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