lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56 /* simple_mkdir() */
  57 #include <lvfs.h>
  58
  59 /*
  60  * struct OBD_{ALLOC,FREE}*()
  61  * OBD_FAIL_CHECK
  62  */
  63 #include <obd_support.h>
  64 /* struct ptlrpc_thread */
  65 #include <lustre_net.h>
  66
  67 /* fid_is_local() */
  68 #include <lustre_fid.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_igif.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75 /* dt_acct_features */
  76 #include <lquota.h>
  77
  78 #ifdef HAVE_LDISKFS_PDO
  79 int ldiskfs_pdo = 1;
  80 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  81                 "ldiskfs with parallel directory operations");
  82 #else
  83 int ldiskfs_pdo = 0;
  84 #endif
  85
  86 static const char dot[] = ".";
  87 static const char dotdot[] = "..";
  88 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  89
  90 static const struct lu_object_operations      osd_lu_obj_ops;
  91 static const struct dt_object_operations      osd_obj_ops;
  92 static const struct dt_object_operations      osd_obj_ea_ops;
  93 static const struct dt_object_operations      osd_obj_otable_it_ops;
  94 static const struct dt_index_operations       osd_index_iam_ops;
  95 static const struct dt_index_operations       osd_index_ea_ops;
  96
  97 static int osd_has_index(const struct osd_object *obj)
  98 {
  99         return obj->oo_dt.do_index_ops != NULL;
 100 }
 101
 102 static int osd_object_invariant(const struct lu_object *l)
 103 {
 104         return osd_invariant(osd_obj(l));
 105 }
 106
 107 #ifdef HAVE_QUOTA_SUPPORT
 108 static inline void
 109 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save, bool is_md)
 110 {
 111         struct md_ucred *uc;
 112         struct cred     *tc;
 113
 114         if (!is_md)
 115                 /* OFD support */
 116                 return;
 117
 118         uc = md_ucred(env);
 119
 120         LASSERT(uc != NULL);
 121
 122         save->oc_uid = current_fsuid();
 123         save->oc_gid = current_fsgid();
 124         save->oc_cap = current_cap();
 125         if ((tc = prepare_creds())) {
 126                 tc->fsuid         = uc->mu_fsuid;
 127                 tc->fsgid         = uc->mu_fsgid;
 128                 commit_creds(tc);
 129         }
 130         /* XXX not suboptimal */
 131         cfs_curproc_cap_unpack(uc->mu_cap);
 132 }
 133
 134 static inline void
 135 osd_pop_ctxt(struct osd_ctxt *save, bool is_md)
 136 {
 137         struct cred *tc;
 138
 139         if (!is_md)
 140                 /* OFD support */
 141                 return;
 142
 143         if ((tc = prepare_creds())) {
 144                 tc->fsuid         = save->oc_uid;
 145                 tc->fsgid         = save->oc_gid;
 146                 tc->cap_effective = save->oc_cap;
 147                 commit_creds(tc);
 148         }
 149 }
 150 #endif
 151
 152 /*
 153  * Concurrency: doesn't matter
 154  */
 155 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 156 {
 157         return osd_oti_get(env)->oti_r_locks > 0;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't matter
 162  */
 163 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 164 {
 165         struct osd_thread_info *oti = osd_oti_get(env);
 166         return oti->oti_w_locks > 0 && o->oo_owner == env;
 167 }
 168
 169 /*
 170  * Concurrency: doesn't access mutable data
 171  */
 172 static int osd_root_get(const struct lu_env *env,
 173                         struct dt_device *dev, struct lu_fid *f)
 174 {
 175         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 176         return 0;
 177 }
 178
 179 /*
 180  * OSD object methods.
 181  */
 182
 183 /*
 184  * Concurrency: no concurrent access is possible that early in object
 185  * life-cycle.
 186  */
 187 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 188                                           const struct lu_object_header *hdr,
 189                                           struct lu_device *d)
 190 {
 191         struct osd_object *mo;
 192
 193         OBD_ALLOC_PTR(mo);
 194         if (mo != NULL) {
 195                 struct lu_object *l;
 196
 197                 l = &mo->oo_dt.do_lu;
 198                 dt_object_init(&mo->oo_dt, NULL, d);
 199                 if (osd_dev(d)->od_iop_mode)
 200                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 201                 else
 202                         mo->oo_dt.do_ops = &osd_obj_ops;
 203
 204                 l->lo_ops = &osd_lu_obj_ops;
 205                 cfs_init_rwsem(&mo->oo_sem);
 206                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 207                 cfs_spin_lock_init(&mo->oo_guard);
 208                 return l;
 209         } else {
 210                 return NULL;
 211         }
 212 }
 213
 214 static int osd_get_lma(struct inode *inode, struct dentry *dentry,
 215                        struct lustre_mdt_attrs *lma)
 216 {
 217         int rc;
 218
 219         dentry->d_inode = inode;
 220         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
 221                                    sizeof(*lma));
 222         if (rc > 0) {
 223                 /* Check LMA compatibility */
 224                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 225                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 226                               "%lx/%#x\n",
 227                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 228                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 229                                                         ~LMA_INCOMPAT_SUPP);
 230                         rc = -ENOSYS;
 231                 } else {
 232                         lustre_lma_swab(lma);
 233                         rc = 0;
 234                 }
 235         } else if (rc == 0) {
 236                 rc = -ENODATA;
 237         }
 238
 239         return rc;
 240 }
 241
 242 /*
 243  * retrieve object from backend ext fs.
 244  **/
 245 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 246                        struct osd_inode_id *id)
 247 {
 248         struct inode *inode = NULL;
 249
 250         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 251         if (IS_ERR(inode)) {
 252                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 253                        id->oii_ino, PTR_ERR(inode));
 254         } else if (id->oii_gen != OSD_OII_NOGEN &&
 255                    inode->i_generation != id->oii_gen) {
 256                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 257                        "gen1 = %u\n",
 258                        id->oii_ino, id->oii_gen, inode->i_generation);
 259                 iput(inode);
 260                 inode = ERR_PTR(-ESTALE);
 261         } else if (inode->i_nlink == 0) {
 262                 /* due to parallel readdir and unlink,
 263                 * we can have dead inode here. */
 264                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 265                 make_bad_inode(inode);
 266                 iput(inode);
 267                 inode = ERR_PTR(-ESTALE);
 268         } else if (is_bad_inode(inode)) {
 269                 CWARN("%.16s: bad inode: ino = %u\n",
 270                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 271                 iput(inode);
 272                 inode = ERR_PTR(-ENOENT);
 273         } else {
 274                 if (id->oii_gen == OSD_OII_NOGEN)
 275                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 276
 277                 /* Do not update file c/mtime in ldiskfs.
 278                  * NB: we don't have any lock to protect this because we don't
 279                  * have reference on osd_object now, but contention with
 280                  * another lookup + attr_set can't happen in the tiny window
 281                  * between if (...) and set S_NOCMTIME. */
 282                 if (!(inode->i_flags & S_NOCMTIME))
 283                         inode->i_flags |= S_NOCMTIME;
 284         }
 285         return inode;
 286 }
 287
 288 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 289                            struct osd_inode_id *id, struct lu_fid *fid)
 290 {
 291         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 292         struct inode            *inode;
 293         int                      rc;
 294
 295         inode = osd_iget(info, dev, id);
 296         if (IS_ERR(inode))
 297                 return inode;
 298
 299         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 300         if (rc == 0) {
 301                 *fid = lma->lma_self_fid;
 302         } else if (rc == -ENODATA) {
 303                 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
 304         } else {
 305                 iput(inode);
 306                 inode = ERR_PTR(rc);
 307         }
 308         return inode;
 309 }
 310
 311 static struct inode *
 312 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 313                 struct osd_inode_id *id, const struct lu_fid *fid)
 314 {
 315         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 316         struct inode            *inode;
 317         int                      rc;
 318
 319         inode = osd_iget(info, dev, id);
 320         if (IS_ERR(inode))
 321                 return inode;
 322
 323         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 324         if (rc != 0) {
 325                 if (rc == -ENODATA) {
 326                         CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
 327                                inode->i_ino, PFID(fid));
 328                         rc = -EREMCHG;
 329                 }
 330                 iput(inode);
 331                 return ERR_PTR(rc);
 332         }
 333
 334         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 335                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 336                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 337                 iput(inode);
 338                 return ERR_PTR(EREMCHG);
 339         }
 340         return inode;
 341 }
 342
 343 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 344                           const struct lu_fid *fid,
 345                           const struct lu_object_conf *conf)
 346 {
 347         struct osd_thread_info *info;
 348         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 349         struct osd_device      *dev;
 350         struct osd_idmap_cache *oic;
 351         struct osd_inode_id    *id;
 352         struct inode           *inode;
 353         struct osd_scrub       *scrub;
 354         struct scrub_file      *sf;
 355         int                     result;
 356         int                     verify = 0;
 357         ENTRY;
 358
 359         LINVRNT(osd_invariant(obj));
 360         LASSERT(obj->oo_inode == NULL);
 361         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 362
 363         dev = osd_dev(ldev);
 364         scrub = &dev->od_scrub;
 365         sf = &scrub->os_file;
 366         info = osd_oti_get(env);
 367         LASSERT(info);
 368         oic = &info->oti_cache;
 369         id  = &oic->oic_lid;
 370
 371         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 372                 RETURN(-ENOENT);
 373
 374         if (fid_is_norm(fid)) {
 375                 /* Search order: 1. per-thread cache. */
 376                 if (lu_fid_eq(fid, &oic->oic_fid)) {
 377                         goto iget;
 378                 } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 379                         /* Search order: 2. OI scrub pending list. */
 380                         result = osd_oii_lookup(dev, fid, id);
 381                         if (result == 0)
 382                                 goto iget;
 383                 }
 384
 385                 if (sf->sf_flags & SF_INCONSISTENT)
 386                         verify = 1;
 387         }
 388
 389         /*
 390          * Objects are created as locking anchors or place holders for objects
 391          * yet to be created. No need to osd_oi_lookup() at here because FID
 392          * shouldn't never be re-used, if it's really a duplicate FID from
 393          * unexpected reason, we should be able to detect it later by calling
 394          * do_create->osd_oi_insert()
 395          */
 396         if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
 397                 GOTO(out, result = 0);
 398
 399         /* Search order: 3. OI files. */
 400         result = osd_oi_lookup(info, dev, fid, id);
 401         if (result == -ENOENT) {
 402                 if (!fid_is_norm(fid) ||
 403                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 404                                       sf->sf_oi_bitmap))
 405                         GOTO(out, result = 0);
 406
 407                 goto trigger;
 408         }
 409
 410         if (result != 0)
 411                 GOTO(out, result);
 412
 413 iget:
 414         if (verify == 0)
 415                 inode = osd_iget(info, dev, id);
 416         else
 417                 inode = osd_iget_verify(info, dev, id, fid);
 418         if (IS_ERR(inode)) {
 419                 result = PTR_ERR(inode);
 420                 if (result == -ENOENT || result == -ESTALE) {
 421                         fid_zero(&oic->oic_fid);
 422                         result = 0;
 423                 } else if (result == -EREMCHG) {
 424
 425 trigger:
 426                         if (thread_is_running(&scrub->os_thread)) {
 427                                 result = -EINPROGRESS;
 428                         } else if (!scrub->os_no_scrub) {
 429                                 result = osd_scrub_start(dev);
 430                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 431                                                "for "DFID", rc = %d [1]\n",
 432                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 433                                                s_volume_name,PFID(fid), result);
 434                                 if (result == 0 || result == -EALREADY)
 435                                         result = -EINPROGRESS;
 436                                 else
 437                                         result = -EREMCHG;
 438                         }
 439                 }
 440
 441                 GOTO(out, result);
 442         }
 443
 444         obj->oo_inode = inode;
 445         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 446         if (dev->od_iop_mode) {
 447                 obj->oo_compat_dot_created = 1;
 448                 obj->oo_compat_dotdot_created = 1;
 449         }
 450
 451         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 452                 GOTO(out, result = 0);
 453
 454         LASSERT(obj->oo_hl_head == NULL);
 455         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 456         if (obj->oo_hl_head == NULL) {
 457                 obj->oo_inode = NULL;
 458                 iput(inode);
 459                 GOTO(out, result = -ENOMEM);
 460         }
 461         GOTO(out, result = 0);
 462
 463 out:
 464         LINVRNT(osd_invariant(obj));
 465         return result;
 466 }
 467
 468 /*
 469  * Concurrency: shouldn't matter.
 470  */
 471 static void osd_object_init0(struct osd_object *obj)
 472 {
 473         LASSERT(obj->oo_inode != NULL);
 474         obj->oo_dt.do_body_ops = &osd_body_ops;
 475         obj->oo_dt.do_lu.lo_header->loh_attr |=
 476                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 477 }
 478
 479 /*
 480  * Concurrency: no concurrent access is possible that early in object
 481  * life-cycle.
 482  */
 483 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 484                            const struct lu_object_conf *conf)
 485 {
 486         struct osd_object *obj = osd_obj(l);
 487         int result;
 488
 489         LINVRNT(osd_invariant(obj));
 490
 491         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 492         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 493         if (result == 0) {
 494                 if (obj->oo_inode != NULL) {
 495                         osd_object_init0(obj);
 496                 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 497                         obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 498                         /* LFSCK iterator object is special without inode */
 499                         l->lo_header->loh_attr |= LOHA_EXISTS;
 500                 }
 501         }
 502         LINVRNT(osd_invariant(obj));
 503         return result;
 504 }
 505
 506 /*
 507  * Concurrency: no concurrent access is possible that late in object
 508  * life-cycle.
 509  */
 510 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 511 {
 512         struct osd_object *obj = osd_obj(l);
 513
 514         LINVRNT(osd_invariant(obj));
 515
 516         dt_object_fini(&obj->oo_dt);
 517         if (obj->oo_hl_head != NULL)
 518                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 519         OBD_FREE_PTR(obj);
 520 }
 521
 522 /*
 523  * Concurrency: no concurrent access is possible that late in object
 524  * life-cycle.
 525  */
 526 static void osd_index_fini(struct osd_object *o)
 527 {
 528         struct iam_container *bag;
 529
 530         if (o->oo_dir != NULL) {
 531                 bag = &o->oo_dir->od_container;
 532                 if (o->oo_inode != NULL) {
 533                         if (bag->ic_object == o->oo_inode)
 534                                 iam_container_fini(bag);
 535                 }
 536                 OBD_FREE_PTR(o->oo_dir);
 537                 o->oo_dir = NULL;
 538         }
 539 }
 540
 541 /*
 542  * Concurrency: no concurrent access is possible that late in object
 543  * life-cycle (for all existing callers, that is. New callers have to provide
 544  * their own locking.)
 545  */
 546 static int osd_inode_unlinked(const struct inode *inode)
 547 {
 548         return inode->i_nlink == 0;
 549 }
 550
 551 enum {
 552         OSD_TXN_OI_DELETE_CREDITS    = 20,
 553         OSD_TXN_INODE_DELETE_CREDITS = 20
 554 };
 555
 556 /*
 557  * Journal
 558  */
 559
 560 #if OSD_THANDLE_STATS
 561 /**
 562  * Set time when the handle is allocated
 563  */
 564 static void osd_th_alloced(struct osd_thandle *oth)
 565 {
 566         oth->oth_alloced = cfs_time_current();
 567 }
 568
 569 /**
 570  * Set time when the handle started
 571  */
 572 static void osd_th_started(struct osd_thandle *oth)
 573 {
 574         oth->oth_started = cfs_time_current();
 575 }
 576
 577 /**
 578  * Helper function to convert time interval to microseconds packed in
 579  * long int (default time units for the counter in "stats" initialized
 580  * by lu_time_init() )
 581  */
 582 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 583 {
 584         struct timeval val;
 585
 586         cfs_duration_usec(cfs_time_sub(end, start), &val);
 587         return val.tv_sec * 1000000 + val.tv_usec;
 588 }
 589
 590 /**
 591  * Check whether the we deal with this handle for too long.
 592  */
 593 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 594                                 cfs_time_t alloced, cfs_time_t started,
 595                                 cfs_time_t closed)
 596 {
 597         cfs_time_t now = cfs_time_current();
 598
 599         LASSERT(dev != NULL);
 600
 601         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 602                             interval_to_usec(alloced, started));
 603         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 604                             interval_to_usec(started, closed));
 605         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 606                             interval_to_usec(closed, now));
 607
 608         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 609                 CWARN("transaction handle %p was open for too long: "
 610                       "now "CFS_TIME_T" ,"
 611                       "alloced "CFS_TIME_T" ,"
 612                       "started "CFS_TIME_T" ,"
 613                       "closed "CFS_TIME_T"\n",
 614                       oth, now, alloced, started, closed);
 615                 libcfs_debug_dumpstack(NULL);
 616         }
 617 }
 618
 619 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 620 {                                                                       \
 621         cfs_time_t __closed = cfs_time_current();                       \
 622         cfs_time_t __alloced = oth->oth_alloced;                        \
 623         cfs_time_t __started = oth->oth_started;                        \
 624                                                                         \
 625         expr;                                                           \
 626         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 627 }
 628
 629 #else /* OSD_THANDLE_STATS */
 630
 631 #define osd_th_alloced(h)                  do {} while(0)
 632 #define osd_th_started(h)                  do {} while(0)
 633 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 634
 635 #endif /* OSD_THANDLE_STATS */
 636
 637 /*
 638  * Concurrency: doesn't access mutable data.
 639  */
 640 static int osd_param_is_sane(const struct osd_device *dev,
 641                              const struct thandle *th)
 642 {
 643         struct osd_thandle *oh;
 644         oh = container_of0(th, struct osd_thandle, ot_super);
 645         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 646 }
 647
 648 /*
 649  * Concurrency: shouldn't matter.
 650  */
 651 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 652 static void osd_trans_commit_cb(struct super_block *sb,
 653                                 struct journal_callback *jcb, int error)
 654 #else
 655 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 656 #endif
 657 {
 658         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 659         struct thandle     *th  = &oh->ot_super;
 660         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 661         struct dt_txn_commit_cb *dcb, *tmp;
 662
 663         LASSERT(oh->ot_handle == NULL);
 664
 665         if (error)
 666                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 667
 668         dt_txn_hook_commit(th);
 669
 670         /* call per-transaction callbacks if any */
 671         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 672                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 673                          "commit callback entry: magic=%x name='%s'\n",
 674                          dcb->dcb_magic, dcb->dcb_name);
 675                 cfs_list_del_init(&dcb->dcb_linkage);
 676                 dcb->dcb_func(NULL, th, dcb, error);
 677         }
 678
 679         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 680         lu_device_put(lud);
 681         th->th_dev = NULL;
 682
 683         lu_context_exit(&th->th_ctx);
 684         lu_context_fini(&th->th_ctx);
 685         OBD_FREE_PTR(oh);
 686 }
 687
 688 static struct thandle *osd_trans_create(const struct lu_env *env,
 689                                         struct dt_device *d)
 690 {
 691         struct osd_thread_info *oti = osd_oti_get(env);
 692         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 693         struct osd_thandle     *oh;
 694         struct thandle         *th;
 695         ENTRY;
 696
 697         /* on pending IO in this thread should left from prev. request */
 698         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 699
 700         th = ERR_PTR(-ENOMEM);
 701         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 702         if (oh != NULL) {
 703                 oh->ot_quota_trans = &oti->oti_quota_trans;
 704                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 705                 th = &oh->ot_super;
 706                 th->th_dev = d;
 707                 th->th_result = 0;
 708                 th->th_tags = LCT_TX_HANDLE;
 709                 oh->ot_credits = 0;
 710                 oti->oti_dev = osd_dt_dev(d);
 711                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 712                 osd_th_alloced(oh);
 713         }
 714         RETURN(th);
 715 }
 716
 717 /*
 718  * Concurrency: shouldn't matter.
 719  */
 720 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 721                     struct thandle *th)
 722 {
 723         struct osd_thread_info *oti = osd_oti_get(env);
 724         struct osd_device  *dev = osd_dt_dev(d);
 725         handle_t           *jh;
 726         struct osd_thandle *oh;
 727         int rc;
 728
 729         ENTRY;
 730
 731         LASSERT(current->journal_info == NULL);
 732
 733         oh = container_of0(th, struct osd_thandle, ot_super);
 734         LASSERT(oh != NULL);
 735         LASSERT(oh->ot_handle == NULL);
 736
 737         rc = dt_txn_hook_start(env, d, th);
 738         if (rc != 0)
 739                 GOTO(out, rc);
 740
 741         if (!osd_param_is_sane(dev, th)) {
 742                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 743                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 744                       oh->ot_credits,
 745                       osd_journal(dev)->j_max_transaction_buffers);
 746                 /* XXX Limit the credits to 'max_transaction_buffers', and
 747                  *     let the underlying filesystem to catch the error if
 748                  *     we really need so many credits.
 749                  *
 750                  *     This should be removed when we can calculate the
 751                  *     credits precisely. */
 752                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 753 #ifdef OSD_TRACK_DECLARES
 754                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 755                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 756                        oh->ot_declare_xattr_set);
 757                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 758                        oh->ot_declare_create, oh->ot_declare_ref_add,
 759                        oh->ot_declare_ref_del, oh->ot_declare_write);
 760                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 761                        oh->ot_declare_insert, oh->ot_declare_delete,
 762                        oh->ot_declare_destroy);
 763 #endif
 764         }
 765
 766         /*
 767          * XXX temporary stuff. Some abstraction layer should
 768          * be used.
 769          */
 770         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 771         osd_th_started(oh);
 772         if (!IS_ERR(jh)) {
 773                 oh->ot_handle = jh;
 774                 LASSERT(oti->oti_txns == 0);
 775                 lu_context_init(&th->th_ctx, th->th_tags);
 776                 lu_context_enter(&th->th_ctx);
 777
 778                 lu_device_get(&d->dd_lu_dev);
 779                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 780                                              "osd-tx", th);
 781                 oti->oti_txns++;
 782                 rc = 0;
 783         } else {
 784                 rc = PTR_ERR(jh);
 785         }
 786 out:
 787         RETURN(rc);
 788 }
 789
 790 /*
 791  * Concurrency: shouldn't matter.
 792  */
 793 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 794 {
 795         int                     rc = 0;
 796         struct osd_thandle     *oh;
 797         struct osd_thread_info *oti = osd_oti_get(env);
 798         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 799         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
 800         ENTRY;
 801
 802         oh = container_of0(th, struct osd_thandle, ot_super);
 803
 804         if (qsd != NULL)
 805                 /* inform the quota slave device that the transaction is
 806                  * stopping */
 807                 qsd_op_end(env, qsd, oh->ot_quota_trans);
 808         oh->ot_quota_trans = NULL;
 809
 810         if (oh->ot_handle != NULL) {
 811                 handle_t *hdl = oh->ot_handle;
 812
 813                 /*
 814                  * add commit callback
 815                  * notice we don't do this in osd_trans_start()
 816                  * as underlying transaction can change during truncate
 817                  */
 818                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 819                                          &oh->ot_jcb);
 820
 821                 LASSERT(oti->oti_txns == 1);
 822                 oti->oti_txns--;
 823                 rc = dt_txn_hook_stop(env, th);
 824                 if (rc != 0)
 825                         CERROR("Failure in transaction hook: %d\n", rc);
 826
 827                 /* hook functions might modify th_sync */
 828                 hdl->h_sync = th->th_sync;
 829
 830                 oh->ot_handle = NULL;
 831                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 832                                   rc = ldiskfs_journal_stop(hdl));
 833                 if (rc != 0)
 834                         CERROR("Failure to stop transaction: %d\n", rc);
 835         } else {
 836                 OBD_FREE_PTR(oh);
 837         }
 838
 839         /* as we want IO to journal and data IO be concurrent, we don't block
 840          * awaiting data IO completion in osd_do_bio(), instead we wait here
 841          * once transaction is submitted to the journal. all reqular requests
 842          * don't do direct IO (except read/write), thus this wait_event becomes
 843          * no-op for them.
 844          *
 845          * IMPORTANT: we have to wait till any IO submited by the thread is
 846          * completed otherwise iobuf may be corrupted by different request
 847          */
 848         cfs_wait_event(iobuf->dr_wait,
 849                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 850         if (!rc)
 851                 rc = iobuf->dr_error;
 852
 853         RETURN(rc);
 854 }
 855
 856 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 857 {
 858         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 859                                                ot_super);
 860
 861         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 862         LASSERT(&dcb->dcb_func != NULL);
 863         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 864
 865         return 0;
 866 }
 867
 868 /*
 869  * Called just before object is freed. Releases all resources except for
 870  * object itself (that is released by osd_object_free()).
 871  *
 872  * Concurrency: no concurrent access is possible that late in object
 873  * life-cycle.
 874  */
 875 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 876 {
 877         struct osd_object *obj   = osd_obj(l);
 878         struct inode      *inode = obj->oo_inode;
 879
 880         LINVRNT(osd_invariant(obj));
 881
 882         /*
 883          * If object is unlinked remove fid->ino mapping from object index.
 884          */
 885
 886         osd_index_fini(obj);
 887         if (inode != NULL) {
 888                 iput(inode);
 889                 obj->oo_inode = NULL;
 890         }
 891 }
 892
 893 /*
 894  * Concurrency: ->loo_object_release() is called under site spin-lock.
 895  */
 896 static void osd_object_release(const struct lu_env *env,
 897                                struct lu_object *l)
 898 {
 899 }
 900
 901 /*
 902  * Concurrency: shouldn't matter.
 903  */
 904 static int osd_object_print(const struct lu_env *env, void *cookie,
 905                             lu_printer_t p, const struct lu_object *l)
 906 {
 907         struct osd_object *o = osd_obj(l);
 908         struct iam_descr  *d;
 909
 910         if (o->oo_dir != NULL)
 911                 d = o->oo_dir->od_container.ic_descr;
 912         else
 913                 d = NULL;
 914         return (*p)(env, cookie,
 915                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 916                     o, o->oo_inode,
 917                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 918                     o->oo_inode ? o->oo_inode->i_generation : 0,
 919                     d ? d->id_ops->id_name : "plain");
 920 }
 921
 922 /*
 923  * Concurrency: shouldn't matter.
 924  */
 925 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 926                struct obd_statfs *sfs)
 927 {
 928         struct osd_device  *osd = osd_dt_dev(d);
 929         struct super_block *sb = osd_sb(osd);
 930         struct kstatfs     *ksfs;
 931         int result = 0;
 932
 933         if (unlikely(osd->od_mnt == NULL))
 934                 return -EINPROGRESS;
 935
 936         /* osd_lproc.c call this without env, allocate ksfs for that case */
 937         if (unlikely(env == NULL)) {
 938                 OBD_ALLOC_PTR(ksfs);
 939                 if (ksfs == NULL)
 940                         return -ENOMEM;
 941         } else {
 942                 ksfs = &osd_oti_get(env)->oti_ksfs;
 943         }
 944
 945         cfs_spin_lock(&osd->od_osfs_lock);
 946         /* cache 1 second */
 947         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 948                 result = sb->s_op->statfs(sb->s_root, ksfs);
 949                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 950                         osd->od_osfs_age = cfs_time_current_64();
 951                         statfs_pack(&osd->od_statfs, ksfs);
 952                 }
 953         }
 954
 955         if (likely(result == 0))
 956                 *sfs = osd->od_statfs;
 957         cfs_spin_unlock(&osd->od_osfs_lock);
 958
 959         if (unlikely(env == NULL))
 960                 OBD_FREE_PTR(ksfs);
 961
 962         return result;
 963 }
 964
 965 /*
 966  * Concurrency: doesn't access mutable data.
 967  */
 968 static void osd_conf_get(const struct lu_env *env,
 969                          const struct dt_device *dev,
 970                          struct dt_device_param *param)
 971 {
 972         struct super_block *sb = osd_sb(osd_dt_dev(dev));
 973
 974         /*
 975          * XXX should be taken from not-yet-existing fs abstraction layer.
 976          */
 977         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
 978         param->ddp_max_name_len = LDISKFS_NAME_LEN;
 979         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
 980         param->ddp_block_shift  = sb->s_blocksize_bits;
 981         param->ddp_mount_type     = LDD_MT_LDISKFS;
 982         param->ddp_mntopts      = 0;
 983         if (test_opt(sb, XATTR_USER))
 984                 param->ddp_mntopts |= MNTOPT_USERXATTR;
 985         if (test_opt(sb, POSIX_ACL))
 986                 param->ddp_mntopts |= MNTOPT_ACL;
 987
 988 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
 989         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
 990                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
 991         else
 992 #endif
 993                 param->ddp_max_ea_size = sb->s_blocksize;
 994
 995 }
 996
 997 /**
 998  * Helper function to get and fill the buffer with input values.
 999  */
1000 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1001 {
1002         struct lu_buf *buf;
1003
1004         buf = &osd_oti_get(env)->oti_buf;
1005         buf->lb_buf = area;
1006         buf->lb_len = len;
1007         return buf;
1008 }
1009
1010 /*
1011  * Concurrency: shouldn't matter.
1012  */
1013 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1014 {
1015         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1016         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1017 }
1018
1019 /**
1020  * Start commit for OSD device.
1021  *
1022  * An implementation of dt_commit_async method for OSD device.
1023  * Asychronously starts underlayng fs sync and thereby a transaction
1024  * commit.
1025  *
1026  * \param env environment
1027  * \param d dt device
1028  *
1029  * \see dt_device_operations
1030  */
1031 static int osd_commit_async(const struct lu_env *env,
1032                             struct dt_device *d)
1033 {
1034         struct super_block *s = osd_sb(osd_dt_dev(d));
1035         ENTRY;
1036
1037         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1038         RETURN(s->s_op->sync_fs(s, 0));
1039 }
1040
1041 /*
1042  * Concurrency: shouldn't matter.
1043  */
1044
1045 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1046 {
1047         struct super_block *sb = osd_sb(osd_dt_dev(d));
1048         int rc;
1049         ENTRY;
1050
1051         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1052
1053         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1054         RETURN(rc);
1055 }
1056
1057 /*
1058  * Concurrency: serialization provided by callers.
1059  */
1060 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1061                               int mode, unsigned long timeout, __u32 alg,
1062                               struct lustre_capa_key *keys)
1063 {
1064         struct osd_device *dev = osd_dt_dev(d);
1065         ENTRY;
1066
1067         dev->od_fl_capa = mode;
1068         dev->od_capa_timeout = timeout;
1069         dev->od_capa_alg = alg;
1070         dev->od_capa_keys = keys;
1071         RETURN(0);
1072 }
1073
1074 /**
1075  * Concurrency: serialization provided by callers.
1076  */
1077 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1078                                struct dt_quota_ctxt *ctxt, void *data)
1079 {
1080         struct obd_device *obd = (void *)ctxt;
1081         struct vfsmount *mnt = (struct vfsmount *)data;
1082         ENTRY;
1083
1084         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1085         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1086         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1087         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1088         obd->obd_lvfs_ctxt.fs = get_ds();
1089
1090         EXIT;
1091 }
1092
1093 /**
1094  * Note: we do not count into QUOTA here.
1095  * If we mount with --data_journal we may need more.
1096  */
1097 const int osd_dto_credits_noquota[DTO_NR] = {
1098         /**
1099          * Insert/Delete.
1100          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1101          * SINGLEDATA_TRANS_BLOCKS(8)
1102          * XXX Note: maybe iam need more, since iam have more level than
1103          *           EXT3 htree.
1104          */
1105         [DTO_INDEX_INSERT]  = 16,
1106         [DTO_INDEX_DELETE]  = 16,
1107         /**
1108          * Used for OI scrub
1109          */
1110         [DTO_INDEX_UPDATE]  = 16,
1111         /**
1112          * Create a object. The same as create object in EXT3.
1113          * DATA_TRANS_BLOCKS(14) +
1114          * INDEX_EXTRA_BLOCKS(8) +
1115          * 3(inode bits, groups, GDT)
1116          */
1117         [DTO_OBJECT_CREATE] = 25,
1118         /**
1119          * XXX: real credits to be fixed
1120          */
1121         [DTO_OBJECT_DELETE] = 25,
1122         /**
1123          * Attr set credits (inode)
1124          */
1125         [DTO_ATTR_SET_BASE] = 1,
1126         /**
1127          * Xattr set. The same as xattr of EXT3.
1128          * DATA_TRANS_BLOCKS(14)
1129          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1130          * are also counted in. Do not know why?
1131          */
1132         [DTO_XATTR_SET]     = 14,
1133         [DTO_LOG_REC]       = 14,
1134         /**
1135          * credits for inode change during write.
1136          */
1137         [DTO_WRITE_BASE]    = 3,
1138         /**
1139          * credits for single block write.
1140          */
1141         [DTO_WRITE_BLOCK]   = 14,
1142         /**
1143          * Attr set credits for chown.
1144          * This is extra credits for setattr, and it is null without quota
1145          */
1146         [DTO_ATTR_SET_CHOWN]= 0
1147 };
1148
1149 static const struct dt_device_operations osd_dt_ops = {
1150         .dt_root_get       = osd_root_get,
1151         .dt_statfs         = osd_statfs,
1152         .dt_trans_create   = osd_trans_create,
1153         .dt_trans_start    = osd_trans_start,
1154         .dt_trans_stop     = osd_trans_stop,
1155         .dt_trans_cb_add   = osd_trans_cb_add,
1156         .dt_conf_get       = osd_conf_get,
1157         .dt_sync           = osd_sync,
1158         .dt_ro             = osd_ro,
1159         .dt_commit_async   = osd_commit_async,
1160         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1161         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1162 };
1163
1164 static void osd_object_read_lock(const struct lu_env *env,
1165                                  struct dt_object *dt, unsigned role)
1166 {
1167         struct osd_object *obj = osd_dt_obj(dt);
1168         struct osd_thread_info *oti = osd_oti_get(env);
1169
1170         LINVRNT(osd_invariant(obj));
1171
1172         LASSERT(obj->oo_owner != env);
1173         cfs_down_read_nested(&obj->oo_sem, role);
1174
1175         LASSERT(obj->oo_owner == NULL);
1176         oti->oti_r_locks++;
1177 }
1178
1179 static void osd_object_write_lock(const struct lu_env *env,
1180                                   struct dt_object *dt, unsigned role)
1181 {
1182         struct osd_object *obj = osd_dt_obj(dt);
1183         struct osd_thread_info *oti = osd_oti_get(env);
1184
1185         LINVRNT(osd_invariant(obj));
1186
1187         LASSERT(obj->oo_owner != env);
1188         cfs_down_write_nested(&obj->oo_sem, role);
1189
1190         LASSERT(obj->oo_owner == NULL);
1191         obj->oo_owner = env;
1192         oti->oti_w_locks++;
1193 }
1194
1195 static void osd_object_read_unlock(const struct lu_env *env,
1196                                    struct dt_object *dt)
1197 {
1198         struct osd_object *obj = osd_dt_obj(dt);
1199         struct osd_thread_info *oti = osd_oti_get(env);
1200
1201         LINVRNT(osd_invariant(obj));
1202
1203         LASSERT(oti->oti_r_locks > 0);
1204         oti->oti_r_locks--;
1205         cfs_up_read(&obj->oo_sem);
1206 }
1207
1208 static void osd_object_write_unlock(const struct lu_env *env,
1209                                     struct dt_object *dt)
1210 {
1211         struct osd_object *obj = osd_dt_obj(dt);
1212         struct osd_thread_info *oti = osd_oti_get(env);
1213
1214         LINVRNT(osd_invariant(obj));
1215
1216         LASSERT(obj->oo_owner == env);
1217         LASSERT(oti->oti_w_locks > 0);
1218         oti->oti_w_locks--;
1219         obj->oo_owner = NULL;
1220         cfs_up_write(&obj->oo_sem);
1221 }
1222
1223 static int osd_object_write_locked(const struct lu_env *env,
1224                                    struct dt_object *dt)
1225 {
1226         struct osd_object *obj = osd_dt_obj(dt);
1227
1228         LINVRNT(osd_invariant(obj));
1229
1230         return obj->oo_owner == env;
1231 }
1232
1233 static int capa_is_sane(const struct lu_env *env,
1234                         struct osd_device *dev,
1235                         struct lustre_capa *capa,
1236                         struct lustre_capa_key *keys)
1237 {
1238         struct osd_thread_info *oti = osd_oti_get(env);
1239         struct lustre_capa *tcapa = &oti->oti_capa;
1240         struct obd_capa *oc;
1241         int i, rc = 0;
1242         ENTRY;
1243
1244         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1245         if (oc) {
1246                 if (capa_is_expired(oc)) {
1247                         DEBUG_CAPA(D_ERROR, capa, "expired");
1248                         rc = -ESTALE;
1249                 }
1250                 capa_put(oc);
1251                 RETURN(rc);
1252         }
1253
1254         if (capa_is_expired_sec(capa)) {
1255                 DEBUG_CAPA(D_ERROR, capa, "expired");
1256                 RETURN(-ESTALE);
1257         }
1258
1259         cfs_spin_lock(&capa_lock);
1260         for (i = 0; i < 2; i++) {
1261                 if (keys[i].lk_keyid == capa->lc_keyid) {
1262                         oti->oti_capa_key = keys[i];
1263                         break;
1264                 }
1265         }
1266         cfs_spin_unlock(&capa_lock);
1267
1268         if (i == 2) {
1269                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1270                 RETURN(-ESTALE);
1271         }
1272
1273         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1274         if (rc)
1275                 RETURN(rc);
1276
1277         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1278                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1279                 RETURN(-EACCES);
1280         }
1281
1282         oc = capa_add(dev->od_capa_hash, capa);
1283         capa_put(oc);
1284
1285         RETURN(0);
1286 }
1287
1288 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1289                     struct lustre_capa *capa, __u64 opc)
1290 {
1291         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1292         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1293         struct md_capainfo *ci;
1294         int rc;
1295
1296         if (!dev->od_fl_capa)
1297                 return 0;
1298
1299         if (capa == BYPASS_CAPA)
1300                 return 0;
1301
1302         ci = md_capainfo(env);
1303         if (unlikely(!ci))
1304                 return 0;
1305
1306         if (ci->mc_auth == LC_ID_NONE)
1307                 return 0;
1308
1309         if (!capa) {
1310                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1311                 return -EACCES;
1312         }
1313
1314         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1315                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1316                            PFID(fid));
1317                 return -EACCES;
1318         }
1319
1320         if (!capa_opc_supported(capa, opc)) {
1321                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1322                 return -EACCES;
1323         }
1324
1325         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1326                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1327                 return -EACCES;
1328         }
1329
1330         return 0;
1331 }
1332
1333 static struct timespec *osd_inode_time(const struct lu_env *env,
1334                                        struct inode *inode, __u64 seconds)
1335 {
1336         struct osd_thread_info  *oti = osd_oti_get(env);
1337         struct timespec         *t   = &oti->oti_time;
1338
1339         t->tv_sec = seconds;
1340         t->tv_nsec = 0;
1341         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1342         return t;
1343 }
1344
1345
1346 static void osd_inode_getattr(const struct lu_env *env,
1347                               struct inode *inode, struct lu_attr *attr)
1348 {
1349         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1350                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1351                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1352
1353         attr->la_atime      = LTIME_S(inode->i_atime);
1354         attr->la_mtime      = LTIME_S(inode->i_mtime);
1355         attr->la_ctime      = LTIME_S(inode->i_ctime);
1356         attr->la_mode       = inode->i_mode;
1357         attr->la_size       = i_size_read(inode);
1358         attr->la_blocks     = inode->i_blocks;
1359         attr->la_uid        = inode->i_uid;
1360         attr->la_gid        = inode->i_gid;
1361         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1362         attr->la_nlink      = inode->i_nlink;
1363         attr->la_rdev       = inode->i_rdev;
1364         attr->la_blksize    = 1 << inode->i_blkbits;
1365         attr->la_blkbits    = inode->i_blkbits;
1366 }
1367
1368 static int osd_attr_get(const struct lu_env *env,
1369                         struct dt_object *dt,
1370                         struct lu_attr *attr,
1371                         struct lustre_capa *capa)
1372 {
1373         struct osd_object *obj = osd_dt_obj(dt);
1374
1375         LASSERT(dt_object_exists(dt));
1376         LINVRNT(osd_invariant(obj));
1377
1378         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1379                 return -EACCES;
1380
1381         cfs_spin_lock(&obj->oo_guard);
1382         osd_inode_getattr(env, obj->oo_inode, attr);
1383         cfs_spin_unlock(&obj->oo_guard);
1384         return 0;
1385 }
1386
1387 static int osd_declare_attr_set(const struct lu_env *env,
1388                                 struct dt_object *dt,
1389                                 const struct lu_attr *attr,
1390                                 struct thandle *handle)
1391 {
1392         struct osd_thandle     *oh;
1393         struct osd_object      *obj;
1394         struct osd_thread_info *info = osd_oti_get(env);
1395         struct lquota_id_info  *qi = &info->oti_qi;
1396         long long               bspace;
1397         int                     rc = 0;
1398         bool                    allocated;
1399         ENTRY;
1400
1401         LASSERT(dt != NULL);
1402         LASSERT(handle != NULL);
1403
1404         obj = osd_dt_obj(dt);
1405         LASSERT(osd_invariant(obj));
1406
1407         oh = container_of0(handle, struct osd_thandle, ot_super);
1408         LASSERT(oh->ot_handle == NULL);
1409
1410         OSD_DECLARE_OP(oh, attr_set);
1411         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1412
1413         if (attr == NULL || obj->oo_inode == NULL)
1414                 RETURN(rc);
1415
1416         bspace   = obj->oo_inode->i_blocks;
1417         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1418         bspace   = toqb(bspace);
1419
1420         /* Changing ownership is always preformed by super user, it should not
1421          * fail with EDQUOT.
1422          *
1423          * We still need to call the osd_declare_qid() to calculate the journal
1424          * credits for updating quota accounting files and to trigger quota
1425          * space adjustment once the operation is completed.*/
1426         if ((attr->la_valid & LA_UID) != 0 &&
1427              attr->la_uid != obj->oo_inode->i_uid) {
1428                 qi->lqi_type = USRQUOTA;
1429
1430                 /* inode accounting */
1431                 qi->lqi_is_blk = false;
1432
1433                 /* one more inode for the new owner ... */
1434                 qi->lqi_id.qid_uid = attr->la_uid;
1435                 qi->lqi_space      = 1;
1436                 allocated = (attr->la_uid == 0) ? true : false;
1437                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1438                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1439                         rc = 0;
1440                 if (rc)
1441                         RETURN(rc);
1442
1443                 /* and one less inode for the current uid */
1444                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1445                 qi->lqi_space      = -1;
1446                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1447                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1448                         rc = 0;
1449                 if (rc)
1450                         RETURN(rc);
1451
1452                 /* block accounting */
1453                 qi->lqi_is_blk = true;
1454
1455                 /* more blocks for the new owner ... */
1456                 qi->lqi_id.qid_uid = attr->la_uid;
1457                 qi->lqi_space      = bspace;
1458                 allocated = (attr->la_uid == 0) ? true : false;
1459                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1460                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1461                         rc = 0;
1462                 if (rc)
1463                         RETURN(rc);
1464
1465                 /* and finally less blocks for the current owner */
1466                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1467                 qi->lqi_space      = -bspace;
1468                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1469                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1470                         rc = 0;
1471                 if (rc)
1472                         RETURN(rc);
1473         }
1474
1475         if (attr->la_valid & LA_GID &&
1476             attr->la_gid != obj->oo_inode->i_gid) {
1477                 qi->lqi_type = GRPQUOTA;
1478
1479                 /* inode accounting */
1480                 qi->lqi_is_blk = false;
1481
1482                 /* one more inode for the new group owner ... */
1483                 qi->lqi_id.qid_gid = attr->la_gid;
1484                 qi->lqi_space      = 1;
1485                 allocated = (attr->la_gid == 0) ? true : false;
1486                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1487                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1488                         rc = 0;
1489                 if (rc)
1490                         RETURN(rc);
1491
1492                 /* and one less inode for the current gid */
1493                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1494                 qi->lqi_space      = -1;
1495                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1496                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1497                         rc = 0;
1498                 if (rc)
1499                         RETURN(rc);
1500
1501                 /* block accounting */
1502                 qi->lqi_is_blk = true;
1503
1504                 /* more blocks for the new owner ... */
1505                 qi->lqi_id.qid_gid = attr->la_gid;
1506                 qi->lqi_space      = bspace;
1507                 allocated = (attr->la_gid == 0) ? true : false;
1508                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1509                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1510                         rc = 0;
1511                 if (rc)
1512                         RETURN(rc);
1513
1514                 /* and finally less blocks for the current owner */
1515                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1516                 qi->lqi_space      = -bspace;
1517                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1518                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1519                         rc = 0;
1520                 if (rc)
1521                         RETURN(rc);
1522         }
1523
1524         RETURN(rc);
1525 }
1526
1527 static int osd_inode_setattr(const struct lu_env *env,
1528                              struct inode *inode, const struct lu_attr *attr)
1529 {
1530         __u64 bits;
1531
1532         bits = attr->la_valid;
1533
1534         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1535
1536         if (bits & LA_ATIME)
1537                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1538         if (bits & LA_CTIME)
1539                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1540         if (bits & LA_MTIME)
1541                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1542         if (bits & LA_SIZE) {
1543                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1544                 i_size_write(inode, attr->la_size);
1545         }
1546
1547 #if 0
1548         /* OSD should not change "i_blocks" which is used by quota.
1549          * "i_blocks" should be changed by ldiskfs only. */
1550         if (bits & LA_BLOCKS)
1551                 inode->i_blocks = attr->la_blocks;
1552 #endif
1553         if (bits & LA_MODE)
1554                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1555                         (attr->la_mode & ~S_IFMT);
1556         if (bits & LA_UID)
1557                 inode->i_uid    = attr->la_uid;
1558         if (bits & LA_GID)
1559                 inode->i_gid    = attr->la_gid;
1560         if (bits & LA_NLINK)
1561                 inode->i_nlink  = attr->la_nlink;
1562         if (bits & LA_RDEV)
1563                 inode->i_rdev   = attr->la_rdev;
1564
1565         if (bits & LA_FLAGS) {
1566                 /* always keep S_NOCMTIME */
1567                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1568                                  S_NOCMTIME;
1569         }
1570         return 0;
1571 }
1572
1573 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1574 {
1575         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1576             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1577                 struct iattr    iattr;
1578                 int             rc;
1579
1580                 iattr.ia_valid = 0;
1581                 if (attr->la_valid & LA_UID)
1582                         iattr.ia_valid |= ATTR_UID;
1583                 if (attr->la_valid & LA_GID)
1584                         iattr.ia_valid |= ATTR_GID;
1585                 iattr.ia_uid = attr->la_uid;
1586                 iattr.ia_gid = attr->la_gid;
1587
1588                 rc = ll_vfs_dq_transfer(inode, &iattr);
1589                 if (rc) {
1590                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1591                                "enforcement enabled on the ldiskfs filesystem?",
1592                                inode->i_sb->s_id, rc);
1593                         return rc;
1594                 }
1595         }
1596         return 0;
1597 }
1598
1599 static int osd_attr_set(const struct lu_env *env,
1600                         struct dt_object *dt,
1601                         const struct lu_attr *attr,
1602                         struct thandle *handle,
1603                         struct lustre_capa *capa)
1604 {
1605         struct osd_object *obj = osd_dt_obj(dt);
1606         struct inode      *inode;
1607         int rc;
1608
1609         LASSERT(handle != NULL);
1610         LASSERT(dt_object_exists(dt));
1611         LASSERT(osd_invariant(obj));
1612
1613         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1614                 return -EACCES;
1615
1616         OSD_EXEC_OP(handle, attr_set);
1617
1618         inode = obj->oo_inode;
1619         if (!osd_dt_dev(handle->th_dev)->od_is_md) {
1620                 /* OFD support */
1621                 rc = osd_quota_transfer(inode, attr);
1622                 if (rc)
1623                         return rc;
1624         } else {
1625 #ifdef HAVE_QUOTA_SUPPORT
1626                 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1627                     (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1628                         struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1629                         struct           iattr iattr;
1630                         int              rc;
1631
1632                         iattr.ia_valid = 0;
1633                         if (attr->la_valid & LA_UID)
1634                                 iattr.ia_valid |= ATTR_UID;
1635                         if (attr->la_valid & LA_GID)
1636                                 iattr.ia_valid |= ATTR_GID;
1637                         iattr.ia_uid = attr->la_uid;
1638                         iattr.ia_gid = attr->la_gid;
1639                         osd_push_ctxt(env, save, 1);
1640                         rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1641                         osd_pop_ctxt(save, 1);
1642                         if (rc != 0)
1643                                 return rc;
1644                 }
1645 #endif
1646         }
1647         cfs_spin_lock(&obj->oo_guard);
1648         rc = osd_inode_setattr(env, inode, attr);
1649         cfs_spin_unlock(&obj->oo_guard);
1650
1651         if (!rc)
1652                 inode->i_sb->s_op->dirty_inode(inode);
1653         return rc;
1654 }
1655
1656 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1657                                     struct osd_object *obj,
1658                                     const char *name, const int namelen)
1659 {
1660         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1661 }
1662
1663 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1664                       cfs_umode_t mode,
1665                       struct dt_allocation_hint *hint,
1666                       struct thandle *th)
1667 {
1668         int result;
1669         struct osd_device  *osd = osd_obj2dev(obj);
1670         struct osd_thandle *oth;
1671         struct dt_object   *parent = NULL;
1672         struct inode       *inode;
1673 #ifdef HAVE_QUOTA_SUPPORT
1674         struct osd_ctxt    *save = &info->oti_ctxt;
1675 #endif
1676
1677         LINVRNT(osd_invariant(obj));
1678         LASSERT(obj->oo_inode == NULL);
1679         LASSERT(obj->oo_hl_head == NULL);
1680
1681         if (S_ISDIR(mode) && ldiskfs_pdo) {
1682                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1683                 if (obj->oo_hl_head == NULL)
1684                         return -ENOMEM;
1685         }
1686
1687         oth = container_of(th, struct osd_thandle, ot_super);
1688         LASSERT(oth->ot_handle->h_transaction != NULL);
1689
1690         if (hint && hint->dah_parent)
1691                 parent = hint->dah_parent;
1692
1693 #ifdef HAVE_QUOTA_SUPPORT
1694         osd_push_ctxt(info->oti_env, save, osd_dt_dev(th->th_dev)->od_is_md);
1695 #endif
1696         inode = ldiskfs_create_inode(oth->ot_handle,
1697                                      parent ? osd_dt_obj(parent)->oo_inode :
1698                                               osd_sb(osd)->s_root->d_inode,
1699                                      mode);
1700 #ifdef HAVE_QUOTA_SUPPORT
1701         osd_pop_ctxt(save, osd_dt_dev(th->th_dev)->od_is_md);
1702 #endif
1703         if (!IS_ERR(inode)) {
1704                 /* Do not update file c/mtime in ldiskfs.
1705                  * NB: don't need any lock because no contention at this
1706                  * early stage */
1707                 inode->i_flags |= S_NOCMTIME;
1708                 inode->i_state |= I_LUSTRE_NOSCRUB;
1709                 obj->oo_inode = inode;
1710                 result = 0;
1711         } else {
1712                 if (obj->oo_hl_head != NULL) {
1713                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1714                         obj->oo_hl_head = NULL;
1715                 }
1716                 result = PTR_ERR(inode);
1717         }
1718         LINVRNT(osd_invariant(obj));
1719         return result;
1720 }
1721
1722 enum {
1723         OSD_NAME_LEN = 255
1724 };
1725
1726 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1727                      struct lu_attr *attr,
1728                      struct dt_allocation_hint *hint,
1729                      struct dt_object_format *dof,
1730                      struct thandle *th)
1731 {
1732         int result;
1733         struct osd_thandle *oth;
1734         struct osd_device *osd = osd_obj2dev(obj);
1735         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1736
1737         LASSERT(S_ISDIR(attr->la_mode));
1738
1739         oth = container_of(th, struct osd_thandle, ot_super);
1740         LASSERT(oth->ot_handle->h_transaction != NULL);
1741         result = osd_mkfile(info, obj, mode, hint, th);
1742         if (result == 0 && osd->od_iop_mode == 0) {
1743                 LASSERT(obj->oo_inode != NULL);
1744                 /*
1745                  * XXX uh-oh... call low-level iam function directly.
1746                  */
1747
1748                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1749                                          sizeof (struct osd_fid_pack),
1750                                          oth->ot_handle);
1751         }
1752         return result;
1753 }
1754
1755 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1756                         struct lu_attr *attr,
1757                         struct dt_allocation_hint *hint,
1758                         struct dt_object_format *dof,
1759                         struct thandle *th)
1760 {
1761         int result;
1762         struct osd_thandle *oth;
1763         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1764
1765         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1766
1767         LASSERT(S_ISREG(attr->la_mode));
1768
1769         oth = container_of(th, struct osd_thandle, ot_super);
1770         LASSERT(oth->ot_handle->h_transaction != NULL);
1771
1772         result = osd_mkfile(info, obj, mode, hint, th);
1773         if (result == 0) {
1774                 LASSERT(obj->oo_inode != NULL);
1775                 if (feat->dif_flags & DT_IND_VARKEY)
1776                         result = iam_lvar_create(obj->oo_inode,
1777                                                  feat->dif_keysize_max,
1778                                                  feat->dif_ptrsize,
1779                                                  feat->dif_recsize_max,
1780                                                  oth->ot_handle);
1781                 else
1782                         result = iam_lfix_create(obj->oo_inode,
1783                                                  feat->dif_keysize_max,
1784                                                  feat->dif_ptrsize,
1785                                                  feat->dif_recsize_max,
1786                                                  oth->ot_handle);
1787
1788         }
1789         return result;
1790 }
1791
1792 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1793                      struct lu_attr *attr,
1794                      struct dt_allocation_hint *hint,
1795                      struct dt_object_format *dof,
1796                      struct thandle *th)
1797 {
1798         LASSERT(S_ISREG(attr->la_mode));
1799         return osd_mkfile(info, obj, (attr->la_mode &
1800                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1801 }
1802
1803 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1804                      struct lu_attr *attr,
1805                      struct dt_allocation_hint *hint,
1806                      struct dt_object_format *dof,
1807                      struct thandle *th)
1808 {
1809         LASSERT(S_ISLNK(attr->la_mode));
1810         return osd_mkfile(info, obj, (attr->la_mode &
1811                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1812 }
1813
1814 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1815                      struct lu_attr *attr,
1816                      struct dt_allocation_hint *hint,
1817                      struct dt_object_format *dof,
1818                      struct thandle *th)
1819 {
1820         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1821         int result;
1822
1823         LINVRNT(osd_invariant(obj));
1824         LASSERT(obj->oo_inode == NULL);
1825         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1826                 S_ISFIFO(mode) || S_ISSOCK(mode));
1827
1828         result = osd_mkfile(info, obj, mode, hint, th);
1829         if (result == 0) {
1830                 LASSERT(obj->oo_inode != NULL);
1831                 /*
1832                  * This inode should be marked dirty for i_rdev.  Currently
1833                  * that is done in the osd_attr_init().
1834                  */
1835                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1836         }
1837         LINVRNT(osd_invariant(obj));
1838         return result;
1839 }
1840
1841 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1842                               struct lu_attr *,
1843                               struct dt_allocation_hint *hint,
1844                               struct dt_object_format *dof,
1845                               struct thandle *);
1846
1847 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1848 {
1849         osd_obj_type_f result;
1850
1851         switch (type) {
1852         case DFT_DIR:
1853                 result = osd_mkdir;
1854                 break;
1855         case DFT_REGULAR:
1856                 result = osd_mkreg;
1857                 break;
1858         case DFT_SYM:
1859                 result = osd_mksym;
1860                 break;
1861         case DFT_NODE:
1862                 result = osd_mknod;
1863                 break;
1864         case DFT_INDEX:
1865                 result = osd_mk_index;
1866                 break;
1867
1868         default:
1869                 LBUG();
1870                 break;
1871         }
1872         return result;
1873 }
1874
1875
1876 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1877                         struct dt_object *parent, struct dt_object *child,
1878                         cfs_umode_t child_mode)
1879 {
1880         LASSERT(ah);
1881
1882         memset(ah, 0, sizeof(*ah));
1883         ah->dah_parent = parent;
1884         ah->dah_mode = child_mode;
1885 }
1886
1887 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1888                           struct lu_attr *attr, struct dt_object_format *dof)
1889 {
1890         struct inode   *inode = obj->oo_inode;
1891         __u64           valid = attr->la_valid;
1892         int             result;
1893
1894         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1895
1896         if (dof->dof_type != DFT_NODE)
1897                 attr->la_valid &= ~LA_RDEV;
1898         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1899                 attr->la_valid &= ~LA_ATIME;
1900         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1901                 attr->la_valid &= ~LA_CTIME;
1902         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1903                 attr->la_valid &= ~LA_MTIME;
1904
1905         if (!osd_obj2dev(obj)->od_is_md) {
1906                 /* OFD support */
1907                 result = osd_quota_transfer(inode, attr);
1908                 if (result)
1909                         return;
1910         } else {
1911 #ifdef HAVE_QUOTA_SUPPORT
1912                 attr->la_valid &= ~(LA_UID | LA_GID);
1913 #endif
1914         }
1915
1916         if (attr->la_valid != 0) {
1917                 result = osd_inode_setattr(info->oti_env, inode, attr);
1918                 /*
1919                  * The osd_inode_setattr() should always succeed here.  The
1920                  * only error that could be returned is EDQUOT when we are
1921                  * trying to change the UID or GID of the inode. However, this
1922                  * should not happen since quota enforcement is no longer
1923                  * enabled on ldiskfs (lquota takes care of it).
1924                  */
1925                 LASSERTF(result == 0, "%d", result);
1926                 inode->i_sb->s_op->dirty_inode(inode);
1927         }
1928
1929         attr->la_valid = valid;
1930 }
1931
1932 /**
1933  * Helper function for osd_object_create()
1934  *
1935  * \retval 0, on success
1936  */
1937 static int __osd_object_create(struct osd_thread_info *info,
1938                                struct osd_object *obj, struct lu_attr *attr,
1939                                struct dt_allocation_hint *hint,
1940                                struct dt_object_format *dof,
1941                                struct thandle *th)
1942 {
1943         int     result;
1944         __u32   umask;
1945
1946         /* we drop umask so that permissions we pass are not affected */
1947         umask = current->fs->umask;
1948         current->fs->umask = 0;
1949
1950         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1951                                                   th);
1952         if (result == 0) {
1953                 osd_attr_init(info, obj, attr, dof);
1954                 osd_object_init0(obj);
1955                 /* bz 24037 */
1956                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1957                         unlock_new_inode(obj->oo_inode);
1958         }
1959
1960         /* restore previous umask value */
1961         current->fs->umask = umask;
1962
1963         return result;
1964 }
1965
1966 /**
1967  * Helper function for osd_object_create()
1968  *
1969  * \retval 0, on success
1970  */
1971 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1972                            const struct lu_fid *fid, struct thandle *th)
1973 {
1974         struct osd_thread_info *info = osd_oti_get(env);
1975         struct osd_inode_id    *id   = &info->oti_id;
1976         struct osd_device      *osd  = osd_obj2dev(obj);
1977
1978         LASSERT(obj->oo_inode != NULL);
1979
1980         if (osd->od_is_md) {
1981                 struct md_ucred *uc = md_ucred(env);
1982                 LASSERT(uc != NULL);
1983         }
1984
1985         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1986         return osd_oi_insert(info, osd, fid, id, th);
1987 }
1988
1989 static int osd_declare_object_create(const struct lu_env *env,
1990                                      struct dt_object *dt,
1991                                      struct lu_attr *attr,
1992                                      struct dt_allocation_hint *hint,
1993                                      struct dt_object_format *dof,
1994                                      struct thandle *handle)
1995 {
1996         struct osd_thandle      *oh;
1997         int                      rc;
1998         ENTRY;
1999
2000         LASSERT(handle != NULL);
2001
2002         oh = container_of0(handle, struct osd_thandle, ot_super);
2003         LASSERT(oh->ot_handle == NULL);
2004
2005         OSD_DECLARE_OP(oh, create);
2006         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
2007         /* XXX: So far, only normal fid needs be inserted into the oi,
2008          *      things could be changed later. Revise following code then. */
2009         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
2010                 OSD_DECLARE_OP(oh, insert);
2011                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
2012                 /* Reuse idle OI block may cause additional one OI block
2013                  * to be changed. */
2014                 oh->ot_credits += 1;
2015         }
2016         /* If this is directory, then we expect . and .. to be inserted as
2017          * well. The one directory block always needs to be created for the
2018          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2019          * block), there is no danger of needing a tree for the first block.
2020          */
2021         if (attr && S_ISDIR(attr->la_mode)) {
2022                 OSD_DECLARE_OP(oh, insert);
2023                 OSD_DECLARE_OP(oh, insert);
2024                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
2025         }
2026
2027         if (!attr)
2028                 RETURN(0);
2029
2030         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2031                                    false, false, NULL, false);
2032         RETURN(rc);
2033 }
2034
2035 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2036                              struct lu_attr *attr,
2037                              struct dt_allocation_hint *hint,
2038                              struct dt_object_format *dof,
2039                              struct thandle *th)
2040 {
2041         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2042         struct osd_object      *obj    = osd_dt_obj(dt);
2043         struct osd_thread_info *info   = osd_oti_get(env);
2044         int result;
2045
2046         ENTRY;
2047
2048         LINVRNT(osd_invariant(obj));
2049         LASSERT(!dt_object_exists(dt));
2050         LASSERT(osd_write_locked(env, obj));
2051         LASSERT(th != NULL);
2052
2053         if (unlikely(fid_is_acct(fid)))
2054                 /* Quota files can't be created from the kernel any more,
2055                  * 'tune2fs -O quota' will take care of creating them */
2056                 RETURN(-EPERM);
2057
2058         OSD_EXEC_OP(th, create);
2059
2060         result = __osd_object_create(info, obj, attr, hint, dof, th);
2061         if (result == 0)
2062                 result = __osd_oi_insert(env, obj, fid, th);
2063
2064         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2065         LASSERT(osd_invariant(obj));
2066         RETURN(result);
2067 }
2068
2069 /**
2070  * Called to destroy on-disk representation of the object
2071  *
2072  * Concurrency: must be locked
2073  */
2074 static int osd_declare_object_destroy(const struct lu_env *env,
2075                                       struct dt_object *dt,
2076                                       struct thandle *th)
2077 {
2078         struct osd_object  *obj = osd_dt_obj(dt);
2079         struct inode       *inode = obj->oo_inode;
2080         struct osd_thandle *oh;
2081         int                 rc;
2082         ENTRY;
2083
2084         oh = container_of0(th, struct osd_thandle, ot_super);
2085         LASSERT(oh->ot_handle == NULL);
2086         LASSERT(inode);
2087
2088         OSD_DECLARE_OP(oh, destroy);
2089         OSD_DECLARE_OP(oh, delete);
2090         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
2091         /* XXX: So far, only normal fid needs to be inserted into the OI,
2092          *      so only normal fid needs to be removed from the OI also. */
2093         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
2094                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2095                 /* Recycle idle OI leaf may cause additional three OI blocks
2096                  * to be changed. */
2097                 oh->ot_credits += 3;
2098         }
2099
2100         /* one less inode */
2101         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2102                                    false, true, NULL, false);
2103         if (rc)
2104                 RETURN(rc);
2105         /* data to be truncated */
2106         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, true,
2107                                    true, NULL, false);
2108         RETURN(rc);
2109 }
2110
2111 static int osd_object_destroy(const struct lu_env *env,
2112                               struct dt_object *dt,
2113                               struct thandle *th)
2114 {
2115         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2116         struct osd_object      *obj = osd_dt_obj(dt);
2117         struct inode           *inode = obj->oo_inode;
2118         struct osd_device      *osd = osd_obj2dev(obj);
2119         struct osd_thandle     *oh;
2120         int                     result;
2121         ENTRY;
2122
2123         oh = container_of0(th, struct osd_thandle, ot_super);
2124         LASSERT(oh->ot_handle);
2125         LASSERT(inode);
2126         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2127
2128         if (unlikely(fid_is_acct(fid)))
2129                 RETURN(-EPERM);
2130
2131         /* Parallel control for OI scrub. For most of cases, there is no
2132          * lock contention. So it will not affect unlink performance. */
2133         cfs_mutex_lock(&inode->i_mutex);
2134         if (S_ISDIR(inode->i_mode)) {
2135                 LASSERT(osd_inode_unlinked(inode) ||
2136                         inode->i_nlink == 1);
2137                 cfs_spin_lock(&obj->oo_guard);
2138                 inode->i_nlink = 0;
2139                 cfs_spin_unlock(&obj->oo_guard);
2140                 inode->i_sb->s_op->dirty_inode(inode);
2141         } else {
2142                 LASSERT(osd_inode_unlinked(inode));
2143         }
2144
2145         OSD_EXEC_OP(th, destroy);
2146
2147         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2148         cfs_mutex_unlock(&inode->i_mutex);
2149
2150         /* XXX: add to ext3 orphan list */
2151         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2152
2153         /* not needed in the cache anymore */
2154         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2155
2156         RETURN(0);
2157 }
2158
2159 /**
2160  * Helper function for osd_xattr_set()
2161  */
2162 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2163                            const struct lu_buf *buf, const char *name, int fl)
2164 {
2165         struct osd_object      *obj      = osd_dt_obj(dt);
2166         struct inode           *inode    = obj->oo_inode;
2167         struct osd_thread_info *info     = osd_oti_get(env);
2168         struct dentry          *dentry   = &info->oti_child_dentry;
2169         int                     fs_flags = 0;
2170         int                     rc;
2171
2172         LASSERT(dt_object_exists(dt));
2173         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2174
2175         if (fl & LU_XATTR_REPLACE)
2176                 fs_flags |= XATTR_REPLACE;
2177
2178         if (fl & LU_XATTR_CREATE)
2179                 fs_flags |= XATTR_CREATE;
2180
2181         dentry->d_inode = inode;
2182         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2183                                    buf->lb_len, fs_flags);
2184         return rc;
2185 }
2186
2187 /**
2188  * Put the fid into lustre_mdt_attrs, and then place the structure
2189  * inode's ea. This fid should not be altered during the life time
2190  * of the inode.
2191  *
2192  * \retval +ve, on success
2193  * \retval -ve, on error
2194  *
2195  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2196  */
2197 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2198                           const struct lu_fid *fid)
2199 {
2200         struct osd_thread_info  *info      = osd_oti_get(env);
2201         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2202
2203         lustre_lma_init(mdt_attrs, fid);
2204         lustre_lma_swab(mdt_attrs);
2205         return __osd_xattr_set(env, dt,
2206                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2207                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2208
2209 }
2210
2211 /**
2212  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2213  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2214  * To have compatilibility with 1.8 ldiskfs driver we need to have
2215  * magic number at start of fid data.
2216  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2217  * its inmemory API.
2218  */
2219 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2220                                   const struct dt_rec *fid)
2221 {
2222         param->edp_magic = LDISKFS_LUFID_MAGIC;
2223         param->edp_len =  sizeof(struct lu_fid) + 1;
2224
2225         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2226                       (struct lu_fid *)fid);
2227 }
2228
2229 /**
2230  * Try to read the fid from inode ea into dt_rec, if return value
2231  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2232  *
2233  * \param fid object fid.
2234  *
2235  * \retval 0 on success
2236  */
2237 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2238                           __u32 ino, struct lu_fid *fid,
2239                           struct osd_inode_id *id)
2240 {
2241         struct osd_thread_info *info  = osd_oti_get(env);
2242         struct inode           *inode;
2243         ENTRY;
2244
2245         osd_id_gen(id, ino, OSD_OII_NOGEN);
2246         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2247         if (IS_ERR(inode))
2248                 RETURN(PTR_ERR(inode));
2249
2250         iput(inode);
2251         RETURN(0);
2252 }
2253
2254 /**
2255  * OSD layer object create function for interoperability mode (b11826).
2256  * This is mostly similar to osd_object_create(). Only difference being, fid is
2257  * inserted into inode ea here.
2258  *
2259  * \retval   0, on success
2260  * \retval -ve, on error
2261  */
2262 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2263                                 struct lu_attr *attr,
2264                                 struct dt_allocation_hint *hint,
2265                                 struct dt_object_format *dof,
2266                                 struct thandle *th)
2267 {
2268         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2269         struct osd_object      *obj    = osd_dt_obj(dt);
2270         struct osd_thread_info *info   = osd_oti_get(env);
2271         int                     result;
2272
2273         ENTRY;
2274
2275         LASSERT(osd_invariant(obj));
2276         LASSERT(!dt_object_exists(dt));
2277         LASSERT(osd_write_locked(env, obj));
2278         LASSERT(th != NULL);
2279
2280         if (unlikely(fid_is_acct(fid)))
2281                 /* Quota files can't be created from the kernel any more,
2282                  * 'tune2fs -O quota' will take care of creating them */
2283                 RETURN(-EPERM);
2284
2285         OSD_EXEC_OP(th, create);
2286
2287         result = __osd_object_create(info, obj, attr, hint, dof, th);
2288         /* objects under osd root shld have igif fid, so dont add fid EA */
2289         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2290                 result = osd_ea_fid_set(env, dt, fid);
2291
2292         if (result == 0)
2293                 result = __osd_oi_insert(env, obj, fid, th);
2294
2295         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2296         LINVRNT(osd_invariant(obj));
2297         RETURN(result);
2298 }
2299
2300 static int osd_declare_object_ref_add(const struct lu_env *env,
2301                                       struct dt_object *dt,
2302                                       struct thandle *handle)
2303 {
2304         struct osd_thandle *oh;
2305
2306         /* it's possible that object doesn't exist yet */
2307         LASSERT(handle != NULL);
2308
2309         oh = container_of0(handle, struct osd_thandle, ot_super);
2310         LASSERT(oh->ot_handle == NULL);
2311
2312         OSD_DECLARE_OP(oh, ref_add);
2313         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2314
2315         return 0;
2316 }
2317
2318 /*
2319  * Concurrency: @dt is write locked.
2320  */
2321 static int osd_object_ref_add(const struct lu_env *env,
2322                               struct dt_object *dt, struct thandle *th)
2323 {
2324         struct osd_object *obj = osd_dt_obj(dt);
2325         struct inode      *inode = obj->oo_inode;
2326
2327         LINVRNT(osd_invariant(obj));
2328         LASSERT(dt_object_exists(dt));
2329         LASSERT(osd_write_locked(env, obj));
2330         LASSERT(th != NULL);
2331
2332         OSD_EXEC_OP(th, ref_add);
2333
2334         /*
2335          * DIR_NLINK feature is set for compatibility reasons if:
2336          * 1) nlinks > LDISKFS_LINK_MAX, or
2337          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2338          *
2339          * It is easier to always set this flag (rather than check and set),
2340          * since it has less overhead, and the superblock will be dirtied
2341          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2342          * do not actually care whether this flag is set or not.
2343          */
2344         cfs_spin_lock(&obj->oo_guard);
2345         inode->i_nlink++;
2346         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2347                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2348                     inode->i_nlink == 2)
2349                         inode->i_nlink = 1;
2350         }
2351         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2352         cfs_spin_unlock(&obj->oo_guard);
2353         inode->i_sb->s_op->dirty_inode(inode);
2354         LINVRNT(osd_invariant(obj));
2355
2356         return 0;
2357 }
2358
2359 static int osd_declare_object_ref_del(const struct lu_env *env,
2360                                       struct dt_object *dt,
2361                                       struct thandle *handle)
2362 {
2363         struct osd_thandle *oh;
2364
2365         LASSERT(dt_object_exists(dt));
2366         LASSERT(handle != NULL);
2367
2368         oh = container_of0(handle, struct osd_thandle, ot_super);
2369         LASSERT(oh->ot_handle == NULL);
2370
2371         OSD_DECLARE_OP(oh, ref_del);
2372         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2373
2374         return 0;
2375 }
2376
2377 /*
2378  * Concurrency: @dt is write locked.
2379  */
2380 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2381                               struct thandle *th)
2382 {
2383         struct osd_object *obj = osd_dt_obj(dt);
2384         struct inode      *inode = obj->oo_inode;
2385
2386         LINVRNT(osd_invariant(obj));
2387         LASSERT(dt_object_exists(dt));
2388         LASSERT(osd_write_locked(env, obj));
2389         LASSERT(th != NULL);
2390
2391         OSD_EXEC_OP(th, ref_del);
2392
2393         cfs_spin_lock(&obj->oo_guard);
2394         LASSERT(inode->i_nlink > 0);
2395         inode->i_nlink--;
2396         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2397          * then the nlink count is 1. Don't let it be set to 0 or the directory
2398          * inode will be deleted incorrectly. */
2399         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2400                 inode->i_nlink++;
2401         cfs_spin_unlock(&obj->oo_guard);
2402         inode->i_sb->s_op->dirty_inode(inode);
2403         LINVRNT(osd_invariant(obj));
2404
2405         return 0;
2406 }
2407
2408 /*
2409  * Get the 64-bit version for an inode.
2410  */
2411 static int osd_object_version_get(const struct lu_env *env,
2412                                   struct dt_object *dt, dt_obj_version_t *ver)
2413 {
2414         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2415
2416         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2417                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2418         *ver = LDISKFS_I(inode)->i_fs_version;
2419         return 0;
2420 }
2421
2422 /*
2423  * Concurrency: @dt is read locked.
2424  */
2425 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2426                          struct lu_buf *buf, const char *name,
2427                          struct lustre_capa *capa)
2428 {
2429         struct osd_object      *obj    = osd_dt_obj(dt);
2430         struct inode           *inode  = obj->oo_inode;
2431         struct osd_thread_info *info   = osd_oti_get(env);
2432         struct dentry          *dentry = &info->oti_obj_dentry;
2433
2434         /* version get is not real XATTR but uses xattr API */
2435         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2436                 /* for version we are just using xattr API but change inode
2437                  * field instead */
2438                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2439                 osd_object_version_get(env, dt, buf->lb_buf);
2440                 return sizeof(dt_obj_version_t);
2441         }
2442
2443         LASSERT(dt_object_exists(dt));
2444         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2445
2446         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2447                 return -EACCES;
2448
2449         dentry->d_inode = inode;
2450         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2451 }
2452
2453
2454 static int osd_declare_xattr_set(const struct lu_env *env,
2455                                  struct dt_object *dt,
2456                                  const struct lu_buf *buf, const char *name,
2457                                  int fl, struct thandle *handle)
2458 {
2459         struct osd_thandle *oh;
2460
2461         LASSERT(handle != NULL);
2462
2463         oh = container_of0(handle, struct osd_thandle, ot_super);
2464         LASSERT(oh->ot_handle == NULL);
2465
2466         OSD_DECLARE_OP(oh, xattr_set);
2467         if (strcmp(name, XATTR_NAME_VERSION) == 0)
2468                 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2469         else
2470                 oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2471
2472         return 0;
2473 }
2474
2475 /*
2476  * Set the 64-bit version for object
2477  */
2478 static void osd_object_version_set(const struct lu_env *env,
2479                                    struct dt_object *dt,
2480                                    dt_obj_version_t *new_version)
2481 {
2482         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2483
2484         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2485                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2486
2487         LDISKFS_I(inode)->i_fs_version = *new_version;
2488         /** Version is set after all inode operations are finished,
2489          *  so we should mark it dirty here */
2490         inode->i_sb->s_op->dirty_inode(inode);
2491 }
2492
2493 /*
2494  * Concurrency: @dt is write locked.
2495  */
2496 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2497                          const struct lu_buf *buf, const char *name, int fl,
2498                          struct thandle *handle, struct lustre_capa *capa)
2499 {
2500         LASSERT(handle != NULL);
2501
2502         /* version set is not real XATTR */
2503         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2504                 /* for version we are just using xattr API but change inode
2505                  * field instead */
2506                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2507                 osd_object_version_set(env, dt, buf->lb_buf);
2508                 return sizeof(dt_obj_version_t);
2509         }
2510
2511         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2512                 return -EACCES;
2513
2514         OSD_EXEC_OP(handle, xattr_set);
2515         return __osd_xattr_set(env, dt, buf, name, fl);
2516 }
2517
2518 /*
2519  * Concurrency: @dt is read locked.
2520  */
2521 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2522                           struct lu_buf *buf, struct lustre_capa *capa)
2523 {
2524         struct osd_object      *obj    = osd_dt_obj(dt);
2525         struct inode           *inode  = obj->oo_inode;
2526         struct osd_thread_info *info   = osd_oti_get(env);
2527         struct dentry          *dentry = &info->oti_obj_dentry;
2528
2529         LASSERT(dt_object_exists(dt));
2530         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2531         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2532
2533         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2534                 return -EACCES;
2535
2536         dentry->d_inode = inode;
2537         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2538 }
2539
2540 static int osd_declare_xattr_del(const struct lu_env *env,
2541                                  struct dt_object *dt, const char *name,
2542                                  struct thandle *handle)
2543 {
2544         struct osd_thandle *oh;
2545
2546         LASSERT(dt_object_exists(dt));
2547         LASSERT(handle != NULL);
2548
2549         oh = container_of0(handle, struct osd_thandle, ot_super);
2550         LASSERT(oh->ot_handle == NULL);
2551
2552         OSD_DECLARE_OP(oh, xattr_set);
2553         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2554
2555         return 0;
2556 }
2557
2558 /*
2559  * Concurrency: @dt is write locked.
2560  */
2561 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2562                          const char *name, struct thandle *handle,
2563                          struct lustre_capa *capa)
2564 {
2565         struct osd_object      *obj    = osd_dt_obj(dt);
2566         struct inode           *inode  = obj->oo_inode;
2567         struct osd_thread_info *info   = osd_oti_get(env);
2568         struct dentry          *dentry = &info->oti_obj_dentry;
2569         int                     rc;
2570
2571         LASSERT(dt_object_exists(dt));
2572         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2573         LASSERT(osd_write_locked(env, obj));
2574         LASSERT(handle != NULL);
2575
2576         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2577                 return -EACCES;
2578
2579         OSD_EXEC_OP(handle, xattr_set);
2580
2581         dentry->d_inode = inode;
2582         rc = inode->i_op->removexattr(dentry, name);
2583         return rc;
2584 }
2585
2586 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2587                                      struct dt_object *dt,
2588                                      struct lustre_capa *old,
2589                                      __u64 opc)
2590 {
2591         struct osd_thread_info *info = osd_oti_get(env);
2592         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2593         struct osd_object *obj = osd_dt_obj(dt);
2594         struct osd_device *dev = osd_obj2dev(obj);
2595         struct lustre_capa_key *key = &info->oti_capa_key;
2596         struct lustre_capa *capa = &info->oti_capa;
2597         struct obd_capa *oc;
2598         struct md_capainfo *ci;
2599         int rc;
2600         ENTRY;
2601
2602         if (!dev->od_fl_capa)
2603                 RETURN(ERR_PTR(-ENOENT));
2604
2605         LASSERT(dt_object_exists(dt));
2606         LINVRNT(osd_invariant(obj));
2607
2608         /* renewal sanity check */
2609         if (old && osd_object_auth(env, dt, old, opc))
2610                 RETURN(ERR_PTR(-EACCES));
2611
2612         ci = md_capainfo(env);
2613         if (unlikely(!ci))
2614                 RETURN(ERR_PTR(-ENOENT));
2615
2616         switch (ci->mc_auth) {
2617         case LC_ID_NONE:
2618                 RETURN(NULL);
2619         case LC_ID_PLAIN:
2620                 capa->lc_uid = obj->oo_inode->i_uid;
2621                 capa->lc_gid = obj->oo_inode->i_gid;
2622                 capa->lc_flags = LC_ID_PLAIN;
2623                 break;
2624         case LC_ID_CONVERT: {
2625                 __u32 d[4], s[4];
2626
2627                 s[0] = obj->oo_inode->i_uid;
2628                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2629                 s[2] = obj->oo_inode->i_gid;
2630                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2631                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2632                 if (unlikely(rc))
2633                         RETURN(ERR_PTR(rc));
2634
2635                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2636                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2637                 capa->lc_flags = LC_ID_CONVERT;
2638                 break;
2639         }
2640         default:
2641                 RETURN(ERR_PTR(-EINVAL));
2642         }
2643
2644         capa->lc_fid = *fid;
2645         capa->lc_opc = opc;
2646         capa->lc_flags |= dev->od_capa_alg << 24;
2647         capa->lc_timeout = dev->od_capa_timeout;
2648         capa->lc_expiry = 0;
2649
2650         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2651         if (oc) {
2652                 LASSERT(!capa_is_expired(oc));
2653                 RETURN(oc);
2654         }
2655
2656         cfs_spin_lock(&capa_lock);
2657         *key = dev->od_capa_keys[1];
2658         cfs_spin_unlock(&capa_lock);
2659
2660         capa->lc_keyid = key->lk_keyid;
2661         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2662
2663         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2664         if (rc) {
2665                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2666                 RETURN(ERR_PTR(rc));
2667         }
2668
2669         oc = capa_add(dev->od_capa_hash, capa);
2670         RETURN(oc);
2671 }
2672
2673 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2674 {
2675         struct osd_object       *obj    = osd_dt_obj(dt);
2676         struct inode            *inode  = obj->oo_inode;
2677         struct osd_thread_info  *info   = osd_oti_get(env);
2678         struct dentry           *dentry = &info->oti_obj_dentry;
2679         struct file             *file   = &info->oti_file;
2680         int                     rc;
2681
2682         ENTRY;
2683
2684         dentry->d_inode = inode;
2685         file->f_dentry = dentry;
2686         file->f_mapping = inode->i_mapping;
2687         file->f_op = inode->i_fop;
2688         mutex_lock(&inode->i_mutex);
2689         rc = file->f_op->fsync(file, dentry, 0);
2690         mutex_unlock(&inode->i_mutex);
2691         RETURN(rc);
2692 }
2693
2694 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2695                         void **data)
2696 {
2697         struct osd_object *obj = osd_dt_obj(dt);
2698         ENTRY;
2699
2700         *data = (void *)obj->oo_inode;
2701         RETURN(0);
2702 }
2703
2704 /*
2705  * Index operations.
2706  */
2707
2708 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2709                            const struct dt_index_features *feat)
2710 {
2711         struct iam_descr *descr;
2712
2713         if (osd_object_is_root(o))
2714                 return feat == &dt_directory_features;
2715
2716         LASSERT(o->oo_dir != NULL);
2717
2718         descr = o->oo_dir->od_container.ic_descr;
2719         if (feat == &dt_directory_features) {
2720                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2721                         return 1;
2722                 else
2723                         return 0;
2724         } else {
2725                 return
2726                         feat->dif_keysize_min <= descr->id_key_size &&
2727                         descr->id_key_size <= feat->dif_keysize_max &&
2728                         feat->dif_recsize_min <= descr->id_rec_size &&
2729                         descr->id_rec_size <= feat->dif_recsize_max &&
2730                         !(feat->dif_flags & (DT_IND_VARKEY |
2731                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2732                         ergo(feat->dif_flags & DT_IND_UPDATE,
2733                              1 /* XXX check that object (and file system) is
2734                                 * writable */);
2735         }
2736 }
2737
2738 static int osd_iam_container_init(const struct lu_env *env,
2739                                   struct osd_object *obj,
2740                                   struct osd_directory *dir)
2741 {
2742         struct iam_container *bag = &dir->od_container;
2743         int result;
2744
2745         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2746         if (result != 0)
2747                 return result;
2748
2749         result = iam_container_setup(bag);
2750         if (result == 0)
2751                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2752         else
2753                 iam_container_fini(bag);
2754
2755         return result;
2756 }
2757
2758
2759 /*
2760  * Concurrency: no external locking is necessary.
2761  */
2762 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2763                          const struct dt_index_features *feat)
2764 {
2765         int                      result;
2766         int                      skip_iam = 0;
2767         struct osd_object       *obj = osd_dt_obj(dt);
2768         struct osd_device       *osd = osd_obj2dev(obj);
2769
2770         LINVRNT(osd_invariant(obj));
2771         LASSERT(dt_object_exists(dt));
2772
2773         if (osd_object_is_root(obj)) {
2774                 dt->do_index_ops = &osd_index_ea_ops;
2775                 result = 0;
2776         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2777                 dt->do_index_ops = &osd_index_ea_ops;
2778                 if (S_ISDIR(obj->oo_inode->i_mode))
2779                         result = 0;
2780                 else
2781                         result = -ENOTDIR;
2782                 skip_iam = 1;
2783         } else if (unlikely(feat == &dt_otable_features)) {
2784                 dt->do_index_ops = &osd_otable_ops;
2785                 return 0;
2786         } else if (feat == &dt_acct_features) {
2787                 dt->do_index_ops = &osd_acct_index_ops;
2788                 result = 0;
2789                 skip_iam = 1;
2790         } else if (!osd_has_index(obj)) {
2791                 struct osd_directory *dir;
2792
2793                 OBD_ALLOC_PTR(dir);
2794                 if (dir != NULL) {
2795
2796                         cfs_spin_lock(&obj->oo_guard);
2797                         if (obj->oo_dir == NULL)
2798                                 obj->oo_dir = dir;
2799                         else
2800                                 /*
2801                                  * Concurrent thread allocated container data.
2802                                  */
2803                                 OBD_FREE_PTR(dir);
2804                         cfs_spin_unlock(&obj->oo_guard);
2805                         /*
2806                          * Now, that we have container data, serialize its
2807                          * initialization.
2808                          */
2809                         cfs_down_write(&obj->oo_ext_idx_sem);
2810                         /*
2811                          * recheck under lock.
2812                          */
2813                         if (!osd_has_index(obj))
2814                                 result = osd_iam_container_init(env, obj, dir);
2815                         else
2816                                 result = 0;
2817                         cfs_up_write(&obj->oo_ext_idx_sem);
2818                 } else {
2819                         result = -ENOMEM;
2820                 }
2821         } else {
2822                 result = 0;
2823         }
2824
2825         if (result == 0 && skip_iam == 0) {
2826                 if (!osd_iam_index_probe(env, obj, feat))
2827                         result = -ENOTDIR;
2828         }
2829         LINVRNT(osd_invariant(obj));
2830
2831         if (is_quota_glb_feat(feat))
2832                 result = osd_quota_migration(env, dt, feat);
2833
2834         return result;
2835 }
2836
2837 static int osd_otable_it_attr_get(const struct lu_env *env,
2838                                  struct dt_object *dt,
2839                                  struct lu_attr *attr,
2840                                  struct lustre_capa *capa)
2841 {
2842         attr->la_valid = 0;
2843         return 0;
2844 }
2845
2846 static const struct dt_object_operations osd_obj_ops = {
2847         .do_read_lock         = osd_object_read_lock,
2848         .do_write_lock        = osd_object_write_lock,
2849         .do_read_unlock       = osd_object_read_unlock,
2850         .do_write_unlock      = osd_object_write_unlock,
2851         .do_write_locked      = osd_object_write_locked,
2852         .do_attr_get          = osd_attr_get,
2853         .do_declare_attr_set  = osd_declare_attr_set,
2854         .do_attr_set          = osd_attr_set,
2855         .do_ah_init           = osd_ah_init,
2856         .do_declare_create    = osd_declare_object_create,
2857         .do_create            = osd_object_create,
2858         .do_declare_destroy   = osd_declare_object_destroy,
2859         .do_destroy           = osd_object_destroy,
2860         .do_index_try         = osd_index_try,
2861         .do_declare_ref_add   = osd_declare_object_ref_add,
2862         .do_ref_add           = osd_object_ref_add,
2863         .do_declare_ref_del   = osd_declare_object_ref_del,
2864         .do_ref_del           = osd_object_ref_del,
2865         .do_xattr_get         = osd_xattr_get,
2866         .do_declare_xattr_set = osd_declare_xattr_set,
2867         .do_xattr_set         = osd_xattr_set,
2868         .do_declare_xattr_del = osd_declare_xattr_del,
2869         .do_xattr_del         = osd_xattr_del,
2870         .do_xattr_list        = osd_xattr_list,
2871         .do_capa_get          = osd_capa_get,
2872         .do_object_sync       = osd_object_sync,
2873         .do_data_get          = osd_data_get,
2874 };
2875
2876 /**
2877  * dt_object_operations for interoperability mode
2878  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2879  */
2880 static const struct dt_object_operations osd_obj_ea_ops = {
2881         .do_read_lock         = osd_object_read_lock,
2882         .do_write_lock        = osd_object_write_lock,
2883         .do_read_unlock       = osd_object_read_unlock,
2884         .do_write_unlock      = osd_object_write_unlock,
2885         .do_write_locked      = osd_object_write_locked,
2886         .do_attr_get          = osd_attr_get,
2887         .do_declare_attr_set  = osd_declare_attr_set,
2888         .do_attr_set          = osd_attr_set,
2889         .do_ah_init           = osd_ah_init,
2890         .do_declare_create    = osd_declare_object_create,
2891         .do_create            = osd_object_ea_create,
2892         .do_declare_destroy   = osd_declare_object_destroy,
2893         .do_destroy           = osd_object_destroy,
2894         .do_index_try         = osd_index_try,
2895         .do_declare_ref_add   = osd_declare_object_ref_add,
2896         .do_ref_add           = osd_object_ref_add,
2897         .do_declare_ref_del   = osd_declare_object_ref_del,
2898         .do_ref_del           = osd_object_ref_del,
2899         .do_xattr_get         = osd_xattr_get,
2900         .do_declare_xattr_set = osd_declare_xattr_set,
2901         .do_xattr_set         = osd_xattr_set,
2902         .do_declare_xattr_del = osd_declare_xattr_del,
2903         .do_xattr_del         = osd_xattr_del,
2904         .do_xattr_list        = osd_xattr_list,
2905         .do_capa_get          = osd_capa_get,
2906         .do_object_sync       = osd_object_sync,
2907         .do_data_get          = osd_data_get,
2908 };
2909
2910 static const struct dt_object_operations osd_obj_otable_it_ops = {
2911         .do_attr_get    = osd_otable_it_attr_get,
2912         .do_index_try   = osd_index_try,
2913 };
2914
2915 static int osd_index_declare_iam_delete(const struct lu_env *env,
2916                                         struct dt_object *dt,
2917                                         const struct dt_key *key,
2918                                         struct thandle *handle)
2919 {
2920         struct osd_thandle    *oh;
2921
2922         oh = container_of0(handle, struct osd_thandle, ot_super);
2923         LASSERT(oh->ot_handle == NULL);
2924
2925         OSD_DECLARE_OP(oh, delete);
2926         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2927
2928         return 0;
2929 }
2930
2931 /**
2932  *      delete a (key, value) pair from index \a dt specified by \a key
2933  *
2934  *      \param  dt      osd index object
2935  *      \param  key     key for index
2936  *      \param  rec     record reference
2937  *      \param  handle  transaction handler
2938  *
2939  *      \retval  0  success
2940  *      \retval -ve   failure
2941  */
2942
2943 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2944                                 const struct dt_key *key,
2945                                 struct thandle *handle,
2946                                 struct lustre_capa *capa)
2947 {
2948         struct osd_thread_info *oti = osd_oti_get(env);
2949         struct osd_object      *obj = osd_dt_obj(dt);
2950         struct osd_thandle     *oh;
2951         struct iam_path_descr  *ipd;
2952         struct iam_container   *bag = &obj->oo_dir->od_container;
2953         int                     rc;
2954
2955         ENTRY;
2956
2957         LINVRNT(osd_invariant(obj));
2958         LASSERT(dt_object_exists(dt));
2959         LASSERT(bag->ic_object == obj->oo_inode);
2960         LASSERT(handle != NULL);
2961
2962         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2963                 RETURN(-EACCES);
2964
2965         OSD_EXEC_OP(handle, delete);
2966
2967         ipd = osd_idx_ipd_get(env, bag);
2968         if (unlikely(ipd == NULL))
2969                 RETURN(-ENOMEM);
2970
2971         oh = container_of0(handle, struct osd_thandle, ot_super);
2972         LASSERT(oh->ot_handle != NULL);
2973         LASSERT(oh->ot_handle->h_transaction != NULL);
2974
2975         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
2976                 /* swab quota uid/gid provided by caller */
2977                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
2978                 key = (const struct dt_key *)&oti->oti_quota_id;
2979         }
2980
2981         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2982         osd_ipd_put(env, bag, ipd);
2983         LINVRNT(osd_invariant(obj));
2984         RETURN(rc);
2985 }
2986
2987 static int osd_index_declare_ea_delete(const struct lu_env *env,
2988                                        struct dt_object *dt,
2989                                        const struct dt_key *key,
2990                                        struct thandle *handle)
2991 {
2992         struct osd_thandle *oh;
2993         struct inode       *inode;
2994         int                 rc;
2995         ENTRY;
2996
2997         LASSERT(dt_object_exists(dt));
2998         LASSERT(handle != NULL);
2999
3000         oh = container_of0(handle, struct osd_thandle, ot_super);
3001         LASSERT(oh->ot_handle == NULL);
3002
3003         OSD_DECLARE_OP(oh, delete);
3004         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3005
3006         inode = osd_dt_obj(dt)->oo_inode;
3007         LASSERT(inode);
3008
3009         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3010                                    true, true, NULL, false);
3011         RETURN(rc);
3012 }
3013
3014 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3015                                           struct dt_rec *fid)
3016 {
3017         struct osd_fid_pack *rec;
3018         int                  rc = -ENODATA;
3019
3020         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3021                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3022                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3023         }
3024         RETURN(rc);
3025 }
3026
3027 /**
3028  * Index delete function for interoperability mode (b11826).
3029  * It will remove the directory entry added by osd_index_ea_insert().
3030  * This entry is needed to maintain name->fid mapping.
3031  *
3032  * \param key,  key i.e. file entry to be deleted
3033  *
3034  * \retval   0, on success
3035  * \retval -ve, on error
3036  */
3037 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3038                                const struct dt_key *key,
3039                                struct thandle *handle,
3040                                struct lustre_capa *capa)
3041 {
3042         struct osd_object          *obj    = osd_dt_obj(dt);
3043         struct inode               *dir    = obj->oo_inode;
3044         struct dentry              *dentry;
3045         struct osd_thandle         *oh;
3046         struct ldiskfs_dir_entry_2 *de;
3047         struct buffer_head         *bh;
3048         struct htree_lock          *hlock = NULL;
3049         int                         rc;
3050
3051         ENTRY;
3052
3053         LINVRNT(osd_invariant(obj));
3054         LASSERT(dt_object_exists(dt));
3055         LASSERT(handle != NULL);
3056
3057         OSD_EXEC_OP(handle, delete);
3058
3059         oh = container_of(handle, struct osd_thandle, ot_super);
3060         LASSERT(oh->ot_handle != NULL);
3061         LASSERT(oh->ot_handle->h_transaction != NULL);
3062
3063         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3064                 RETURN(-EACCES);
3065
3066         dentry = osd_child_dentry_get(env, obj,
3067                                       (char *)key, strlen((char *)key));
3068
3069         if (obj->oo_hl_head != NULL) {
3070                 hlock = osd_oti_get(env)->oti_hlock;
3071                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3072                                    dir, LDISKFS_HLOCK_DEL);
3073         } else {
3074                 cfs_down_write(&obj->oo_ext_idx_sem);
3075         }
3076
3077         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3078         if (bh) {
3079                 rc = ldiskfs_delete_entry(oh->ot_handle,
3080                                           dir, de, bh);
3081                 brelse(bh);
3082         } else {
3083                 rc = -ENOENT;
3084         }
3085         if (hlock != NULL)
3086                 ldiskfs_htree_unlock(hlock);
3087         else
3088                 cfs_up_write(&obj->oo_ext_idx_sem);
3089
3090         LASSERT(osd_invariant(obj));
3091         RETURN(rc);
3092 }
3093
3094 /**
3095  *      Lookup index for \a key and copy record to \a rec.
3096  *
3097  *      \param  dt      osd index object
3098  *      \param  key     key for index
3099  *      \param  rec     record reference
3100  *
3101  *      \retval  +ve  success : exact mach
3102  *      \retval  0    return record with key not greater than \a key
3103  *      \retval -ve   failure
3104  */
3105 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3106                                 struct dt_rec *rec, const struct dt_key *key,
3107                                 struct lustre_capa *capa)
3108 {
3109         struct osd_object      *obj = osd_dt_obj(dt);
3110         struct iam_path_descr  *ipd;
3111         struct iam_container   *bag = &obj->oo_dir->od_container;
3112         struct osd_thread_info *oti = osd_oti_get(env);
3113         struct iam_iterator    *it = &oti->oti_idx_it;
3114         struct iam_rec         *iam_rec;
3115         int                     rc;
3116
3117         ENTRY;
3118
3119         LASSERT(osd_invariant(obj));
3120         LASSERT(dt_object_exists(dt));
3121         LASSERT(bag->ic_object == obj->oo_inode);
3122
3123         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3124                 RETURN(-EACCES);
3125
3126         ipd = osd_idx_ipd_get(env, bag);
3127         if (IS_ERR(ipd))
3128                 RETURN(-ENOMEM);
3129
3130         /* got ipd now we can start iterator. */
3131         iam_it_init(it, bag, 0, ipd);
3132
3133         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3134                 /* swab quota uid/gid provided by caller */
3135                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3136                 key = (const struct dt_key *)&oti->oti_quota_id;
3137         }
3138
3139         rc = iam_it_get(it, (struct iam_key *)key);
3140         if (rc >= 0) {
3141                 if (S_ISDIR(obj->oo_inode->i_mode))
3142                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3143                 else
3144                         iam_rec = (struct iam_rec *) rec;
3145
3146                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3147
3148                 if (S_ISDIR(obj->oo_inode->i_mode))
3149                         osd_fid_unpack((struct lu_fid *) rec,
3150                                        (struct osd_fid_pack *)iam_rec);
3151                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3152                         osd_quota_unpack(obj, rec);
3153         }
3154
3155         iam_it_put(it);
3156         iam_it_fini(it);
3157         osd_ipd_put(env, bag, ipd);
3158
3159         LINVRNT(osd_invariant(obj));
3160
3161         RETURN(rc);
3162 }
3163
3164 static int osd_index_declare_iam_insert(const struct lu_env *env,
3165                                         struct dt_object *dt,
3166                                         const struct dt_rec *rec,
3167                                         const struct dt_key *key,
3168                                         struct thandle *handle)
3169 {
3170         struct osd_thandle *oh;
3171
3172         LASSERT(dt_object_exists(dt));
3173         LASSERT(handle != NULL);
3174
3175         oh = container_of0(handle, struct osd_thandle, ot_super);
3176         LASSERT(oh->ot_handle == NULL);
3177
3178         OSD_DECLARE_OP(oh, insert);
3179         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3180
3181         return 0;
3182 }
3183
3184 /**
3185  *      Inserts (key, value) pair in \a dt index object.
3186  *
3187  *      \param  dt      osd index object
3188  *      \param  key     key for index
3189  *      \param  rec     record reference
3190  *      \param  th      transaction handler
3191  *
3192  *      \retval  0  success
3193  *      \retval -ve failure
3194  */
3195 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3196                                 const struct dt_rec *rec,
3197                                 const struct dt_key *key, struct thandle *th,
3198                                 struct lustre_capa *capa, int ignore_quota)
3199 {
3200         struct osd_object     *obj = osd_dt_obj(dt);
3201         struct iam_path_descr *ipd;
3202         struct osd_thandle    *oh;
3203         struct iam_container  *bag = &obj->oo_dir->od_container;
3204 #ifdef HAVE_QUOTA_SUPPORT
3205         cfs_cap_t              save = cfs_curproc_cap_pack();
3206 #endif
3207         struct osd_thread_info *oti = osd_oti_get(env);
3208         struct iam_rec         *iam_rec;
3209         int                     rc;
3210
3211         ENTRY;
3212
3213         LINVRNT(osd_invariant(obj));
3214         LASSERT(dt_object_exists(dt));
3215         LASSERT(bag->ic_object == obj->oo_inode);
3216         LASSERT(th != NULL);
3217
3218         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3219                 RETURN(-EACCES);
3220
3221         OSD_EXEC_OP(th, insert);
3222
3223         ipd = osd_idx_ipd_get(env, bag);
3224         if (unlikely(ipd == NULL))
3225                 RETURN(-ENOMEM);
3226
3227         oh = container_of0(th, struct osd_thandle, ot_super);
3228         LASSERT(oh->ot_handle != NULL);
3229         LASSERT(oh->ot_handle->h_transaction != NULL);
3230 #ifdef HAVE_QUOTA_SUPPORT
3231         if (ignore_quota)
3232                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3233         else
3234                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3235 #endif
3236         if (S_ISDIR(obj->oo_inode->i_mode)) {
3237                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3238                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3239         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3240                 /* pack quota uid/gid */
3241                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3242                 key = (const struct dt_key *)&oti->oti_quota_id;
3243                 /* pack quota record */
3244                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3245                 iam_rec = (struct iam_rec *)rec;
3246         } else {
3247                 iam_rec = (struct iam_rec *)rec;
3248         }
3249
3250         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3251                         iam_rec, ipd);
3252 #ifdef HAVE_QUOTA_SUPPORT
3253         cfs_curproc_cap_unpack(save);
3254 #endif
3255         osd_ipd_put(env, bag, ipd);
3256         LINVRNT(osd_invariant(obj));
3257         RETURN(rc);
3258 }
3259
3260 /**
3261  * Calls ldiskfs_add_entry() to add directory entry
3262  * into the directory. This is required for
3263  * interoperability mode (b11826)
3264  *
3265  * \retval   0, on success
3266  * \retval -ve, on error
3267  */
3268 static int __osd_ea_add_rec(struct osd_thread_info *info,
3269                             struct osd_object *pobj, struct inode  *cinode,
3270                             const char *name, const struct dt_rec *fid,
3271                             struct htree_lock *hlock, struct thandle *th)
3272 {
3273         struct ldiskfs_dentry_param *ldp;
3274         struct dentry               *child;
3275         struct osd_thandle          *oth;
3276         int                          rc;
3277
3278         oth = container_of(th, struct osd_thandle, ot_super);
3279         LASSERT(oth->ot_handle != NULL);
3280         LASSERT(oth->ot_handle->h_transaction != NULL);
3281
3282         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3283
3284         /* XXX: remove fid_is_igif() check here.
3285          * IGIF check is just to handle insertion of .. when it is 'ROOT',
3286          * it is IGIF now but needs FID in dir entry as well for readdir
3287          * to work.
3288          * LU-838 should fix that and remove fid_is_igif() check */
3289         if (fid_is_igif((struct lu_fid *)fid) ||
3290             fid_is_norm((struct lu_fid *)fid)) {
3291                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3292                 osd_get_ldiskfs_dirent_param(ldp, fid);
3293                 child->d_fsdata = (void *)ldp;
3294         } else {
3295                 child->d_fsdata = NULL;
3296         }
3297         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3298
3299         RETURN(rc);
3300 }
3301
3302 /**
3303  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3304  * into the directory.Also sets flags into osd object to
3305  * indicate dot and dotdot are created. This is required for
3306  * interoperability mode (b11826)
3307  *
3308  * \param dir   directory for dot and dotdot fixup.
3309  * \param obj   child object for linking
3310  *
3311  * \retval   0, on success
3312  * \retval -ve, on error
3313  */
3314 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3315                               struct osd_object *dir,
3316                               struct inode  *parent_dir, const char *name,
3317                               const struct dt_rec *dot_fid,
3318                               const struct dt_rec *dot_dot_fid,
3319                               struct thandle *th)
3320 {
3321         struct inode                *inode = dir->oo_inode;
3322         struct ldiskfs_dentry_param *dot_ldp;
3323         struct ldiskfs_dentry_param *dot_dot_ldp;
3324         struct osd_thandle          *oth;
3325         int result = 0;
3326
3327         oth = container_of(th, struct osd_thandle, ot_super);
3328         LASSERT(oth->ot_handle->h_transaction != NULL);
3329         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3330
3331         if (strcmp(name, dot) == 0) {
3332                 if (dir->oo_compat_dot_created) {
3333                         result = -EEXIST;
3334                 } else {
3335                         LASSERT(inode == parent_dir);
3336                         dir->oo_compat_dot_created = 1;
3337                         result = 0;
3338                 }
3339         } else if(strcmp(name, dotdot) == 0) {
3340                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3341                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3342
3343                 if (!dir->oo_compat_dot_created)
3344                         return -EINVAL;
3345                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3346                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3347                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3348                 } else {
3349                         dot_ldp = NULL;
3350                         dot_dot_ldp = NULL;
3351                 }
3352                 /* in case of rename, dotdot is already created */
3353                 if (dir->oo_compat_dotdot_created) {
3354                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3355                                                 dot_dot_fid, NULL, th);
3356                 }
3357
3358                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3359                                                 inode, dot_ldp, dot_dot_ldp);
3360                 if (result == 0)
3361                        dir->oo_compat_dotdot_created = 1;
3362         }
3363
3364         return result;
3365 }
3366
3367
3368 /**
3369  * It will call the appropriate osd_add* function and return the
3370  * value, return by respective functions.
3371  */
3372 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3373                           struct inode *cinode, const char *name,
3374                           const struct dt_rec *fid, struct thandle *th)
3375 {
3376         struct osd_thread_info *info   = osd_oti_get(env);
3377         struct htree_lock      *hlock;
3378         int                     rc;
3379
3380         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3381
3382         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3383                                                    name[2] =='\0'))) {
3384                 if (hlock != NULL) {
3385                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3386                                            pobj->oo_inode, 0);
3387                 } else {
3388                         cfs_down_write(&pobj->oo_ext_idx_sem);
3389                 }
3390                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3391                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3392                                         fid, th);
3393         } else {
3394                 if (hlock != NULL) {
3395                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3396                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3397                 } else {
3398                         cfs_down_write(&pobj->oo_ext_idx_sem);
3399                 }
3400
3401                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3402                                       hlock, th);
3403         }
3404         if (hlock != NULL)
3405                 ldiskfs_htree_unlock(hlock);
3406         else
3407                 cfs_up_write(&pobj->oo_ext_idx_sem);
3408
3409         return rc;
3410 }
3411
3412 static int
3413 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3414                       struct osd_idmap_cache *oic)
3415 {
3416         struct osd_scrub    *scrub = &dev->od_scrub;
3417         struct lu_fid       *fid   = &oic->oic_fid;
3418         struct osd_inode_id *id    = &oti->oti_id;
3419         int                  once  = 0;
3420         int                  rc;
3421         ENTRY;
3422
3423 again:
3424         rc = osd_oi_lookup(oti, dev, fid, id);
3425         if (rc != 0 && rc != -ENOENT)
3426                 RETURN(rc);
3427
3428         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3429                 RETURN(0);
3430
3431         if (thread_is_running(&scrub->os_thread)) {
3432                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3433                 /* There is race condition between osd_oi_lookup and OI scrub.
3434                  * The OI scrub finished just after osd_oi_lookup() failure.
3435                  * Under such case, it is unnecessary to trigger OI scrub again,
3436                  * but try to call osd_oi_lookup() again. */
3437                 if (unlikely(rc == -EAGAIN))
3438                         goto again;
3439
3440                 RETURN(rc);
3441         }
3442
3443         if (!scrub->os_no_scrub && ++once == 1) {
3444                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3445                        PFID(fid));
3446                 rc = osd_scrub_start(dev);
3447                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3448                                ", rc = %d [2]\n",
3449                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3450                                PFID(fid), rc);
3451                 if (rc == 0)
3452                         goto again;
3453         }
3454
3455         RETURN(rc = -EREMCHG);
3456 }
3457
3458 /**
3459  * Calls ->lookup() to find dentry. From dentry get inode and
3460  * read inode's ea to get fid. This is required for  interoperability
3461  * mode (b11826)
3462  *
3463  * \retval   0, on success
3464  * \retval -ve, on error
3465  */
3466 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3467                              struct dt_rec *rec, const struct dt_key *key)
3468 {
3469         struct inode               *dir    = obj->oo_inode;
3470         struct dentry              *dentry;
3471         struct ldiskfs_dir_entry_2 *de;
3472         struct buffer_head         *bh;
3473         struct lu_fid              *fid = (struct lu_fid *) rec;
3474         struct htree_lock          *hlock = NULL;
3475         int                         ino;
3476         int                         rc;
3477
3478         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3479
3480         dentry = osd_child_dentry_get(env, obj,
3481                                       (char *)key, strlen((char *)key));
3482
3483         if (obj->oo_hl_head != NULL) {
3484                 hlock = osd_oti_get(env)->oti_hlock;
3485                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3486                                    dir, LDISKFS_HLOCK_LOOKUP);
3487         } else {
3488                 cfs_down_read(&obj->oo_ext_idx_sem);
3489         }
3490
3491         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3492         if (bh) {
3493                 struct osd_thread_info *oti = osd_oti_get(env);
3494                 struct osd_idmap_cache *oic = &oti->oti_cache;
3495                 struct osd_device *dev = osd_obj2dev(obj);
3496                 struct osd_scrub *scrub = &dev->od_scrub;
3497                 struct scrub_file *sf = &scrub->os_file;
3498
3499                 ino = le32_to_cpu(de->inode);
3500                 rc = osd_get_fid_from_dentry(de, rec);
3501
3502                 /* done with de, release bh */
3503                 brelse(bh);
3504                 if (rc != 0)
3505                         rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3506                 else
3507                         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3508                 if (rc != 0 || !fid_is_norm(fid)) {
3509                         fid_zero(&oic->oic_fid);
3510                         GOTO(out, rc);
3511                 }
3512
3513                 oic->oic_fid = *fid;
3514                 if ((scrub->os_pos_current <= ino) &&
3515                     (sf->sf_flags & SF_INCONSISTENT ||
3516                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3517                                       sf->sf_oi_bitmap)))
3518                         rc = osd_consistency_check(oti, dev, oic);
3519         } else {
3520                 rc = -ENOENT;
3521         }
3522
3523         GOTO(out, rc);
3524
3525 out:
3526         if (hlock != NULL)
3527                 ldiskfs_htree_unlock(hlock);
3528         else
3529                 cfs_up_read(&obj->oo_ext_idx_sem);
3530         return rc;
3531 }
3532
3533 /**
3534  * Find the osd object for given fid.
3535  *
3536  * \param fid need to find the osd object having this fid
3537  *
3538  * \retval osd_object on success
3539  * \retval        -ve on error
3540  */
3541 struct osd_object *osd_object_find(const struct lu_env *env,
3542                                    struct dt_object *dt,
3543                                    const struct lu_fid *fid)
3544 {
3545         struct lu_device  *ludev = dt->do_lu.lo_dev;
3546         struct osd_object *child = NULL;
3547         struct lu_object  *luch;
3548         struct lu_object  *lo;
3549
3550         /*
3551          * at this point topdev might not exist yet
3552          * (i.e. MGS is preparing profiles). so we can
3553          * not rely on topdev and instead lookup with
3554          * our device passed as topdev. this can't work
3555          * if the object isn't cached yet (as osd doesn't
3556          * allocate lu_header). IOW, the object must be
3557          * in the cache, otherwise lu_object_alloc() crashes
3558          * -bzzz
3559          */
3560         luch = lu_object_find_at(env, ludev, fid, NULL);
3561         if (!IS_ERR(luch)) {
3562                 if (lu_object_exists(luch)) {
3563                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3564                         if (lo != NULL)
3565                                 child = osd_obj(lo);
3566                         else
3567                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3568                                                 "lu_object can't be located"
3569                                                 DFID"\n", PFID(fid));
3570
3571                         if (child == NULL) {
3572                                 lu_object_put(env, luch);
3573                                 CERROR("Unable to get osd_object\n");
3574                                 child = ERR_PTR(-ENOENT);
3575                         }
3576                 } else {
3577                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3578                                         "lu_object does not exists "DFID"\n",
3579                                         PFID(fid));
3580                         lu_object_put(env, luch);
3581                         child = ERR_PTR(-ENOENT);
3582                 }
3583         } else
3584                 child = (void *)luch;
3585
3586         return child;
3587 }
3588
3589 /**
3590  * Put the osd object once done with it.
3591  *
3592  * \param obj osd object that needs to be put
3593  */
3594 static inline void osd_object_put(const struct lu_env *env,
3595                                   struct osd_object *obj)
3596 {
3597         lu_object_put(env, &obj->oo_dt.do_lu);
3598 }
3599
3600 static int osd_index_declare_ea_insert(const struct lu_env *env,
3601                                        struct dt_object *dt,
3602                                        const struct dt_rec *rec,
3603                                        const struct dt_key *key,
3604                                        struct thandle *handle)
3605 {
3606         struct osd_thandle *oh;
3607         struct inode       *inode;
3608         int                 rc;
3609         ENTRY;
3610
3611         LASSERT(dt_object_exists(dt));
3612         LASSERT(handle != NULL);
3613
3614         oh = container_of0(handle, struct osd_thandle, ot_super);
3615         LASSERT(oh->ot_handle == NULL);
3616
3617         OSD_DECLARE_OP(oh, insert);
3618         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3619
3620         inode = osd_dt_obj(dt)->oo_inode;
3621         LASSERT(inode);
3622
3623         /* We ignore block quota on meta pool (MDTs), so needn't
3624          * calculate how many blocks will be consumed by this index
3625          * insert */
3626         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3627                                    true, true, NULL, false);
3628         RETURN(rc);
3629 }
3630
3631 /**
3632  * Index add function for interoperability mode (b11826).
3633  * It will add the directory entry.This entry is needed to
3634  * maintain name->fid mapping.
3635  *
3636  * \param key it is key i.e. file entry to be inserted
3637  * \param rec it is value of given key i.e. fid
3638  *
3639  * \retval   0, on success
3640  * \retval -ve, on error
3641  */
3642 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3643                                const struct dt_rec *rec,
3644                                const struct dt_key *key, struct thandle *th,
3645                                struct lustre_capa *capa, int ignore_quota)
3646 {
3647         struct osd_object *obj   = osd_dt_obj(dt);
3648         struct lu_fid     *fid   = (struct lu_fid *) rec;
3649         const char        *name  = (const char *)key;
3650         struct osd_object *child;
3651 #ifdef HAVE_QUOTA_SUPPORT
3652         cfs_cap_t          save  = cfs_curproc_cap_pack();
3653 #endif
3654         int                rc;
3655
3656         ENTRY;
3657
3658         LASSERT(osd_invariant(obj));
3659         LASSERT(dt_object_exists(dt));
3660         LASSERT(th != NULL);
3661
3662         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3663                 RETURN(-EACCES);
3664
3665         child = osd_object_find(env, dt, fid);
3666         if (!IS_ERR(child)) {
3667 #ifdef HAVE_QUOTA_SUPPORT
3668                 if (ignore_quota)
3669                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3670                 else
3671                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3672 #endif
3673                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3674 #ifdef HAVE_QUOTA_SUPPORT
3675                 cfs_curproc_cap_unpack(save);
3676 #endif
3677                 osd_object_put(env, child);
3678         } else {
3679                 rc = PTR_ERR(child);
3680         }
3681
3682         LASSERT(osd_invariant(obj));
3683         RETURN(rc);
3684 }
3685
3686 /**
3687  *  Initialize osd Iterator for given osd index object.
3688  *
3689  *  \param  dt      osd index object
3690  */
3691
3692 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3693                                      struct dt_object *dt,
3694                                      __u32 unused,
3695                                      struct lustre_capa *capa)
3696 {
3697         struct osd_it_iam      *it;
3698         struct osd_thread_info *oti = osd_oti_get(env);
3699         struct osd_object      *obj = osd_dt_obj(dt);
3700         struct lu_object       *lo  = &dt->do_lu;
3701         struct iam_path_descr  *ipd;
3702         struct iam_container   *bag = &obj->oo_dir->od_container;
3703
3704         LASSERT(lu_object_exists(lo));
3705
3706         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3707                 return ERR_PTR(-EACCES);
3708
3709         it = &oti->oti_it;
3710         ipd = osd_it_ipd_get(env, bag);
3711         if (likely(ipd != NULL)) {
3712                 it->oi_obj = obj;
3713                 it->oi_ipd = ipd;
3714                 lu_object_get(lo);
3715                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3716                 return (struct dt_it *)it;
3717         }
3718         return ERR_PTR(-ENOMEM);
3719 }
3720
3721 /**
3722  * free given Iterator.
3723  */
3724
3725 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3726 {
3727         struct osd_it_iam *it = (struct osd_it_iam *)di;
3728         struct osd_object *obj = it->oi_obj;
3729
3730         iam_it_fini(&it->oi_it);
3731         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3732         lu_object_put(env, &obj->oo_dt.do_lu);
3733 }
3734
3735 /**
3736  *  Move Iterator to record specified by \a key
3737  *
3738  *  \param  di      osd iterator
3739  *  \param  key     key for index
3740  *
3741  *  \retval +ve  di points to record with least key not larger than key
3742  *  \retval  0   di points to exact matched key
3743  *  \retval -ve  failure
3744  */
3745
3746 static int osd_it_iam_get(const struct lu_env *env,
3747                           struct dt_it *di, const struct dt_key *key)
3748 {
3749         struct osd_thread_info  *oti = osd_oti_get(env);
3750         struct osd_it_iam       *it = (struct osd_it_iam *)di;
3751
3752         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3753                 /* swab quota uid/gid */
3754                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3755                 key = (struct dt_key *)&oti->oti_quota_id;
3756         }
3757
3758         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3759 }
3760
3761 /**
3762  *  Release Iterator
3763  *
3764  *  \param  di      osd iterator
3765  */
3766
3767 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3768 {
3769         struct osd_it_iam *it = (struct osd_it_iam *)di;
3770
3771         iam_it_put(&it->oi_it);
3772 }
3773
3774 /**
3775  *  Move iterator by one record
3776  *
3777  *  \param  di      osd iterator
3778  *
3779  *  \retval +1   end of container reached
3780  *  \retval  0   success
3781  *  \retval -ve  failure
3782  */
3783
3784 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3785 {
3786         struct osd_it_iam *it = (struct osd_it_iam *)di;
3787
3788         return iam_it_next(&it->oi_it);
3789 }
3790
3791 /**
3792  * Return pointer to the key under iterator.
3793  */
3794
3795 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3796                                  const struct dt_it *di)
3797 {
3798         struct osd_thread_info *oti = osd_oti_get(env);
3799         struct osd_it_iam      *it = (struct osd_it_iam *)di;
3800         struct osd_object      *obj = it->oi_obj;
3801         struct dt_key          *key;
3802
3803         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3804
3805         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3806                 /* swab quota uid/gid */
3807                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3808                 key = (struct dt_key *)&oti->oti_quota_id;
3809         }
3810
3811         return key;
3812 }
3813
3814 /**
3815  * Return size of key under iterator (in bytes)
3816  */
3817
3818 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3819 {
3820         struct osd_it_iam *it = (struct osd_it_iam *)di;
3821
3822         return iam_it_key_size(&it->oi_it);
3823 }
3824
3825 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3826                                        int len, __u16 type)
3827 {
3828         struct luda_type *lt;
3829         const unsigned    align = sizeof(struct luda_type) - 1;
3830
3831         /* check if file type is required */
3832         if (attr & LUDA_TYPE) {
3833                         len = (len + align) & ~align;
3834
3835                         lt = (void *) ent->lde_name + len;
3836                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3837                         ent->lde_attrs |= LUDA_TYPE;
3838         }
3839
3840         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3841 }
3842
3843 /**
3844  * build lu direct from backend fs dirent.
3845  */
3846
3847 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3848                                       struct lu_fid *fid, __u64 offset,
3849                                       char *name, __u16 namelen,
3850                                       __u16 type, __u32 attr)
3851 {
3852         fid_cpu_to_le(&ent->lde_fid, fid);
3853         ent->lde_attrs = LUDA_FID;
3854
3855         ent->lde_hash = cpu_to_le64(offset);
3856         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3857
3858         strncpy(ent->lde_name, name, namelen);
3859         ent->lde_namelen = cpu_to_le16(namelen);
3860
3861         /* append lustre attributes */
3862         osd_it_append_attrs(ent, attr, namelen, type);
3863 }
3864
3865 /**
3866  * Return pointer to the record under iterator.
3867  */
3868 static int osd_it_iam_rec(const struct lu_env *env,
3869                           const struct dt_it *di,
3870                           struct dt_rec *dtrec, __u32 attr)
3871 {
3872         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
3873         struct osd_thread_info *info = osd_oti_get(env);
3874         ENTRY;
3875
3876         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3877                 const struct osd_fid_pack *rec;
3878                 struct lu_fid             *fid = &info->oti_fid;
3879                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
3880                 char                      *name;
3881                 int                        namelen;
3882                 __u64                      hash;
3883                 int                        rc;
3884
3885                 name = (char *)iam_it_key_get(&it->oi_it);
3886                 if (IS_ERR(name))
3887                         RETURN(PTR_ERR(name));
3888
3889                 namelen = iam_it_key_size(&it->oi_it);
3890
3891                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3892                 if (IS_ERR(rec))
3893                         RETURN(PTR_ERR(rec));
3894
3895                 rc = osd_fid_unpack(fid, rec);
3896                 if (rc)
3897                         RETURN(rc);
3898
3899                 hash = iam_it_store(&it->oi_it);
3900
3901                 /* IAM does not store object type in IAM index (dir) */
3902                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3903                                    0, LUDA_FID);
3904         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3905                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3906                            (struct iam_rec *)dtrec);
3907                 osd_quota_unpack(it->oi_obj, dtrec);
3908         } else {
3909                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3910                            (struct iam_rec *)dtrec);
3911         }
3912
3913         RETURN(0);
3914 }
3915
3916 /**
3917  * Returns cookie for current Iterator position.
3918  */
3919 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3920 {
3921         struct osd_it_iam *it = (struct osd_it_iam *)di;
3922
3923         return iam_it_store(&it->oi_it);
3924 }
3925
3926 /**
3927  * Restore iterator from cookie.
3928  *
3929  * \param  di      osd iterator
3930  * \param  hash    Iterator location cookie
3931  *
3932  * \retval +ve  di points to record with least key not larger than key.
3933  * \retval  0   di points to exact matched key
3934  * \retval -ve  failure
3935  */
3936
3937 static int osd_it_iam_load(const struct lu_env *env,
3938                            const struct dt_it *di, __u64 hash)
3939 {
3940         struct osd_it_iam *it = (struct osd_it_iam *)di;
3941
3942         return iam_it_load(&it->oi_it, hash);
3943 }
3944
3945 static const struct dt_index_operations osd_index_iam_ops = {
3946         .dio_lookup         = osd_index_iam_lookup,
3947         .dio_declare_insert = osd_index_declare_iam_insert,
3948         .dio_insert         = osd_index_iam_insert,
3949         .dio_declare_delete = osd_index_declare_iam_delete,
3950         .dio_delete         = osd_index_iam_delete,
3951         .dio_it     = {
3952                 .init     = osd_it_iam_init,
3953                 .fini     = osd_it_iam_fini,
3954                 .get      = osd_it_iam_get,
3955                 .put      = osd_it_iam_put,
3956                 .next     = osd_it_iam_next,
3957                 .key      = osd_it_iam_key,
3958                 .key_size = osd_it_iam_key_size,
3959                 .rec      = osd_it_iam_rec,
3960                 .store    = osd_it_iam_store,
3961                 .load     = osd_it_iam_load
3962         }
3963 };
3964
3965 /**
3966  * Creates or initializes iterator context.
3967  *
3968  * \retval struct osd_it_ea, iterator structure on success
3969  *
3970  */
3971 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3972                                     struct dt_object *dt,
3973                                     __u32 attr,
3974                                     struct lustre_capa *capa)
3975 {
3976         struct osd_object       *obj  = osd_dt_obj(dt);
3977         struct osd_thread_info  *info = osd_oti_get(env);
3978         struct osd_it_ea        *it   = &info->oti_it_ea;
3979         struct lu_object        *lo   = &dt->do_lu;
3980         struct dentry           *obj_dentry = &info->oti_it_dentry;
3981         ENTRY;
3982         LASSERT(lu_object_exists(lo));
3983
3984         obj_dentry->d_inode = obj->oo_inode;
3985         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3986         obj_dentry->d_name.hash = 0;
3987
3988         it->oie_rd_dirent       = 0;
3989         it->oie_it_dirent       = 0;
3990         it->oie_dirent          = NULL;
3991         it->oie_buf             = info->oti_it_ea_buf;
3992         it->oie_obj             = obj;
3993         it->oie_file.f_pos      = 0;
3994         it->oie_file.f_dentry   = obj_dentry;
3995         if (attr & LUDA_64BITHASH)
3996                 it->oie_file.f_mode |= FMODE_64BITHASH;
3997         else
3998                 it->oie_file.f_mode |= FMODE_32BITHASH;
3999         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4000         it->oie_file.f_op         = obj->oo_inode->i_fop;
4001         it->oie_file.private_data = NULL;
4002         lu_object_get(lo);
4003         RETURN((struct dt_it *) it);
4004 }
4005
4006 /**
4007  * Destroy or finishes iterator context.
4008  *
4009  * \param di iterator structure to be destroyed
4010  */
4011 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4012 {
4013         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4014         struct osd_object    *obj  = it->oie_obj;
4015         struct inode       *inode  = obj->oo_inode;
4016
4017         ENTRY;
4018         it->oie_file.f_op->release(inode, &it->oie_file);
4019         lu_object_put(env, &obj->oo_dt.do_lu);
4020         EXIT;
4021 }
4022
4023 /**
4024  * It position the iterator at given key, so that next lookup continues from
4025  * that key Or it is similar to dio_it->load() but based on a key,
4026  * rather than file position.
4027  *
4028  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4029  * to the beginning.
4030  *
4031  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4032  */
4033 static int osd_it_ea_get(const struct lu_env *env,
4034                          struct dt_it *di, const struct dt_key *key)
4035 {
4036         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4037
4038         ENTRY;
4039         LASSERT(((const char *)key)[0] == '\0');
4040         it->oie_file.f_pos      = 0;
4041         it->oie_rd_dirent       = 0;
4042         it->oie_it_dirent       = 0;
4043         it->oie_dirent          = NULL;
4044
4045         RETURN(+1);
4046 }
4047
4048 /**
4049  * Does nothing
4050  */
4051 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4052 {
4053 }
4054
4055 /**
4056  * It is called internally by ->readdir(). It fills the
4057  * iterator's in-memory data structure with required
4058  * information i.e. name, namelen, rec_size etc.
4059  *
4060  * \param buf in which information to be filled in.
4061  * \param name name of the file in given dir
4062  *
4063  * \retval 0 on success
4064  * \retval 1 on buffer full
4065  */
4066 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4067                                loff_t offset, __u64 ino,
4068                                unsigned d_type)
4069 {
4070         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4071         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4072         struct lu_fid           *fid  = &ent->oied_fid;
4073         struct osd_fid_pack     *rec;
4074         ENTRY;
4075
4076         /* this should never happen */
4077         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4078                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4079                 RETURN(-EIO);
4080         }
4081
4082         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4083             OSD_IT_EA_BUFSIZE)
4084                 RETURN(1);
4085
4086         if (d_type & LDISKFS_DIRENT_LUFID) {
4087                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4088
4089                 if (osd_fid_unpack(fid, rec) != 0)
4090                         fid_zero(fid);
4091
4092                 d_type &= ~LDISKFS_DIRENT_LUFID;
4093         } else {
4094                 fid_zero(fid);
4095         }
4096
4097         ent->oied_ino     = ino;
4098         ent->oied_off     = offset;
4099         ent->oied_namelen = namelen;
4100         ent->oied_type    = d_type;
4101
4102         memcpy(ent->oied_name, name, namelen);
4103
4104         it->oie_rd_dirent++;
4105         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4106         RETURN(0);
4107 }
4108
4109 /**
4110  * Calls ->readdir() to load a directory entry at a time
4111  * and stored it in iterator's in-memory data structure.
4112  *
4113  * \param di iterator's in memory structure
4114  *
4115  * \retval   0 on success
4116  * \retval -ve on error
4117  */
4118 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4119                                const struct dt_it *di)
4120 {
4121         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4122         struct osd_object  *obj   = it->oie_obj;
4123         struct inode       *inode = obj->oo_inode;
4124         struct htree_lock  *hlock = NULL;
4125         int                 result = 0;
4126
4127         ENTRY;
4128         it->oie_dirent = it->oie_buf;
4129         it->oie_rd_dirent = 0;
4130
4131         if (obj->oo_hl_head != NULL) {
4132                 hlock = osd_oti_get(env)->oti_hlock;
4133                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4134                                    inode, LDISKFS_HLOCK_READDIR);
4135         } else {
4136                 cfs_down_read(&obj->oo_ext_idx_sem);
4137         }
4138
4139         result = inode->i_fop->readdir(&it->oie_file, it,
4140                                        (filldir_t) osd_ldiskfs_filldir);
4141
4142         if (hlock != NULL)
4143                 ldiskfs_htree_unlock(hlock);
4144         else
4145                 cfs_up_read(&obj->oo_ext_idx_sem);
4146
4147         if (it->oie_rd_dirent == 0) {
4148                 result = -EIO;
4149         } else {
4150                 it->oie_dirent = it->oie_buf;
4151                 it->oie_it_dirent = 1;
4152         }
4153
4154         RETURN(result);
4155 }
4156
4157 /**
4158  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4159  * to load a directory entry at a time and stored it in
4160  * iterator's in-memory data structure.
4161  *
4162  * \param di iterator's in memory structure
4163  *
4164  * \retval +ve iterator reached to end
4165  * \retval   0 iterator not reached to end
4166  * \retval -ve on error
4167  */
4168 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4169 {
4170         struct osd_it_ea *it = (struct osd_it_ea *)di;
4171         int rc;
4172
4173         ENTRY;
4174
4175         if (it->oie_it_dirent < it->oie_rd_dirent) {
4176                 it->oie_dirent =
4177                         (void *) it->oie_dirent +
4178                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4179                                        it->oie_dirent->oied_namelen);
4180                 it->oie_it_dirent++;
4181                 RETURN(0);
4182         } else {
4183                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4184                         rc = +1;
4185                 else
4186                         rc = osd_ldiskfs_it_fill(env, di);
4187         }
4188
4189         RETURN(rc);
4190 }
4191
4192 /**
4193  * Returns the key at current position from iterator's in memory structure.
4194  *
4195  * \param di iterator's in memory structure
4196  *
4197  * \retval key i.e. struct dt_key on success
4198  */
4199 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4200                                     const struct dt_it *di)
4201 {
4202         struct osd_it_ea *it = (struct osd_it_ea *)di;
4203
4204         return (struct dt_key *)it->oie_dirent->oied_name;
4205 }
4206
4207 /**
4208  * Returns the key's size at current position from iterator's in memory structure.
4209  *
4210  * \param di iterator's in memory structure
4211  *
4212  * \retval key_size i.e. struct dt_key on success
4213  */
4214 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4215 {
4216         struct osd_it_ea *it = (struct osd_it_ea *)di;
4217
4218         return it->oie_dirent->oied_namelen;
4219 }
4220
4221
4222 /**
4223  * Returns the value (i.e. fid/igif) at current position from iterator's
4224  * in memory structure.
4225  *
4226  * \param di struct osd_it_ea, iterator's in memory structure
4227  * \param attr attr requested for dirent.
4228  * \param lde lustre dirent
4229  *
4230  * \retval   0 no error and \param lde has correct lustre dirent.
4231  * \retval -ve on error
4232  */
4233 static inline int osd_it_ea_rec(const struct lu_env *env,
4234                                 const struct dt_it *di,
4235                                 struct dt_rec *dtrec, __u32 attr)
4236 {
4237         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4238         struct osd_object      *obj   = it->oie_obj;
4239         struct osd_device      *dev   = osd_obj2dev(obj);
4240         struct osd_scrub       *scrub = &dev->od_scrub;
4241         struct scrub_file      *sf    = &scrub->os_file;
4242         struct osd_thread_info *oti   = osd_oti_get(env);
4243         struct osd_idmap_cache *oic   = &oti->oti_cache;
4244         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4245         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4246         __u32                   ino   = it->oie_dirent->oied_ino;
4247         int                     rc    = 0;
4248         ENTRY;
4249
4250         if (!fid_is_sane(fid)) {
4251                 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4252                 if (rc != 0)
4253                         RETURN(rc);
4254         } else {
4255                 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4256         }
4257
4258         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4259                            it->oie_dirent->oied_name,
4260                            it->oie_dirent->oied_namelen,
4261                            it->oie_dirent->oied_type, attr);
4262         if (!fid_is_norm(fid)) {
4263                 fid_zero(&oic->oic_fid);
4264                 RETURN(0);
4265         }
4266
4267         oic->oic_fid = *fid;
4268         if ((scrub->os_pos_current <= ino) &&
4269             (sf->sf_flags & SF_INCONSISTENT ||
4270              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4271                 rc = osd_consistency_check(oti, dev, oic);
4272
4273         RETURN(rc);
4274 }
4275
4276 /**
4277  * Returns a cookie for current position of the iterator head, so that
4278  * user can use this cookie to load/start the iterator next time.
4279  *
4280  * \param di iterator's in memory structure
4281  *
4282  * \retval cookie for current position, on success
4283  */
4284 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4285 {
4286         struct osd_it_ea *it = (struct osd_it_ea *)di;
4287
4288         return it->oie_dirent->oied_off;
4289 }
4290
4291 /**
4292  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4293  * to load a directory entry at a time and stored it i inn,
4294  * in iterator's in-memory data structure.
4295  *
4296  * \param di struct osd_it_ea, iterator's in memory structure
4297  *
4298  * \retval +ve on success
4299  * \retval -ve on error
4300  */
4301 static int osd_it_ea_load(const struct lu_env *env,
4302                           const struct dt_it *di, __u64 hash)
4303 {
4304         struct osd_it_ea *it = (struct osd_it_ea *)di;
4305         int rc;
4306
4307         ENTRY;
4308         it->oie_file.f_pos = hash;
4309
4310         rc =  osd_ldiskfs_it_fill(env, di);
4311         if (rc == 0)
4312                 rc = +1;
4313
4314         RETURN(rc);
4315 }
4316
4317 /**
4318  * Index lookup function for interoperability mode (b11826).
4319  *
4320  * \param key,  key i.e. file name to be searched
4321  *
4322  * \retval +ve, on success
4323  * \retval -ve, on error
4324  */
4325 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4326                                struct dt_rec *rec, const struct dt_key *key,
4327                                struct lustre_capa *capa)
4328 {
4329         struct osd_object *obj = osd_dt_obj(dt);
4330         int rc = 0;
4331
4332         ENTRY;
4333
4334         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4335         LINVRNT(osd_invariant(obj));
4336
4337         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4338                 return -EACCES;
4339
4340         rc = osd_ea_lookup_rec(env, obj, rec, key);
4341
4342         if (rc == 0)
4343                 rc = +1;
4344         RETURN(rc);
4345 }
4346
4347 /**
4348  * Index and Iterator operations for interoperability
4349  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4350  */
4351 static const struct dt_index_operations osd_index_ea_ops = {
4352         .dio_lookup         = osd_index_ea_lookup,
4353         .dio_declare_insert = osd_index_declare_ea_insert,
4354         .dio_insert         = osd_index_ea_insert,
4355         .dio_declare_delete = osd_index_declare_ea_delete,
4356         .dio_delete         = osd_index_ea_delete,
4357         .dio_it     = {
4358                 .init     = osd_it_ea_init,
4359                 .fini     = osd_it_ea_fini,
4360                 .get      = osd_it_ea_get,
4361                 .put      = osd_it_ea_put,
4362                 .next     = osd_it_ea_next,
4363                 .key      = osd_it_ea_key,
4364                 .key_size = osd_it_ea_key_size,
4365                 .rec      = osd_it_ea_rec,
4366                 .store    = osd_it_ea_store,
4367                 .load     = osd_it_ea_load
4368         }
4369 };
4370
4371 static void *osd_key_init(const struct lu_context *ctx,
4372                           struct lu_context_key *key)
4373 {
4374         struct osd_thread_info *info;
4375
4376         OBD_ALLOC_PTR(info);
4377         if (info == NULL)
4378                 return ERR_PTR(-ENOMEM);
4379
4380         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4381         if (info->oti_it_ea_buf == NULL)
4382                 goto out_free_info;
4383
4384         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4385
4386         info->oti_hlock = ldiskfs_htree_lock_alloc();
4387         if (info->oti_hlock == NULL)
4388                 goto out_free_ea;
4389
4390         return info;
4391
4392  out_free_ea:
4393         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4394  out_free_info:
4395         OBD_FREE_PTR(info);
4396         return ERR_PTR(-ENOMEM);
4397 }
4398
4399 static void osd_key_fini(const struct lu_context *ctx,
4400                          struct lu_context_key *key, void* data)
4401 {
4402         struct osd_thread_info *info = data;
4403
4404         if (info->oti_hlock != NULL)
4405                 ldiskfs_htree_lock_free(info->oti_hlock);
4406         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4407         OBD_FREE_PTR(info);
4408 }
4409
4410 static void osd_key_exit(const struct lu_context *ctx,
4411                          struct lu_context_key *key, void *data)
4412 {
4413         struct osd_thread_info *info = data;
4414
4415         LASSERT(info->oti_r_locks == 0);
4416         LASSERT(info->oti_w_locks == 0);
4417         LASSERT(info->oti_txns    == 0);
4418 }
4419
4420 /* type constructor/destructor: osd_type_init, osd_type_fini */
4421 LU_TYPE_INIT_FINI(osd, &osd_key);
4422
4423 struct lu_context_key osd_key = {
4424         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4425         .lct_init = osd_key_init,
4426         .lct_fini = osd_key_fini,
4427         .lct_exit = osd_key_exit
4428 };
4429
4430
4431 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4432                            const char *name, struct lu_device *next)
4433 {
4434         struct osd_device *osd = osd_dev(d);
4435
4436         strncpy(osd->od_svname, name, MAX_OBD_NAME);
4437         return osd_procfs_init(osd, name);
4438 }
4439
4440 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4441 {
4442         ENTRY;
4443
4444         osd_scrub_cleanup(env, o);
4445
4446         if (o->od_fsops) {
4447                 fsfilt_put_ops(o->od_fsops);
4448                 o->od_fsops = NULL;
4449         }
4450
4451         /* shutdown quota slave instance associated with the device */
4452         if (o->od_quota_slave != NULL) {
4453                 qsd_fini(env, o->od_quota_slave);
4454                 o->od_quota_slave = NULL;
4455         }
4456
4457         RETURN(0);
4458 }
4459
4460 static int osd_mount(const struct lu_env *env,
4461                      struct osd_device *o, struct lustre_cfg *cfg)
4462 {
4463         const char              *name  = lustre_cfg_string(cfg, 0);
4464         const char              *dev  = lustre_cfg_string(cfg, 1);
4465         const char              *opts;
4466         unsigned long            page, s_flags, lmd_flags = 0;
4467         struct page             *__page;
4468         struct file_system_type *type;
4469         char                    *options = NULL;
4470         char                    *str;
4471         int                       rc = 0;
4472         ENTRY;
4473
4474         if (o->od_mnt != NULL)
4475                 RETURN(0);
4476
4477         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4478         if (o->od_fsops == NULL) {
4479                 CERROR("Can't find fsfilt_ldiskfs\n");
4480                 RETURN(-ENOTSUPP);
4481         }
4482
4483         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4484         if (__page == NULL)
4485                 GOTO(out, rc = -ENOMEM);
4486
4487         str = lustre_cfg_string(cfg, 2);
4488         s_flags = simple_strtoul(str, NULL, 0);
4489         str = strstr(str, ":");
4490         if (str)
4491                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4492         opts = lustre_cfg_string(cfg, 3);
4493         page = (unsigned long)cfs_page_address(__page);
4494         options = (char *)page;
4495         *options = '\0';
4496         if (opts == NULL)
4497                 strcat(options, "user_xattr,acl");
4498         else
4499                 strcat(options, opts);
4500
4501         /* Glom up mount options */
4502         if (*options != '\0')
4503                 strcat(options, ",");
4504         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4505
4506         type = get_fs_type("ldiskfs");
4507         if (!type) {
4508                 CERROR("%s: cannot find ldiskfs module\n", name);
4509                 GOTO(out, rc = -ENODEV);
4510         }
4511
4512         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4513         cfs_module_put(type->owner);
4514
4515         if (IS_ERR(o->od_mnt)) {
4516                 rc = PTR_ERR(o->od_mnt);
4517                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4518                 o->od_mnt = NULL;
4519                 GOTO(out, rc);
4520         }
4521
4522         if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4523                 CERROR("%s: underlying device %s is marked as read-only. "
4524                        "Setup failed\n", name, dev);
4525                 mntput(o->od_mnt);
4526                 o->od_mnt = NULL;
4527                 GOTO(out, rc = -EROFS);
4528         }
4529
4530         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4531             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4532                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4533                 mntput(o->od_mnt);
4534                 o->od_mnt = NULL;
4535                 GOTO(out, rc = -EINVAL);
4536         }
4537
4538         if (lmd_flags & LMD_FLG_IAM) {
4539                 o->od_iop_mode = 0;
4540                 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name);
4541         } else
4542                 o->od_iop_mode = 1;
4543         if (lmd_flags & LMD_FLG_NOSCRUB)
4544                 o->od_scrub.os_no_scrub = 1;
4545
4546 out:
4547         if (__page)
4548                 OBD_PAGE_FREE(__page);
4549         if (rc)
4550                 fsfilt_put_ops(o->od_fsops);
4551
4552         RETURN(rc);
4553 }
4554
4555 static struct lu_device *osd_device_fini(const struct lu_env *env,
4556                                          struct lu_device *d)
4557 {
4558         int rc;
4559         ENTRY;
4560
4561         rc = osd_shutdown(env, osd_dev(d));
4562
4563         osd_compat_fini(osd_dev(d));
4564
4565         shrink_dcache_sb(osd_sb(osd_dev(d)));
4566         osd_sync(env, lu2dt_dev(d));
4567
4568         rc = osd_procfs_fini(osd_dev(d));
4569         if (rc) {
4570                 CERROR("proc fini error %d \n", rc);
4571                 RETURN (ERR_PTR(rc));
4572         }
4573
4574         if (osd_dev(d)->od_mnt) {
4575                 mntput(osd_dev(d)->od_mnt);
4576                 osd_dev(d)->od_mnt = NULL;
4577         }
4578
4579         RETURN(NULL);
4580 }
4581
4582 static int osd_device_init0(const struct lu_env *env,
4583                             struct osd_device *o,
4584                             struct lustre_cfg *cfg)
4585 {
4586         struct lu_device        *l = osd2lu_dev(o);
4587         struct osd_thread_info *info;
4588         int                     rc;
4589
4590         /* if the module was re-loaded, env can loose its keys */
4591         rc = lu_env_refill((struct lu_env *) env);
4592         if (rc)
4593                 GOTO(out, rc);
4594         info = osd_oti_get(env);
4595         LASSERT(info);
4596
4597         l->ld_ops = &osd_lu_ops;
4598         o->od_dt_dev.dd_ops = &osd_dt_ops;
4599
4600         cfs_spin_lock_init(&o->od_osfs_lock);
4601         cfs_mutex_init(&o->od_otable_mutex);
4602         o->od_osfs_age = cfs_time_shift_64(-1000);
4603
4604         o->od_capa_hash = init_capa_hash();
4605         if (o->od_capa_hash == NULL)
4606                 GOTO(out, rc = -ENOMEM);
4607
4608         o->od_read_cache = 1;
4609         o->od_writethrough_cache = 1;
4610
4611         rc = osd_mount(env, o, cfg);
4612         if (rc)
4613                 GOTO(out_capa, rc);
4614
4615         /* setup scrub, including OI files initialization */
4616         rc = osd_scrub_setup(env, o);
4617         if (rc < 0)
4618                 GOTO(out_mnt, rc);
4619
4620         strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4621                         sizeof(o->od_svname) - 1);
4622
4623         if (strstr(o->od_svname, "-OST")) {
4624                 rc = osd_compat_init(o);
4625                 if (rc != 0)
4626                         GOTO(out_mnt, rc);
4627         }
4628
4629         rc = lu_site_init(&o->od_site, l);
4630         if (rc)
4631                 GOTO(out_compat, rc);
4632
4633         rc = lu_site_init_finish(&o->od_site);
4634         if (rc)
4635                 GOTO(out_compat, rc);
4636
4637         rc = osd_procfs_init(o, o->od_svname);
4638         if (rc != 0) {
4639                 CERROR("%s: can't initialize procfs: rc = %d\n",
4640                        o->od_svname, rc);
4641                 GOTO(out_compat, rc);
4642         }
4643
4644         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4645
4646         RETURN(0);
4647 out_compat:
4648         osd_compat_fini(o);
4649 out_mnt:
4650         osd_oi_fini(info, o);
4651         osd_shutdown(env, o);
4652         mntput(o->od_mnt);
4653         o->od_mnt = NULL;
4654 out_capa:
4655         cleanup_capa_hash(o->od_capa_hash);
4656 out:
4657         RETURN(rc);
4658 }
4659
4660 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4661                                           struct lu_device_type *t,
4662                                           struct lustre_cfg *cfg)
4663 {
4664         struct osd_device *o;
4665         int                rc;
4666
4667         OBD_ALLOC_PTR(o);
4668         if (o == NULL)
4669                 return ERR_PTR(-ENOMEM);
4670
4671         rc = dt_device_init(&o->od_dt_dev, t);
4672         if (rc == 0) {
4673                 rc = osd_device_init0(env, o, cfg);
4674                 if (rc)
4675                         dt_device_fini(&o->od_dt_dev);
4676         }
4677
4678         if (unlikely(rc != 0))
4679                 OBD_FREE_PTR(o);
4680
4681         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4682 }
4683
4684 static struct lu_device *osd_device_free(const struct lu_env *env,
4685                                          struct lu_device *d)
4686 {
4687         struct osd_device *o = osd_dev(d);
4688         ENTRY;
4689
4690         cleanup_capa_hash(o->od_capa_hash);
4691         /* XXX: make osd top device in order to release reference */
4692         d->ld_site->ls_top_dev = d;
4693         lu_site_purge(env, d->ld_site, -1);
4694         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4695                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4696                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4697         }
4698         lu_site_fini(&o->od_site);
4699         dt_device_fini(&o->od_dt_dev);
4700         OBD_FREE_PTR(o);
4701         RETURN(NULL);
4702 }
4703
4704 static int osd_process_config(const struct lu_env *env,
4705                               struct lu_device *d, struct lustre_cfg *cfg)
4706 {
4707         struct osd_device *o = osd_dev(d);
4708         int err;
4709         ENTRY;
4710
4711         switch(cfg->lcfg_command) {
4712         case LCFG_SETUP:
4713                 err = osd_mount(env, o, cfg);
4714                 break;
4715         case LCFG_CLEANUP:
4716                 lu_dev_del_linkage(d->ld_site, d);
4717                 err = osd_shutdown(env, o);
4718                 break;
4719         default:
4720                 err = -ENOSYS;
4721         }
4722
4723         RETURN(err);
4724 }
4725
4726 static int osd_recovery_complete(const struct lu_env *env,
4727                                  struct lu_device *d)
4728 {
4729         RETURN(0);
4730 }
4731
4732 /*
4733  * we use exports to track all osd users
4734  */
4735 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4736                            struct obd_device *obd, struct obd_uuid *cluuid,
4737                            struct obd_connect_data *data, void *localdata)
4738 {
4739         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
4740         struct lustre_handle  conn;
4741         int                   rc;
4742         ENTRY;
4743
4744         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4745
4746         rc = class_connect(&conn, obd, cluuid);
4747         if (rc)
4748                 RETURN(rc);
4749
4750         *exp = class_conn2export(&conn);
4751
4752         cfs_spin_lock(&osd->od_osfs_lock);
4753         osd->od_connects++;
4754         cfs_spin_unlock(&osd->od_osfs_lock);
4755
4756         RETURN(0);
4757 }
4758
4759 /*
4760  * once last export (we don't count self-export) disappeared
4761  * osd can be released
4762  */
4763 static int osd_obd_disconnect(struct obd_export *exp)
4764 {
4765         struct obd_device *obd = exp->exp_obd;
4766         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4767         int                rc, release = 0;
4768         ENTRY;
4769
4770         /* Only disconnect the underlying layers on the final disconnect. */
4771         cfs_spin_lock(&osd->od_osfs_lock);
4772         osd->od_connects--;
4773         if (osd->od_connects == 0)
4774                 release = 1;
4775         cfs_spin_unlock(&osd->od_osfs_lock);
4776
4777         rc = class_disconnect(exp); /* bz 9811 */
4778
4779         if (rc == 0 && release)
4780                 class_manual_cleanup(obd);
4781         RETURN(rc);
4782 }
4783
4784 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4785                        struct lu_device *dev)
4786 {
4787         struct osd_device *osd = osd_dev(dev);
4788         int                result = 0;
4789         ENTRY;
4790
4791 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 55, 0)
4792         /* Unfortunately, the current MDD implementation relies on some specific
4793          * code to be executed in the OSD layer. Since OFD now also uses the OSD
4794          * module, we need a way to skip the metadata-specific code when running
4795          * with OFD.
4796          * The hack here is to check the type of the parent device which is
4797          * either MD (i.e. MDD device) with the current MDT stack or DT (i.e.
4798          * OFD device) on an OST. As a reminder, obdfilter does not use the OSD
4799          * layer and still relies on lvfs. This hack won't work any more when
4800          * LOD is landed since LOD is of DT type.
4801          * This code should be removed once the orion MDT changes (LOD/OSP, ...)
4802          * have been landed */
4803         osd->od_is_md = lu_device_is_md(pdev);
4804 #else
4805 #warning "all is_md checks must be removed from osd-ldiskfs"
4806 #endif
4807
4808         if (osd->od_is_md) {
4809                 /* 1. setup local objects */
4810                 result = llo_local_objects_setup(env, lu2md_dev(pdev),
4811                                                  lu2dt_dev(dev));
4812                 if (result)
4813                         RETURN(result);
4814         }
4815
4816         /* 2. setup quota slave instance */
4817         osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
4818                                        osd->od_proc_entry);
4819         if (IS_ERR(osd->od_quota_slave)) {
4820                 result = PTR_ERR(osd->od_quota_slave);
4821                 osd->od_quota_slave = NULL;
4822         }
4823
4824         RETURN(result);
4825 }
4826
4827 static const struct lu_object_operations osd_lu_obj_ops = {
4828         .loo_object_init      = osd_object_init,
4829         .loo_object_delete    = osd_object_delete,
4830         .loo_object_release   = osd_object_release,
4831         .loo_object_free      = osd_object_free,
4832         .loo_object_print     = osd_object_print,
4833         .loo_object_invariant = osd_object_invariant
4834 };
4835
4836 const struct lu_device_operations osd_lu_ops = {
4837         .ldo_object_alloc      = osd_object_alloc,
4838         .ldo_process_config    = osd_process_config,
4839         .ldo_recovery_complete = osd_recovery_complete,
4840         .ldo_prepare           = osd_prepare,
4841 };
4842
4843 static const struct lu_device_type_operations osd_device_type_ops = {
4844         .ldto_init = osd_type_init,
4845         .ldto_fini = osd_type_fini,
4846
4847         .ldto_start = osd_type_start,
4848         .ldto_stop  = osd_type_stop,
4849
4850         .ldto_device_alloc = osd_device_alloc,
4851         .ldto_device_free  = osd_device_free,
4852
4853         .ldto_device_init    = osd_device_init,
4854         .ldto_device_fini    = osd_device_fini
4855 };
4856
4857 struct lu_device_type osd_device_type = {
4858         .ldt_tags     = LU_DEVICE_DT,
4859         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
4860         .ldt_ops      = &osd_device_type_ops,
4861         .ldt_ctx_tags = LCT_LOCAL,
4862 };
4863
4864 /*
4865  * lprocfs legacy support.
4866  */
4867 static struct obd_ops osd_obd_device_ops = {
4868         .o_owner = THIS_MODULE,
4869         .o_connect      = osd_obd_connect,
4870         .o_disconnect   = osd_obd_disconnect
4871 };
4872
4873 static int __init osd_mod_init(void)
4874 {
4875         struct lprocfs_static_vars lvars;
4876
4877         osd_oi_mod_init();
4878         lprocfs_osd_init_vars(&lvars);
4879         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4880                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4881 }
4882
4883 static void __exit osd_mod_exit(void)
4884 {
4885         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4886 }
4887
4888 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4889 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4890 MODULE_LICENSE("GPL");
4891
4892 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);