= size;
3026                 buf += size;
3027         }
3028         if (bh)
3029                 brelse(bh);
3030
3031         /* correct in-core and on-disk sizes */
3032         if (new_size > i_size_read(inode)) {
3033                 spin_lock(&inode->i_lock);
3034                 if (new_size > i_size_read(inode))
3035                         i_size_write(inode, new_size);
3036                 if (i_size_read(inode) > LDISKFS_I(inode)->i_disksize) {
3037                         LDISKFS_I(inode)->i_disksize = i_size_read(inode);
3038                         dirty_inode = 1;
3039                 }
3040                 spin_unlock(&inode->i_lock);
3041                 if (dirty_inode)
3042                         inode->i_sb->s_op->dirty_inode(inode);
3043         }
3044
3045         if (err == 0)
3046                 *offs = offset;
3047         return err;
3048 }
3049
3050 static ssize_t osd_declare_write(const struct lu_env *env, struct dt_object *dt,
3051                                  const loff_t size, loff_t pos,
3052                                  struct thandle *handle)
3053 {
3054         struct osd_thandle *oh;
3055         int credits;
3056
3057         LASSERT(handle != NULL);
3058
3059         oh = container_of0(handle, struct osd_thandle, ot_super);
3060         LASSERT(oh->ot_handle == NULL);
3061
3062         /* XXX: size == 0 or INT_MAX indicating a catalog header update or
3063          *      llog write, see comment in mdd_declare_llog_record().
3064          *
3065          *      This hack should be removed in 2.3
3066          */
3067         if (size == DECLARE_LLOG_REWRITE)
3068                 credits = 2;
3069         else if (size == DECLARE_LLOG_WRITE)
3070                 credits = 6;
3071         else
3072                 credits = osd_dto_credits_noquota[DTO_WRITE_BLOCK];
3073
3074         OSD_DECLARE_OP(oh, write);
3075         oh->ot_credits += credits;
3076
3077         if (osd_dt_obj(dt)->oo_inode == NULL)
3078                 return 0;
3079
3080         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3081                         osd_dt_obj(dt)->oo_inode);
3082         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3083                         osd_dt_obj(dt)->oo_inode);
3084         return 0;
3085 }
3086
3087 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
3088                          const struct lu_buf *buf, loff_t *pos,
3089                          struct thandle *handle, struct lustre_capa *capa,
3090                          int ignore_quota)
3091 {
3092         struct osd_object  *obj   = osd_dt_obj(dt);
3093         struct inode       *inode = obj->oo_inode;
3094         struct osd_thandle *oh;
3095         ssize_t            result = 0;
3096 #ifdef HAVE_QUOTA_SUPPORT
3097         cfs_cap_t           save = cfs_curproc_cap_pack();
3098 #endif
3099
3100         LASSERT(handle != NULL);
3101
3102         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
3103                 RETURN(-EACCES);
3104
3105         oh = container_of(handle, struct osd_thandle, ot_super);
3106         LASSERT(oh->ot_handle->h_transaction != NULL);
3107 #ifdef HAVE_QUOTA_SUPPORT
3108         if (ignore_quota)
3109                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3110         else
3111                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3112 #endif
3113         /* Write small symlink to inode body as we need to maintain correct
3114          * on-disk symlinks for ldiskfs.
3115          */
3116         if(S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
3117            (buf->lb_len < sizeof (LDISKFS_I(inode)->i_data)))
3118                 result = osd_ldiskfs_writelink(inode, buf->lb_buf, buf->lb_len);
3119         else
3120                 result = osd_ldiskfs_write_record(inode, buf->lb_buf,
3121                                                   buf->lb_len, pos,
3122                                                   oh->ot_handle);
3123 #ifdef HAVE_QUOTA_SUPPORT
3124         cfs_curproc_cap_unpack(save);
3125 #endif
3126         if (result == 0)
3127                 result = buf->lb_len;
3128         return result;
3129 }
3130
3131 /*
3132  * in some cases we may need declare methods for objects being created
3133  * e.g., when we create symlink
3134  */
3135 static const struct dt_body_operations osd_body_ops_new = {
3136         .dbo_declare_write = osd_declare_write,
3137 };
3138
3139 static const struct dt_body_operations osd_body_ops = {
3140         .dbo_read          = osd_read,
3141         .dbo_declare_write = osd_declare_write,
3142         .dbo_write         = osd_write
3143 };
3144
3145 static int osd_index_declare_iam_delete(const struct lu_env *env,
3146                                         struct dt_object *dt,
3147                                         const struct dt_key *key,
3148                                         struct thandle *handle)
3149 {
3150         struct osd_thandle    *oh;
3151
3152         oh = container_of0(handle, struct osd_thandle, ot_super);
3153         LASSERT(oh->ot_handle == NULL);
3154
3155         OSD_DECLARE_OP(oh, delete);
3156         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3157
3158         return 0;
3159 }
3160
3161 /**
3162  *      delete a (key, value) pair from index \a dt specified by \a key
3163  *
3164  *      \param  dt      osd index object
3165  *      \param  key     key for index
3166  *      \param  rec     record reference
3167  *      \param  handle  transaction handler
3168  *
3169  *      \retval  0  success
3170  *      \retval -ve   failure
3171  */
3172
3173 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3174                                 const struct dt_key *key, struct thandle *handle,
3175                                 struct lustre_capa *capa)
3176 {
3177         struct osd_object     *obj = osd_dt_obj(dt);
3178         struct osd_thandle    *oh;
3179         struct iam_path_descr *ipd;
3180         struct iam_container  *bag = &obj->oo_dir->od_container;
3181         int rc;
3182
3183         ENTRY;
3184
3185         LINVRNT(osd_invariant(obj));
3186         LASSERT(dt_object_exists(dt));
3187         LASSERT(bag->ic_object == obj->oo_inode);
3188         LASSERT(handle != NULL);
3189
3190         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3191                 RETURN(-EACCES);
3192
3193         OSD_EXEC_OP(handle, delete);
3194
3195         ipd = osd_idx_ipd_get(env, bag);
3196         if (unlikely(ipd == NULL))
3197                 RETURN(-ENOMEM);
3198
3199         oh = container_of0(handle, struct osd_thandle, ot_super);
3200         LASSERT(oh->ot_handle != NULL);
3201         LASSERT(oh->ot_handle->h_transaction != NULL);
3202
3203         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3204         osd_ipd_put(env, bag, ipd);
3205         LINVRNT(osd_invariant(obj));
3206         RETURN(rc);
3207 }
3208
3209 static int osd_index_declare_ea_delete(const struct lu_env *env,
3210                                        struct dt_object *dt,
3211                                        const struct dt_key *key,
3212                                        struct thandle *handle)
3213 {
3214         struct osd_thandle *oh;
3215
3216         LASSERT(dt_object_exists(dt));
3217         LASSERT(handle != NULL);
3218
3219         oh = container_of0(handle, struct osd_thandle, ot_super);
3220         LASSERT(oh->ot_handle == NULL);
3221
3222         OSD_DECLARE_OP(oh, delete);
3223         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3224
3225         LASSERT(osd_dt_obj(dt)->oo_inode);
3226         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3227                         osd_dt_obj(dt)->oo_inode);
3228         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3229                         osd_dt_obj(dt)->oo_inode);
3230
3231         return 0;
3232 }
3233
3234 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3235                                           struct dt_rec *fid)
3236 {
3237         struct osd_fid_pack *rec;
3238         int rc = -ENODATA;
3239
3240         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3241                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3242                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3243         }
3244         RETURN(rc);
3245 }
3246
3247 /**
3248  * Index delete function for interoperability mode (b11826).
3249  * It will remove the directory entry added by osd_index_ea_insert().
3250  * This entry is needed to maintain name->fid mapping.
3251  *
3252  * \param key,  key i.e. file entry to be deleted
3253  *
3254  * \retval   0, on success
3255  * \retval -ve, on error
3256  */
3257 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3258                                const struct dt_key *key, struct thandle *handle,
3259                                struct lustre_capa *capa)
3260 {
3261         struct osd_object          *obj    = osd_dt_obj(dt);
3262         struct inode               *dir    = obj->oo_inode;
3263         struct dentry              *dentry;
3264         struct osd_thandle         *oh;
3265         struct ldiskfs_dir_entry_2 *de;
3266         struct buffer_head         *bh;
3267         struct htree_lock          *hlock = NULL;
3268
3269         int rc;
3270
3271         ENTRY;
3272
3273         LINVRNT(osd_invariant(obj));
3274         LASSERT(dt_object_exists(dt));
3275         LASSERT(handle != NULL);
3276
3277         OSD_EXEC_OP(handle, delete);
3278
3279         oh = container_of(handle, struct osd_thandle, ot_super);
3280         LASSERT(oh->ot_handle != NULL);
3281         LASSERT(oh->ot_handle->h_transaction != NULL);
3282
3283         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3284                 RETURN(-EACCES);
3285
3286         dentry = osd_child_dentry_get(env, obj,
3287                                       (char *)key, strlen((char *)key));
3288
3289         if (obj->oo_hl_head != NULL) {
3290                 hlock = osd_oti_get(env)->oti_hlock;
3291                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3292                                    dir, LDISKFS_HLOCK_DEL);
3293         } else {
3294                 cfs_down_write(&obj->oo_ext_idx_sem);
3295         }
3296
3297         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3298         if (bh) {
3299                 rc = ldiskfs_delete_entry(oh->ot_handle,
3300                                           dir, de, bh);
3301                 brelse(bh);
3302         } else {
3303                 rc = -ENOENT;
3304         }
3305         if (hlock != NULL)
3306                 ldiskfs_htree_unlock(hlock);
3307         else
3308                 cfs_up_write(&obj->oo_ext_idx_sem);
3309
3310         LASSERT(osd_invariant(obj));
3311         RETURN(rc);
3312 }
3313
3314 /**
3315  *      Lookup index for \a key and copy record to \a rec.
3316  *
3317  *      \param  dt      osd index object
3318  *      \param  key     key for index
3319  *      \param  rec     record reference
3320  *
3321  *      \retval  +ve  success : exact mach
3322  *      \retval  0    return record with key not greater than \a key
3323  *      \retval -ve   failure
3324  */
3325 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3326                                 struct dt_rec *rec, const struct dt_key *key,
3327                                 struct lustre_capa *capa)
3328 {
3329         struct osd_object     *obj = osd_dt_obj(dt);
3330         struct iam_path_descr *ipd;
3331         struct iam_container  *bag = &obj->oo_dir->od_container;
3332         struct osd_thread_info *oti = osd_oti_get(env);
3333         struct iam_iterator    *it = &oti->oti_idx_it;
3334         struct iam_rec *iam_rec;
3335         int rc;
3336         ENTRY;
3337
3338         LASSERT(osd_invariant(obj));
3339         LASSERT(dt_object_exists(dt));
3340         LASSERT(bag->ic_object == obj->oo_inode);
3341
3342         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3343                 RETURN(-EACCES);
3344
3345         ipd = osd_idx_ipd_get(env, bag);
3346         if (IS_ERR(ipd))
3347                 RETURN(-ENOMEM);
3348
3349         /* got ipd now we can start iterator. */
3350         iam_it_init(it, bag, 0, ipd);
3351
3352         rc = iam_it_get(it, (struct iam_key *)key);
3353         if (rc >= 0) {
3354                 if (S_ISDIR(obj->oo_inode->i_mode))
3355                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3356                 else
3357                         iam_rec = (struct iam_rec *) rec;
3358
3359                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3360                 if (S_ISDIR(obj->oo_inode->i_mode))
3361                         osd_fid_unpack((struct lu_fid *) rec,
3362                                        (struct osd_fid_pack *)iam_rec);
3363         }
3364         iam_it_put(it);
3365         iam_it_fini(it);
3366         osd_ipd_put(env, bag, ipd);
3367
3368         LINVRNT(osd_invariant(obj));
3369
3370         RETURN(rc);
3371 }
3372
3373 static int osd_index_declare_iam_insert(const struct lu_env *env,
3374                                         struct dt_object *dt,
3375                                         const struct dt_rec *rec,
3376                                         const struct dt_key *key,
3377                                         struct thandle *handle)
3378 {
3379         struct osd_thandle *oh;
3380
3381         LASSERT(dt_object_exists(dt));
3382         LASSERT(handle != NULL);
3383
3384         oh = container_of0(handle, struct osd_thandle, ot_super);
3385         LASSERT(oh->ot_handle == NULL);
3386
3387         OSD_DECLARE_OP(oh, insert);
3388         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3389
3390         return 0;
3391 }
3392
3393 /**
3394  *      Inserts (key, value) pair in \a dt index object.
3395  *
3396  *      \param  dt      osd index object
3397  *      \param  key     key for index
3398  *      \param  rec     record reference
3399  *      \param  th      transaction handler
3400  *
3401  *      \retval  0  success
3402  *      \retval -ve failure
3403  */
3404 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3405                                 const struct dt_rec *rec, const struct dt_key *key,
3406                                 struct thandle *th, struct lustre_capa *capa,
3407                                 int ignore_quota)
3408 {
3409         struct osd_object     *obj = osd_dt_obj(dt);
3410         struct iam_path_descr *ipd;
3411         struct osd_thandle    *oh;
3412         struct iam_container  *bag = &obj->oo_dir->od_container;
3413 #ifdef HAVE_QUOTA_SUPPORT
3414         cfs_cap_t              save = cfs_curproc_cap_pack();
3415 #endif
3416         struct osd_thread_info *oti = osd_oti_get(env);
3417         struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_ldp;
3418         int rc;
3419
3420         ENTRY;
3421
3422         LINVRNT(osd_invariant(obj));
3423         LASSERT(dt_object_exists(dt));
3424         LASSERT(bag->ic_object == obj->oo_inode);
3425         LASSERT(th != NULL);
3426
3427         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3428                 return -EACCES;
3429
3430         OSD_EXEC_OP(th, insert);
3431
3432         ipd = osd_idx_ipd_get(env, bag);
3433         if (unlikely(ipd == NULL))
3434                 RETURN(-ENOMEM);
3435
3436         oh = container_of0(th, struct osd_thandle, ot_super);
3437         LASSERT(oh->ot_handle != NULL);
3438         LASSERT(oh->ot_handle->h_transaction != NULL);
3439 #ifdef HAVE_QUOTA_SUPPORT
3440         if (ignore_quota)
3441                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3442         else
3443                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3444 #endif
3445         if (S_ISDIR(obj->oo_inode->i_mode))
3446                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3447         else
3448                 iam_rec = (struct iam_rec *) rec;
3449         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3450                         iam_rec, ipd);
3451 #ifdef HAVE_QUOTA_SUPPORT
3452         cfs_curproc_cap_unpack(save);
3453 #endif
3454         osd_ipd_put(env, bag, ipd);
3455         LINVRNT(osd_invariant(obj));
3456         RETURN(rc);
3457 }
3458
3459 /**
3460  * Calls ldiskfs_add_entry() to add directory entry
3461  * into the directory. This is required for
3462  * interoperability mode (b11826)
3463  *
3464  * \retval   0, on success
3465  * \retval -ve, on error
3466  */
3467 static int __osd_ea_add_rec(struct osd_thread_info *info,
3468                             struct osd_object *pobj,
3469                             struct inode  *cinode,
3470                             const char *name,
3471                             const struct dt_rec *fid,
3472                             struct htree_lock *hlock,
3473                             struct thandle *th)
3474 {
3475         struct ldiskfs_dentry_param *ldp;
3476         struct dentry      *child;
3477         struct osd_thandle *oth;
3478         int rc;
3479
3480         oth = container_of(th, struct osd_thandle, ot_super);
3481         LASSERT(oth->ot_handle != NULL);
3482         LASSERT(oth->ot_handle->h_transaction != NULL);
3483
3484         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3485
3486         if (fid_is_igif((struct lu_fid *)fid) ||
3487             fid_is_norm((struct lu_fid *)fid)) {
3488                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3489                 osd_get_ldiskfs_dirent_param(ldp, fid);
3490                 child->d_fsdata = (void*) ldp;
3491         } else
3492                 child->d_fsdata = NULL;
3493         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3494
3495         RETURN(rc);
3496 }
3497
3498 /**
3499  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3500  * into the directory.Also sets flags into osd object to
3501  * indicate dot and dotdot are created. This is required for
3502  * interoperability mode (b11826)
3503  *
3504  * \param dir   directory for dot and dotdot fixup.
3505  * \param obj   child object for linking
3506  *
3507  * \retval   0, on success
3508  * \retval -ve, on error
3509  */
3510 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3511                               struct osd_object *dir,
3512                               struct inode  *parent_dir, const char *name,
3513                               const struct dt_rec *dot_fid,
3514                               const struct dt_rec *dot_dot_fid,
3515                               struct thandle *th)
3516 {
3517         struct inode            *inode  = dir->oo_inode;
3518         struct ldiskfs_dentry_param *dot_ldp;
3519         struct ldiskfs_dentry_param *dot_dot_ldp;
3520         struct osd_thandle      *oth;
3521         int result = 0;
3522
3523         oth = container_of(th, struct osd_thandle, ot_super);
3524         LASSERT(oth->ot_handle->h_transaction != NULL);
3525         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3526
3527         if (strcmp(name, dot) == 0) {
3528                 if (dir->oo_compat_dot_created) {
3529                         result = -EEXIST;
3530                 } else {
3531                         LASSERT(inode == parent_dir);
3532                         dir->oo_compat_dot_created = 1;
3533                         result = 0;
3534                 }
3535         } else if(strcmp(name, dotdot) == 0) {
3536                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3537                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3538
3539                 if (!dir->oo_compat_dot_created)
3540                         return -EINVAL;
3541                 if (fid_seq((struct lu_fid *)dot_fid) >= FID_SEQ_NORMAL) {
3542                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3543                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3544                 } else {
3545                         dot_ldp = NULL;
3546                         dot_dot_ldp = NULL;
3547                 }
3548                 /* in case of rename, dotdot is already created */
3549                 if (dir->oo_compat_dotdot_created) {
3550                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3551                                                 dot_dot_fid, NULL, th);
3552                 }
3553
3554                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3555                                                 inode, dot_ldp, dot_dot_ldp);
3556                 if (result == 0)
3557                        dir->oo_compat_dotdot_created = 1;
3558         }
3559
3560         return result;
3561 }
3562
3563
3564 /**
3565  * It will call the appropriate osd_add* function and return the
3566  * value, return by respective functions.
3567  */
3568 static int osd_ea_add_rec(const struct lu_env *env,
3569                           struct osd_object *pobj,
3570                           struct inode *cinode,
3571                           const char *name,
3572                           const struct dt_rec *fid,
3573                           struct thandle *th)
3574 {
3575         struct osd_thread_info    *info   = osd_oti_get(env);
3576         struct htree_lock         *hlock;
3577         int rc;
3578
3579         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3580
3581         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3582                                                    name[2] =='\0'))) {
3583                 if (hlock != NULL) {
3584                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3585                                            pobj->oo_inode, 0);
3586                 } else {
3587                         cfs_down_write(&pobj->oo_ext_idx_sem);
3588                 }
3589                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3590                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3591                                         fid, th);
3592         } else {
3593                 if (hlock != NULL) {
3594                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3595                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3596                 } else {
3597                         cfs_down_write(&pobj->oo_ext_idx_sem);
3598                 }
3599
3600                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3601                                       hlock, th);
3602         }
3603         if (hlock != NULL)
3604                 ldiskfs_htree_unlock(hlock);
3605         else
3606                 cfs_up_write(&pobj->oo_ext_idx_sem);
3607
3608         return rc;
3609 }
3610
3611 /**
3612  * Calls ->lookup() to find dentry. From dentry get inode and
3613  * read inode's ea to get fid. This is required for  interoperability
3614  * mode (b11826)
3615  *
3616  * \retval   0, on success
3617  * \retval -ve, on error
3618  */
3619 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3620                              struct dt_rec *rec, const struct dt_key *key)
3621 {
3622         struct inode               *dir    = obj->oo_inode;
3623         struct dentry              *dentry;
3624         struct ldiskfs_dir_entry_2 *de;
3625         struct buffer_head         *bh;
3626         struct lu_fid              *fid = (struct lu_fid *) rec;
3627         struct htree_lock          *hlock = NULL;
3628         int ino;
3629         int rc;
3630
3631         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3632
3633         dentry = osd_child_dentry_get(env, obj,
3634                                       (char *)key, strlen((char *)key));
3635
3636         if (obj->oo_hl_head != NULL) {
3637                 hlock = osd_oti_get(env)->oti_hlock;
3638                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3639                                    dir, LDISKFS_HLOCK_LOOKUP);
3640         } else {
3641                 cfs_down_read(&obj->oo_ext_idx_sem);
3642         }
3643
3644         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3645         if (bh) {
3646                 ino = le32_to_cpu(de->inode);
3647                 rc = osd_get_fid_from_dentry(de, rec);
3648
3649                 /* done with de, release bh */
3650                 brelse(bh);
3651                 if (rc != 0)
3652                         rc = osd_ea_fid_get(env, obj, ino, fid);
3653         } else {
3654                 rc = -ENOENT;
3655         }
3656
3657         if (hlock != NULL)
3658                 ldiskfs_htree_unlock(hlock);
3659         else
3660                 cfs_up_read(&obj->oo_ext_idx_sem);
3661         RETURN (rc);
3662 }
3663
3664 /**
3665  * Find the osd object for given fid.
3666  *
3667  * \param fid need to find the osd object having this fid
3668  *
3669  * \retval osd_object on success
3670  * \retval        -ve on error
3671  */
3672 struct osd_object *osd_object_find(const struct lu_env *env,
3673                                    struct dt_object *dt,
3674                                    const struct lu_fid *fid)
3675 {
3676         struct lu_device         *ludev = dt->do_lu.lo_dev;
3677         struct osd_object        *child = NULL;
3678         struct lu_object         *luch;
3679         struct lu_object         *lo;
3680
3681         luch = lu_object_find(env, ludev, fid, NULL);
3682         if (!IS_ERR(luch)) {
3683                 if (lu_object_exists(luch)) {
3684                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3685                         if (lo != NULL)
3686                                 child = osd_obj(lo);
3687                         else
3688                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3689                                                 "lu_object can't be located"
3690                                                 ""DFID"\n", PFID(fid));
3691
3692                         if (child == NULL) {
3693                                 lu_object_put(env, luch);
3694                                 CERROR("Unable to get osd_object\n");
3695                                 child = ERR_PTR(-ENOENT);
3696                         }
3697                 } else {
3698                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3699                                         "lu_object does not exists "DFID"\n",
3700                                         PFID(fid));
3701                         child = ERR_PTR(-ENOENT);
3702                 }
3703         } else
3704                 child = (void *)luch;
3705
3706         return child;
3707 }
3708
3709 /**
3710  * Put the osd object once done with it.
3711  *
3712  * \param obj osd object that needs to be put
3713  */
3714 static inline void osd_object_put(const struct lu_env *env,
3715                                   struct osd_object *obj)
3716 {
3717         lu_object_put(env, &obj->oo_dt.do_lu);
3718 }
3719
3720 static int osd_index_declare_ea_insert(const struct lu_env *env,
3721                                        struct dt_object *dt,
3722                                        const struct dt_rec *rec,
3723                                        const struct dt_key *key,
3724                                        struct thandle *handle)
3725 {
3726         struct osd_thandle *oh;
3727
3728         LASSERT(dt_object_exists(dt));
3729         LASSERT(handle != NULL);
3730
3731         oh = container_of0(handle, struct osd_thandle, ot_super);
3732         LASSERT(oh->ot_handle == NULL);
3733
3734         OSD_DECLARE_OP(oh, insert);
3735         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3736
3737         LASSERT(osd_dt_obj(dt)->oo_inode);
3738         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3739                         osd_dt_obj(dt)->oo_inode);
3740         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3741                         osd_dt_obj(dt)->oo_inode);
3742
3743         return 0;
3744 }
3745
3746 /**
3747  * Index add function for interoperability mode (b11826).
3748  * It will add the directory entry.This entry is needed to
3749  * maintain name->fid mapping.
3750  *
3751  * \param key it is key i.e. file entry to be inserted
3752  * \param rec it is value of given key i.e. fid
3753  *
3754  * \retval   0, on success
3755  * \retval -ve, on error
3756  */
3757 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3758                                const struct dt_rec *rec,
3759                                const struct dt_key *key, struct thandle *th,
3760                                struct lustre_capa *capa, int ignore_quota)
3761 {
3762         struct osd_object        *obj   = osd_dt_obj(dt);
3763         struct lu_fid            *fid   = (struct lu_fid *) rec;
3764         const char               *name  = (const char *)key;
3765         struct osd_object        *child;
3766 #ifdef HAVE_QUOTA_SUPPORT
3767         cfs_cap_t                 save  = cfs_curproc_cap_pack();
3768 #endif
3769         int rc;
3770
3771         ENTRY;
3772
3773         LASSERT(osd_invariant(obj));
3774         LASSERT(dt_object_exists(dt));
3775         LASSERT(th != NULL);
3776
3777         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3778                 RETURN(-EACCES);
3779
3780         child = osd_object_find(env, dt, fid);
3781         if (!IS_ERR(child)) {
3782 #ifdef HAVE_QUOTA_SUPPORT
3783                 if (ignore_quota)
3784                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3785                 else
3786                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3787 #endif
3788                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3789 #ifdef HAVE_QUOTA_SUPPORT
3790                 cfs_curproc_cap_unpack(save);
3791 #endif
3792                 osd_object_put(env, child);
3793         } else {
3794                 rc = PTR_ERR(child);
3795         }
3796
3797         LASSERT(osd_invariant(obj));
3798         RETURN(rc);
3799 }
3800
3801 /**
3802  *  Initialize osd Iterator for given osd index object.
3803  *
3804  *  \param  dt      osd index object
3805  */
3806
3807 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3808                                      struct dt_object *dt,
3809                                      __u32 unused,
3810                                      struct lustre_capa *capa)
3811 {
3812         struct osd_it_iam         *it;
3813         struct osd_thread_info *oti = osd_oti_get(env);
3814         struct osd_object     *obj = osd_dt_obj(dt);
3815         struct lu_object      *lo  = &dt->do_lu;
3816         struct iam_path_descr *ipd;
3817         struct iam_container  *bag = &obj->oo_dir->od_container;
3818
3819         LASSERT(lu_object_exists(lo));
3820
3821         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3822                 return ERR_PTR(-EACCES);
3823
3824         it = &oti->oti_it;
3825         ipd = osd_it_ipd_get(env, bag);
3826         if (likely(ipd != NULL)) {
3827                 it->oi_obj = obj;
3828                 it->oi_ipd = ipd;
3829                 lu_object_get(lo);
3830                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3831                 return (struct dt_it *)it;
3832         }
3833         return ERR_PTR(-ENOMEM);
3834 }
3835
3836 /**
3837  * free given Iterator.
3838  */
3839
3840 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3841 {
3842         struct osd_it_iam     *it = (struct osd_it_iam *)di;
3843         struct osd_object *obj = it->oi_obj;
3844
3845         iam_it_fini(&it->oi_it);
3846         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3847         lu_object_put(env, &obj->oo_dt.do_lu);
3848 }
3849
3850 /**
3851  *  Move Iterator to record specified by \a key
3852  *
3853  *  \param  di      osd iterator
3854  *  \param  key     key for index
3855  *
3856  *  \retval +ve  di points to record with least key not larger than key
3857  *  \retval  0   di points to exact matched key
3858  *  \retval -ve  failure
3859  */
3860
3861 static int osd_it_iam_get(const struct lu_env *env,
3862                       struct dt_it *di, const struct dt_key *key)
3863 {
3864         struct osd_it_iam *it = (struct osd_it_iam *)di;
3865
3866         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3867 }
3868
3869 /**
3870  *  Release Iterator
3871  *
3872  *  \param  di      osd iterator
3873  */
3874
3875 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3876 {
3877         struct osd_it_iam *it = (struct osd_it_iam *)di;
3878
3879         iam_it_put(&it->oi_it);
3880 }
3881
3882 /**
3883  *  Move iterator by one record
3884  *
3885  *  \param  di      osd iterator
3886  *
3887  *  \retval +1   end of container reached
3888  *  \retval  0   success
3889  *  \retval -ve  failure
3890  */
3891
3892 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3893 {
3894         struct osd_it_iam *it = (struct osd_it_iam *)di;
3895
3896         return iam_it_next(&it->oi_it);
3897 }
3898
3899 /**
3900  * Return pointer to the key under iterator.
3901  */
3902
3903 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3904                                  const struct dt_it *di)
3905 {
3906         struct osd_it_iam *it = (struct osd_it_iam *)di;
3907
3908         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3909 }
3910
3911 /**
3912  * Return size of key under iterator (in bytes)
3913  */
3914
3915 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3916 {
3917         struct osd_it_iam *it = (struct osd_it_iam *)di;
3918
3919         return iam_it_key_size(&it->oi_it);
3920 }
3921
3922 static inline void osd_it_append_attrs(struct lu_dirent*ent,
3923                                        __u32 attr,
3924                                        int len,
3925                                        __u16 type)
3926 {
3927         struct luda_type        *lt;
3928         const unsigned           align = sizeof(struct luda_type) - 1;
3929
3930         /* check if file type is required */
3931         if (attr & LUDA_TYPE) {
3932                         len = (len + align) & ~align;
3933
3934                         lt = (void *) ent->lde_name + len;
3935                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3936                         ent->lde_attrs |= LUDA_TYPE;
3937         }
3938
3939         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3940 }
3941
3942 /**
3943  * build lu direct from backend fs dirent.
3944  */
3945
3946 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3947                                       struct lu_fid *fid,
3948                                       __u64 offset,
3949                                       char *name,
3950                                       __u16 namelen,
3951                                       __u16 type,
3952                                       __u32 attr)
3953 {
3954         fid_cpu_to_le(&ent->lde_fid, fid);
3955         ent->lde_attrs = LUDA_FID;
3956
3957         ent->lde_hash = cpu_to_le64(offset);
3958         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3959
3960         strncpy(ent->lde_name, name, namelen);
3961         ent->lde_namelen = cpu_to_le16(namelen);
3962
3963         /* append lustre attributes */
3964         osd_it_append_attrs(ent, attr, namelen, type);
3965 }
3966
3967 /**
3968  * Return pointer to the record under iterator.
3969  */
3970 static int osd_it_iam_rec(const struct lu_env *env,
3971                           const struct dt_it *di,
3972                           struct dt_rec *dtrec,
3973                           __u32 attr)
3974 {
3975         struct osd_it_iam *it        = (struct osd_it_iam *)di;
3976         struct osd_thread_info *info = osd_oti_get(env);
3977         struct lu_fid     *fid       = &info->oti_fid;
3978         const struct osd_fid_pack *rec;
3979         struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3980         char *name;
3981         int namelen;
3982         __u64 hash;
3983         int rc;
3984
3985         name = (char *)iam_it_key_get(&it->oi_it);
3986         if (IS_ERR(name))
3987                 RETURN(PTR_ERR(name));
3988
3989         namelen = iam_it_key_size(&it->oi_it);
3990
3991         rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
3992         if (IS_ERR(rec))
3993                 RETURN(PTR_ERR(rec));
3994
3995         rc = osd_fid_unpack(fid, rec);
3996         if (rc)
3997                 RETURN(rc);
3998
3999         hash = iam_it_store(&it->oi_it);
4000
4001         /* IAM does not store object type in IAM index (dir) */
4002         osd_it_pack_dirent(lde, fid, hash, name, namelen,
4003                            0, LUDA_FID);
4004
4005         return 0;
4006 }
4007
4008 /**
4009  * Returns cookie for current Iterator position.
4010  */
4011 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4012 {
4013         struct osd_it_iam *it = (struct osd_it_iam *)di;
4014
4015         return iam_it_store(&it->oi_it);
4016 }
4017
4018 /**
4019  * Restore iterator from cookie.
4020  *
4021  * \param  di      osd iterator
4022  * \param  hash    Iterator location cookie
4023  *
4024  * \retval +ve  di points to record with least key not larger than key.
4025  * \retval  0   di points to exact matched key
4026  * \retval -ve  failure
4027  */
4028
4029 static int osd_it_iam_load(const struct lu_env *env,
4030                        const struct dt_it *di, __u64 hash)
4031 {
4032         struct osd_it_iam *it = (struct osd_it_iam *)di;
4033
4034         return iam_it_load(&it->oi_it, hash);
4035 }
4036
4037 static const struct dt_index_operations osd_index_iam_ops = {
4038         .dio_lookup         = osd_index_iam_lookup,
4039         .dio_declare_insert = osd_index_declare_iam_insert,
4040         .dio_insert         = osd_index_iam_insert,
4041         .dio_declare_delete = osd_index_declare_iam_delete,
4042         .dio_delete         = osd_index_iam_delete,
4043         .dio_it     = {
4044                 .init     = osd_it_iam_init,
4045                 .fini     = osd_it_iam_fini,
4046                 .get      = osd_it_iam_get,
4047                 .put      = osd_it_iam_put,
4048                 .next     = osd_it_iam_next,
4049                 .key      = osd_it_iam_key,
4050                 .key_size = osd_it_iam_key_size,
4051                 .rec      = osd_it_iam_rec,
4052                 .store    = osd_it_iam_store,
4053                 .load     = osd_it_iam_load
4054         }
4055 };
4056
4057 /**
4058  * Creates or initializes iterator context.
4059  *
4060  * \retval struct osd_it_ea, iterator structure on success
4061  *
4062  */
4063 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4064                                     struct dt_object *dt,
4065                                     __u32 attr,
4066                                     struct lustre_capa *capa)
4067 {
4068         struct osd_object       *obj  = osd_dt_obj(dt);
4069         struct osd_thread_info  *info = osd_oti_get(env);
4070         struct osd_it_ea        *it   = &info->oti_it_ea;
4071         struct lu_object        *lo   = &dt->do_lu;
4072         struct dentry           *obj_dentry = &info->oti_it_dentry;
4073         ENTRY;
4074         LASSERT(lu_object_exists(lo));
4075
4076         obj_dentry->d_inode = obj->oo_inode;
4077         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4078         obj_dentry->d_name.hash = 0;
4079
4080         it->oie_rd_dirent       = 0;
4081         it->oie_it_dirent       = 0;
4082         it->oie_dirent          = NULL;
4083         it->oie_buf             = info->oti_it_ea_buf;
4084         it->oie_obj             = obj;
4085         it->oie_file.f_pos      = 0;
4086         it->oie_file.f_dentry   = obj_dentry;
4087         if (attr & LUDA_64BITHASH)
4088                 it->oie_file.f_flags = O_64BITHASH;
4089         else
4090                 it->oie_file.f_flags = O_32BITHASH;
4091         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4092         it->oie_file.f_op         = obj->oo_inode->i_fop;
4093         it->oie_file.private_data = NULL;
4094         lu_object_get(lo);
4095         RETURN((struct dt_it *) it);
4096 }
4097
4098 /**
4099  * Destroy or finishes iterator context.
4100  *
4101  * \param di iterator structure to be destroyed
4102  */
4103 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4104 {
4105         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4106         struct osd_object    *obj  = it->oie_obj;
4107         struct inode       *inode  = obj->oo_inode;
4108
4109         ENTRY;
4110         it->oie_file.f_op->release(inode, &it->oie_file);
4111         lu_object_put(env, &obj->oo_dt.do_lu);
4112         EXIT;
4113 }
4114
4115 /**
4116  * It position the iterator at given key, so that next lookup continues from
4117  * that key Or it is similar to dio_it->load() but based on a key,
4118  * rather than file position.
4119  *
4120  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4121  * to the beginning.
4122  *
4123  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4124  */
4125 static int osd_it_ea_get(const struct lu_env *env,
4126                          struct dt_it *di, const struct dt_key *key)
4127 {
4128         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4129
4130         ENTRY;
4131         LASSERT(((const char *)key)[0] == '\0');
4132         it->oie_file.f_pos      = 0;
4133         it->oie_rd_dirent       = 0;
4134         it->oie_it_dirent       = 0;
4135         it->oie_dirent          = NULL;
4136
4137         RETURN(+1);
4138 }
4139
4140 /**
4141  * Does nothing
4142  */
4143 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4144 {
4145 }
4146
4147 /**
4148  * It is called internally by ->readdir(). It fills the
4149  * iterator's in-memory data structure with required
4150  * information i.e. name, namelen, rec_size etc.
4151  *
4152  * \param buf in which information to be filled in.
4153  * \param name name of the file in given dir
4154  *
4155  * \retval 0 on success
4156  * \retval 1 on buffer full
4157  */
4158 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4159                                loff_t offset, __u64 ino,
4160                                unsigned d_type)
4161 {
4162         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4163         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4164         struct lu_fid           *fid  = &ent->oied_fid;
4165         struct osd_fid_pack     *rec;
4166         ENTRY;
4167
4168         /* this should never happen */
4169         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4170                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4171                 RETURN(-EIO);
4172         }
4173
4174         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4175             OSD_IT_EA_BUFSIZE)
4176                 RETURN(1);
4177
4178         if (d_type & LDISKFS_DIRENT_LUFID) {
4179                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4180
4181                 if (osd_fid_unpack(fid, rec) != 0)
4182                         fid_zero(fid);
4183
4184                 d_type &= ~LDISKFS_DIRENT_LUFID;
4185         } else {
4186                 fid_zero(fid);
4187         }
4188
4189         ent->oied_ino     = ino;
4190         ent->oied_off     = offset;
4191         ent->oied_namelen = namelen;
4192         ent->oied_type    = d_type;
4193
4194         memcpy(ent->oied_name, name, namelen);
4195
4196         it->oie_rd_dirent++;
4197         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4198         RETURN(0);
4199 }
4200
4201 /**
4202  * Calls ->readdir() to load a directory entry at a time
4203  * and stored it in iterator's in-memory data structure.
4204  *
4205  * \param di iterator's in memory structure
4206  *
4207  * \retval   0 on success
4208  * \retval -ve on error
4209  */
4210 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4211                                const struct dt_it *di)
4212 {
4213         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4214         struct osd_object  *obj   = it->oie_obj;
4215         struct inode       *inode = obj->oo_inode;
4216         struct htree_lock  *hlock = NULL;
4217         int                 result = 0;
4218
4219         ENTRY;
4220         it->oie_dirent = it->oie_buf;
4221         it->oie_rd_dirent = 0;
4222
4223         if (obj->oo_hl_head != NULL) {
4224                 hlock = osd_oti_get(env)->oti_hlock;
4225                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4226                                    inode, LDISKFS_HLOCK_READDIR);
4227         } else {
4228                 cfs_down_read(&obj->oo_ext_idx_sem);
4229         }
4230
4231         result = inode->i_fop->readdir(&it->oie_file, it,
4232                                        (filldir_t) osd_ldiskfs_filldir);
4233
4234         if (hlock != NULL)
4235                 ldiskfs_htree_unlock(hlock);
4236         else
4237                 cfs_up_read(&obj->oo_ext_idx_sem);
4238
4239         if (it->oie_rd_dirent == 0) {
4240                 result = -EIO;
4241         } else {
4242                 it->oie_dirent = it->oie_buf;
4243                 it->oie_it_dirent = 1;
4244         }
4245
4246         RETURN(result);
4247 }
4248
4249 /**
4250  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4251  * to load a directory entry at a time and stored it in
4252  * iterator's in-memory data structure.
4253  *
4254  * \param di iterator's in memory structure
4255  *
4256  * \retval +ve iterator reached to end
4257  * \retval   0 iterator not reached to end
4258  * \retval -ve on error
4259  */
4260 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4261 {
4262         struct osd_it_ea *it = (struct osd_it_ea *)di;
4263         int rc;
4264
4265         ENTRY;
4266
4267         if (it->oie_it_dirent < it->oie_rd_dirent) {
4268                 it->oie_dirent =
4269                         (void *) it->oie_dirent +
4270                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4271                                        it->oie_dirent->oied_namelen);
4272                 it->oie_it_dirent++;
4273                 RETURN(0);
4274         } else {
4275                 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
4276                         rc = +1;
4277                 else
4278                         rc = osd_ldiskfs_it_fill(env, di);
4279         }
4280
4281         RETURN(rc);
4282 }
4283
4284 /**
4285  * Returns the key at current position from iterator's in memory structure.
4286  *
4287  * \param di iterator's in memory structure
4288  *
4289  * \retval key i.e. struct dt_key on success
4290  */
4291 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4292                                     const struct dt_it *di)
4293 {
4294         struct osd_it_ea *it = (struct osd_it_ea *)di;
4295         ENTRY;
4296         RETURN((struct dt_key *)it->oie_dirent->oied_name);
4297 }
4298
4299 /**
4300  * Returns the key's size at current position from iterator's in memory structure.
4301  *
4302  * \param di iterator's in memory structure
4303  *
4304  * \retval key_size i.e. struct dt_key on success
4305  */
4306 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4307 {
4308         struct osd_it_ea *it = (struct osd_it_ea *)di;
4309         ENTRY;
4310         RETURN(it->oie_dirent->oied_namelen);
4311 }
4312
4313
4314 /**
4315  * Returns the value (i.e. fid/igif) at current position from iterator's
4316  * in memory structure.
4317  *
4318  * \param di struct osd_it_ea, iterator's in memory structure
4319  * \param attr attr requested for dirent.
4320  * \param lde lustre dirent
4321  *
4322  * \retval   0 no error and \param lde has correct lustre dirent.
4323  * \retval -ve on error
4324  */
4325 static inline int osd_it_ea_rec(const struct lu_env *env,
4326                                 const struct dt_it *di,
4327                                 struct dt_rec *dtrec,
4328                                 __u32 attr)
4329 {
4330         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4331         struct osd_object       *obj    = it->oie_obj;
4332         struct lu_fid           *fid    = &it->oie_dirent->oied_fid;
4333         struct lu_dirent        *lde    = (struct lu_dirent *)dtrec;
4334         int    rc = 0;
4335
4336         ENTRY;
4337
4338         if (!fid_is_sane(fid))
4339                 rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
4340
4341         if (rc == 0)
4342                 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4343                                    it->oie_dirent->oied_name,
4344                                    it->oie_dirent->oied_namelen,
4345                                    it->oie_dirent->oied_type,
4346                                    attr);
4347         RETURN(rc);
4348 }
4349
4350 /**
4351  * Returns a cookie for current position of the iterator head, so that
4352  * user can use this cookie to load/start the iterator next time.
4353  *
4354  * \param di iterator's in memory structure
4355  *
4356  * \retval cookie for current position, on success
4357  */
4358 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4359 {
4360         struct osd_it_ea *it = (struct osd_it_ea *)di;
4361         ENTRY;
4362         RETURN(it->oie_dirent->oied_off);
4363 }
4364
4365 /**
4366  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4367  * to load a directory entry at a time and stored it i inn,
4368  * in iterator's in-memory data structure.
4369  *
4370  * \param di struct osd_it_ea, iterator's in memory structure
4371  *
4372  * \retval +ve on success
4373  * \retval -ve on error
4374  */
4375 static int osd_it_ea_load(const struct lu_env *env,
4376                           const struct dt_it *di, __u64 hash)
4377 {
4378         struct osd_it_ea *it = (struct osd_it_ea *)di;
4379         int rc;
4380
4381         ENTRY;
4382         it->oie_file.f_pos = hash;
4383
4384         rc =  osd_ldiskfs_it_fill(env, di);
4385         if (rc == 0)
4386                 rc = +1;
4387
4388         RETURN(rc);
4389 }
4390
4391 /**
4392  * Index lookup function for interoperability mode (b11826).
4393  *
4394  * \param key,  key i.e. file name to be searched
4395  *
4396  * \retval +ve, on success
4397  * \retval -ve, on error
4398  */
4399 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4400                                struct dt_rec *rec, const struct dt_key *key,
4401                                struct lustre_capa *capa)
4402 {
4403         struct osd_object *obj = osd_dt_obj(dt);
4404         int rc = 0;
4405
4406         ENTRY;
4407
4408         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4409         LINVRNT(osd_invariant(obj));
4410
4411         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4412                 return -EACCES;
4413
4414         rc = osd_ea_lookup_rec(env, obj, rec, key);
4415
4416         if (rc == 0)
4417                 rc = +1;
4418         RETURN(rc);
4419 }
4420
4421 /**
4422  * Index and Iterator operations for interoperability
4423  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4424  */
4425 static const struct dt_index_operations osd_index_ea_ops = {
4426         .dio_lookup         = osd_index_ea_lookup,
4427         .dio_declare_insert = osd_index_declare_ea_insert,
4428         .dio_insert         = osd_index_ea_insert,
4429         .dio_declare_delete = osd_index_declare_ea_delete,
4430         .dio_delete         = osd_index_ea_delete,
4431         .dio_it     = {
4432                 .init     = osd_it_ea_init,
4433                 .fini     = osd_it_ea_fini,
4434                 .get      = osd_it_ea_get,
4435                 .put      = osd_it_ea_put,
4436                 .next     = osd_it_ea_next,
4437                 .key      = osd_it_ea_key,
4438                 .key_size = osd_it_ea_key_size,
4439                 .rec      = osd_it_ea_rec,
4440                 .store    = osd_it_ea_store,
4441                 .load     = osd_it_ea_load
4442         }
4443 };
4444
4445 static void *osd_key_init(const struct lu_context *ctx,
4446                           struct lu_context_key *key)
4447 {
4448         struct osd_thread_info *info;
4449
4450         OBD_ALLOC_PTR(info);
4451         if (info == NULL)
4452                 return ERR_PTR(-ENOMEM);
4453
4454         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4455         if (info->oti_it_ea_buf == NULL)
4456                 goto out_free_info;
4457
4458         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4459
4460         info->oti_hlock = ldiskfs_htree_lock_alloc();
4461         if (info->oti_hlock == NULL)
4462                 goto out_free_ea;
4463
4464         return info;
4465
4466  out_free_ea:
4467         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4468  out_free_info:
4469         OBD_FREE_PTR(info);
4470         return ERR_PTR(-ENOMEM);
4471 }
4472
4473 static void osd_key_fini(const struct lu_context *ctx,
4474                          struct lu_context_key *key, void* data)
4475 {
4476         struct osd_thread_info *info = data;
4477
4478         if (info->oti_hlock != NULL)
4479                 ldiskfs_htree_lock_free(info->oti_hlock);
4480         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4481         OBD_FREE_PTR(info);
4482 }
4483
4484 static void osd_key_exit(const struct lu_context *ctx,
4485                          struct lu_context_key *key, void *data)
4486 {
4487         struct osd_thread_info *info = data;
4488
4489         LASSERT(info->oti_r_locks == 0);
4490         LASSERT(info->oti_w_locks == 0);
4491         LASSERT(info->oti_txns    == 0);
4492 }
4493
4494 /* type constructor/destructor: osd_type_init, osd_type_fini */
4495 LU_TYPE_INIT_FINI(osd, &osd_key);
4496
4497 static struct lu_context_key osd_key = {
4498         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
4499         .lct_init = osd_key_init,
4500         .lct_fini = osd_key_fini,
4501         .lct_exit = osd_key_exit
4502 };
4503
4504
4505 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4506                            const char *name, struct lu_device *next)
4507 {
4508         return osd_procfs_init(osd_dev(d), name);
4509 }
4510
4511 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4512 {
4513         struct osd_thread_info *info = osd_oti_get(env);
4514         ENTRY;
4515         if (o->od_obj_area != NULL) {
4516                 lu_object_put(env, &o->od_obj_area->do_lu);
4517                 o->od_obj_area = NULL;
4518         }
4519         if (o->od_oi_table != NULL)
4520                 osd_oi_fini(info, &o->od_oi_table, o->od_oi_count);
4521
4522         RETURN(0);
4523 }
4524
4525 static int osd_mount(const struct lu_env *env,
4526                      struct osd_device *o, struct lustre_cfg *cfg)
4527 {
4528         struct lustre_mount_info *lmi;
4529         const char               *dev  = lustre_cfg_string(cfg, 0);
4530         struct lustre_disk_data  *ldd;
4531         struct lustre_sb_info    *lsi;
4532
4533         ENTRY;
4534         if (o->od_mount != NULL) {
4535                 CERROR("Already mounted (%s)\n", dev);
4536                 RETURN(-EEXIST);
4537         }
4538
4539         /* get mount */
4540         lmi = server_get_mount(dev);
4541         if (lmi == NULL) {
4542                 CERROR("Cannot get mount info for %s!\n", dev);
4543                 RETURN(-EFAULT);
4544         }
4545
4546         LASSERT(lmi != NULL);
4547         /* save lustre_mount_info in dt_device */
4548         o->od_mount = lmi;
4549
4550         lsi = s2lsi(lmi->lmi_sb);
4551         ldd = lsi->lsi_ldd;
4552
4553         if (ldd->ldd_flags & LDD_F_IAM_DIR) {
4554                 o->od_iop_mode = 0;
4555                 LCONSOLE_WARN("OSD: IAM mode enabled\n");
4556         } else
4557                 o->od_iop_mode = 1;
4558
4559         o->od_obj_area = NULL;
4560         RETURN(0);
4561 }
4562
4563 static struct lu_device *osd_device_fini(const struct lu_env *env,
4564                                          struct lu_device *d)
4565 {
4566         int rc;
4567         ENTRY;
4568
4569         shrink_dcache_sb(osd_sb(osd_dev(d)));
4570         osd_sync(env, lu2dt_dev(d));
4571
4572         rc = osd_procfs_fini(osd_dev(d));
4573         if (rc) {
4574                 CERROR("proc fini error %d \n", rc);
4575                 RETURN (ERR_PTR(rc));
4576         }
4577
4578         if (osd_dev(d)->od_mount)
4579                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
4580                                  osd_dev(d)->od_mount->lmi_mnt);
4581         osd_dev(d)->od_mount = NULL;
4582
4583         RETURN(NULL);
4584 }
4585
4586 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4587                                           struct lu_device_type *t,
4588                                           struct lustre_cfg *cfg)
4589 {
4590         struct lu_device  *l;
4591         struct osd_device *o;
4592
4593         OBD_ALLOC_PTR(o);
4594         if (o != NULL) {
4595                 int result;
4596
4597                 result = dt_device_init(&o->od_dt_dev, t);
4598                 if (result == 0) {
4599                         l = osd2lu_dev(o);
4600                         l->ld_ops = &osd_lu_ops;
4601                         o->od_dt_dev.dd_ops = &osd_dt_ops;
4602                         cfs_spin_lock_init(&o->od_osfs_lock);
4603                         o->od_osfs_age = cfs_time_shift_64(-1000);
4604                         o->od_capa_hash = init_capa_hash();
4605                         if (o->od_capa_hash == NULL) {
4606                                 dt_device_fini(&o->od_dt_dev);
4607                                 l = ERR_PTR(-ENOMEM);
4608                         }
4609                 } else
4610                         l = ERR_PTR(result);
4611
4612                 if (IS_ERR(l))
4613                         OBD_FREE_PTR(o);
4614         } else
4615                 l = ERR_PTR(-ENOMEM);
4616         return l;
4617 }
4618
4619 static struct lu_device *osd_device_free(const struct lu_env *env,
4620                                          struct lu_device *d)
4621 {
4622         struct osd_device *o = osd_dev(d);
4623         ENTRY;
4624
4625         cleanup_capa_hash(o->od_capa_hash);
4626         dt_device_fini(&o->od_dt_dev);
4627         OBD_FREE_PTR(o);
4628         RETURN(NULL);
4629 }
4630
4631 static int osd_process_config(const struct lu_env *env,
4632                               struct lu_device *d, struct lustre_cfg *cfg)
4633 {
4634         struct osd_device *o = osd_dev(d);
4635         int err;
4636         ENTRY;
4637
4638         switch(cfg->lcfg_command) {
4639         case LCFG_SETUP:
4640                 err = osd_mount(env, o, cfg);
4641                 break;
4642         case LCFG_CLEANUP:
4643                 err = osd_shutdown(env, o);
4644                 break;
4645         default:
4646                 err = -ENOSYS;
4647         }
4648
4649         RETURN(err);
4650 }
4651
4652 static int osd_recovery_complete(const struct lu_env *env,
4653                                  struct lu_device *d)
4654 {
4655         RETURN(0);
4656 }
4657
4658 static int osd_prepare(const struct lu_env *env,
4659                        struct lu_device *pdev,
4660                        struct lu_device *dev)
4661 {
4662         struct osd_device *osd = osd_dev(dev);
4663         struct lustre_sb_info *lsi;
4664         struct lustre_disk_data *ldd;
4665         struct lustre_mount_info  *lmi;
4666         struct osd_thread_info *oti = osd_oti_get(env);
4667         struct dt_object *d;
4668         int result;
4669
4670         ENTRY;
4671         /* 1. initialize oi before any file create or file open */
4672         result = osd_oi_init(oti, &osd->od_oi_table,
4673                              &osd->od_dt_dev, lu2md_dev(pdev));
4674         if (result < 0)
4675                 RETURN(result);
4676
4677         LASSERT(result > 0);
4678         osd->od_oi_count = result;
4679
4680         lmi = osd->od_mount;
4681         lsi = s2lsi(lmi->lmi_sb);
4682         ldd = lsi->lsi_ldd;
4683
4684         /* 2. setup local objects */
4685         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4686         if (result)
4687                 goto out;
4688
4689         /* 3. open remote object dir */
4690         d = dt_store_open(env, lu2dt_dev(dev), "",
4691                           remote_obj_dir, &oti->oti_fid);
4692         if (!IS_ERR(d)) {
4693                 osd->od_obj_area = d;
4694                 result = 0;
4695         } else {
4696                 result = PTR_ERR(d);
4697                 osd->od_obj_area = NULL;
4698         }
4699
4700 out:
4701         RETURN(result);
4702 }
4703
4704 static const struct lu_object_operations osd_lu_obj_ops = {
4705         .loo_object_init      = osd_object_init,
4706         .loo_object_delete    = osd_object_delete,
4707         .loo_object_release   = osd_object_release,
4708         .loo_object_free      = osd_object_free,
4709         .loo_object_print     = osd_object_print,
4710         .loo_object_invariant = osd_object_invariant
4711 };
4712
4713 static const struct lu_device_operations osd_lu_ops = {
4714         .ldo_object_alloc      = osd_object_alloc,
4715         .ldo_process_config    = osd_process_config,
4716         .ldo_recovery_complete = osd_recovery_complete,
4717         .ldo_prepare           = osd_prepare,
4718 };
4719
4720 static const struct lu_device_type_operations osd_device_type_ops = {
4721         .ldto_init = osd_type_init,
4722         .ldto_fini = osd_type_fini,
4723
4724         .ldto_start = osd_type_start,
4725         .ldto_stop  = osd_type_stop,
4726
4727         .ldto_device_alloc = osd_device_alloc,
4728         .ldto_device_free  = osd_device_free,
4729
4730         .ldto_device_init    = osd_device_init,
4731         .ldto_device_fini    = osd_device_fini
4732 };
4733
4734 static struct lu_device_type osd_device_type = {
4735         .ldt_tags     = LU_DEVICE_DT,
4736         .ldt_name     = LUSTRE_OSD_NAME,
4737         .ldt_ops      = &osd_device_type_ops,
4738         .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4739 };
4740
4741 /*
4742  * lprocfs legacy support.
4743  */
4744 static struct obd_ops osd_obd_device_ops = {
4745         .o_owner = THIS_MODULE
4746 };
4747
4748 static struct lu_local_obj_desc llod_osd_rem_obj_dir = {
4749         .llod_name      = remote_obj_dir,
4750         .llod_oid       = OSD_REM_OBJ_DIR_OID,
4751         .llod_is_index  = 1,
4752         .llod_feat      = &dt_directory_features,
4753 };
4754
4755 static int __init osd_mod_init(void)
4756 {
4757         struct lprocfs_static_vars lvars;
4758
4759         osd_oi_mod_init();
4760         llo_local_obj_register(&llod_osd_rem_obj_dir);
4761         lprocfs_osd_init_vars(&lvars);
4762         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4763                                    LUSTRE_OSD_NAME, &osd_device_type);
4764 }
4765
4766 static void __exit osd_mod_exit(void)
4767 {
4768         llo_local_obj_unregister(&llod_osd_rem_obj_dir);
4769         class_unregister_type(LUSTRE_OSD_NAME);
4770 }
4771
4772 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4773 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4774 MODULE_LICENSE("GPL");
4775
4776 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);