lustre/osd/osd_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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  */
  42
  43 #ifndef EXPORT_SYMTAB
  44 # define EXPORT_SYMTAB
  45 #endif
  46 #define DEBUG_SUBSYSTEM S_MDS
  47
  48 #include <linux/module.h>
  49
  50 /* LUSTRE_VERSION_CODE */
  51 #include <lustre_ver.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/types.h>
  54 /* prerequisite for linux/xattr.h */
  55 #include <linux/fs.h>
  56 /* XATTR_{REPLACE,CREATE} */
  57 #include <linux/xattr.h>
  58 /* simple_mkdir() */
  59 #include <lvfs.h>
  60
  61 /*
  62  * struct OBD_{ALLOC,FREE}*()
  63  * OBD_FAIL_CHECK
  64  */
  65 #include <obd_support.h>
  66 /* struct ptlrpc_thread */
  67 #include <lustre_net.h>
  68
  69 /* fid_is_local() */
  70 #include <lustre_fid.h>
  71
  72 #include "osd_internal.h"
  73 #include "osd_igif.h"
  74
  75 /* llo_* api support */
  76 #include <md_object.h>
  77
  78 static const char dot[] = ".";
  79 static const char dotdot[] = "..";
  80 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  81
  82 struct osd_directory {
  83         struct iam_container od_container;
  84         struct iam_descr     od_descr;
  85 };
  86
  87 struct osd_object {
  88         struct dt_object       oo_dt;
  89         /**
  90          * Inode for file system object represented by this osd_object. This
  91          * inode is pinned for the whole duration of lu_object life.
  92          *
  93          * Not modified concurrently (either setup early during object
  94          * creation, or assigned by osd_object_create() under write lock).
  95          */
  96         struct inode          *oo_inode;
  97         /**
  98          * to protect index ops.
  99          */
 100         struct rw_semaphore    oo_ext_idx_sem;
 101         struct rw_semaphore    oo_sem;
 102         struct osd_directory  *oo_dir;
 103         /** protects inode attributes. */
 104         spinlock_t             oo_guard;
 105         /**
 106          * Following two members are used to indicate the presence of dot and
 107          * dotdot in the given directory. This is required for interop mode
 108          * (b11826).
 109          */
 110         int oo_compat_dot_created;
 111         int oo_compat_dotdot_created;
 112
 113         const struct lu_env   *oo_owner;
 114 #ifdef CONFIG_LOCKDEP
 115         struct lockdep_map     oo_dep_map;
 116 #endif
 117 };
 118
 119 static const struct lu_object_operations      osd_lu_obj_ops;
 120 static const struct lu_device_operations      osd_lu_ops;
 121 static       struct lu_context_key            osd_key;
 122 static const struct dt_object_operations      osd_obj_ops;
 123 static const struct dt_object_operations      osd_obj_ea_ops;
 124 static const struct dt_body_operations        osd_body_ops;
 125 static const struct dt_index_operations       osd_index_iam_ops;
 126 static const struct dt_index_operations       osd_index_ea_ops;
 127
 128 struct osd_thandle {
 129         struct thandle          ot_super;
 130         handle_t               *ot_handle;
 131         struct journal_callback ot_jcb;
 132         /* Link to the device, for debugging. */
 133         struct lu_ref_link     *ot_dev_link;
 134
 135 };
 136
 137 /*
 138  * Helpers.
 139  */
 140 static int lu_device_is_osd(const struct lu_device *d)
 141 {
 142         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
 143 }
 144
 145 static struct osd_device *osd_dt_dev(const struct dt_device *d)
 146 {
 147         LASSERT(lu_device_is_osd(&d->dd_lu_dev));
 148         return container_of0(d, struct osd_device, od_dt_dev);
 149 }
 150
 151 static struct osd_device *osd_dev(const struct lu_device *d)
 152 {
 153         LASSERT(lu_device_is_osd(d));
 154         return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
 155 }
 156
 157 static struct osd_device *osd_obj2dev(const struct osd_object *o)
 158 {
 159         return osd_dev(o->oo_dt.do_lu.lo_dev);
 160 }
 161
 162 static struct super_block *osd_sb(const struct osd_device *dev)
 163 {
 164         return dev->od_mount->lmi_mnt->mnt_sb;
 165 }
 166
 167 static int osd_object_is_root(const struct osd_object *obj)
 168 {
 169         return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
 170 }
 171
 172 static struct osd_object *osd_obj(const struct lu_object *o)
 173 {
 174         LASSERT(lu_device_is_osd(o->lo_dev));
 175         return container_of0(o, struct osd_object, oo_dt.do_lu);
 176 }
 177
 178 static struct osd_object *osd_dt_obj(const struct dt_object *d)
 179 {
 180         return osd_obj(&d->do_lu);
 181 }
 182
 183 static struct lu_device *osd2lu_dev(struct osd_device *osd)
 184 {
 185         return &osd->od_dt_dev.dd_lu_dev;
 186 }
 187
 188 static journal_t *osd_journal(const struct osd_device *dev)
 189 {
 190         return LDISKFS_SB(osd_sb(dev))->s_journal;
 191 }
 192
 193 static int osd_has_index(const struct osd_object *obj)
 194 {
 195         return obj->oo_dt.do_index_ops != NULL;
 196 }
 197
 198 static int osd_object_invariant(const struct lu_object *l)
 199 {
 200         return osd_invariant(osd_obj(l));
 201 }
 202
 203 #ifdef HAVE_QUOTA_SUPPORT
 204 static inline void
 205 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
 206 {
 207         struct md_ucred    *uc = md_ucred(env);
 208
 209         LASSERT(uc != NULL);
 210
 211         save->oc_uid = current->fsuid;
 212         save->oc_gid = current->fsgid;
 213         save->oc_cap = current->cap_effective;
 214         current->fsuid         = uc->mu_fsuid;
 215         current->fsgid         = uc->mu_fsgid;
 216         current->cap_effective = uc->mu_cap;
 217 }
 218
 219 static inline void
 220 osd_pop_ctxt(struct osd_ctxt *save)
 221 {
 222         current->fsuid         = save->oc_uid;
 223         current->fsgid         = save->oc_gid;
 224         current->cap_effective = save->oc_cap;
 225 }
 226 #endif
 227
 228 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
 229 {
 230         return lu_context_key_get(&env->le_ctx, &osd_key);
 231 }
 232
 233 /*
 234  * Concurrency: doesn't matter
 235  */
 236 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 237 {
 238         return osd_oti_get(env)->oti_r_locks > 0;
 239 }
 240
 241 /*
 242  * Concurrency: doesn't matter
 243  */
 244 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 245 {
 246         struct osd_thread_info *oti = osd_oti_get(env);
 247         return oti->oti_w_locks > 0 && o->oo_owner == env;
 248 }
 249
 250 /*
 251  * Concurrency: doesn't access mutable data
 252  */
 253 static int osd_root_get(const struct lu_env *env,
 254                         struct dt_device *dev, struct lu_fid *f)
 255 {
 256         struct inode *inode;
 257
 258         inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
 259         lu_igif_build(f, inode->i_ino, inode->i_generation);
 260         return 0;
 261 }
 262
 263 /*
 264  * OSD object methods.
 265  */
 266
 267 /*
 268  * Concurrency: no concurrent access is possible that early in object
 269  * life-cycle.
 270  */
 271 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 272                                           const struct lu_object_header *hdr,
 273                                           struct lu_device *d)
 274 {
 275         struct osd_object *mo;
 276
 277         OBD_ALLOC_PTR(mo);
 278         if (mo != NULL) {
 279                 struct lu_object *l;
 280
 281                 l = &mo->oo_dt.do_lu;
 282                 dt_object_init(&mo->oo_dt, NULL, d);
 283                 if (osd_dev(d)->od_iop_mode)
 284                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 285                 else
 286                         mo->oo_dt.do_ops = &osd_obj_ops;
 287
 288                 l->lo_ops = &osd_lu_obj_ops;
 289                 init_rwsem(&mo->oo_sem);
 290                 init_rwsem(&mo->oo_ext_idx_sem);
 291                 spin_lock_init(&mo->oo_guard);
 292                 return l;
 293         } else
 294                 return NULL;
 295 }
 296
 297 /*
 298  * retrieve object from backend ext fs.
 299  **/
 300 static struct inode *osd_iget(struct osd_thread_info *info,
 301                               struct osd_device *dev,
 302                               const struct osd_inode_id *id)
 303 {
 304         struct inode *inode = NULL;
 305
 306 #ifdef HAVE_EXT4_LDISKFS
 307         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 308         if (IS_ERR(inode))
 309         /* Newer kernels return an error instead of a NULL pointer */
 310                 inode = NULL;
 311 #else
 312         inode = iget(osd_sb(dev), id->oii_ino);
 313 #endif
 314         if (inode == NULL) {
 315                 CERROR("no inode\n");
 316                 inode = ERR_PTR(-EACCES);
 317         } else if (id->oii_gen != OSD_OII_NOGEN &&
 318                    inode->i_generation != id->oii_gen) {
 319                 iput(inode);
 320                 inode = ERR_PTR(-ESTALE);
 321         } else if (inode->i_nlink == 0) {
 322                 /* due to parallel readdir and unlink,
 323                 * we can have dead inode here. */
 324                 CWARN("stale inode\n");
 325                 make_bad_inode(inode);
 326                 iput(inode);
 327                 inode = ERR_PTR(-ESTALE);
 328         } else if (is_bad_inode(inode)) {
 329                 CERROR("bad inode %lx\n",inode->i_ino);
 330                 iput(inode);
 331                 inode = ERR_PTR(-ENOENT);
 332         }
 333         return inode;
 334 }
 335
 336 static int osd_fid_lookup(const struct lu_env *env,
 337                           struct osd_object *obj, const struct lu_fid *fid)
 338 {
 339         struct osd_thread_info *info;
 340         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
 341         struct osd_device      *dev;
 342         struct osd_inode_id    *id;
 343         struct osd_oi          *oi;
 344         struct inode           *inode;
 345         int                     result;
 346
 347         LINVRNT(osd_invariant(obj));
 348         LASSERT(obj->oo_inode == NULL);
 349         LASSERT(fid_is_sane(fid));
 350         /*
 351          * This assertion checks that osd layer sees only local
 352          * fids. Unfortunately it is somewhat expensive (does a
 353          * cache-lookup). Disabling it for production/acceptance-testing.
 354          */
 355         LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
 356
 357         ENTRY;
 358
 359         info = osd_oti_get(env);
 360         dev  = osd_dev(ldev);
 361         id   = &info->oti_id;
 362         oi   = &dev->od_oi;
 363
 364         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 365                 RETURN(-ENOENT);
 366
 367         result = osd_oi_lookup(info, oi, fid, id);
 368         if (result == 0) {
 369                 inode = osd_iget(info, dev, id);
 370                 if (!IS_ERR(inode)) {
 371                         obj->oo_inode = inode;
 372                         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 373                         if (dev->od_iop_mode) {
 374                                 obj->oo_compat_dot_created = 1;
 375                                 obj->oo_compat_dotdot_created = 1;
 376                         }
 377                         result = 0;
 378                 } else
 379                         /*
 380                          * If fid wasn't found in oi, inode-less object is
 381                          * created, for which lu_object_exists() returns
 382                          * false. This is used in a (frequent) case when
 383                          * objects are created as locking anchors or
 384                          * place holders for objects yet to be created.
 385                          */
 386                         result = PTR_ERR(inode);
 387         } else if (result == -ENOENT)
 388                 result = 0;
 389         LINVRNT(osd_invariant(obj));
 390
 391         RETURN(result);
 392 }
 393
 394 /*
 395  * Concurrency: shouldn't matter.
 396  */
 397 static void osd_object_init0(struct osd_object *obj)
 398 {
 399         LASSERT(obj->oo_inode != NULL);
 400         obj->oo_dt.do_body_ops = &osd_body_ops;
 401         obj->oo_dt.do_lu.lo_header->loh_attr |=
 402                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 403 }
 404
 405 /*
 406  * Concurrency: no concurrent access is possible that early in object
 407  * life-cycle.
 408  */
 409 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 410                            const struct lu_object_conf *unused)
 411 {
 412         struct osd_object *obj = osd_obj(l);
 413         int result;
 414
 415         LINVRNT(osd_invariant(obj));
 416
 417         result = osd_fid_lookup(env, obj, lu_object_fid(l));
 418         if (result == 0) {
 419                 if (obj->oo_inode != NULL)
 420                         osd_object_init0(obj);
 421         }
 422         LINVRNT(osd_invariant(obj));
 423         return result;
 424 }
 425
 426 /*
 427  * Concurrency: no concurrent access is possible that late in object
 428  * life-cycle.
 429  */
 430 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 431 {
 432         struct osd_object *obj = osd_obj(l);
 433
 434         LINVRNT(osd_invariant(obj));
 435
 436         dt_object_fini(&obj->oo_dt);
 437         OBD_FREE_PTR(obj);
 438 }
 439
 440 /**
 441  * IAM Iterator
 442  */
 443 static struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
 444                                              const struct iam_container *bag)
 445 {
 446         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 447                                            osd_oti_get(env)->oti_it_ipd);
 448 }
 449
 450 static struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
 451                                               const struct iam_container *bag)
 452 {
 453         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 454                                            osd_oti_get(env)->oti_idx_ipd);
 455 }
 456
 457 static void osd_ipd_put(const struct lu_env *env,
 458                         const struct iam_container *bag,
 459                         struct iam_path_descr *ipd)
 460 {
 461         bag->ic_descr->id_ops->id_ipd_free(ipd);
 462 }
 463
 464 /*
 465  * Concurrency: no concurrent access is possible that late in object
 466  * life-cycle.
 467  */
 468 static void osd_index_fini(struct osd_object *o)
 469 {
 470         struct iam_container *bag;
 471
 472         if (o->oo_dir != NULL) {
 473                 bag = &o->oo_dir->od_container;
 474                 if (o->oo_inode != NULL) {
 475                         if (bag->ic_object == o->oo_inode)
 476                                 iam_container_fini(bag);
 477                 }
 478                 OBD_FREE_PTR(o->oo_dir);
 479                 o->oo_dir = NULL;
 480         }
 481 }
 482
 483 /*
 484  * Concurrency: no concurrent access is possible that late in object
 485  * life-cycle (for all existing callers, that is. New callers have to provide
 486  * their own locking.)
 487  */
 488 static int osd_inode_unlinked(const struct inode *inode)
 489 {
 490         return inode->i_nlink == 0;
 491 }
 492
 493 enum {
 494         OSD_TXN_OI_DELETE_CREDITS    = 20,
 495         OSD_TXN_INODE_DELETE_CREDITS = 20
 496 };
 497
 498 /*
 499  * Journal
 500  */
 501
 502 /*
 503  * Concurrency: doesn't access mutable data.
 504  */
 505 static int osd_param_is_sane(const struct osd_device *dev,
 506                              const struct txn_param *param)
 507 {
 508         return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
 509 }
 510
 511 /*
 512  * Concurrency: shouldn't matter.
 513  */
 514 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 515 {
 516         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 517         struct thandle     *th  = &oh->ot_super;
 518         struct dt_device   *dev = th->th_dev;
 519         struct lu_device   *lud = &dev->dd_lu_dev;
 520
 521         LASSERT(dev != NULL);
 522         LASSERT(oh->ot_handle == NULL);
 523
 524         if (error) {
 525                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 526         } else {
 527                 struct lu_env *env = &osd_dt_dev(dev)->od_env_for_commit;
 528                 /*
 529                  * This od_env_for_commit is only for commit usage.  see
 530                  * "struct dt_device"
 531                  */
 532                 lu_context_enter(&env->le_ctx);
 533                 dt_txn_hook_commit(env, th);
 534                 lu_context_exit(&env->le_ctx);
 535         }
 536
 537         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 538         lu_device_put(lud);
 539         th->th_dev = NULL;
 540
 541         lu_context_exit(&th->th_ctx);
 542         lu_context_fini(&th->th_ctx);
 543         OBD_FREE_PTR(oh);
 544 }
 545
 546 /*
 547  * Concurrency: shouldn't matter.
 548  */
 549 static struct thandle *osd_trans_start(const struct lu_env *env,
 550                                        struct dt_device *d,
 551                                        struct txn_param *p)
 552 {
 553         struct osd_device  *dev = osd_dt_dev(d);
 554         handle_t           *jh;
 555         struct osd_thandle *oh;
 556         struct thandle     *th;
 557         int hook_res;
 558
 559         ENTRY;
 560
 561         hook_res = dt_txn_hook_start(env, d, p);
 562         if (hook_res != 0)
 563                 RETURN(ERR_PTR(hook_res));
 564
 565         if (osd_param_is_sane(dev, p)) {
 566                 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 567                 if (oh != NULL) {
 568                         struct osd_thread_info *oti = osd_oti_get(env);
 569
 570                         /*
 571                          * XXX temporary stuff. Some abstraction layer should
 572                          * be used.
 573                          */
 574
 575                         jh = ldiskfs_journal_start_sb(osd_sb(dev), p->tp_credits);
 576                         if (!IS_ERR(jh)) {
 577                                 oh->ot_handle = jh;
 578                                 th = &oh->ot_super;
 579                                 th->th_dev = d;
 580                                 th->th_result = 0;
 581                                 jh->h_sync = p->tp_sync;
 582                                 lu_device_get(&d->dd_lu_dev);
 583                                 oh->ot_dev_link = lu_ref_add
 584                                         (&d->dd_lu_dev.ld_reference,
 585                                          "osd-tx", th);
 586                                 /* add commit callback */
 587                                 lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
 588                                 lu_context_enter(&th->th_ctx);
 589                                 osd_journal_callback_set(jh, osd_trans_commit_cb,
 590                                                          (struct journal_callback *)&oh->ot_jcb);
 591                                         LASSERT(oti->oti_txns == 0);
 592                                         LASSERT(oti->oti_r_locks == 0);
 593                                         LASSERT(oti->oti_w_locks == 0);
 594                                         oti->oti_txns++;
 595                         } else {
 596                                 OBD_FREE_PTR(oh);
 597                                 th = (void *)jh;
 598                         }
 599                 } else
 600                         th = ERR_PTR(-ENOMEM);
 601         } else {
 602                 CERROR("Invalid transaction parameters\n");
 603                 th = ERR_PTR(-EINVAL);
 604         }
 605
 606         RETURN(th);
 607 }
 608
 609 /*
 610  * Concurrency: shouldn't matter.
 611  */
 612 static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
 613 {
 614         int result;
 615         struct osd_thandle *oh;
 616         struct osd_thread_info *oti = osd_oti_get(env);
 617
 618         ENTRY;
 619
 620         oh = container_of0(th, struct osd_thandle, ot_super);
 621         if (oh->ot_handle != NULL) {
 622                 handle_t *hdl = oh->ot_handle;
 623
 624                 LASSERT(oti->oti_txns == 1);
 625                 oti->oti_txns--;
 626                 LASSERT(oti->oti_r_locks == 0);
 627                 LASSERT(oti->oti_w_locks == 0);
 628                 result = dt_txn_hook_stop(env, th);
 629                 if (result != 0)
 630                         CERROR("Failure in transaction hook: %d\n", result);
 631                 oh->ot_handle = NULL;
 632                 result = ldiskfs_journal_stop(hdl);
 633                 if (result != 0)
 634                         CERROR("Failure to stop transaction: %d\n", result);
 635         }
 636         EXIT;
 637 }
 638
 639 /*
 640  * Concurrency: no concurrent access is possible that late in object
 641  * life-cycle.
 642  */
 643 static int osd_inode_remove(const struct lu_env *env, struct osd_object *obj)
 644 {
 645         const struct lu_fid    *fid = lu_object_fid(&obj->oo_dt.do_lu);
 646         struct osd_device      *osd = osd_obj2dev(obj);
 647         struct osd_thread_info *oti = osd_oti_get(env);
 648         struct txn_param       *prm = &oti->oti_txn;
 649         struct lu_env          *env_del_obj = &oti->oti_obj_delete_tx_env;
 650         struct thandle         *th;
 651         int result;
 652
 653         lu_env_init(env_del_obj, LCT_DT_THREAD);
 654         txn_param_init(prm, OSD_TXN_OI_DELETE_CREDITS +
 655                             OSD_TXN_INODE_DELETE_CREDITS);
 656         th = osd_trans_start(env_del_obj, &osd->od_dt_dev, prm);
 657         if (!IS_ERR(th)) {
 658                 result = osd_oi_delete(osd_oti_get(env_del_obj),
 659                                        &osd->od_oi, fid, th);
 660                 osd_trans_stop(env_del_obj, th);
 661         } else
 662                 result = PTR_ERR(th);
 663
 664         lu_env_fini(env_del_obj);
 665         return result;
 666 }
 667
 668 /*
 669  * Called just before object is freed. Releases all resources except for
 670  * object itself (that is released by osd_object_free()).
 671  *
 672  * Concurrency: no concurrent access is possible that late in object
 673  * life-cycle.
 674  */
 675 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 676 {
 677         struct osd_object *obj   = osd_obj(l);
 678         struct inode      *inode = obj->oo_inode;
 679
 680         LINVRNT(osd_invariant(obj));
 681
 682         /*
 683          * If object is unlinked remove fid->ino mapping from object index.
 684          */
 685
 686         osd_index_fini(obj);
 687         if (inode != NULL) {
 688                 int result;
 689
 690                 if (osd_inode_unlinked(inode)) {
 691                         result = osd_inode_remove(env, obj);
 692                         if (result != 0)
 693                                 LU_OBJECT_DEBUG(D_ERROR, env, l,
 694                                                 "Failed to cleanup: %d\n",
 695                                                 result);
 696                 }
 697
 698                 iput(inode);
 699                 obj->oo_inode = NULL;
 700         }
 701 }
 702
 703 /*
 704  * Concurrency: ->loo_object_release() is called under site spin-lock.
 705  */
 706 static void osd_object_release(const struct lu_env *env,
 707                                struct lu_object *l)
 708 {
 709         struct osd_object *o = osd_obj(l);
 710
 711         LASSERT(!lu_object_is_dying(l->lo_header));
 712         if (o->oo_inode != NULL && osd_inode_unlinked(o->oo_inode))
 713                 set_bit(LU_OBJECT_HEARD_BANSHEE, &l->lo_header->loh_flags);
 714 }
 715
 716 /*
 717  * Concurrency: shouldn't matter.
 718  */
 719 static int osd_object_print(const struct lu_env *env, void *cookie,
 720                             lu_printer_t p, const struct lu_object *l)
 721 {
 722         struct osd_object *o = osd_obj(l);
 723         struct iam_descr  *d;
 724
 725         if (o->oo_dir != NULL)
 726                 d = o->oo_dir->od_container.ic_descr;
 727         else
 728                 d = NULL;
 729         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 730                     o, o->oo_inode,
 731                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 732                     o->oo_inode ? o->oo_inode->i_generation : 0,
 733                     d ? d->id_ops->id_name : "plain");
 734 }
 735
 736 /*
 737  * Concurrency: shouldn't matter.
 738  */
 739 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 740                struct kstatfs *sfs)
 741 {
 742         struct osd_device *osd = osd_dt_dev(d);
 743         struct super_block *sb = osd_sb(osd);
 744         int result = 0;
 745
 746         spin_lock(&osd->od_osfs_lock);
 747         /* cache 1 second */
 748         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 749                 result = ll_do_statfs(sb, &osd->od_kstatfs);
 750                 if (likely(result == 0)) /* N.B. statfs can't really fail */
 751                         osd->od_osfs_age = cfs_time_current_64();
 752         }
 753
 754         if (likely(result == 0))
 755                 *sfs = osd->od_kstatfs;
 756         spin_unlock(&osd->od_osfs_lock);
 757
 758         return result;
 759 }
 760
 761 /*
 762  * Concurrency: doesn't access mutable data.
 763  */
 764 static void osd_conf_get(const struct lu_env *env,
 765                          const struct dt_device *dev,
 766                          struct dt_device_param *param)
 767 {
 768         /*
 769          * XXX should be taken from not-yet-existing fs abstraction layer.
 770          */
 771         param->ddp_max_name_len  = LDISKFS_NAME_LEN;
 772         param->ddp_max_nlink     = LDISKFS_LINK_MAX;
 773         param->ddp_block_shift   = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
 774 }
 775
 776 /**
 777  * Helper function to get and fill the buffer with input values.
 778  */
 779 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
 780 {
 781         struct lu_buf *buf;
 782
 783         buf = &osd_oti_get(env)->oti_buf;
 784         buf->lb_buf = area;
 785         buf->lb_len = len;
 786         return buf;
 787 }
 788
 789 /*
 790  * Concurrency: shouldn't matter.
 791  */
 792 static int osd_sync(const struct lu_env *env, struct dt_device *d)
 793 {
 794         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
 795         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
 796 }
 797
 798 /**
 799  * Start commit for OSD device.
 800  *
 801  * An implementation of dt_commit_async method for OSD device.
 802  * Asychronously starts underlayng fs sync and thereby a transaction
 803  * commit.
 804  *
 805  * \param env environment
 806  * \param d dt device
 807  *
 808  * \see dt_device_operations
 809  */
 810 static int osd_commit_async(const struct lu_env *env,
 811                             struct dt_device *d)
 812 {
 813         struct super_block *s = osd_sb(osd_dt_dev(d));
 814         ENTRY;
 815
 816         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
 817         RETURN(s->s_op->sync_fs(s, 0));
 818 }
 819
 820 /*
 821  * Concurrency: shouldn't matter.
 822  */
 823 lvfs_sbdev_type fsfilt_ldiskfs_journal_sbdev(struct super_block *);
 824
 825 static void osd_ro(const struct lu_env *env, struct dt_device *d)
 826 {
 827         ENTRY;
 828
 829         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
 830
 831         __lvfs_set_rdonly(lvfs_sbdev(osd_sb(osd_dt_dev(d))),
 832                           fsfilt_ldiskfs_journal_sbdev(osd_sb(osd_dt_dev(d))));
 833         EXIT;
 834 }
 835
 836
 837 /*
 838  * Concurrency: serialization provided by callers.
 839  */
 840 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
 841                               int mode, unsigned long timeout, __u32 alg,
 842                               struct lustre_capa_key *keys)
 843 {
 844         struct osd_device *dev = osd_dt_dev(d);
 845         ENTRY;
 846
 847         dev->od_fl_capa = mode;
 848         dev->od_capa_timeout = timeout;
 849         dev->od_capa_alg = alg;
 850         dev->od_capa_keys = keys;
 851         RETURN(0);
 852 }
 853
 854 /**
 855  * Concurrency: serialization provided by callers.
 856  */
 857 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
 858                                struct dt_quota_ctxt *ctxt, void *data)
 859 {
 860         struct obd_device *obd = (void *)ctxt;
 861         struct vfsmount *mnt = (struct vfsmount *)data;
 862         ENTRY;
 863
 864         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
 865         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
 866         obd->obd_lvfs_ctxt.pwdmnt = mnt;
 867         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
 868         obd->obd_lvfs_ctxt.fs = get_ds();
 869
 870         EXIT;
 871 }
 872
 873 /**
 874  * Note: we do not count into QUOTA here.
 875  * If we mount with --data_journal we may need more.
 876  */
 877 static const int osd_dto_credits_noquota[DTO_NR] = {
 878         /**
 879          * Insert/Delete.
 880          * INDEX_EXTRA_TRANS_BLOCKS(8) +
 881          * SINGLEDATA_TRANS_BLOCKS(8)
 882          * XXX Note: maybe iam need more, since iam have more level than
 883          *           EXT3 htree.
 884          */
 885         [DTO_INDEX_INSERT]  = 16,
 886         [DTO_INDEX_DELETE]  = 16,
 887         /**
 888          * Unused now
 889          */
 890         [DTO_IDNEX_UPDATE]  = 16,
 891         /**
 892          * Create a object. The same as create object in EXT3.
 893          * DATA_TRANS_BLOCKS(14) +
 894          * INDEX_EXTRA_BLOCKS(8) +
 895          * 3(inode bits, groups, GDT)
 896          */
 897         [DTO_OBJECT_CREATE] = 25,
 898         /**
 899          * Unused now
 900          */
 901         [DTO_OBJECT_DELETE] = 25,
 902         /**
 903          * Attr set credits.
 904          * 3(inode bits, group, GDT)
 905          */
 906         [DTO_ATTR_SET_BASE] = 3,
 907         /**
 908          * Xattr set. The same as xattr of EXT3.
 909          * DATA_TRANS_BLOCKS(14)
 910          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
 911          * are also counted in. Do not know why?
 912          */
 913         [DTO_XATTR_SET]     = 14,
 914         [DTO_LOG_REC]       = 14,
 915         /**
 916          * creadits for inode change during write.
 917          */
 918         [DTO_WRITE_BASE]    = 3,
 919         /**
 920          * credits for single block write.
 921          */
 922         [DTO_WRITE_BLOCK]   = 14,
 923         /**
 924          * Attr set credits for chown.
 925          * This is extra credits for setattr, and it is null without quota
 926          */
 927         [DTO_ATTR_SET_CHOWN]= 0
 928 };
 929
 930 /**
 931  * Note: we count into QUOTA here.
 932  * If we mount with --data_journal we may need more.
 933  */
 934 static const int osd_dto_credits_quota[DTO_NR] = {
 935         /**
 936          * INDEX_EXTRA_TRANS_BLOCKS(8) +
 937          * SINGLEDATA_TRANS_BLOCKS(8) +
 938          * 2 * QUOTA_TRANS_BLOCKS(2)
 939          */
 940         [DTO_INDEX_INSERT]  = 20,
 941         /**
 942          * INDEX_EXTRA_TRANS_BLOCKS(8) +
 943          * SINGLEDATA_TRANS_BLOCKS(8) +
 944          * 2 * QUOTA_TRANS_BLOCKS(2)
 945          */
 946         [DTO_INDEX_DELETE]  = 20,
 947         /**
 948          * Unused now.
 949          */
 950         [DTO_IDNEX_UPDATE]  = 16,
 951         /*
 952          * Create a object. Same as create object in EXT3 filesystem.
 953          * DATA_TRANS_BLOCKS(16) +
 954          * INDEX_EXTRA_BLOCKS(8) +
 955          * 3(inode bits, groups, GDT) +
 956          * 2 * QUOTA_INIT_BLOCKS(25)
 957          */
 958         [DTO_OBJECT_CREATE] = 77,
 959         /*
 960          * Unused now.
 961          * DATA_TRANS_BLOCKS(16) +
 962          * INDEX_EXTRA_BLOCKS(8) +
 963          * 3(inode bits, groups, GDT) +
 964          * QUOTA(?)
 965          */
 966         [DTO_OBJECT_DELETE] = 27,
 967         /**
 968          * Attr set credits.
 969          * 3 (inode bit, group, GDT) +
 970          */
 971         [DTO_ATTR_SET_BASE] = 3,
 972         /**
 973          * Xattr set. The same as xattr of EXT3.
 974          * DATA_TRANS_BLOCKS(16)
 975          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS are
 976          *           also counted in. Do not know why?
 977          */
 978         [DTO_XATTR_SET]     = 16,
 979         [DTO_LOG_REC]       = 16,
 980         /**
 981          * creadits for inode change during write.
 982          */
 983         [DTO_WRITE_BASE]    = 3,
 984         /**
 985          * credits for single block write.
 986          */
 987         [DTO_WRITE_BLOCK]   = 16,
 988         /**
 989          * Attr set credits for chown.
 990          * It is added to already set setattr credits
 991          * 2 * QUOTA_INIT_BLOCKS(25) +
 992          * 2 * QUOTA_DEL_BLOCKS(9)
 993          */
 994         [DTO_ATTR_SET_CHOWN]= 68,
 995 };
 996
 997 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
 998                           enum dt_txn_op op)
 999 {
1000         LASSERT(ARRAY_SIZE(osd_dto_credits_noquota) ==
1001                 ARRAY_SIZE(osd_dto_credits_quota));
1002         LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits_noquota));
1003 #ifdef HAVE_QUOTA_SUPPORT
1004         if (test_opt(osd_sb(osd_dt_dev(d)), QUOTA))
1005                 return osd_dto_credits_quota[op];
1006         else
1007 #endif
1008                 return osd_dto_credits_noquota[op];
1009 }
1010
1011 static const struct dt_device_operations osd_dt_ops = {
1012         .dt_root_get       = osd_root_get,
1013         .dt_statfs         = osd_statfs,
1014         .dt_trans_start    = osd_trans_start,
1015         .dt_trans_stop     = osd_trans_stop,
1016         .dt_conf_get       = osd_conf_get,
1017         .dt_sync           = osd_sync,
1018         .dt_ro             = osd_ro,
1019         .dt_commit_async   = osd_commit_async,
1020         .dt_credit_get     = osd_credit_get,
1021         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1022         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1023 };
1024
1025 static void osd_object_read_lock(const struct lu_env *env,
1026                                  struct dt_object *dt, unsigned role)
1027 {
1028         struct osd_object *obj = osd_dt_obj(dt);
1029         struct osd_thread_info *oti = osd_oti_get(env);
1030
1031         LINVRNT(osd_invariant(obj));
1032
1033         LASSERT(obj->oo_owner != env);
1034         down_read_nested(&obj->oo_sem, role);
1035
1036         LASSERT(obj->oo_owner == NULL);
1037         oti->oti_r_locks++;
1038 }
1039
1040 static void osd_object_write_lock(const struct lu_env *env,
1041                                   struct dt_object *dt, unsigned role)
1042 {
1043         struct osd_object *obj = osd_dt_obj(dt);
1044         struct osd_thread_info *oti = osd_oti_get(env);
1045
1046         LINVRNT(osd_invariant(obj));
1047
1048         LASSERT(obj->oo_owner != env);
1049         down_write_nested(&obj->oo_sem, role);
1050
1051         LASSERT(obj->oo_owner == NULL);
1052         obj->oo_owner = env;
1053         oti->oti_w_locks++;
1054 }
1055
1056 static void osd_object_read_unlock(const struct lu_env *env,
1057                                    struct dt_object *dt)
1058 {
1059         struct osd_object *obj = osd_dt_obj(dt);
1060         struct osd_thread_info *oti = osd_oti_get(env);
1061
1062         LINVRNT(osd_invariant(obj));
1063
1064         LASSERT(oti->oti_r_locks > 0);
1065         oti->oti_r_locks--;
1066         up_read(&obj->oo_sem);
1067 }
1068
1069 static void osd_object_write_unlock(const struct lu_env *env,
1070                                     struct dt_object *dt)
1071 {
1072         struct osd_object *obj = osd_dt_obj(dt);
1073         struct osd_thread_info *oti = osd_oti_get(env);
1074
1075         LINVRNT(osd_invariant(obj));
1076
1077         LASSERT(obj->oo_owner == env);
1078         LASSERT(oti->oti_w_locks > 0);
1079         oti->oti_w_locks--;
1080         obj->oo_owner = NULL;
1081         up_write(&obj->oo_sem);
1082 }
1083
1084 static int osd_object_write_locked(const struct lu_env *env,
1085                                    struct dt_object *dt)
1086 {
1087         struct osd_object *obj = osd_dt_obj(dt);
1088
1089         LINVRNT(osd_invariant(obj));
1090
1091         return obj->oo_owner == env;
1092 }
1093
1094 static int capa_is_sane(const struct lu_env *env,
1095                         struct osd_device *dev,
1096                         struct lustre_capa *capa,
1097                         struct lustre_capa_key *keys)
1098 {
1099         struct osd_thread_info *oti = osd_oti_get(env);
1100         struct lustre_capa *tcapa = &oti->oti_capa;
1101         struct obd_capa *oc;
1102         int i, rc = 0;
1103         ENTRY;
1104
1105         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1106         if (oc) {
1107                 if (capa_is_expired(oc)) {
1108                         DEBUG_CAPA(D_ERROR, capa, "expired");
1109                         rc = -ESTALE;
1110                 }
1111                 capa_put(oc);
1112                 RETURN(rc);
1113         }
1114
1115         if (capa_is_expired_sec(capa)) {
1116                 DEBUG_CAPA(D_ERROR, capa, "expired");
1117                 RETURN(-ESTALE);
1118         }
1119
1120         spin_lock(&capa_lock);
1121         for (i = 0; i < 2; i++) {
1122                 if (keys[i].lk_keyid == capa->lc_keyid) {
1123                         oti->oti_capa_key = keys[i];
1124                         break;
1125                 }
1126         }
1127         spin_unlock(&capa_lock);
1128
1129         if (i == 2) {
1130                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1131                 RETURN(-ESTALE);
1132         }
1133
1134         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1135         if (rc)
1136                 RETURN(rc);
1137
1138         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1139                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1140                 RETURN(-EACCES);
1141         }
1142
1143         oc = capa_add(dev->od_capa_hash, capa);
1144         capa_put(oc);
1145
1146         RETURN(0);
1147 }
1148
1149 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1150                            struct lustre_capa *capa, __u64 opc)
1151 {
1152         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1153         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1154         struct md_capainfo *ci;
1155         int rc;
1156
1157         if (!dev->od_fl_capa)
1158                 return 0;
1159
1160         if (capa == BYPASS_CAPA)
1161                 return 0;
1162
1163         ci = md_capainfo(env);
1164         if (unlikely(!ci))
1165                 return 0;
1166
1167         if (ci->mc_auth == LC_ID_NONE)
1168                 return 0;
1169
1170         if (!capa) {
1171                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1172                 return -EACCES;
1173         }
1174
1175         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1176                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1177                            PFID(fid));
1178                 return -EACCES;
1179         }
1180
1181         if (!capa_opc_supported(capa, opc)) {
1182                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1183                 return -EACCES;
1184         }
1185
1186         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1187                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1188                 return -EACCES;
1189         }
1190
1191         return 0;
1192 }
1193
1194 static struct timespec *osd_inode_time(const struct lu_env *env,
1195                                        struct inode *inode, __u64 seconds)
1196 {
1197         struct osd_thread_info *oti = osd_oti_get(env);
1198         struct timespec        *t   = &oti->oti_time;
1199
1200         t->tv_sec  = seconds;
1201         t->tv_nsec = 0;
1202         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1203         return t;
1204 }
1205
1206
1207 static void osd_inode_getattr(const struct lu_env *env,
1208                               struct inode *inode, struct lu_attr *attr)
1209 {
1210         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1211                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1212                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1213
1214         attr->la_atime      = LTIME_S(inode->i_atime);
1215         attr->la_mtime      = LTIME_S(inode->i_mtime);
1216         attr->la_ctime      = LTIME_S(inode->i_ctime);
1217         attr->la_mode       = inode->i_mode;
1218         attr->la_size       = i_size_read(inode);
1219         attr->la_blocks     = inode->i_blocks;
1220         attr->la_uid        = inode->i_uid;
1221         attr->la_gid        = inode->i_gid;
1222         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1223         attr->la_nlink      = inode->i_nlink;
1224         attr->la_rdev       = inode->i_rdev;
1225         attr->la_blksize    = ll_inode_blksize(inode);
1226         attr->la_blkbits    = inode->i_blkbits;
1227 }
1228
1229 static int osd_attr_get(const struct lu_env *env,
1230                         struct dt_object *dt,
1231                         struct lu_attr *attr,
1232                         struct lustre_capa *capa)
1233 {
1234         struct osd_object *obj = osd_dt_obj(dt);
1235
1236         LASSERT(dt_object_exists(dt));
1237         LINVRNT(osd_invariant(obj));
1238
1239         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1240                 return -EACCES;
1241
1242         spin_lock(&obj->oo_guard);
1243         osd_inode_getattr(env, obj->oo_inode, attr);
1244         spin_unlock(&obj->oo_guard);
1245         return 0;
1246 }
1247
1248 static int osd_inode_setattr(const struct lu_env *env,
1249                              struct inode *inode, const struct lu_attr *attr)
1250 {
1251         __u64 bits;
1252
1253         bits = attr->la_valid;
1254
1255         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1256
1257 #ifdef HAVE_QUOTA_SUPPORT
1258         if ((bits & LA_UID && attr->la_uid != inode->i_uid) ||
1259             (bits & LA_GID && attr->la_gid != inode->i_gid)) {
1260                 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1261                 struct iattr iattr;
1262                 int rc;
1263
1264                 iattr.ia_valid = 0;
1265                 if (bits & LA_UID)
1266                         iattr.ia_valid |= ATTR_UID;
1267                 if (bits & LA_GID)
1268                         iattr.ia_valid |= ATTR_GID;
1269                 iattr.ia_uid = attr->la_uid;
1270                 iattr.ia_gid = attr->la_gid;
1271                 osd_push_ctxt(env, save);
1272                 rc = DQUOT_TRANSFER(inode, &iattr) ? -EDQUOT : 0;
1273                 osd_pop_ctxt(save);
1274                 if (rc != 0)
1275                         return rc;
1276         }
1277 #endif
1278
1279         if (bits & LA_ATIME)
1280                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1281         if (bits & LA_CTIME)
1282                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1283         if (bits & LA_MTIME)
1284                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1285         if (bits & LA_SIZE) {
1286                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1287                 i_size_write(inode, attr->la_size);
1288         }
1289
1290         /* OSD should not change "i_blocks" which is used by quota.
1291          * "i_blocks" should be changed by ldiskfs only.
1292          * Enable this assignment for SOM purpose now, until it is
1293          * stored in SOM EA. */
1294         if (bits & LA_BLOCKS)
1295                 inode->i_blocks = attr->la_blocks;
1296
1297         if (bits & LA_MODE)
1298                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1299                         (attr->la_mode & ~S_IFMT);
1300         if (bits & LA_UID)
1301                 inode->i_uid    = attr->la_uid;
1302         if (bits & LA_GID)
1303                 inode->i_gid    = attr->la_gid;
1304         if (bits & LA_NLINK)
1305                 inode->i_nlink  = attr->la_nlink;
1306         if (bits & LA_RDEV)
1307                 inode->i_rdev   = attr->la_rdev;
1308
1309         if (bits & LA_FLAGS) {
1310                 struct ldiskfs_inode_info *li = LDISKFS_I(inode);
1311
1312                 li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
1313                         (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
1314         }
1315         return 0;
1316 }
1317
1318 static int osd_attr_set(const struct lu_env *env,
1319                         struct dt_object *dt,
1320                         const struct lu_attr *attr,
1321                         struct thandle *handle,
1322                         struct lustre_capa *capa)
1323 {
1324         struct osd_object *obj = osd_dt_obj(dt);
1325         int rc;
1326
1327         LASSERT(handle != NULL);
1328         LASSERT(dt_object_exists(dt));
1329         LASSERT(osd_invariant(obj));
1330
1331         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1332                 return -EACCES;
1333
1334         spin_lock(&obj->oo_guard);
1335         rc = osd_inode_setattr(env, obj->oo_inode, attr);
1336         spin_unlock(&obj->oo_guard);
1337
1338         if (!rc)
1339                 mark_inode_dirty(obj->oo_inode);
1340         return rc;
1341 }
1342
1343 /*
1344  * Object creation.
1345  *
1346  * XXX temporary solution.
1347  */
1348 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1349                           struct lu_attr *attr, struct thandle *th)
1350 {
1351         return 0;
1352 }
1353
1354 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1355                            struct lu_attr *attr, struct thandle *th)
1356 {
1357         osd_object_init0(obj);
1358         return 0;
1359 }
1360
1361 static struct dentry * osd_child_dentry_get(const struct lu_env *env,
1362                                             struct osd_object *obj,
1363                                             const char *name,
1364                                             const int namelen)
1365 {
1366         struct osd_thread_info *info   = osd_oti_get(env);
1367         struct dentry *child_dentry = &info->oti_child_dentry;
1368         struct dentry *obj_dentry = &info->oti_obj_dentry;
1369
1370         obj_dentry->d_inode = obj->oo_inode;
1371         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
1372         obj_dentry->d_name.hash = 0;
1373
1374         child_dentry->d_name.hash = 0;
1375         child_dentry->d_parent = obj_dentry;
1376         child_dentry->d_name.name = name;
1377         child_dentry->d_name.len = namelen;
1378         return child_dentry;
1379 }
1380
1381
1382 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1383                       umode_t mode,
1384                       struct dt_allocation_hint *hint,
1385                       struct thandle *th)
1386 {
1387         int result;
1388         struct osd_device  *osd = osd_obj2dev(obj);
1389         struct osd_thandle *oth;
1390         struct dt_object   *parent;
1391         struct inode       *inode;
1392 #ifdef HAVE_QUOTA_SUPPORT
1393         struct osd_ctxt    *save = &info->oti_ctxt;
1394 #endif
1395
1396         LINVRNT(osd_invariant(obj));
1397         LASSERT(obj->oo_inode == NULL);
1398
1399         oth = container_of(th, struct osd_thandle, ot_super);
1400         LASSERT(oth->ot_handle->h_transaction != NULL);
1401
1402         if (hint && hint->dah_parent)
1403                 parent = hint->dah_parent;
1404         else
1405                 parent = osd->od_obj_area;
1406
1407         LASSERT(parent != NULL);
1408         LASSERT(osd_dt_obj(parent)->oo_inode->i_op != NULL);
1409
1410 #ifdef HAVE_QUOTA_SUPPORT
1411         osd_push_ctxt(info->oti_env, save);
1412 #endif
1413         inode = ldiskfs_create_inode(oth->ot_handle,
1414                                      osd_dt_obj(parent)->oo_inode, mode);
1415 #ifdef HAVE_QUOTA_SUPPORT
1416         osd_pop_ctxt(save);
1417 #endif
1418         if (!IS_ERR(inode)) {
1419                 obj->oo_inode = inode;
1420                 result = 0;
1421         } else
1422                 result = PTR_ERR(inode);
1423         LINVRNT(osd_invariant(obj));
1424         return result;
1425 }
1426
1427 enum {
1428         OSD_NAME_LEN = 255
1429 };
1430
1431 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1432                      struct lu_attr *attr,
1433                      struct dt_allocation_hint *hint,
1434                      struct dt_object_format *dof,
1435                      struct thandle *th)
1436 {
1437         int result;
1438         struct osd_thandle *oth;
1439         struct osd_device *osd = osd_obj2dev(obj);
1440         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1441
1442         LASSERT(S_ISDIR(attr->la_mode));
1443
1444         oth = container_of(th, struct osd_thandle, ot_super);
1445         LASSERT(oth->ot_handle->h_transaction != NULL);
1446         result = osd_mkfile(info, obj, mode, hint, th);
1447         if (result == 0 && osd->od_iop_mode == 0) {
1448                 LASSERT(obj->oo_inode != NULL);
1449                 /*
1450                  * XXX uh-oh... call low-level iam function directly.
1451                  */
1452
1453                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1454                                          sizeof (struct osd_fid_pack),
1455                                          oth->ot_handle);
1456         }
1457         return result;
1458 }
1459
1460 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1461                         struct lu_attr *attr,
1462                         struct dt_allocation_hint *hint,
1463                         struct dt_object_format *dof,
1464                         struct thandle *th)
1465 {
1466         int result;
1467         struct osd_thandle *oth;
1468         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1469
1470         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1471
1472         LASSERT(S_ISREG(attr->la_mode));
1473
1474         oth = container_of(th, struct osd_thandle, ot_super);
1475         LASSERT(oth->ot_handle->h_transaction != NULL);
1476
1477         result = osd_mkfile(info, obj, mode, hint, th);
1478         if (result == 0) {
1479                 LASSERT(obj->oo_inode != NULL);
1480                 if (feat->dif_flags & DT_IND_VARKEY)
1481                         result = iam_lvar_create(obj->oo_inode,
1482                                                  feat->dif_keysize_max,
1483                                                  feat->dif_ptrsize,
1484                                                  feat->dif_recsize_max,
1485                                                  oth->ot_handle);
1486                 else
1487                         result = iam_lfix_create(obj->oo_inode,
1488                                                  feat->dif_keysize_max,
1489                                                  feat->dif_ptrsize,
1490                                                  feat->dif_recsize_max,
1491                                                  oth->ot_handle);
1492
1493         }
1494         return result;
1495 }
1496
1497 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1498                      struct lu_attr *attr,
1499                      struct dt_allocation_hint *hint,
1500                      struct dt_object_format *dof,
1501                      struct thandle *th)
1502 {
1503         LASSERT(S_ISREG(attr->la_mode));
1504         return osd_mkfile(info, obj, (attr->la_mode &
1505                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1506 }
1507
1508 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1509                      struct lu_attr *attr,
1510                      struct dt_allocation_hint *hint,
1511                      struct dt_object_format *dof,
1512                      struct thandle *th)
1513 {
1514         LASSERT(S_ISLNK(attr->la_mode));
1515         return osd_mkfile(info, obj, (attr->la_mode &
1516                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1517 }
1518
1519 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1520                      struct lu_attr *attr,
1521                      struct dt_allocation_hint *hint,
1522                      struct dt_object_format *dof,
1523                      struct thandle *th)
1524 {
1525         umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1526         int result;
1527
1528         LINVRNT(osd_invariant(obj));
1529         LASSERT(obj->oo_inode == NULL);
1530         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1531                 S_ISFIFO(mode) || S_ISSOCK(mode));
1532
1533         result = osd_mkfile(info, obj, mode, hint, th);
1534         if (result == 0) {
1535                 LASSERT(obj->oo_inode != NULL);
1536                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1537         }
1538         LINVRNT(osd_invariant(obj));
1539         return result;
1540 }
1541
1542 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1543                               struct lu_attr *,
1544                               struct dt_allocation_hint *hint,
1545                               struct dt_object_format *dof,
1546                               struct thandle *);
1547
1548 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1549 {
1550         osd_obj_type_f result;
1551
1552         switch (type) {
1553         case DFT_DIR:
1554                 result = osd_mkdir;
1555                 break;
1556         case DFT_REGULAR:
1557                 result = osd_mkreg;
1558                 break;
1559         case DFT_SYM:
1560                 result = osd_mksym;
1561                 break;
1562         case DFT_NODE:
1563                 result = osd_mknod;
1564                 break;
1565         case DFT_INDEX:
1566                 result = osd_mk_index;
1567                 break;
1568
1569         default:
1570                 LBUG();
1571                 break;
1572         }
1573         return result;
1574 }
1575
1576
1577 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1578                         struct dt_object *parent, umode_t child_mode)
1579 {
1580         LASSERT(ah);
1581
1582         memset(ah, 0, sizeof(*ah));
1583         ah->dah_parent = parent;
1584         ah->dah_mode = child_mode;
1585 }
1586
1587 /**
1588  * Helper function for osd_object_create()
1589  *
1590  * \retval 0, on success
1591  */
1592 static int __osd_object_create(struct osd_thread_info *info,
1593                                struct osd_object *obj, struct lu_attr *attr,
1594                                struct dt_allocation_hint *hint,
1595                                struct dt_object_format *dof,
1596                                struct thandle *th)
1597 {
1598
1599         int result;
1600
1601         result = osd_create_pre(info, obj, attr, th);
1602         if (result == 0) {
1603                 result = osd_create_type_f(dof->dof_type)(info, obj,
1604                                            attr, hint, dof, th);
1605                 if (result == 0)
1606                         result = osd_create_post(info, obj, attr, th);
1607         }
1608         return result;
1609 }
1610
1611 /**
1612  * Helper function for osd_object_create()
1613  *
1614  * \retval 0, on success
1615  */
1616 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1617                            const struct lu_fid *fid, struct thandle *th)
1618 {
1619         struct osd_thread_info *info = osd_oti_get(env);
1620         struct osd_inode_id    *id   = &info->oti_id;
1621         struct osd_device      *osd  = osd_obj2dev(obj);
1622         struct md_ucred        *uc   = md_ucred(env);
1623
1624         LASSERT(obj->oo_inode != NULL);
1625         LASSERT(uc != NULL);
1626
1627         id->oii_ino = obj->oo_inode->i_ino;
1628         id->oii_gen = obj->oo_inode->i_generation;
1629
1630         return osd_oi_insert(info, &osd->od_oi, fid, id, th,
1631                              uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
1632 }
1633
1634 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1635                              struct lu_attr *attr,
1636                              struct dt_allocation_hint *hint,
1637                              struct dt_object_format *dof,
1638                              struct thandle *th)
1639 {
1640         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1641         struct osd_object      *obj    = osd_dt_obj(dt);
1642         struct osd_thread_info *info   = osd_oti_get(env);
1643         int result;
1644
1645         ENTRY;
1646
1647         LINVRNT(osd_invariant(obj));
1648         LASSERT(!dt_object_exists(dt));
1649         LASSERT(osd_write_locked(env, obj));
1650         LASSERT(th != NULL);
1651
1652         result = __osd_object_create(info, obj, attr, hint, dof, th);
1653         if (result == 0)
1654                 result = __osd_oi_insert(env, obj, fid, th);
1655
1656         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1657         LASSERT(osd_invariant(obj));
1658         RETURN(result);
1659 }
1660
1661 /**
1662  * Helper function for osd_xattr_set()
1663  */
1664 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1665                            const struct lu_buf *buf, const char *name, int fl)
1666 {
1667         struct osd_object      *obj      = osd_dt_obj(dt);
1668         struct inode           *inode    = obj->oo_inode;
1669         struct osd_thread_info *info     = osd_oti_get(env);
1670         struct dentry          *dentry   = &info->oti_child_dentry;
1671         struct timespec        *t        = &info->oti_time;
1672         int                     fs_flags = 0;
1673         int  rc;
1674
1675         LASSERT(dt_object_exists(dt));
1676         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1677         LASSERT(osd_write_locked(env, obj));
1678
1679         if (fl & LU_XATTR_REPLACE)
1680                 fs_flags |= XATTR_REPLACE;
1681
1682         if (fl & LU_XATTR_CREATE)
1683                 fs_flags |= XATTR_CREATE;
1684
1685         dentry->d_inode = inode;
1686         *t = inode->i_ctime;
1687         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
1688                                    buf->lb_len, fs_flags);
1689         /* ctime should not be updated with server-side time. */
1690         spin_lock(&obj->oo_guard);
1691         inode->i_ctime = *t;
1692         spin_unlock(&obj->oo_guard);
1693         mark_inode_dirty(inode);
1694         return rc;
1695 }
1696
1697 /**
1698  * Put the fid into lustre_mdt_attrs, and then place the structure
1699  * inode's ea. This fid should not be altered during the life time
1700  * of the inode.
1701  *
1702  * \retval +ve, on success
1703  * \retval -ve, on error
1704  *
1705  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
1706  */
1707 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
1708                           const struct lu_fid *fid)
1709 {
1710         struct osd_thread_info  *info      = osd_oti_get(env);
1711         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1712
1713         lustre_lma_init(mdt_attrs, fid);
1714         lustre_lma_swab(mdt_attrs);
1715         return __osd_xattr_set(env, dt,
1716                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
1717                                XATTR_NAME_LMA, LU_XATTR_CREATE);
1718
1719 }
1720
1721 /**
1722  * Helper function to form igif
1723  */
1724 static inline void osd_igif_get(const struct lu_env *env, struct inode  *inode,
1725                                 struct lu_fid *fid)
1726 {
1727         lu_igif_build(fid, inode->i_ino, inode->i_generation);
1728 }
1729
1730 /**
1731  * Helper function to pack the fid
1732  */
1733 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1734                   struct lu_fid *befider)
1735 {
1736         fid_cpu_to_be(befider, (struct lu_fid *)fid);
1737         memcpy(pack->fp_area, befider, sizeof(*befider));
1738         pack->fp_len =  sizeof(*befider) + 1;
1739 }
1740
1741 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1742 {
1743         int result;
1744
1745         result = 0;
1746         switch (pack->fp_len) {
1747         case sizeof *fid + 1:
1748                 memcpy(fid, pack->fp_area, sizeof *fid);
1749                 fid_be_to_cpu(fid, fid);
1750                 break;
1751         default:
1752                 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1753                 result = -EIO;
1754         }
1755         return result;
1756 }
1757
1758 /**
1759  * Try to read the fid from inode ea into dt_rec, if return value
1760  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
1761  *
1762  * \param fid object fid.
1763  *
1764  * \retval 0 on success
1765  */
1766 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
1767                           __u32 ino, struct lu_fid *fid)
1768 {
1769         struct osd_thread_info  *info      = osd_oti_get(env);
1770         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1771         struct lu_device        *ldev   = obj->oo_dt.do_lu.lo_dev;
1772         struct dentry           *dentry = &info->oti_child_dentry;
1773         struct osd_inode_id     *id     = &info->oti_id;
1774         struct osd_device       *dev;
1775         struct inode            *inode;
1776         int                      rc;
1777
1778         ENTRY;
1779         dev  = osd_dev(ldev);
1780
1781         id->oii_ino = ino;
1782         id->oii_gen = OSD_OII_NOGEN;
1783
1784         inode = osd_iget(info, dev, id);
1785         if (IS_ERR(inode)) {
1786                 rc = PTR_ERR(inode);
1787                 GOTO(out,rc);
1788         }
1789         dentry->d_inode = inode;
1790
1791         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1792         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
1793                                    sizeof *mdt_attrs);
1794
1795         /* Check LMA compatibility */
1796         if (rc > 0 &&
1797             (mdt_attrs->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP))) {
1798                 CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
1799                       inode->i_ino, le32_to_cpu(mdt_attrs->lma_incompat) &
1800                       ~LMA_INCOMPAT_SUPP);
1801                 return -ENOSYS;
1802         }
1803
1804         if (rc > 0) {
1805                 lustre_lma_swab(mdt_attrs);
1806                 memcpy(fid, &mdt_attrs->lma_self_fid, sizeof(*mdt_attrs));
1807                 rc = 0;
1808         } else if (rc == -ENODATA) {
1809                 osd_igif_get(env, inode, fid);
1810                 rc = 0;
1811         }
1812         iput(inode);
1813
1814 out:
1815         RETURN(rc);
1816 }
1817
1818 /**
1819  * OSD layer object create function for interoperability mode (b11826).
1820  * This is mostly similar to osd_object_create(). Only difference being, fid is
1821  * inserted into inode ea here.
1822  *
1823  * \retval   0, on success
1824  * \retval -ve, on error
1825  */
1826 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
1827                              struct lu_attr *attr,
1828                              struct dt_allocation_hint *hint,
1829                              struct dt_object_format *dof,
1830                              struct thandle *th)
1831 {
1832         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1833         struct osd_object      *obj    = osd_dt_obj(dt);
1834         struct osd_thread_info *info   = osd_oti_get(env);
1835         int result;
1836         int is_root = 0;
1837
1838         ENTRY;
1839
1840         LASSERT(osd_invariant(obj));
1841         LASSERT(!dt_object_exists(dt));
1842         LASSERT(osd_write_locked(env, obj));
1843         LASSERT(th != NULL);
1844
1845         result = __osd_object_create(info, obj, attr, hint, dof, th);
1846
1847         if (hint && hint->dah_parent)
1848                 is_root = osd_object_is_root(osd_dt_obj(hint->dah_parent));
1849
1850         /* objects under osd root shld have igif fid, so dont add fid EA */
1851         if (result == 0 && is_root == 0)
1852                 result = osd_ea_fid_set(env, dt, fid);
1853
1854         if (result == 0)
1855                 result = __osd_oi_insert(env, obj, fid, th);
1856
1857         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1858         LINVRNT(osd_invariant(obj));
1859         RETURN(result);
1860 }
1861
1862 /*
1863  * Concurrency: @dt is write locked.
1864  */
1865 static void osd_object_ref_add(const struct lu_env *env,
1866                                struct dt_object *dt,
1867                                struct thandle *th)
1868 {
1869         struct osd_object *obj = osd_dt_obj(dt);
1870         struct inode *inode = obj->oo_inode;
1871
1872         LINVRNT(osd_invariant(obj));
1873         LASSERT(dt_object_exists(dt));
1874         LASSERT(osd_write_locked(env, obj));
1875         LASSERT(th != NULL);
1876
1877         spin_lock(&obj->oo_guard);
1878         LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
1879         inode->i_nlink++;
1880         spin_unlock(&obj->oo_guard);
1881         mark_inode_dirty(inode);
1882         LINVRNT(osd_invariant(obj));
1883 }
1884
1885 /*
1886  * Concurrency: @dt is write locked.
1887  */
1888 static void osd_object_ref_del(const struct lu_env *env,
1889                                struct dt_object *dt,
1890                                struct thandle *th)
1891 {
1892         struct osd_object *obj = osd_dt_obj(dt);
1893         struct inode *inode = obj->oo_inode;
1894
1895         LINVRNT(osd_invariant(obj));
1896         LASSERT(dt_object_exists(dt));
1897         LASSERT(osd_write_locked(env, obj));
1898         LASSERT(th != NULL);
1899
1900         spin_lock(&obj->oo_guard);
1901         LASSERT(inode->i_nlink > 0);
1902         inode->i_nlink--;
1903         spin_unlock(&obj->oo_guard);
1904         mark_inode_dirty(inode);
1905         LINVRNT(osd_invariant(obj));
1906 }
1907
1908 /*
1909  * Concurrency: @dt is read locked.
1910  */
1911 static int osd_xattr_get(const struct lu_env *env,
1912                          struct dt_object *dt,
1913                          struct lu_buf *buf,
1914                          const char *name,
1915                          struct lustre_capa *capa)
1916 {
1917         struct osd_object      *obj    = osd_dt_obj(dt);
1918         struct inode           *inode  = obj->oo_inode;
1919         struct osd_thread_info *info   = osd_oti_get(env);
1920         struct dentry          *dentry = &info->oti_obj_dentry;
1921
1922         LASSERT(dt_object_exists(dt));
1923         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1924         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1925
1926         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1927                 return -EACCES;
1928
1929         dentry->d_inode = inode;
1930         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
1931 }
1932
1933 /*
1934  * Concurrency: @dt is write locked.
1935  */
1936 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1937                          const struct lu_buf *buf, const char *name, int fl,
1938                          struct thandle *handle, struct lustre_capa *capa)
1939 {
1940         LASSERT(handle != NULL);
1941
1942         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1943                 return -EACCES;
1944
1945         return __osd_xattr_set(env, dt, buf, name, fl);
1946 }
1947
1948 /*
1949  * Concurrency: @dt is read locked.
1950  */
1951 static int osd_xattr_list(const struct lu_env *env,
1952                           struct dt_object *dt,
1953                           struct lu_buf *buf,
1954                           struct lustre_capa *capa)
1955 {
1956         struct osd_object      *obj    = osd_dt_obj(dt);
1957         struct inode           *inode  = obj->oo_inode;
1958         struct osd_thread_info *info   = osd_oti_get(env);
1959         struct dentry          *dentry = &info->oti_obj_dentry;
1960
1961         LASSERT(dt_object_exists(dt));
1962         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
1963         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1964
1965         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1966                 return -EACCES;
1967
1968         dentry->d_inode = inode;
1969         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
1970 }
1971
1972 /*
1973  * Concurrency: @dt is write locked.
1974  */
1975 static int osd_xattr_del(const struct lu_env *env,
1976                          struct dt_object *dt,
1977                          const char *name,
1978                          struct thandle *handle,
1979                          struct lustre_capa *capa)
1980 {
1981         struct osd_object      *obj    = osd_dt_obj(dt);
1982         struct inode           *inode  = obj->oo_inode;
1983         struct osd_thread_info *info   = osd_oti_get(env);
1984         struct dentry          *dentry = &info->oti_obj_dentry;
1985         struct timespec        *t      = &info->oti_time;
1986         int                     rc;
1987
1988         LASSERT(dt_object_exists(dt));
1989         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
1990         LASSERT(osd_write_locked(env, obj));
1991         LASSERT(handle != NULL);
1992
1993         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1994                 return -EACCES;
1995
1996         dentry->d_inode = inode;
1997         *t = inode->i_ctime;
1998         rc = inode->i_op->removexattr(dentry, name);
1999         /* ctime should not be updated with server-side time. */
2000         spin_lock(&obj->oo_guard);
2001         inode->i_ctime = *t;
2002         spin_unlock(&obj->oo_guard);
2003         mark_inode_dirty(inode);
2004         return rc;
2005 }
2006
2007 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2008                                      struct dt_object *dt,
2009                                      struct lustre_capa *old,
2010                                      __u64 opc)
2011 {
2012         struct osd_thread_info *info = osd_oti_get(env);
2013         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2014         struct osd_object *obj = osd_dt_obj(dt);
2015         struct osd_device *dev = osd_obj2dev(obj);
2016         struct lustre_capa_key *key = &info->oti_capa_key;
2017         struct lustre_capa *capa = &info->oti_capa;
2018         struct obd_capa *oc;
2019         struct md_capainfo *ci;
2020         int rc;
2021         ENTRY;
2022
2023         if (!dev->od_fl_capa)
2024                 RETURN(ERR_PTR(-ENOENT));
2025
2026         LASSERT(dt_object_exists(dt));
2027         LINVRNT(osd_invariant(obj));
2028
2029         /* renewal sanity check */
2030         if (old && osd_object_auth(env, dt, old, opc))
2031                 RETURN(ERR_PTR(-EACCES));
2032
2033         ci = md_capainfo(env);
2034         if (unlikely(!ci))
2035                 RETURN(ERR_PTR(-ENOENT));
2036
2037         switch (ci->mc_auth) {
2038         case LC_ID_NONE:
2039                 RETURN(NULL);
2040         case LC_ID_PLAIN:
2041                 capa->lc_uid = obj->oo_inode->i_uid;
2042                 capa->lc_gid = obj->oo_inode->i_gid;
2043                 capa->lc_flags = LC_ID_PLAIN;
2044                 break;
2045         case LC_ID_CONVERT: {
2046                 __u32 d[4], s[4];
2047
2048                 s[0] = obj->oo_inode->i_uid;
2049                 get_random_bytes(&(s[1]), sizeof(__u32));
2050                 s[2] = obj->oo_inode->i_gid;
2051                 get_random_bytes(&(s[3]), sizeof(__u32));
2052                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2053                 if (unlikely(rc))
2054                         RETURN(ERR_PTR(rc));
2055
2056                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2057                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2058                 capa->lc_flags = LC_ID_CONVERT;
2059                 break;
2060         }
2061         default:
2062                 RETURN(ERR_PTR(-EINVAL));
2063         }
2064
2065         capa->lc_fid = *fid;
2066         capa->lc_opc = opc;
2067         capa->lc_flags |= dev->od_capa_alg << 24;
2068         capa->lc_timeout = dev->od_capa_timeout;
2069         capa->lc_expiry = 0;
2070
2071         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2072         if (oc) {
2073                 LASSERT(!capa_is_expired(oc));
2074                 RETURN(oc);
2075         }
2076
2077         spin_lock(&capa_lock);
2078         *key = dev->od_capa_keys[1];
2079         spin_unlock(&capa_lock);
2080
2081         capa->lc_keyid = key->lk_keyid;
2082         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2083
2084         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2085         if (rc) {
2086                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2087                 RETURN(ERR_PTR(rc));
2088         }
2089
2090         oc = capa_add(dev->od_capa_hash, capa);
2091         RETURN(oc);
2092 }
2093
2094 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2095 {
2096         int rc;
2097         struct osd_object      *obj    = osd_dt_obj(dt);
2098         struct inode           *inode  = obj->oo_inode;
2099         struct osd_thread_info *info   = osd_oti_get(env);
2100         struct dentry          *dentry = &info->oti_obj_dentry;
2101         struct file            *file   = &info->oti_file;
2102         ENTRY;
2103
2104         dentry->d_inode = inode;
2105         file->f_dentry = dentry;
2106         file->f_mapping = inode->i_mapping;
2107         file->f_op = inode->i_fop;
2108         LOCK_INODE_MUTEX(inode);
2109         rc = file->f_op->fsync(file, dentry, 0);
2110         UNLOCK_INODE_MUTEX(inode);
2111         RETURN(rc);
2112 }
2113
2114 /*
2115  * Get the 64-bit version for an inode.
2116  */
2117 static dt_obj_version_t osd_object_version_get(const struct lu_env *env,
2118                                                struct dt_object *dt)
2119 {
2120         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2121
2122         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
2123                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2124         return LDISKFS_I(inode)->i_fs_version;
2125 }
2126
2127 /*
2128  * Set the 64-bit version and return the old version.
2129  */
2130 static void osd_object_version_set(const struct lu_env *env, struct dt_object *dt,
2131                                    dt_obj_version_t new_version)
2132 {
2133         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2134
2135         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2136                new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2137         LDISKFS_I(inode)->i_fs_version = new_version;
2138         /** Version is set after all inode operations are finished,
2139          *  so we should mark it dirty here */
2140         inode->i_sb->s_op->dirty_inode(inode);
2141 }
2142
2143 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2144                         void **data)
2145 {
2146         struct osd_object *obj = osd_dt_obj(dt);
2147         ENTRY;
2148
2149         *data = (void *)obj->oo_inode;
2150         RETURN(0);
2151 }
2152
2153 /*
2154  * Index operations.
2155  */
2156
2157 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2158                            const struct dt_index_features *feat)
2159 {
2160         struct iam_descr *descr;
2161
2162         if (osd_object_is_root(o))
2163                 return feat == &dt_directory_features;
2164
2165         LASSERT(o->oo_dir != NULL);
2166
2167         descr = o->oo_dir->od_container.ic_descr;
2168         if (feat == &dt_directory_features) {
2169                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2170                         return 1;
2171                 else
2172                         return 0;
2173         } else {
2174                 return
2175                         feat->dif_keysize_min <= descr->id_key_size &&
2176                         descr->id_key_size <= feat->dif_keysize_max &&
2177                         feat->dif_recsize_min <= descr->id_rec_size &&
2178                         descr->id_rec_size <= feat->dif_recsize_max &&
2179                         !(feat->dif_flags & (DT_IND_VARKEY |
2180                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2181                         ergo(feat->dif_flags & DT_IND_UPDATE,
2182                              1 /* XXX check that object (and file system) is
2183                                 * writable */);
2184         }
2185 }
2186
2187 static int osd_iam_container_init(const struct lu_env *env,
2188                                   struct osd_object *obj,
2189                                   struct osd_directory *dir)
2190 {
2191         int result;
2192         struct iam_container *bag;
2193
2194         bag    = &dir->od_container;
2195         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2196         if (result == 0) {
2197                 result = iam_container_setup(bag);
2198                 if (result == 0)
2199                         obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2200                 else
2201                         iam_container_fini(bag);
2202         }
2203         return result;
2204 }
2205
2206
2207 /*
2208  * Concurrency: no external locking is necessary.
2209  */
2210 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2211                          const struct dt_index_features *feat)
2212 {
2213         int result;
2214         int ea_dir = 0;
2215         struct osd_object *obj = osd_dt_obj(dt);
2216         struct osd_device *osd = osd_obj2dev(obj);
2217
2218         LINVRNT(osd_invariant(obj));
2219         LASSERT(dt_object_exists(dt));
2220
2221         if (osd_object_is_root(obj)) {
2222                 dt->do_index_ops = &osd_index_ea_ops;
2223                 result = 0;
2224         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2225                 dt->do_index_ops = &osd_index_ea_ops;
2226                 if (S_ISDIR(obj->oo_inode->i_mode))
2227                         result = 0;
2228                 else
2229                         result = -ENOTDIR;
2230                 ea_dir = 1;
2231         } else if (!osd_has_index(obj)) {
2232                 struct osd_directory *dir;
2233
2234                 OBD_ALLOC_PTR(dir);
2235                 if (dir != NULL) {
2236
2237                         spin_lock(&obj->oo_guard);
2238                         if (obj->oo_dir == NULL)
2239                                 obj->oo_dir = dir;
2240                         else
2241                                 /*
2242                                  * Concurrent thread allocated container data.
2243                                  */
2244                                 OBD_FREE_PTR(dir);
2245                         spin_unlock(&obj->oo_guard);
2246                         /*
2247                          * Now, that we have container data, serialize its
2248                          * initialization.
2249                          */
2250                         down_write(&obj->oo_ext_idx_sem);
2251                         /*
2252                          * recheck under lock.
2253                          */
2254                         if (!osd_has_index(obj))
2255                                 result = osd_iam_container_init(env, obj, dir);
2256                         else
2257                                 result = 0;
2258                         up_write(&obj->oo_ext_idx_sem);
2259                 } else
2260                         result = -ENOMEM;
2261         } else
2262                 result = 0;
2263
2264         if (result == 0 && ea_dir == 0) {
2265                 if (!osd_iam_index_probe(env, obj, feat))
2266                         result = -ENOTDIR;
2267         }
2268         LINVRNT(osd_invariant(obj));
2269
2270         return result;
2271 }
2272
2273 static const struct dt_object_operations osd_obj_ops = {
2274         .do_read_lock    = osd_object_read_lock,
2275         .do_write_lock   = osd_object_write_lock,
2276         .do_read_unlock  = osd_object_read_unlock,
2277         .do_write_unlock = osd_object_write_unlock,
2278         .do_write_locked = osd_object_write_locked,
2279         .do_attr_get     = osd_attr_get,
2280         .do_attr_set     = osd_attr_set,
2281         .do_ah_init      = osd_ah_init,
2282         .do_create       = osd_object_create,
2283         .do_index_try    = osd_index_try,
2284         .do_ref_add      = osd_object_ref_add,
2285         .do_ref_del      = osd_object_ref_del,
2286         .do_xattr_get    = osd_xattr_get,
2287         .do_xattr_set    = osd_xattr_set,
2288         .do_xattr_del    = osd_xattr_del,
2289         .do_xattr_list   = osd_xattr_list,
2290         .do_capa_get     = osd_capa_get,
2291         .do_object_sync  = osd_object_sync,
2292         .do_version_get  = osd_object_version_get,
2293         .do_version_set  = osd_object_version_set,
2294         .do_data_get     = osd_data_get,
2295 };
2296
2297 /**
2298  * dt_object_operations for interoperability mode
2299  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2300  */
2301 static const struct dt_object_operations osd_obj_ea_ops = {
2302         .do_read_lock    = osd_object_read_lock,
2303         .do_write_lock   = osd_object_write_lock,
2304         .do_read_unlock  = osd_object_read_unlock,
2305         .do_write_unlock = osd_object_write_unlock,
2306         .do_write_locked = osd_object_write_locked,
2307         .do_attr_get     = osd_attr_get,
2308         .do_attr_set     = osd_attr_set,
2309         .do_ah_init      = osd_ah_init,
2310         .do_create       = osd_object_ea_create,
2311         .do_index_try    = osd_index_try,
2312         .do_ref_add      = osd_object_ref_add,
2313         .do_ref_del      = osd_object_ref_del,
2314         .do_xattr_get    = osd_xattr_get,
2315         .do_xattr_set    = osd_xattr_set,
2316         .do_xattr_del    = osd_xattr_del,
2317         .do_xattr_list   = osd_xattr_list,
2318         .do_capa_get     = osd_capa_get,
2319         .do_object_sync  = osd_object_sync,
2320         .do_version_get  = osd_object_version_get,
2321         .do_version_set  = osd_object_version_set,
2322         .do_data_get     = osd_data_get,
2323 };
2324
2325 /*
2326  * Body operations.
2327  */
2328
2329 /*
2330  * XXX: Another layering violation for now.
2331  *
2332  * We don't want to use ->f_op->read methods, because generic file write
2333  *
2334  *         - serializes on ->i_sem, and
2335  *
2336  *         - does a lot of extra work like balance_dirty_pages(),
2337  *
2338  * which doesn't work for globally shared files like /last-received.
2339  */
2340 int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
2341 int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
2342                                 loff_t *offs, handle_t *handle);
2343
2344 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
2345                         struct lu_buf *buf, loff_t *pos,
2346                         struct lustre_capa *capa)
2347 {
2348         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2349
2350         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2351                 RETURN(-EACCES);
2352
2353         return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
2354 }
2355
2356 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
2357                          const struct lu_buf *buf, loff_t *pos,
2358                          struct thandle *handle, struct lustre_capa *capa,
2359                          int ignore_quota)
2360 {
2361         struct inode       *inode = osd_dt_obj(dt)->oo_inode;
2362         struct osd_thandle *oh;
2363         ssize_t             result;
2364 #ifdef HAVE_QUOTA_SUPPORT
2365         cfs_cap_t           save = current->cap_effective;
2366 #endif
2367
2368         LASSERT(handle != NULL);
2369
2370         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
2371                 RETURN(-EACCES);
2372
2373         oh = container_of(handle, struct osd_thandle, ot_super);
2374         LASSERT(oh->ot_handle->h_transaction != NULL);
2375 #ifdef HAVE_QUOTA_SUPPORT
2376         if (ignore_quota)
2377                 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2378         else
2379                 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2380 #endif
2381         result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
2382                                              pos, oh->ot_handle);
2383 #ifdef HAVE_QUOTA_SUPPORT
2384         current->cap_effective = save;
2385 #endif
2386         if (result == 0)
2387                 result = buf->lb_len;
2388         return result;
2389 }
2390
2391 static const struct dt_body_operations osd_body_ops = {
2392         .dbo_read  = osd_read,
2393         .dbo_write = osd_write
2394 };
2395
2396
2397 /**
2398  *      delete a (key, value) pair from index \a dt specified by \a key
2399  *
2400  *      \param  dt      osd index object
2401  *      \param  key     key for index
2402  *      \param  rec     record reference
2403  *      \param  handle  transaction handler
2404  *
2405  *      \retval  0  success
2406  *      \retval -ve   failure
2407  */
2408
2409 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2410                                 const struct dt_key *key, struct thandle *handle,
2411                                 struct lustre_capa *capa)
2412 {
2413         struct osd_object     *obj = osd_dt_obj(dt);
2414         struct osd_thandle    *oh;
2415         struct iam_path_descr *ipd;
2416         struct iam_container  *bag = &obj->oo_dir->od_container;
2417         int rc;
2418
2419         ENTRY;
2420
2421         LINVRNT(osd_invariant(obj));
2422         LASSERT(dt_object_exists(dt));
2423         LASSERT(bag->ic_object == obj->oo_inode);
2424         LASSERT(handle != NULL);
2425
2426         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2427                 RETURN(-EACCES);
2428
2429         ipd = osd_idx_ipd_get(env, bag);
2430         if (unlikely(ipd == NULL))
2431                 RETURN(-ENOMEM);
2432
2433         oh = container_of0(handle, struct osd_thandle, ot_super);
2434         LASSERT(oh->ot_handle != NULL);
2435         LASSERT(oh->ot_handle->h_transaction != NULL);
2436
2437         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2438         osd_ipd_put(env, bag, ipd);
2439         LINVRNT(osd_invariant(obj));
2440         RETURN(rc);
2441 }
2442
2443 /**
2444  * Index delete function for interoperability mode (b11826).
2445  * It will remove the directory entry added by osd_index_ea_insert().
2446  * This entry is needed to maintain name->fid mapping.
2447  *
2448  * \param key,  key i.e. file entry to be deleted
2449  *
2450  * \retval   0, on success
2451  * \retval -ve, on error
2452  */
2453 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2454                                const struct dt_key *key, struct thandle *handle,
2455                                struct lustre_capa *capa)
2456 {
2457         struct osd_object          *obj    = osd_dt_obj(dt);
2458         struct inode               *dir    = obj->oo_inode;
2459         struct dentry              *dentry;
2460         struct osd_thandle         *oh;
2461         struct ldiskfs_dir_entry_2 *de;
2462         struct buffer_head         *bh;
2463
2464         int rc;
2465
2466         ENTRY;
2467
2468         LINVRNT(osd_invariant(obj));
2469         LASSERT(dt_object_exists(dt));
2470         LASSERT(handle != NULL);
2471
2472         oh = container_of(handle, struct osd_thandle, ot_super);
2473         LASSERT(oh->ot_handle != NULL);
2474         LASSERT(oh->ot_handle->h_transaction != NULL);
2475
2476         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2477                 RETURN(-EACCES);
2478
2479         dentry = osd_child_dentry_get(env, obj,
2480                                       (char *)key, strlen((char *)key));
2481
2482         down_write(&obj->oo_ext_idx_sem);
2483         bh = ll_ldiskfs_find_entry(dir, dentry, &de);
2484         if (bh) {
2485                 struct osd_thread_info *oti = osd_oti_get(env);
2486                 struct timespec *ctime = &oti->oti_time;
2487                 struct timespec *mtime = &oti->oti_time2;
2488
2489                 *ctime = dir->i_ctime;
2490                 *mtime = dir->i_mtime;
2491                 rc = ldiskfs_delete_entry(oh->ot_handle,
2492                                 dir, de, bh);
2493                 /* xtime should not be updated with server-side time. */
2494                 spin_lock(&obj->oo_guard);
2495                 dir->i_ctime = *ctime;
2496                 dir->i_mtime = *mtime;
2497                 spin_unlock(&obj->oo_guard);
2498                 mark_inode_dirty(dir);
2499                 brelse(bh);
2500         } else
2501                 rc = -ENOENT;
2502
2503         up_write(&obj->oo_ext_idx_sem);
2504         LASSERT(osd_invariant(obj));
2505         RETURN(rc);
2506 }
2507
2508 /**
2509  *      Lookup index for \a key and copy record to \a rec.
2510  *
2511  *      \param  dt      osd index object
2512  *      \param  key     key for index
2513  *      \param  rec     record reference
2514  *
2515  *      \retval  +ve  success : exact mach
2516  *      \retval  0    return record with key not greater than \a key
2517  *      \retval -ve   failure
2518  */
2519 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
2520                                 struct dt_rec *rec, const struct dt_key *key,
2521                                 struct lustre_capa *capa)
2522 {
2523         struct osd_object     *obj = osd_dt_obj(dt);
2524         struct iam_path_descr *ipd;
2525         struct iam_container  *bag = &obj->oo_dir->od_container;
2526         struct osd_thread_info *oti = osd_oti_get(env);
2527         struct iam_iterator    *it = &oti->oti_idx_it;
2528         struct iam_rec *iam_rec;
2529         int rc;
2530         ENTRY;
2531
2532         LASSERT(osd_invariant(obj));
2533         LASSERT(dt_object_exists(dt));
2534         LASSERT(bag->ic_object == obj->oo_inode);
2535
2536         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2537                 RETURN(-EACCES);
2538
2539         ipd = osd_idx_ipd_get(env, bag);
2540         if (IS_ERR(ipd))
2541                 RETURN(-ENOMEM);
2542
2543         /* got ipd now we can start iterator. */
2544         iam_it_init(it, bag, 0, ipd);
2545
2546         rc = iam_it_get(it, (struct iam_key *)key);
2547         if (rc >= 0) {
2548                 if (S_ISDIR(obj->oo_inode->i_mode))
2549                         iam_rec = (struct iam_rec *)oti->oti_fid_packed;
2550                 else
2551                         iam_rec = (struct iam_rec *) rec;
2552
2553                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
2554                 if (S_ISDIR(obj->oo_inode->i_mode))
2555                         osd_fid_unpack((struct lu_fid *) rec,
2556                                        (struct osd_fid_pack *)iam_rec);
2557         }
2558         iam_it_put(it);
2559         iam_it_fini(it);
2560         osd_ipd_put(env, bag, ipd);
2561
2562         LINVRNT(osd_invariant(obj));
2563
2564         RETURN(rc);
2565 }
2566
2567 /**
2568  *      Inserts (key, value) pair in \a dt index object.
2569  *
2570  *      \param  dt      osd index object
2571  *      \param  key     key for index
2572  *      \param  rec     record reference
2573  *      \param  th      transaction handler
2574  *
2575  *      \retval  0  success
2576  *      \retval -ve failure
2577  */
2578 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
2579                                 const struct dt_rec *rec, const struct dt_key *key,
2580                                 struct thandle *th, struct lustre_capa *capa,
2581                                 int ignore_quota)
2582 {
2583         struct osd_object     *obj = osd_dt_obj(dt);
2584         struct iam_path_descr *ipd;
2585         struct osd_thandle    *oh;
2586         struct iam_container  *bag = &obj->oo_dir->od_container;
2587 #ifdef HAVE_QUOTA_SUPPORT
2588         cfs_cap_t              save = current->cap_effective;
2589 #endif
2590         struct osd_thread_info *oti = osd_oti_get(env);
2591         struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_fid_packed;
2592         int rc;
2593
2594         ENTRY;
2595
2596         LINVRNT(osd_invariant(obj));
2597         LASSERT(dt_object_exists(dt));
2598         LASSERT(bag->ic_object == obj->oo_inode);
2599         LASSERT(th != NULL);
2600
2601         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2602                 return -EACCES;
2603
2604         ipd = osd_idx_ipd_get(env, bag);
2605         if (unlikely(ipd == NULL))
2606                 RETURN(-ENOMEM);
2607
2608         oh = container_of0(th, struct osd_thandle, ot_super);
2609         LASSERT(oh->ot_handle != NULL);
2610         LASSERT(oh->ot_handle->h_transaction != NULL);
2611 #ifdef HAVE_QUOTA_SUPPORT
2612         if (ignore_quota)
2613                 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2614         else
2615                 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2616 #endif
2617         if (S_ISDIR(obj->oo_inode->i_mode))
2618                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
2619         else
2620                 iam_rec = (struct iam_rec *) rec;
2621         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
2622                         iam_rec, ipd);
2623 #ifdef HAVE_QUOTA_SUPPORT
2624         current->cap_effective = save;
2625 #endif
2626         osd_ipd_put(env, bag, ipd);
2627         LINVRNT(osd_invariant(obj));
2628         RETURN(rc);
2629 }
2630
2631 /**
2632  * Calls ldiskfs_add_entry() to add directory entry
2633  * into the directory. This is required for
2634  * interoperability mode (b11826)
2635  *
2636  * \retval   0, on success
2637  * \retval -ve, on error
2638  */
2639 static int __osd_ea_add_rec(struct osd_thread_info *info,
2640                             struct osd_object *pobj,
2641                             struct osd_object *cobj,
2642                             const char *name,
2643                             struct thandle *th)
2644 {
2645         struct dentry      *child;
2646         struct osd_thandle *oth;
2647         struct inode       *cinode  = cobj->oo_inode;
2648         int rc;
2649
2650         oth = container_of(th, struct osd_thandle, ot_super);
2651         LASSERT(oth->ot_handle != NULL);
2652         LASSERT(oth->ot_handle->h_transaction != NULL);
2653
2654         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
2655         rc = ldiskfs_add_entry(oth->ot_handle, child, cinode);
2656
2657         RETURN(rc);
2658 }
2659
2660 /**
2661  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
2662  * into the directory.Also sets flags into osd object to
2663  * indicate dot and dotdot are created. This is required for
2664  * interoperability mode (b11826)
2665  *
2666  * \param dir   directory for dot and dotdot fixup.
2667  * \param obj   child object for linking
2668  *
2669  * \retval   0, on success
2670  * \retval -ve, on error
2671  */
2672 static int osd_add_dot_dotdot(struct osd_thread_info *info,
2673                               struct osd_object *dir,
2674                               struct osd_object *obj, const char *name,
2675                               struct thandle *th)
2676 {
2677         struct inode            *parent_dir   = obj->oo_inode;
2678         struct inode            *inode  = dir->oo_inode;
2679         struct osd_thandle      *oth;
2680         int result = 0;
2681
2682         oth = container_of(th, struct osd_thandle, ot_super);
2683         LASSERT(oth->ot_handle->h_transaction != NULL);
2684         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
2685
2686         if (strcmp(name, dot) == 0) {
2687                 if (dir->oo_compat_dot_created) {
2688                         result = -EEXIST;
2689                 } else {
2690                         LASSERT(obj == dir);
2691                         dir->oo_compat_dot_created = 1;
2692                         result = 0;
2693                 }
2694         } else if(strcmp(name, dotdot) == 0) {
2695                 if (!dir->oo_compat_dot_created)
2696                         return -EINVAL;
2697                 if (dir->oo_compat_dotdot_created)
2698                         return __osd_ea_add_rec(info, dir, obj, name, th);
2699
2700                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, inode);
2701                 if (result == 0)
2702                        dir->oo_compat_dotdot_created = 1;
2703         }
2704
2705         return result;
2706 }
2707
2708
2709 /**
2710  * It will call the appropriate osd_add* function and return the
2711  * value, return by respective functions.
2712  */
2713 static int osd_ea_add_rec(const struct lu_env *env,
2714                           struct osd_object *pobj,
2715                           struct osd_object *cobj,
2716                           const char *name,
2717                           struct thandle *th)
2718 {
2719         struct osd_thread_info    *info   = osd_oti_get(env);
2720         int rc;
2721
2722         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
2723                                                    name[2] =='\0')))
2724                 rc = osd_add_dot_dotdot(info, pobj, cobj, name, th);
2725         else
2726                 rc = __osd_ea_add_rec(info, pobj, cobj, name, th);
2727
2728         return rc;
2729 }
2730
2731 /**
2732  * Calls ->lookup() to find dentry. From dentry get inode and
2733  * read inode's ea to get fid. This is required for  interoperability
2734  * mode (b11826)
2735  *
2736  * \retval   0, on success
2737  * \retval -ve, on error
2738  */
2739 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
2740                              struct dt_rec *rec, const struct dt_key *key)
2741 {
2742         struct inode               *dir    = obj->oo_inode;
2743         struct dentry              *dentry;
2744         struct ldiskfs_dir_entry_2 *de;
2745         struct buffer_head         *bh;
2746         struct lu_fid              *fid = (struct lu_fid *) rec;
2747         int ino;
2748         int rc;
2749
2750         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2751
2752         dentry = osd_child_dentry_get(env, obj,
2753                                       (char *)key, strlen((char *)key));
2754
2755         down_read(&obj->oo_ext_idx_sem);
2756         bh = ll_ldiskfs_find_entry(dir, dentry, &de);
2757         if (bh) {
2758                 ino = le32_to_cpu(de->inode);
2759                 brelse(bh);
2760                 rc = osd_ea_fid_get(env, obj, ino, fid);
2761         } else
2762                 rc = -ENOENT;
2763
2764         up_read(&obj->oo_ext_idx_sem);
2765         RETURN (rc);
2766 }
2767
2768 /**
2769  * Find the osd object for given fid.
2770  *
2771  * \param fid need to find the osd object having this fid
2772  *
2773  * \retval osd_object on success
2774  * \retval        -ve on error
2775  */
2776 struct osd_object *osd_object_find(const struct lu_env *env,
2777                                    struct dt_object *dt,
2778                                    const struct lu_fid *fid)
2779 {
2780         struct lu_device         *ludev = dt->do_lu.lo_dev;
2781         struct osd_object        *child = NULL;
2782         struct lu_object         *luch;
2783         struct lu_object         *lo;
2784
2785         luch = lu_object_find(env, ludev, fid, NULL);
2786         if (!IS_ERR(luch)) {
2787                 if (lu_object_exists(luch)) {
2788                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
2789                         if (lo != NULL)
2790                                 child = osd_obj(lo);
2791                         else
2792                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
2793                                                 "lu_object can't be located"
2794                                                 ""DFID"\n", PFID(fid));
2795
2796                         if (child == NULL) {
2797                                 lu_object_put(env, luch);
2798                                 CERROR("Unable to get osd_object\n");
2799                                 child = ERR_PTR(-ENOENT);
2800                         }
2801                 } else {
2802                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
2803                                         "lu_object does not exists "DFID"\n",
2804                                         PFID(fid));
2805                         child = ERR_PTR(-ENOENT);
2806                 }
2807         } else
2808                 child = (void *)luch;
2809
2810         return child;
2811 }
2812
2813 /**
2814  * Put the osd object once done with it.
2815  *
2816  * \param obj osd object that needs to be put
2817  */
2818 static inline void osd_object_put(const struct lu_env *env,
2819                                   struct osd_object *obj)
2820 {
2821         lu_object_put(env, &obj->oo_dt.do_lu);
2822 }
2823
2824 /**
2825  * Index add function for interoperability mode (b11826).
2826  * It will add the directory entry.This entry is needed to
2827  * maintain name->fid mapping.
2828  *
2829  * \param key it is key i.e. file entry to be inserted
2830  * \param rec it is value of given key i.e. fid
2831  *
2832  * \retval   0, on success
2833  * \retval -ve, on error
2834  */
2835 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
2836                                const struct dt_rec *rec,
2837                                const struct dt_key *key, struct thandle *th,
2838                                struct lustre_capa *capa, int ignore_quota)
2839 {
2840         struct osd_object        *obj   = osd_dt_obj(dt);
2841         struct lu_fid            *fid   = (struct lu_fid *) rec;
2842         const char               *name  = (const char *)key;
2843         struct osd_object        *child;
2844 #ifdef HAVE_QUOTA_SUPPORT
2845         cfs_cap_t                 save  = current->cap_effective;
2846 #endif
2847         int rc;
2848
2849         ENTRY;
2850
2851         LASSERT(osd_invariant(obj));
2852         LASSERT(dt_object_exists(dt));
2853         LASSERT(th != NULL);
2854
2855         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2856                 RETURN(-EACCES);
2857
2858         child = osd_object_find(env, dt, fid);
2859         if (!IS_ERR(child)) {
2860                 struct inode *inode = obj->oo_inode;
2861                 struct osd_thread_info *oti = osd_oti_get(env);
2862                 struct timespec *ctime = &oti->oti_time;
2863                 struct timespec *mtime = &oti->oti_time2;
2864
2865                 *ctime = inode->i_ctime;
2866                 *mtime = inode->i_mtime;
2867 #ifdef HAVE_QUOTA_SUPPORT
2868                 if (ignore_quota)
2869                         current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2870                 else
2871                         current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2872 #endif
2873                 down_write(&obj->oo_ext_idx_sem);
2874                 rc = osd_ea_add_rec(env, obj, child, name, th);
2875                 up_write(&obj->oo_ext_idx_sem);
2876 #ifdef HAVE_QUOTA_SUPPORT
2877                 current->cap_effective = save;
2878 #endif
2879                 osd_object_put(env, child);
2880                 /* xtime should not be updated with server-side time. */
2881                 spin_lock(&obj->oo_guard);
2882                 inode->i_ctime = *ctime;
2883                 inode->i_mtime = *mtime;
2884                 spin_unlock(&obj->oo_guard);
2885                 mark_inode_dirty(inode);
2886         } else {
2887                 rc = PTR_ERR(child);
2888         }
2889
2890         LASSERT(osd_invariant(obj));
2891         RETURN(rc);
2892 }
2893
2894 /**
2895  *  Initialize osd Iterator for given osd index object.
2896  *
2897  *  \param  dt      osd index object
2898  */
2899
2900 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
2901                                  struct dt_object *dt,
2902                                  struct lustre_capa *capa)
2903 {
2904         struct osd_it_iam         *it;
2905         struct osd_thread_info *oti = osd_oti_get(env);
2906         struct osd_object     *obj = osd_dt_obj(dt);
2907         struct lu_object      *lo  = &dt->do_lu;
2908         struct iam_path_descr *ipd;
2909         struct iam_container  *bag = &obj->oo_dir->od_container;
2910
2911         LASSERT(lu_object_exists(lo));
2912
2913         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2914                 return ERR_PTR(-EACCES);
2915
2916         it = &oti->oti_it;
2917         ipd = osd_it_ipd_get(env, bag);
2918         if (likely(ipd != NULL)) {
2919                 it->oi_obj = obj;
2920                 it->oi_ipd = ipd;
2921                 lu_object_get(lo);
2922                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
2923                 return (struct dt_it *)it;
2924         }
2925         return ERR_PTR(-ENOMEM);
2926 }
2927
2928 /**
2929  * free given Iterator.
2930  */
2931
2932 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
2933 {
2934         struct osd_it_iam     *it = (struct osd_it_iam *)di;
2935         struct osd_object *obj = it->oi_obj;
2936
2937         iam_it_fini(&it->oi_it);
2938         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
2939         lu_object_put(env, &obj->oo_dt.do_lu);
2940 }
2941
2942 /**
2943  *  Move Iterator to record specified by \a key
2944  *
2945  *  \param  di      osd iterator
2946  *  \param  key     key for index
2947  *
2948  *  \retval +ve  di points to record with least key not larger than key
2949  *  \retval  0   di points to exact matched key
2950  *  \retval -ve  failure
2951  */
2952
2953 static int osd_it_iam_get(const struct lu_env *env,
2954                       struct dt_it *di, const struct dt_key *key)
2955 {
2956         struct osd_it_iam *it = (struct osd_it_iam *)di;
2957
2958         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
2959 }
2960
2961 /**
2962  *  Release Iterator
2963  *
2964  *  \param  di      osd iterator
2965  */
2966
2967 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
2968 {
2969         struct osd_it_iam *it = (struct osd_it_iam *)di;
2970
2971         iam_it_put(&it->oi_it);
2972 }
2973
2974 /**
2975  *  Move iterator by one record
2976  *
2977  *  \param  di      osd iterator
2978  *
2979  *  \retval +1   end of container reached
2980  *  \retval  0   success
2981  *  \retval -ve  failure
2982  */
2983
2984 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
2985 {
2986         struct osd_it_iam *it = (struct osd_it_iam *)di;
2987
2988         return iam_it_next(&it->oi_it);
2989 }
2990
2991 /**
2992  * Return pointer to the key under iterator.
2993  */
2994
2995 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
2996                                  const struct dt_it *di)
2997 {
2998         struct osd_it_iam *it = (struct osd_it_iam *)di;
2999
3000         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3001 }
3002
3003 /**
3004  * Return size of key under iterator (in bytes)
3005  */
3006
3007 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3008 {
3009         struct osd_it_iam *it = (struct osd_it_iam *)di;
3010
3011         return iam_it_key_size(&it->oi_it);
3012 }
3013
3014 static inline void osd_it_append_attrs(struct lu_dirent*ent,
3015                                        __u32 attr,
3016                                        int len,
3017                                        __u16 type)
3018 {
3019         struct luda_type        *lt;
3020         const unsigned           align = sizeof(struct luda_type) - 1;
3021
3022         /* check if file type is required */
3023         if (attr & LUDA_TYPE) {
3024                         len = (len + align) & ~align;
3025
3026                         lt = (void *) ent->lde_name + len;
3027                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3028                         ent->lde_attrs |= LUDA_TYPE;
3029         }
3030
3031         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3032 }
3033
3034 /**
3035  * build lu direct from backend fs dirent.
3036  */
3037
3038 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3039                                       struct lu_fid *fid,
3040                                       __u64 offset,
3041                                       char *name,
3042                                       __u16 namelen,
3043                                       __u16 type,
3044                                       __u32 attr)
3045 {
3046         fid_cpu_to_le(&ent->lde_fid, fid);
3047         ent->lde_attrs = LUDA_FID;
3048
3049         ent->lde_hash = cpu_to_le64(offset);
3050         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3051
3052         strncpy(ent->lde_name, name, namelen);
3053         ent->lde_namelen = cpu_to_le16(namelen);
3054
3055         /* append lustre attributes */
3056         osd_it_append_attrs(ent, attr, namelen, type);
3057 }
3058
3059 /**
3060  * Return pointer to the record under iterator.
3061  */
3062 static int osd_it_iam_rec(const struct lu_env *env,
3063                           const struct dt_it *di,
3064                           struct lu_dirent *lde,
3065                           __u32 attr)
3066 {
3067         struct osd_it_iam *it        = (struct osd_it_iam *)di;
3068         struct osd_thread_info *info = osd_oti_get(env);
3069         struct lu_fid     *fid       = &info->oti_fid;
3070         const struct osd_fid_pack *rec;
3071         char *name;
3072         int namelen;
3073         __u64 hash;
3074         int rc;
3075
3076         name = (char *)iam_it_key_get(&it->oi_it);
3077         if (IS_ERR(name))
3078                 RETURN(PTR_ERR(name));
3079
3080         namelen = iam_it_key_size(&it->oi_it);
3081
3082         rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
3083         if (IS_ERR(rec))
3084                 RETURN(PTR_ERR(rec));
3085
3086         rc = osd_fid_unpack(fid, rec);
3087         if (rc)
3088                 RETURN(rc);
3089
3090         hash = iam_it_store(&it->oi_it);
3091
3092         /* IAM does not store object type in IAM index (dir) */
3093         osd_it_pack_dirent(lde, fid, hash, name, namelen,
3094                            0, LUDA_FID);
3095
3096         return 0;
3097 }
3098
3099 /**
3100  * Returns cookie for current Iterator position.
3101  */
3102 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3103 {
3104         struct osd_it_iam *it = (struct osd_it_iam *)di;
3105
3106         return iam_it_store(&it->oi_it);
3107 }
3108
3109 /**
3110  * Restore iterator from cookie.
3111  *
3112  * \param  di      osd iterator
3113  * \param  hash    Iterator location cookie
3114  *
3115  * \retval +ve  di points to record with least key not larger than key.
3116  * \retval  0   di points to exact matched key
3117  * \retval -ve  failure
3118  */
3119
3120 static int osd_it_iam_load(const struct lu_env *env,
3121                        const struct dt_it *di, __u64 hash)
3122 {
3123         struct osd_it_iam *it = (struct osd_it_iam *)di;
3124
3125         return iam_it_load(&it->oi_it, hash);
3126 }
3127
3128 static const struct dt_index_operations osd_index_iam_ops = {
3129         .dio_lookup = osd_index_iam_lookup,
3130         .dio_insert = osd_index_iam_insert,
3131         .dio_delete = osd_index_iam_delete,
3132         .dio_it     = {
3133                 .init     = osd_it_iam_init,
3134                 .fini     = osd_it_iam_fini,
3135                 .get      = osd_it_iam_get,
3136                 .put      = osd_it_iam_put,
3137                 .next     = osd_it_iam_next,
3138                 .key      = osd_it_iam_key,
3139                 .key_size = osd_it_iam_key_size,
3140                 .rec      = osd_it_iam_rec,
3141                 .store    = osd_it_iam_store,
3142                 .load     = osd_it_iam_load
3143         }
3144 };
3145
3146 /**
3147  * Creates or initializes iterator context.
3148  *
3149  * \retval struct osd_it_ea, iterator structure on success
3150  *
3151  */
3152 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3153                                     struct dt_object *dt,
3154                                     struct lustre_capa *capa)
3155 {
3156         struct osd_object       *obj  = osd_dt_obj(dt);
3157         struct osd_thread_info  *info = osd_oti_get(env);
3158         struct osd_it_ea        *it   = &info->oti_it_ea;
3159         struct lu_object        *lo   = &dt->do_lu;
3160         struct dentry           *obj_dentry = &info->oti_it_dentry;
3161         ENTRY;
3162         LASSERT(lu_object_exists(lo));
3163
3164         obj_dentry->d_inode = obj->oo_inode;
3165         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3166         obj_dentry->d_name.hash = 0;
3167
3168         it->oie_rd_dirent       = 0;
3169         it->oie_it_dirent       = 0;
3170         it->oie_dirent          = NULL;
3171         it->oie_buf             = info->oti_it_ea_buf;
3172         it->oie_obj             = obj;
3173         it->oie_file.f_pos      = 0;
3174         it->oie_file.f_dentry   = obj_dentry;
3175         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
3176         it->oie_file.f_op         = obj->oo_inode->i_fop;
3177         it->oie_file.private_data = NULL;
3178         lu_object_get(lo);
3179         RETURN((struct dt_it *) it);
3180 }
3181
3182 /**
3183  * Destroy or finishes iterator context.
3184  *
3185  * \param di iterator structure to be destroyed
3186  */
3187 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3188 {
3189         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3190         struct osd_object    *obj  = it->oie_obj;
3191         struct inode       *inode  = obj->oo_inode;
3192
3193         ENTRY;
3194         it->oie_file.f_op->release(inode, &it->oie_file);
3195         lu_object_put(env, &obj->oo_dt.do_lu);
3196         EXIT;
3197 }
3198
3199 /**
3200  * It position the iterator at given key, so that next lookup continues from
3201  * that key Or it is similar to dio_it->load() but based on a key,
3202  * rather than file position.
3203  *
3204  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3205  * to the beginning.
3206  *
3207  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3208  */
3209 static int osd_it_ea_get(const struct lu_env *env,
3210                          struct dt_it *di, const struct dt_key *key)
3211 {
3212         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3213
3214         ENTRY;
3215         LASSERT(((const char *)key)[0] == '\0');
3216         it->oie_file.f_pos      = 0;
3217         it->oie_rd_dirent       = 0;
3218         it->oie_it_dirent       = 0;
3219         it->oie_dirent          = NULL;
3220
3221         RETURN(+1);
3222 }
3223
3224 /**
3225  * Does nothing
3226  */
3227 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3228 {
3229 }
3230
3231 /**
3232  * It is called internally by ->readdir(). It fills the
3233  * iterator's in-memory data structure with required
3234  * information i.e. name, namelen, rec_size etc.
3235  *
3236  * \param buf in which information to be filled in.
3237  * \param name name of the file in given dir
3238  *
3239  * \retval 0 on success
3240  * \retval 1 on buffer full
3241  */
3242 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3243                                loff_t offset, __u64 ino,
3244                                unsigned d_type)
3245 {
3246         struct osd_it_ea        *it = (struct osd_it_ea *)buf;
3247         struct osd_it_ea_dirent *ent = it->oie_dirent;
3248         ENTRY;
3249
3250         /* this should never happen */
3251         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3252                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3253                 RETURN(-EIO);
3254         }
3255
3256         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3257             OSD_IT_EA_BUFSIZE)
3258                 RETURN(1);
3259
3260         ent->oied_ino     = ino;
3261         ent->oied_off     = offset;
3262         ent->oied_namelen = namelen;
3263         ent->oied_type    = d_type;
3264
3265         memcpy(ent->oied_name, name, namelen);
3266
3267         it->oie_rd_dirent++;
3268         it->oie_dirent = (void *) ent + size_round(sizeof(*ent) + namelen);
3269         RETURN(0);
3270 }
3271
3272 /**
3273  * Calls ->readdir() to load a directory entry at a time
3274  * and stored it in iterator's in-memory data structure.
3275  *
3276  * \param di iterator's in memory structure
3277  *
3278  * \retval   0 on success
3279  * \retval -ve on error
3280  */
3281 static int osd_ldiskfs_it_fill(const struct dt_it *di)
3282 {
3283         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
3284         struct osd_object  *obj   = it->oie_obj;
3285         struct inode       *inode = obj->oo_inode;
3286         int                result = 0;
3287
3288         ENTRY;
3289         it->oie_dirent = it->oie_buf;
3290         it->oie_rd_dirent = 0;
3291
3292         down_read(&obj->oo_ext_idx_sem);
3293         result = inode->i_fop->readdir(&it->oie_file, it,
3294                                        (filldir_t) osd_ldiskfs_filldir);
3295
3296         up_read(&obj->oo_ext_idx_sem);
3297
3298         if (it->oie_rd_dirent == 0) {
3299                 result = -EIO;
3300         } else {
3301                 it->oie_dirent = it->oie_buf;
3302                 it->oie_it_dirent = 1;
3303         }
3304
3305         RETURN(result);
3306 }
3307
3308 /**
3309  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3310  * to load a directory entry at a time and stored it in
3311  * iterator's in-memory data structure.
3312  *
3313  * \param di iterator's in memory structure
3314  *
3315  * \retval +ve iterator reached to end
3316  * \retval   0 iterator not reached to end
3317  * \retval -ve on error
3318  */
3319 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
3320 {
3321         struct osd_it_ea *it = (struct osd_it_ea *)di;
3322         int rc;
3323
3324         ENTRY;
3325
3326         if (it->oie_it_dirent < it->oie_rd_dirent) {
3327                 it->oie_dirent = (void *) it->oie_dirent +
3328                                  size_round(sizeof(struct osd_it_ea_dirent) +
3329                                             it->oie_dirent->oied_namelen);
3330                 it->oie_it_dirent++;
3331                 RETURN(0);
3332         } else {
3333                 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
3334                         rc = +1;
3335                 else
3336                         rc = osd_ldiskfs_it_fill(di);
3337         }
3338
3339         RETURN(rc);
3340 }
3341
3342 /**
3343  * Returns the key at current position from iterator's in memory structure.
3344  *
3345  * \param di iterator's in memory structure
3346  *
3347  * \retval key i.e. struct dt_key on success
3348  */
3349 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
3350                                     const struct dt_it *di)
3351 {
3352         struct osd_it_ea *it = (struct osd_it_ea *)di;
3353         ENTRY;
3354         RETURN((struct dt_key *)it->oie_dirent->oied_name);
3355 }
3356
3357 /**
3358  * Returns the key's size at current position from iterator's in memory structure.
3359  *
3360  * \param di iterator's in memory structure
3361  *
3362  * \retval key_size i.e. struct dt_key on success
3363  */
3364 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
3365 {
3366         struct osd_it_ea *it = (struct osd_it_ea *)di;
3367         ENTRY;
3368         RETURN(it->oie_dirent->oied_namelen);
3369 }
3370
3371
3372 /**
3373  * Returns the value (i.e. fid/igif) at current position from iterator's
3374  * in memory structure.
3375  *
3376  * \param di struct osd_it_ea, iterator's in memory structure
3377  * \param attr attr requested for dirent.
3378  * \param lde lustre dirent
3379  *
3380  * \retval   0 no error and \param lde has correct lustre dirent.
3381  * \retval -ve on error
3382  */
3383 static inline int osd_it_ea_rec(const struct lu_env *env,
3384                                 const struct dt_it *di,
3385                                 struct lu_dirent *lde,
3386                                 __u32 attr)
3387 {
3388         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
3389         struct osd_object       *obj    = it->oie_obj;
3390         struct osd_thread_info  *info   = osd_oti_get(env);
3391         struct lu_fid           *fid       = &info->oti_fid;
3392         int                      rc;
3393
3394         ENTRY;
3395
3396         rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
3397
3398         if (rc == 0)
3399                 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
3400                                    it->oie_dirent->oied_name,
3401                                    it->oie_dirent->oied_namelen,
3402                                    it->oie_dirent->oied_type,
3403                                    attr);
3404         RETURN(rc);
3405 }
3406
3407 /**
3408  * Returns a cookie for current position of the iterator head, so that
3409  * user can use this cookie to load/start the iterator next time.
3410  *
3411  * \param di iterator's in memory structure
3412  *
3413  * \retval cookie for current position, on success
3414  */
3415 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
3416 {
3417         struct osd_it_ea *it = (struct osd_it_ea *)di;
3418         ENTRY;
3419         RETURN(it->oie_dirent->oied_off);
3420 }
3421
3422 /**
3423  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3424  * to load a directory entry at a time and stored it i inn,
3425  * in iterator's in-memory data structure.
3426  *
3427  * \param di struct osd_it_ea, iterator's in memory structure
3428  *
3429  * \retval +ve on success
3430  * \retval -ve on error
3431  */
3432 static int osd_it_ea_load(const struct lu_env *env,
3433                           const struct dt_it *di, __u64 hash)
3434 {
3435         struct osd_it_ea *it = (struct osd_it_ea *)di;
3436         int rc;
3437
3438         ENTRY;
3439         it->oie_file.f_pos = hash;
3440
3441         rc =  osd_ldiskfs_it_fill(di);
3442         if (rc == 0)
3443                 rc = +1;
3444
3445         RETURN(rc);
3446 }
3447
3448 /**
3449  * Index lookup function for interoperability mode (b11826).
3450  *
3451  * \param key,  key i.e. file name to be searched
3452  *
3453  * \retval +ve, on success
3454  * \retval -ve, on error
3455  */
3456 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
3457                                struct dt_rec *rec, const struct dt_key *key,
3458                                struct lustre_capa *capa)
3459 {
3460         struct osd_object *obj = osd_dt_obj(dt);
3461         int rc = 0;
3462
3463         ENTRY;
3464
3465         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
3466         LINVRNT(osd_invariant(obj));
3467
3468         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3469                 return -EACCES;
3470
3471         rc = osd_ea_lookup_rec(env, obj, rec, key);
3472
3473         if (rc == 0)
3474                 rc = +1;
3475         RETURN(rc);
3476 }
3477
3478 /**
3479  * Index and Iterator operations for interoperability
3480  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3481  */
3482 static const struct dt_index_operations osd_index_ea_ops = {
3483         .dio_lookup = osd_index_ea_lookup,
3484         .dio_insert = osd_index_ea_insert,
3485         .dio_delete = osd_index_ea_delete,
3486         .dio_it     = {
3487                 .init     = osd_it_ea_init,
3488                 .fini     = osd_it_ea_fini,
3489                 .get      = osd_it_ea_get,
3490                 .put      = osd_it_ea_put,
3491                 .next     = osd_it_ea_next,
3492                 .key      = osd_it_ea_key,
3493                 .key_size = osd_it_ea_key_size,
3494                 .rec      = osd_it_ea_rec,
3495                 .store    = osd_it_ea_store,
3496                 .load     = osd_it_ea_load
3497         }
3498 };
3499
3500 static void *osd_key_init(const struct lu_context *ctx,
3501                           struct lu_context_key *key)
3502 {
3503         struct osd_thread_info *info;
3504
3505         OBD_ALLOC_PTR(info);
3506         if (info != NULL) {
3507                 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3508                 if (info->oti_it_ea_buf != NULL) {
3509                         info->oti_env = container_of(ctx, struct lu_env,
3510                                                      le_ctx);
3511                 } else {
3512                         OBD_FREE_PTR(info);
3513                         info = ERR_PTR(-ENOMEM);
3514                 }
3515         } else {
3516                 info = ERR_PTR(-ENOMEM);
3517         }
3518         return info;
3519 }
3520
3521 static void osd_key_fini(const struct lu_context *ctx,
3522                          struct lu_context_key *key, void* data)
3523 {
3524         struct osd_thread_info *info = data;
3525
3526         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3527         OBD_FREE_PTR(info);
3528 }
3529
3530 static void osd_key_exit(const struct lu_context *ctx,
3531                          struct lu_context_key *key, void *data)
3532 {
3533         struct osd_thread_info *info = data;
3534
3535         LASSERT(info->oti_r_locks == 0);
3536         LASSERT(info->oti_w_locks == 0);
3537         LASSERT(info->oti_txns    == 0);
3538 }
3539
3540 /* type constructor/destructor: osd_type_init, osd_type_fini */
3541 LU_TYPE_INIT_FINI(osd, &osd_key);
3542
3543 static struct lu_context_key osd_key = {
3544         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
3545         .lct_init = osd_key_init,
3546         .lct_fini = osd_key_fini,
3547         .lct_exit = osd_key_exit
3548 };
3549
3550
3551 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
3552                            const char *name, struct lu_device *next)
3553 {
3554         int rc;
3555         struct lu_context *ctx;
3556
3557         /* context for commit hooks */
3558         ctx = &osd_dev(d)->od_env_for_commit.le_ctx;
3559         rc = lu_context_init(ctx, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
3560         if (rc == 0) {
3561                 rc = osd_procfs_init(osd_dev(d), name);
3562                 ctx->lc_cookie = 0x3;
3563         }
3564         return rc;
3565 }
3566
3567 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
3568 {
3569         struct osd_thread_info *info = osd_oti_get(env);
3570         ENTRY;
3571         if (o->od_obj_area != NULL) {
3572                 lu_object_put(env, &o->od_obj_area->do_lu);
3573                 o->od_obj_area = NULL;
3574         }
3575         osd_oi_fini(info, &o->od_oi);
3576
3577         RETURN(0);
3578 }
3579
3580 static int osd_mount(const struct lu_env *env,
3581                      struct osd_device *o, struct lustre_cfg *cfg)
3582 {
3583         struct lustre_mount_info *lmi;
3584         const char               *dev  = lustre_cfg_string(cfg, 0);
3585         struct lustre_disk_data  *ldd;
3586         struct lustre_sb_info    *lsi;
3587
3588         ENTRY;
3589         if (o->od_mount != NULL) {
3590                 CERROR("Already mounted (%s)\n", dev);
3591                 RETURN(-EEXIST);
3592         }
3593
3594         /* get mount */
3595         lmi = server_get_mount(dev);
3596         if (lmi == NULL) {
3597                 CERROR("Cannot get mount info for %s!\n", dev);
3598                 RETURN(-EFAULT);
3599         }
3600
3601         LASSERT(lmi != NULL);
3602         /* save lustre_mount_info in dt_device */
3603         o->od_mount = lmi;
3604
3605         lsi = s2lsi(lmi->lmi_sb);
3606         ldd = lsi->lsi_ldd;
3607
3608         if (ldd->ldd_flags & LDD_F_IAM_DIR) {
3609                 o->od_iop_mode = 0;
3610                 LCONSOLE_WARN("OSD: IAM mode enabled\n");
3611         } else
3612                 o->od_iop_mode = 1;
3613
3614         o->od_obj_area = NULL;
3615         RETURN(0);
3616 }
3617
3618 static struct lu_device *osd_device_fini(const struct lu_env *env,
3619                                          struct lu_device *d)
3620 {
3621         int rc;
3622         ENTRY;
3623
3624         shrink_dcache_sb(osd_sb(osd_dev(d)));
3625         osd_sync(env, lu2dt_dev(d));
3626
3627         rc = osd_procfs_fini(osd_dev(d));
3628         if (rc) {
3629                 CERROR("proc fini error %d \n", rc);
3630                 RETURN (ERR_PTR(rc));
3631         }
3632
3633         if (osd_dev(d)->od_mount)
3634                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
3635                                  osd_dev(d)->od_mount->lmi_mnt);
3636         osd_dev(d)->od_mount = NULL;
3637
3638         lu_context_fini(&osd_dev(d)->od_env_for_commit.le_ctx);
3639         RETURN(NULL);
3640 }
3641
3642 static struct lu_device *osd_device_alloc(const struct lu_env *env,
3643                                           struct lu_device_type *t,
3644                                           struct lustre_cfg *cfg)
3645 {
3646         struct lu_device  *l;
3647         struct osd_device *o;
3648
3649         OBD_ALLOC_PTR(o);
3650         if (o != NULL) {
3651                 int result;
3652
3653                 result = dt_device_init(&o->od_dt_dev, t);
3654                 if (result == 0) {
3655                         l = osd2lu_dev(o);
3656                         l->ld_ops = &osd_lu_ops;
3657                         o->od_dt_dev.dd_ops = &osd_dt_ops;
3658                         spin_lock_init(&o->od_osfs_lock);
3659                         o->od_osfs_age = cfs_time_shift_64(-1000);
3660                         o->od_capa_hash = init_capa_hash();
3661                         if (o->od_capa_hash == NULL) {
3662                                 dt_device_fini(&o->od_dt_dev);
3663                                 l = ERR_PTR(-ENOMEM);
3664                         }
3665                 } else
3666                         l = ERR_PTR(result);
3667
3668                 if (IS_ERR(l))
3669                         OBD_FREE_PTR(o);
3670         } else
3671                 l = ERR_PTR(-ENOMEM);
3672         return l;
3673 }
3674
3675 static struct lu_device *osd_device_free(const struct lu_env *env,
3676                                          struct lu_device *d)
3677 {
3678         struct osd_device *o = osd_dev(d);
3679         ENTRY;
3680
3681         cleanup_capa_hash(o->od_capa_hash);
3682         dt_device_fini(&o->od_dt_dev);
3683         OBD_FREE_PTR(o);
3684         RETURN(NULL);
3685 }
3686
3687 static int osd_process_config(const struct lu_env *env,
3688                               struct lu_device *d, struct lustre_cfg *cfg)
3689 {
3690         struct osd_device *o = osd_dev(d);
3691         int err;
3692         ENTRY;
3693
3694         switch(cfg->lcfg_command) {
3695         case LCFG_SETUP:
3696                 err = osd_mount(env, o, cfg);
3697                 break;
3698         case LCFG_CLEANUP:
3699                 err = osd_shutdown(env, o);
3700                 break;
3701         default:
3702                 err = -ENOSYS;
3703         }
3704
3705         RETURN(err);
3706 }
3707
3708 static int osd_recovery_complete(const struct lu_env *env,
3709                                  struct lu_device *d)
3710 {
3711         RETURN(0);
3712 }
3713
3714 static int osd_prepare(const struct lu_env *env,
3715                        struct lu_device *pdev,
3716                        struct lu_device *dev)
3717 {
3718         struct osd_device *osd = osd_dev(dev);
3719         struct lustre_sb_info *lsi;
3720         struct lustre_disk_data *ldd;
3721         struct lustre_mount_info  *lmi;
3722         struct osd_thread_info *oti = osd_oti_get(env);
3723         struct dt_object *d;
3724         int result;
3725
3726         ENTRY;
3727         /* 1. initialize oi before any file create or file open */
3728         result = osd_oi_init(oti, &osd->od_oi,
3729                              &osd->od_dt_dev, lu2md_dev(pdev));
3730         if (result != 0)
3731                 RETURN(result);
3732
3733         lmi = osd->od_mount;
3734         lsi = s2lsi(lmi->lmi_sb);
3735         ldd = lsi->lsi_ldd;
3736
3737         /* 2. setup local objects */
3738         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
3739         if (result)
3740                 goto out;
3741
3742         /* 3. open remote object dir */
3743         d = dt_store_open(env, lu2dt_dev(dev), "",
3744                           remote_obj_dir, &oti->oti_fid);
3745         if (!IS_ERR(d)) {
3746                 osd->od_obj_area = d;
3747                 result = 0;
3748         } else {
3749                 result = PTR_ERR(d);
3750                 osd->od_obj_area = NULL;
3751         }
3752
3753 out:
3754         RETURN(result);
3755 }
3756
3757 static const struct lu_object_operations osd_lu_obj_ops = {
3758         .loo_object_init      = osd_object_init,
3759         .loo_object_delete    = osd_object_delete,
3760         .loo_object_release   = osd_object_release,
3761         .loo_object_free      = osd_object_free,
3762         .loo_object_print     = osd_object_print,
3763         .loo_object_invariant = osd_object_invariant
3764 };
3765
3766 static const struct lu_device_operations osd_lu_ops = {
3767         .ldo_object_alloc      = osd_object_alloc,
3768         .ldo_process_config    = osd_process_config,
3769         .ldo_recovery_complete = osd_recovery_complete,
3770         .ldo_prepare           = osd_prepare,
3771 };
3772
3773 static const struct lu_device_type_operations osd_device_type_ops = {
3774         .ldto_init = osd_type_init,
3775         .ldto_fini = osd_type_fini,
3776
3777         .ldto_start = osd_type_start,
3778         .ldto_stop  = osd_type_stop,
3779
3780         .ldto_device_alloc = osd_device_alloc,
3781         .ldto_device_free  = osd_device_free,
3782
3783         .ldto_device_init    = osd_device_init,
3784         .ldto_device_fini    = osd_device_fini
3785 };
3786
3787 static struct lu_device_type osd_device_type = {
3788         .ldt_tags     = LU_DEVICE_DT,
3789         .ldt_name     = LUSTRE_OSD_NAME,
3790         .ldt_ops      = &osd_device_type_ops,
3791         .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3792 };
3793
3794 /*
3795  * lprocfs legacy support.
3796  */
3797 static struct obd_ops osd_obd_device_ops = {
3798         .o_owner = THIS_MODULE
3799 };
3800
3801 static struct lu_local_obj_desc llod_osd_rem_obj_dir = {
3802         .llod_name      = remote_obj_dir,
3803         .llod_oid       = OSD_REM_OBJ_DIR_OID,
3804         .llod_is_index  = 1,
3805         .llod_feat      = &dt_directory_features,
3806 };
3807
3808 static int __init osd_mod_init(void)
3809 {
3810         struct lprocfs_static_vars lvars;
3811
3812         osd_oi_mod_init();
3813         llo_local_obj_register(&llod_osd_rem_obj_dir);
3814         lprocfs_osd_init_vars(&lvars);
3815         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
3816                                    LUSTRE_OSD_NAME, &osd_device_type);
3817 }
3818
3819 static void __exit osd_mod_exit(void)
3820 {
3821         llo_local_obj_unregister(&llod_osd_rem_obj_dir);
3822         class_unregister_type(LUSTRE_OSD_NAME);
3823 }
3824
3825 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3826 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
3827 MODULE_LICENSE("GPL");
3828
3829 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);