lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65
  66 #include "osd_internal.h"
  67
  68 /* llo_* api support */
  69 #include <md_object.h>
  70 #include <lustre_quota.h>
  71
  72 int ldiskfs_pdo = 1;
  73 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  74                 "ldiskfs with parallel directory operations");
  75
  76 int ldiskfs_track_declares_assert;
  77 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  78                 "LBUG during tracking of declares");
  79
  80 static const char dot[] = ".";
  81 static const char dotdot[] = "..";
  82 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  83
  84 static const struct lu_object_operations      osd_lu_obj_ops;
  85 static const struct dt_object_operations      osd_obj_ops;
  86 static const struct dt_object_operations      osd_obj_ea_ops;
  87 static const struct dt_object_operations      osd_obj_otable_it_ops;
  88 static const struct dt_index_operations       osd_index_iam_ops;
  89 static const struct dt_index_operations       osd_index_ea_ops;
  90
  91 int osd_trans_declare_op2rb[] = {
  92         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
  93         [OSD_OT_PUNCH]          = OSD_OT_MAX,
  94         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
  95         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
  96         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
  97         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
  98         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
  99         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 100         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 101         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 102         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 103 };
 104
 105 static int osd_has_index(const struct osd_object *obj)
 106 {
 107         return obj->oo_dt.do_index_ops != NULL;
 108 }
 109
 110 static int osd_object_invariant(const struct lu_object *l)
 111 {
 112         return osd_invariant(osd_obj(l));
 113 }
 114
 115 /*
 116  * Concurrency: doesn't matter
 117  */
 118 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 119 {
 120         return osd_oti_get(env)->oti_r_locks > 0;
 121 }
 122
 123 /*
 124  * Concurrency: doesn't matter
 125  */
 126 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 127 {
 128         struct osd_thread_info *oti = osd_oti_get(env);
 129         return oti->oti_w_locks > 0 && o->oo_owner == env;
 130 }
 131
 132 /*
 133  * Concurrency: doesn't access mutable data
 134  */
 135 static int osd_root_get(const struct lu_env *env,
 136                         struct dt_device *dev, struct lu_fid *f)
 137 {
 138         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 139         return 0;
 140 }
 141
 142 /*
 143  * OSD object methods.
 144  */
 145
 146 /*
 147  * Concurrency: no concurrent access is possible that early in object
 148  * life-cycle.
 149  */
 150 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 151                                           const struct lu_object_header *hdr,
 152                                           struct lu_device *d)
 153 {
 154         struct osd_object *mo;
 155
 156         OBD_ALLOC_PTR(mo);
 157         if (mo != NULL) {
 158                 struct lu_object *l;
 159
 160                 l = &mo->oo_dt.do_lu;
 161                 dt_object_init(&mo->oo_dt, NULL, d);
 162                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 163                 l->lo_ops = &osd_lu_obj_ops;
 164                 init_rwsem(&mo->oo_sem);
 165                 init_rwsem(&mo->oo_ext_idx_sem);
 166                 spin_lock_init(&mo->oo_guard);
 167                 return l;
 168         } else {
 169                 return NULL;
 170         }
 171 }
 172
 173 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
 174                                   const char *name, void *buf, int len)
 175 {
 176         dentry->d_inode = inode;
 177         dentry->d_sb = inode->i_sb;
 178         return inode->i_op->getxattr(dentry, name, buf, len);
 179 }
 180
 181 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 182                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 183 {
 184         int rc;
 185
 186         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 187         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 188                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 189         if (rc > 0) {
 190                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 191                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 192                 rc = 0;
 193                 lustre_lma_swab(lma);
 194                 /* Check LMA compatibility */
 195                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 196                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 197                               "for fid = "DFID", ino = %lu\n",
 198                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 199                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 200                               PFID(&lma->lma_self_fid), inode->i_ino);
 201                         rc = -EOPNOTSUPP;
 202                 }
 203         } else if (rc == 0) {
 204                 rc = -ENODATA;
 205         }
 206
 207         return rc;
 208 }
 209
 210 /*
 211  * retrieve object from backend ext fs.
 212  **/
 213 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 214                        struct osd_inode_id *id)
 215 {
 216         struct inode *inode = NULL;
 217
 218         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 219         if (IS_ERR(inode)) {
 220                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 221                        id->oii_ino, PTR_ERR(inode));
 222         } else if (id->oii_gen != OSD_OII_NOGEN &&
 223                    inode->i_generation != id->oii_gen) {
 224                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 225                        "gen1 = %u\n",
 226                        id->oii_ino, id->oii_gen, inode->i_generation);
 227                 iput(inode);
 228                 inode = ERR_PTR(-ESTALE);
 229         } else if (inode->i_nlink == 0) {
 230                 /* due to parallel readdir and unlink,
 231                 * we can have dead inode here. */
 232                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 233                 make_bad_inode(inode);
 234                 iput(inode);
 235                 inode = ERR_PTR(-ESTALE);
 236         } else if (is_bad_inode(inode)) {
 237                 CWARN("%.16s: bad inode: ino = %u\n",
 238                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 239                 iput(inode);
 240                 inode = ERR_PTR(-ENOENT);
 241         } else {
 242                 if (id->oii_gen == OSD_OII_NOGEN)
 243                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 244
 245                 /* Do not update file c/mtime in ldiskfs.
 246                  * NB: we don't have any lock to protect this because we don't
 247                  * have reference on osd_object now, but contention with
 248                  * another lookup + attr_set can't happen in the tiny window
 249                  * between if (...) and set S_NOCMTIME. */
 250                 if (!(inode->i_flags & S_NOCMTIME))
 251                         inode->i_flags |= S_NOCMTIME;
 252         }
 253         return inode;
 254 }
 255
 256 static struct inode *
 257 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 258              struct osd_inode_id *id, struct lu_fid *fid)
 259 {
 260         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 261         struct inode            *inode;
 262         int                      rc;
 263
 264         inode = osd_iget(info, dev, id);
 265         if (IS_ERR(inode))
 266                 return inode;
 267
 268         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 269         if (rc == 0) {
 270                 *fid = lma->lma_self_fid;
 271         } else if (rc == -ENODATA) {
 272                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 273                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 274                 else
 275                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 276         } else {
 277                 iput(inode);
 278                 inode = ERR_PTR(rc);
 279         }
 280         return inode;
 281 }
 282
 283 static struct inode *
 284 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 285                 struct osd_inode_id *id, const struct lu_fid *fid)
 286 {
 287         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 288         struct inode            *inode;
 289         int                      rc;
 290
 291         inode = osd_iget(info, dev, id);
 292         if (IS_ERR(inode))
 293                 return inode;
 294
 295         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 296         if (rc == -ENODATA)
 297                 return inode;
 298
 299         if (rc != 0) {
 300                 iput(inode);
 301                 return ERR_PTR(rc);
 302         }
 303
 304         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 305                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 306                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 307                 iput(inode);
 308                 return ERR_PTR(-EREMCHG);
 309         }
 310
 311         return inode;
 312 }
 313
 314 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 315                           const struct lu_fid *fid,
 316                           const struct lu_object_conf *conf)
 317 {
 318         struct osd_thread_info *info;
 319         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 320         struct osd_device      *dev;
 321         struct osd_idmap_cache *oic;
 322         struct osd_inode_id    *id;
 323         struct inode           *inode;
 324         struct osd_scrub       *scrub;
 325         struct scrub_file      *sf;
 326         int                     result;
 327         int                     verify = 0;
 328         ENTRY;
 329
 330         LINVRNT(osd_invariant(obj));
 331         LASSERT(obj->oo_inode == NULL);
 332         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 333
 334         dev = osd_dev(ldev);
 335         scrub = &dev->od_scrub;
 336         sf = &scrub->os_file;
 337         info = osd_oti_get(env);
 338         LASSERT(info);
 339         oic = &info->oti_cache;
 340
 341         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 342                 RETURN(-ENOENT);
 343
 344         /* Search order: 1. per-thread cache. */
 345         if (lu_fid_eq(fid, &oic->oic_fid)) {
 346                 id = &oic->oic_lid;
 347                 goto iget;
 348         }
 349
 350         id = &info->oti_id;
 351         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 352                 /* Search order: 2. OI scrub pending list. */
 353                 result = osd_oii_lookup(dev, fid, id);
 354                 if (result == 0)
 355                         goto iget;
 356         }
 357
 358         if (sf->sf_flags & SF_INCONSISTENT)
 359                 verify = 1;
 360
 361         /*
 362          * Objects are created as locking anchors or place holders for objects
 363          * yet to be created. No need to osd_oi_lookup() at here because FID
 364          * shouldn't never be re-used, if it's really a duplicate FID from
 365          * unexpected reason, we should be able to detect it later by calling
 366          * do_create->osd_oi_insert()
 367          */
 368         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 369                 GOTO(out, result = 0);
 370
 371         /* Search order: 3. OI files. */
 372         result = osd_oi_lookup(info, dev, fid, id, true);
 373         if (result == -ENOENT) {
 374                 if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
 375                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 376                                       sf->sf_oi_bitmap))
 377                         GOTO(out, result = 0);
 378
 379                 goto trigger;
 380         }
 381
 382         if (result != 0)
 383                 GOTO(out, result);
 384
 385 iget:
 386         if (verify == 0)
 387                 inode = osd_iget(info, dev, id);
 388         else
 389                 inode = osd_iget_verify(info, dev, id, fid);
 390         if (IS_ERR(inode)) {
 391                 result = PTR_ERR(inode);
 392                 if (result == -ENOENT || result == -ESTALE) {
 393                         fid_zero(&oic->oic_fid);
 394                         result = 0;
 395                 } else if (result == -EREMCHG) {
 396
 397 trigger:
 398                         if (thread_is_running(&scrub->os_thread)) {
 399                                 result = -EINPROGRESS;
 400                         } else if (!dev->od_noscrub) {
 401                                 result = osd_scrub_start(dev);
 402                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 403                                                "for "DFID", rc = %d [1]\n",
 404                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 405                                                s_volume_name,PFID(fid), result);
 406                                 if (result == 0 || result == -EALREADY)
 407                                         result = -EINPROGRESS;
 408                                 else
 409                                         result = -EREMCHG;
 410                         }
 411                 }
 412
 413                 GOTO(out, result);
 414         }
 415
 416         obj->oo_inode = inode;
 417         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 418
 419         obj->oo_compat_dot_created = 1;
 420         obj->oo_compat_dotdot_created = 1;
 421
 422         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 423                 GOTO(out, result = 0);
 424
 425         LASSERT(obj->oo_hl_head == NULL);
 426         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 427         if (obj->oo_hl_head == NULL) {
 428                 obj->oo_inode = NULL;
 429                 iput(inode);
 430                 GOTO(out, result = -ENOMEM);
 431         }
 432         GOTO(out, result = 0);
 433
 434 out:
 435         LINVRNT(osd_invariant(obj));
 436         return result;
 437 }
 438
 439 /*
 440  * Concurrency: shouldn't matter.
 441  */
 442 static void osd_object_init0(struct osd_object *obj)
 443 {
 444         LASSERT(obj->oo_inode != NULL);
 445         obj->oo_dt.do_body_ops = &osd_body_ops;
 446         obj->oo_dt.do_lu.lo_header->loh_attr |=
 447                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 448 }
 449
 450 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 451 {
 452         struct osd_thread_info  *info   = osd_oti_get(env);
 453         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 454         int                     rc;
 455         ENTRY;
 456
 457         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 458         rc = __osd_xattr_get(obj->oo_inode, &info->oti_obj_dentry,
 459                              XATTR_NAME_LMA, info->oti_mdt_attrs_old,
 460                              LMA_OLD_SIZE);
 461         if (rc > 0) {
 462                 rc = 0;
 463                 lustre_lma_swab(lma);
 464                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 465                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 466                         rc = -EOPNOTSUPP;
 467                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 468                               "fid = "DFID", ino = %lu: rc = %d\n",
 469                               osd_obj2dev(obj)->od_svname,
 470                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 471                               PFID(lu_object_fid(&obj->oo_dt.do_lu)),
 472                               obj->oo_inode->i_ino, rc);
 473                 }
 474                 if (unlikely(!lu_fid_eq(lu_object_fid(&obj->oo_dt.do_lu),
 475                                         &lma->lma_self_fid))) {
 476                         CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 477                                osd_obj2dev(obj)->od_svname,
 478                                PFID(lu_object_fid(&obj->oo_dt.do_lu)),
 479                                PFID(&lma->lma_self_fid));
 480                         if (obj->oo_inode != NULL) {
 481                                 iput(obj->oo_inode);
 482                                 obj->oo_inode = NULL;
 483                         }
 484                         rc = -ESTALE;
 485                 }
 486         } else if (rc == -ENODATA) {
 487                 /* haven't initialize LMA xattr */
 488                 rc = 0;
 489         }
 490
 491         RETURN(rc);
 492 }
 493
 494 /*
 495  * Concurrency: no concurrent access is possible that early in object
 496  * life-cycle.
 497  */
 498 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 499                            const struct lu_object_conf *conf)
 500 {
 501         struct osd_object *obj = osd_obj(l);
 502         int result;
 503
 504         LINVRNT(osd_invariant(obj));
 505
 506         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 507                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 508                 l->lo_header->loh_attr |= LOHA_EXISTS;
 509                 return 0;
 510         }
 511
 512         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 513         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 514         if (result == 0 && obj->oo_inode != NULL) {
 515                 result = osd_check_lma(env, obj);
 516                 if (result != 0)
 517                         return result;
 518
 519                 osd_object_init0(obj);
 520         }
 521
 522         LINVRNT(osd_invariant(obj));
 523         return result;
 524 }
 525
 526 /*
 527  * Concurrency: no concurrent access is possible that late in object
 528  * life-cycle.
 529  */
 530 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 531 {
 532         struct osd_object *obj = osd_obj(l);
 533
 534         LINVRNT(osd_invariant(obj));
 535
 536         dt_object_fini(&obj->oo_dt);
 537         if (obj->oo_hl_head != NULL)
 538                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 539         OBD_FREE_PTR(obj);
 540 }
 541
 542 /*
 543  * Concurrency: no concurrent access is possible that late in object
 544  * life-cycle.
 545  */
 546 static void osd_index_fini(struct osd_object *o)
 547 {
 548         struct iam_container *bag;
 549
 550         if (o->oo_dir != NULL) {
 551                 bag = &o->oo_dir->od_container;
 552                 if (o->oo_inode != NULL) {
 553                         if (bag->ic_object == o->oo_inode)
 554                                 iam_container_fini(bag);
 555                 }
 556                 OBD_FREE_PTR(o->oo_dir);
 557                 o->oo_dir = NULL;
 558         }
 559 }
 560
 561 /*
 562  * Concurrency: no concurrent access is possible that late in object
 563  * life-cycle (for all existing callers, that is. New callers have to provide
 564  * their own locking.)
 565  */
 566 static int osd_inode_unlinked(const struct inode *inode)
 567 {
 568         return inode->i_nlink == 0;
 569 }
 570
 571 enum {
 572         OSD_TXN_OI_DELETE_CREDITS    = 20,
 573         OSD_TXN_INODE_DELETE_CREDITS = 20
 574 };
 575
 576 /*
 577  * Journal
 578  */
 579
 580 #if OSD_THANDLE_STATS
 581 /**
 582  * Set time when the handle is allocated
 583  */
 584 static void osd_th_alloced(struct osd_thandle *oth)
 585 {
 586         oth->oth_alloced = cfs_time_current();
 587 }
 588
 589 /**
 590  * Set time when the handle started
 591  */
 592 static void osd_th_started(struct osd_thandle *oth)
 593 {
 594         oth->oth_started = cfs_time_current();
 595 }
 596
 597 /**
 598  * Helper function to convert time interval to microseconds packed in
 599  * long int.
 600  */
 601 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 602 {
 603         struct timeval val;
 604
 605         cfs_duration_usec(cfs_time_sub(end, start), &val);
 606         return val.tv_sec * 1000000 + val.tv_usec;
 607 }
 608
 609 /**
 610  * Check whether the we deal with this handle for too long.
 611  */
 612 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 613                                 cfs_time_t alloced, cfs_time_t started,
 614                                 cfs_time_t closed)
 615 {
 616         cfs_time_t now = cfs_time_current();
 617
 618         LASSERT(dev != NULL);
 619
 620         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 621                             interval_to_usec(alloced, started));
 622         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 623                             interval_to_usec(started, closed));
 624         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 625                             interval_to_usec(closed, now));
 626
 627         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 628                 CWARN("transaction handle %p was open for too long: "
 629                       "now "CFS_TIME_T" ,"
 630                       "alloced "CFS_TIME_T" ,"
 631                       "started "CFS_TIME_T" ,"
 632                       "closed "CFS_TIME_T"\n",
 633                       oth, now, alloced, started, closed);
 634                 libcfs_debug_dumpstack(NULL);
 635         }
 636 }
 637
 638 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 639 {                                                                       \
 640         cfs_time_t __closed = cfs_time_current();                       \
 641         cfs_time_t __alloced = oth->oth_alloced;                        \
 642         cfs_time_t __started = oth->oth_started;                        \
 643                                                                         \
 644         expr;                                                           \
 645         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 646 }
 647
 648 #else /* OSD_THANDLE_STATS */
 649
 650 #define osd_th_alloced(h)                  do {} while(0)
 651 #define osd_th_started(h)                  do {} while(0)
 652 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 653
 654 #endif /* OSD_THANDLE_STATS */
 655
 656 /*
 657  * Concurrency: doesn't access mutable data.
 658  */
 659 static int osd_param_is_not_sane(const struct osd_device *dev,
 660                                  const struct thandle *th)
 661 {
 662         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 663
 664         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 665 }
 666
 667 /*
 668  * Concurrency: shouldn't matter.
 669  */
 670 static void osd_trans_commit_cb(struct super_block *sb,
 671                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 672 {
 673         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 674         struct thandle     *th  = &oh->ot_super;
 675         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 676         struct dt_txn_commit_cb *dcb, *tmp;
 677
 678         LASSERT(oh->ot_handle == NULL);
 679
 680         if (error)
 681                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 682
 683         dt_txn_hook_commit(th);
 684
 685         /* call per-transaction callbacks if any */
 686         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 687                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 688                          "commit callback entry: magic=%x name='%s'\n",
 689                          dcb->dcb_magic, dcb->dcb_name);
 690                 cfs_list_del_init(&dcb->dcb_linkage);
 691                 dcb->dcb_func(NULL, th, dcb, error);
 692         }
 693
 694         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 695         lu_device_put(lud);
 696         th->th_dev = NULL;
 697
 698         lu_context_exit(&th->th_ctx);
 699         lu_context_fini(&th->th_ctx);
 700         OBD_FREE_PTR(oh);
 701 }
 702
 703 static struct thandle *osd_trans_create(const struct lu_env *env,
 704                                         struct dt_device *d)
 705 {
 706         struct osd_thread_info *oti = osd_oti_get(env);
 707         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 708         struct osd_thandle     *oh;
 709         struct thandle         *th;
 710         ENTRY;
 711
 712         /* on pending IO in this thread should left from prev. request */
 713         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 714
 715         th = ERR_PTR(-ENOMEM);
 716         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 717         if (oh != NULL) {
 718                 oh->ot_quota_trans = &oti->oti_quota_trans;
 719                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 720                 th = &oh->ot_super;
 721                 th->th_dev = d;
 722                 th->th_result = 0;
 723                 th->th_tags = LCT_TX_HANDLE;
 724                 oh->ot_credits = 0;
 725                 oti->oti_dev = osd_dt_dev(d);
 726                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 727                 osd_th_alloced(oh);
 728
 729                 memset(oti->oti_declare_ops, 0,
 730                                         sizeof(oti->oti_declare_ops));
 731                 memset(oti->oti_declare_ops_rb, 0,
 732                                         sizeof(oti->oti_declare_ops_rb));
 733                 memset(oti->oti_declare_ops_cred, 0,
 734                                         sizeof(oti->oti_declare_ops_cred));
 735                 oti->oti_rollback = false;
 736         }
 737         RETURN(th);
 738 }
 739
 740 /*
 741  * Concurrency: shouldn't matter.
 742  */
 743 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 744                     struct thandle *th)
 745 {
 746         struct osd_thread_info *oti = osd_oti_get(env);
 747         struct osd_device  *dev = osd_dt_dev(d);
 748         handle_t           *jh;
 749         struct osd_thandle *oh;
 750         int rc;
 751
 752         ENTRY;
 753
 754         LASSERT(current->journal_info == NULL);
 755
 756         oh = container_of0(th, struct osd_thandle, ot_super);
 757         LASSERT(oh != NULL);
 758         LASSERT(oh->ot_handle == NULL);
 759
 760         rc = dt_txn_hook_start(env, d, th);
 761         if (rc != 0)
 762                 GOTO(out, rc);
 763
 764         if (unlikely(osd_param_is_not_sane(dev, th))) {
 765                 static unsigned long last_printed;
 766                 static int last_credits;
 767
 768                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 769                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 770                       oh->ot_credits,
 771                       osd_journal(dev)->j_max_transaction_buffers);
 772                 CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
 773                       oti->oti_declare_ops[OSD_OT_CREATE],
 774                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
 775                       oti->oti_declare_ops[OSD_OT_DELETE],
 776                       oti->oti_declare_ops_cred[OSD_OT_DELETE],
 777                       oti->oti_declare_ops[OSD_OT_DESTROY],
 778                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 779                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
 780                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
 781                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
 782                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
 783                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
 784                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
 785                       oti->oti_declare_ops[OSD_OT_WRITE],
 786                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
 787                       oti->oti_declare_ops[OSD_OT_PUNCH],
 788                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
 789                       oti->oti_declare_ops[OSD_OT_QUOTA],
 790                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
 791                 CWARN("  insert: %u/%u, delete: %u/%u\n",
 792                       oti->oti_declare_ops[OSD_OT_INSERT],
 793                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
 794                       oti->oti_declare_ops[OSD_OT_DESTROY],
 795                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 796                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
 797                       oti->oti_declare_ops[OSD_OT_REF_ADD],
 798                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
 799                       oti->oti_declare_ops[OSD_OT_REF_DEL],
 800                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
 801
 802                 if (last_credits != oh->ot_credits &&
 803                     time_after(jiffies, last_printed + 60 * HZ)) {
 804                         libcfs_debug_dumpstack(NULL);
 805                         last_credits = oh->ot_credits;
 806                         last_printed = jiffies;
 807                 }
 808                 /* XXX Limit the credits to 'max_transaction_buffers', and
 809                  *     let the underlying filesystem to catch the error if
 810                  *     we really need so many credits.
 811                  *
 812                  *     This should be removed when we can calculate the
 813                  *     credits precisely. */
 814                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 815         }
 816
 817         /*
 818          * XXX temporary stuff. Some abstraction layer should
 819          * be used.
 820          */
 821         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 822         osd_th_started(oh);
 823         if (!IS_ERR(jh)) {
 824                 oh->ot_handle = jh;
 825                 LASSERT(oti->oti_txns == 0);
 826                 lu_context_init(&th->th_ctx, th->th_tags);
 827                 lu_context_enter(&th->th_ctx);
 828
 829                 lu_device_get(&d->dd_lu_dev);
 830                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 831                                              "osd-tx", th);
 832                 oti->oti_txns++;
 833                 rc = 0;
 834         } else {
 835                 rc = PTR_ERR(jh);
 836         }
 837 out:
 838         RETURN(rc);
 839 }
 840
 841 /*
 842  * Concurrency: shouldn't matter.
 843  */
 844 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 845 {
 846         int                     rc = 0;
 847         struct osd_thandle     *oh;
 848         struct osd_thread_info *oti = osd_oti_get(env);
 849         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 850         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
 851         ENTRY;
 852
 853         oh = container_of0(th, struct osd_thandle, ot_super);
 854
 855         if (qsd != NULL)
 856                 /* inform the quota slave device that the transaction is
 857                  * stopping */
 858                 qsd_op_end(env, qsd, oh->ot_quota_trans);
 859         oh->ot_quota_trans = NULL;
 860
 861         if (oh->ot_handle != NULL) {
 862                 handle_t *hdl = oh->ot_handle;
 863
 864                 /*
 865                  * add commit callback
 866                  * notice we don't do this in osd_trans_start()
 867                  * as underlying transaction can change during truncate
 868                  */
 869                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
 870                                          &oh->ot_jcb);
 871
 872                 LASSERT(oti->oti_txns == 1);
 873                 oti->oti_txns--;
 874                 rc = dt_txn_hook_stop(env, th);
 875                 if (rc != 0)
 876                         CERROR("Failure in transaction hook: %d\n", rc);
 877
 878                 /* hook functions might modify th_sync */
 879                 hdl->h_sync = th->th_sync;
 880
 881                 oh->ot_handle = NULL;
 882                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 883                                   rc = ldiskfs_journal_stop(hdl));
 884                 if (rc != 0)
 885                         CERROR("Failure to stop transaction: %d\n", rc);
 886         } else {
 887                 OBD_FREE_PTR(oh);
 888         }
 889
 890         /* as we want IO to journal and data IO be concurrent, we don't block
 891          * awaiting data IO completion in osd_do_bio(), instead we wait here
 892          * once transaction is submitted to the journal. all reqular requests
 893          * don't do direct IO (except read/write), thus this wait_event becomes
 894          * no-op for them.
 895          *
 896          * IMPORTANT: we have to wait till any IO submited by the thread is
 897          * completed otherwise iobuf may be corrupted by different request
 898          */
 899         cfs_wait_event(iobuf->dr_wait,
 900                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 901         if (!rc)
 902                 rc = iobuf->dr_error;
 903
 904         RETURN(rc);
 905 }
 906
 907 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 908 {
 909         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 910                                                ot_super);
 911
 912         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 913         LASSERT(&dcb->dcb_func != NULL);
 914         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 915
 916         return 0;
 917 }
 918
 919 /*
 920  * Called just before object is freed. Releases all resources except for
 921  * object itself (that is released by osd_object_free()).
 922  *
 923  * Concurrency: no concurrent access is possible that late in object
 924  * life-cycle.
 925  */
 926 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 927 {
 928         struct osd_object *obj   = osd_obj(l);
 929         struct inode      *inode = obj->oo_inode;
 930
 931         LINVRNT(osd_invariant(obj));
 932
 933         /*
 934          * If object is unlinked remove fid->ino mapping from object index.
 935          */
 936
 937         osd_index_fini(obj);
 938         if (inode != NULL) {
 939                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
 940                 qid_t                    uid = inode->i_uid;
 941                 qid_t                    gid = inode->i_gid;
 942
 943                 iput(inode);
 944                 obj->oo_inode = NULL;
 945
 946                 if (qsd != NULL) {
 947                         struct osd_thread_info  *info = osd_oti_get(env);
 948                         struct lquota_id_info   *qi = &info->oti_qi;
 949
 950                         /* Release granted quota to master if necessary */
 951                         qi->lqi_id.qid_uid = uid;
 952                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
 953
 954                         qi->lqi_id.qid_uid = gid;
 955                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
 956                 }
 957         }
 958 }
 959
 960 /*
 961  * Concurrency: ->loo_object_release() is called under site spin-lock.
 962  */
 963 static void osd_object_release(const struct lu_env *env,
 964                                struct lu_object *l)
 965 {
 966 }
 967
 968 /*
 969  * Concurrency: shouldn't matter.
 970  */
 971 static int osd_object_print(const struct lu_env *env, void *cookie,
 972                             lu_printer_t p, const struct lu_object *l)
 973 {
 974         struct osd_object *o = osd_obj(l);
 975         struct iam_descr  *d;
 976
 977         if (o->oo_dir != NULL)
 978                 d = o->oo_dir->od_container.ic_descr;
 979         else
 980                 d = NULL;
 981         return (*p)(env, cookie,
 982                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 983                     o, o->oo_inode,
 984                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 985                     o->oo_inode ? o->oo_inode->i_generation : 0,
 986                     d ? d->id_ops->id_name : "plain");
 987 }
 988
 989 /*
 990  * Concurrency: shouldn't matter.
 991  */
 992 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 993                struct obd_statfs *sfs)
 994 {
 995         struct osd_device  *osd = osd_dt_dev(d);
 996         struct super_block *sb = osd_sb(osd);
 997         struct kstatfs     *ksfs;
 998         int result = 0;
 999
1000         if (unlikely(osd->od_mnt == NULL))
1001                 return -EINPROGRESS;
1002
1003         /* osd_lproc.c call this without env, allocate ksfs for that case */
1004         if (unlikely(env == NULL)) {
1005                 OBD_ALLOC_PTR(ksfs);
1006                 if (ksfs == NULL)
1007                         return -ENOMEM;
1008         } else {
1009                 ksfs = &osd_oti_get(env)->oti_ksfs;
1010         }
1011
1012         spin_lock(&osd->od_osfs_lock);
1013         /* cache 1 second */
1014         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
1015                 result = sb->s_op->statfs(sb->s_root, ksfs);
1016                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1017                         osd->od_osfs_age = cfs_time_current_64();
1018                         statfs_pack(&osd->od_statfs, ksfs);
1019                         if (sb->s_flags & MS_RDONLY)
1020                                 sfs->os_state = OS_STATE_READONLY;
1021                 }
1022         }
1023
1024         if (likely(result == 0))
1025                 *sfs = osd->od_statfs;
1026         spin_unlock(&osd->od_osfs_lock);
1027
1028         if (unlikely(env == NULL))
1029                 OBD_FREE_PTR(ksfs);
1030
1031         return result;
1032 }
1033
1034 /**
1035  * Estimate space needed for file creations. We assume the largest filename
1036  * which is 2^64 - 1, hence a filename of 20 chars.
1037  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1038  */
1039 #ifdef __LDISKFS_DIR_REC_LEN
1040 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1041 #else
1042 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1043 #endif
1044
1045 /*
1046  * Concurrency: doesn't access mutable data.
1047  */
1048 static void osd_conf_get(const struct lu_env *env,
1049                          const struct dt_device *dev,
1050                          struct dt_device_param *param)
1051 {
1052         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1053
1054         /*
1055          * XXX should be taken from not-yet-existing fs abstraction layer.
1056          */
1057         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1058         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1059         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1060         param->ddp_block_shift  = sb->s_blocksize_bits;
1061         param->ddp_mount_type     = LDD_MT_LDISKFS;
1062         param->ddp_maxbytes       = sb->s_maxbytes;
1063         /* Overhead estimate should be fairly accurate, so we really take a tiny
1064          * error margin which also avoids fragmenting the filesystem too much */
1065         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1066         /* inode are statically allocated, so per-inode space consumption
1067          * is the space consumed by the directory entry */
1068         param->ddp_inodespace     = PER_OBJ_USAGE;
1069         /* per-fragment overhead to be used by the client code */
1070         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1071         param->ddp_mntopts      = 0;
1072         if (test_opt(sb, XATTR_USER))
1073                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1074         if (test_opt(sb, POSIX_ACL))
1075                 param->ddp_mntopts |= MNTOPT_ACL;
1076
1077 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1078         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1079                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1080         else
1081 #endif
1082                 param->ddp_max_ea_size = sb->s_blocksize;
1083
1084 }
1085
1086 /*
1087  * Concurrency: shouldn't matter.
1088  */
1089 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1090 {
1091         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1092         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1093 }
1094
1095 /**
1096  * Start commit for OSD device.
1097  *
1098  * An implementation of dt_commit_async method for OSD device.
1099  * Asychronously starts underlayng fs sync and thereby a transaction
1100  * commit.
1101  *
1102  * \param env environment
1103  * \param d dt device
1104  *
1105  * \see dt_device_operations
1106  */
1107 static int osd_commit_async(const struct lu_env *env,
1108                             struct dt_device *d)
1109 {
1110         struct super_block *s = osd_sb(osd_dt_dev(d));
1111         ENTRY;
1112
1113         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1114         RETURN(s->s_op->sync_fs(s, 0));
1115 }
1116
1117 /*
1118  * Concurrency: shouldn't matter.
1119  */
1120
1121 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1122 {
1123         struct super_block *sb = osd_sb(osd_dt_dev(d));
1124         struct block_device *dev = sb->s_bdev;
1125 #ifdef HAVE_DEV_SET_RDONLY
1126         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1127         int rc = 0;
1128 #else
1129         int rc = -EOPNOTSUPP;
1130 #endif
1131         ENTRY;
1132
1133 #ifdef HAVE_DEV_SET_RDONLY
1134         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1135
1136         if (jdev && (jdev != dev)) {
1137                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1138                        (long)jdev);
1139                 dev_set_rdonly(jdev);
1140         }
1141         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1142         dev_set_rdonly(dev);
1143 #else
1144         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1145                osd_dt_dev(d)->od_svname, (long)dev, rc);
1146 #endif
1147         RETURN(rc);
1148 }
1149
1150 /*
1151  * Concurrency: serialization provided by callers.
1152  */
1153 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1154                               int mode, unsigned long timeout, __u32 alg,
1155                               struct lustre_capa_key *keys)
1156 {
1157         struct osd_device *dev = osd_dt_dev(d);
1158         ENTRY;
1159
1160         dev->od_fl_capa = mode;
1161         dev->od_capa_timeout = timeout;
1162         dev->od_capa_alg = alg;
1163         dev->od_capa_keys = keys;
1164         RETURN(0);
1165 }
1166
1167 /**
1168  * Note: we do not count into QUOTA here.
1169  * If we mount with --data_journal we may need more.
1170  */
1171 const int osd_dto_credits_noquota[DTO_NR] = {
1172         /**
1173          * Insert/Delete.
1174          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1175          * SINGLEDATA_TRANS_BLOCKS(8)
1176          * XXX Note: maybe iam need more, since iam have more level than
1177          *           EXT3 htree.
1178          */
1179         [DTO_INDEX_INSERT]  = 16,
1180         [DTO_INDEX_DELETE]  = 16,
1181         /**
1182          * Used for OI scrub
1183          */
1184         [DTO_INDEX_UPDATE]  = 16,
1185         /**
1186          * Create a object. The same as create object in EXT3.
1187          * DATA_TRANS_BLOCKS(14) +
1188          * INDEX_EXTRA_BLOCKS(8) +
1189          * 3(inode bits, groups, GDT)
1190          */
1191         [DTO_OBJECT_CREATE] = 25,
1192         /**
1193          * XXX: real credits to be fixed
1194          */
1195         [DTO_OBJECT_DELETE] = 25,
1196         /**
1197          * Attr set credits (inode)
1198          */
1199         [DTO_ATTR_SET_BASE] = 1,
1200         /**
1201          * Xattr set. The same as xattr of EXT3.
1202          * DATA_TRANS_BLOCKS(14)
1203          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1204          * are also counted in. Do not know why?
1205          */
1206         [DTO_XATTR_SET]     = 14,
1207         [DTO_LOG_REC]       = 14,
1208         /**
1209          * credits for inode change during write.
1210          */
1211         [DTO_WRITE_BASE]    = 3,
1212         /**
1213          * credits for single block write.
1214          */
1215         [DTO_WRITE_BLOCK]   = 14,
1216         /**
1217          * Attr set credits for chown.
1218          * This is extra credits for setattr, and it is null without quota
1219          */
1220         [DTO_ATTR_SET_CHOWN]= 0
1221 };
1222
1223 static const struct dt_device_operations osd_dt_ops = {
1224         .dt_root_get       = osd_root_get,
1225         .dt_statfs         = osd_statfs,
1226         .dt_trans_create   = osd_trans_create,
1227         .dt_trans_start    = osd_trans_start,
1228         .dt_trans_stop     = osd_trans_stop,
1229         .dt_trans_cb_add   = osd_trans_cb_add,
1230         .dt_conf_get       = osd_conf_get,
1231         .dt_sync           = osd_sync,
1232         .dt_ro             = osd_ro,
1233         .dt_commit_async   = osd_commit_async,
1234         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1235 };
1236
1237 static void osd_object_read_lock(const struct lu_env *env,
1238                                  struct dt_object *dt, unsigned role)
1239 {
1240         struct osd_object *obj = osd_dt_obj(dt);
1241         struct osd_thread_info *oti = osd_oti_get(env);
1242
1243         LINVRNT(osd_invariant(obj));
1244
1245         LASSERT(obj->oo_owner != env);
1246         down_read_nested(&obj->oo_sem, role);
1247
1248         LASSERT(obj->oo_owner == NULL);
1249         oti->oti_r_locks++;
1250 }
1251
1252 static void osd_object_write_lock(const struct lu_env *env,
1253                                   struct dt_object *dt, unsigned role)
1254 {
1255         struct osd_object *obj = osd_dt_obj(dt);
1256         struct osd_thread_info *oti = osd_oti_get(env);
1257
1258         LINVRNT(osd_invariant(obj));
1259
1260         LASSERT(obj->oo_owner != env);
1261         down_write_nested(&obj->oo_sem, role);
1262
1263         LASSERT(obj->oo_owner == NULL);
1264         obj->oo_owner = env;
1265         oti->oti_w_locks++;
1266 }
1267
1268 static void osd_object_read_unlock(const struct lu_env *env,
1269                                    struct dt_object *dt)
1270 {
1271         struct osd_object *obj = osd_dt_obj(dt);
1272         struct osd_thread_info *oti = osd_oti_get(env);
1273
1274         LINVRNT(osd_invariant(obj));
1275
1276         LASSERT(oti->oti_r_locks > 0);
1277         oti->oti_r_locks--;
1278         up_read(&obj->oo_sem);
1279 }
1280
1281 static void osd_object_write_unlock(const struct lu_env *env,
1282                                     struct dt_object *dt)
1283 {
1284         struct osd_object *obj = osd_dt_obj(dt);
1285         struct osd_thread_info *oti = osd_oti_get(env);
1286
1287         LINVRNT(osd_invariant(obj));
1288
1289         LASSERT(obj->oo_owner == env);
1290         LASSERT(oti->oti_w_locks > 0);
1291         oti->oti_w_locks--;
1292         obj->oo_owner = NULL;
1293         up_write(&obj->oo_sem);
1294 }
1295
1296 static int osd_object_write_locked(const struct lu_env *env,
1297                                    struct dt_object *dt)
1298 {
1299         struct osd_object *obj = osd_dt_obj(dt);
1300
1301         LINVRNT(osd_invariant(obj));
1302
1303         return obj->oo_owner == env;
1304 }
1305
1306 static int capa_is_sane(const struct lu_env *env,
1307                         struct osd_device *dev,
1308                         struct lustre_capa *capa,
1309                         struct lustre_capa_key *keys)
1310 {
1311         struct osd_thread_info *oti = osd_oti_get(env);
1312         struct lustre_capa *tcapa = &oti->oti_capa;
1313         struct obd_capa *oc;
1314         int i, rc = 0;
1315         ENTRY;
1316
1317         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1318         if (oc) {
1319                 if (capa_is_expired(oc)) {
1320                         DEBUG_CAPA(D_ERROR, capa, "expired");
1321                         rc = -ESTALE;
1322                 }
1323                 capa_put(oc);
1324                 RETURN(rc);
1325         }
1326
1327         if (capa_is_expired_sec(capa)) {
1328                 DEBUG_CAPA(D_ERROR, capa, "expired");
1329                 RETURN(-ESTALE);
1330         }
1331
1332         spin_lock(&capa_lock);
1333         for (i = 0; i < 2; i++) {
1334                 if (keys[i].lk_keyid == capa->lc_keyid) {
1335                         oti->oti_capa_key = keys[i];
1336                         break;
1337                 }
1338         }
1339         spin_unlock(&capa_lock);
1340
1341         if (i == 2) {
1342                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1343                 RETURN(-ESTALE);
1344         }
1345
1346         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1347         if (rc)
1348                 RETURN(rc);
1349
1350         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1351                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1352                 RETURN(-EACCES);
1353         }
1354
1355         oc = capa_add(dev->od_capa_hash, capa);
1356         capa_put(oc);
1357
1358         RETURN(0);
1359 }
1360
1361 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1362                     struct lustre_capa *capa, __u64 opc)
1363 {
1364         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1365         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1366         struct md_capainfo *ci;
1367         int rc;
1368
1369         if (!dev->od_fl_capa)
1370                 return 0;
1371
1372         if (capa == BYPASS_CAPA)
1373                 return 0;
1374
1375         ci = md_capainfo(env);
1376         if (unlikely(!ci))
1377                 return 0;
1378
1379         if (ci->mc_auth == LC_ID_NONE)
1380                 return 0;
1381
1382         if (!capa) {
1383                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1384                 return -EACCES;
1385         }
1386
1387         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1388                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1389                            PFID(fid));
1390                 return -EACCES;
1391         }
1392
1393         if (!capa_opc_supported(capa, opc)) {
1394                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1395                 return -EACCES;
1396         }
1397
1398         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1399                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1400                 return -EACCES;
1401         }
1402
1403         return 0;
1404 }
1405
1406 static struct timespec *osd_inode_time(const struct lu_env *env,
1407                                        struct inode *inode, __u64 seconds)
1408 {
1409         struct osd_thread_info  *oti = osd_oti_get(env);
1410         struct timespec         *t   = &oti->oti_time;
1411
1412         t->tv_sec = seconds;
1413         t->tv_nsec = 0;
1414         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1415         return t;
1416 }
1417
1418
1419 static void osd_inode_getattr(const struct lu_env *env,
1420                               struct inode *inode, struct lu_attr *attr)
1421 {
1422         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1423                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1424                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1425                                LA_TYPE;
1426
1427         attr->la_atime      = LTIME_S(inode->i_atime);
1428         attr->la_mtime      = LTIME_S(inode->i_mtime);
1429         attr->la_ctime      = LTIME_S(inode->i_ctime);
1430         attr->la_mode       = inode->i_mode;
1431         attr->la_size       = i_size_read(inode);
1432         attr->la_blocks     = inode->i_blocks;
1433         attr->la_uid        = inode->i_uid;
1434         attr->la_gid        = inode->i_gid;
1435         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1436         attr->la_nlink      = inode->i_nlink;
1437         attr->la_rdev       = inode->i_rdev;
1438         attr->la_blksize    = 1 << inode->i_blkbits;
1439         attr->la_blkbits    = inode->i_blkbits;
1440 }
1441
1442 static int osd_attr_get(const struct lu_env *env,
1443                         struct dt_object *dt,
1444                         struct lu_attr *attr,
1445                         struct lustre_capa *capa)
1446 {
1447         struct osd_object *obj = osd_dt_obj(dt);
1448
1449         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1450         LINVRNT(osd_invariant(obj));
1451
1452         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1453                 return -EACCES;
1454
1455         spin_lock(&obj->oo_guard);
1456         osd_inode_getattr(env, obj->oo_inode, attr);
1457         spin_unlock(&obj->oo_guard);
1458         return 0;
1459 }
1460
1461 static int osd_declare_attr_set(const struct lu_env *env,
1462                                 struct dt_object *dt,
1463                                 const struct lu_attr *attr,
1464                                 struct thandle *handle)
1465 {
1466         struct osd_thandle     *oh;
1467         struct osd_object      *obj;
1468         struct osd_thread_info *info = osd_oti_get(env);
1469         struct lquota_id_info  *qi = &info->oti_qi;
1470         long long               bspace;
1471         int                     rc = 0;
1472         bool                    allocated;
1473         ENTRY;
1474
1475         LASSERT(dt != NULL);
1476         LASSERT(handle != NULL);
1477
1478         obj = osd_dt_obj(dt);
1479         LASSERT(osd_invariant(obj));
1480
1481         oh = container_of0(handle, struct osd_thandle, ot_super);
1482         LASSERT(oh->ot_handle == NULL);
1483
1484         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1485                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1486
1487         if (attr == NULL || obj->oo_inode == NULL)
1488                 RETURN(rc);
1489
1490         bspace   = obj->oo_inode->i_blocks;
1491         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1492         bspace   = toqb(bspace);
1493
1494         /* Changing ownership is always preformed by super user, it should not
1495          * fail with EDQUOT.
1496          *
1497          * We still need to call the osd_declare_qid() to calculate the journal
1498          * credits for updating quota accounting files and to trigger quota
1499          * space adjustment once the operation is completed.*/
1500         if ((attr->la_valid & LA_UID) != 0 &&
1501              attr->la_uid != obj->oo_inode->i_uid) {
1502                 qi->lqi_type = USRQUOTA;
1503
1504                 /* inode accounting */
1505                 qi->lqi_is_blk = false;
1506
1507                 /* one more inode for the new owner ... */
1508                 qi->lqi_id.qid_uid = attr->la_uid;
1509                 qi->lqi_space      = 1;
1510                 allocated = (attr->la_uid == 0) ? true : false;
1511                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1512                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1513                         rc = 0;
1514                 if (rc)
1515                         RETURN(rc);
1516
1517                 /* and one less inode for the current uid */
1518                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1519                 qi->lqi_space      = -1;
1520                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1521                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1522                         rc = 0;
1523                 if (rc)
1524                         RETURN(rc);
1525
1526                 /* block accounting */
1527                 qi->lqi_is_blk = true;
1528
1529                 /* more blocks for the new owner ... */
1530                 qi->lqi_id.qid_uid = attr->la_uid;
1531                 qi->lqi_space      = bspace;
1532                 allocated = (attr->la_uid == 0) ? true : false;
1533                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1534                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1535                         rc = 0;
1536                 if (rc)
1537                         RETURN(rc);
1538
1539                 /* and finally less blocks for the current owner */
1540                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1541                 qi->lqi_space      = -bspace;
1542                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1543                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1544                         rc = 0;
1545                 if (rc)
1546                         RETURN(rc);
1547         }
1548
1549         if (attr->la_valid & LA_GID &&
1550             attr->la_gid != obj->oo_inode->i_gid) {
1551                 qi->lqi_type = GRPQUOTA;
1552
1553                 /* inode accounting */
1554                 qi->lqi_is_blk = false;
1555
1556                 /* one more inode for the new group owner ... */
1557                 qi->lqi_id.qid_gid = attr->la_gid;
1558                 qi->lqi_space      = 1;
1559                 allocated = (attr->la_gid == 0) ? true : false;
1560                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1561                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1562                         rc = 0;
1563                 if (rc)
1564                         RETURN(rc);
1565
1566                 /* and one less inode for the current gid */
1567                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1568                 qi->lqi_space      = -1;
1569                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1570                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1571                         rc = 0;
1572                 if (rc)
1573                         RETURN(rc);
1574
1575                 /* block accounting */
1576                 qi->lqi_is_blk = true;
1577
1578                 /* more blocks for the new owner ... */
1579                 qi->lqi_id.qid_gid = attr->la_gid;
1580                 qi->lqi_space      = bspace;
1581                 allocated = (attr->la_gid == 0) ? true : false;
1582                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1583                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1584                         rc = 0;
1585                 if (rc)
1586                         RETURN(rc);
1587
1588                 /* and finally less blocks for the current owner */
1589                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1590                 qi->lqi_space      = -bspace;
1591                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1592                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1593                         rc = 0;
1594                 if (rc)
1595                         RETURN(rc);
1596         }
1597
1598         RETURN(rc);
1599 }
1600
1601 static int osd_inode_setattr(const struct lu_env *env,
1602                              struct inode *inode, const struct lu_attr *attr)
1603 {
1604         __u64 bits;
1605
1606         bits = attr->la_valid;
1607
1608         if (bits & LA_ATIME)
1609                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1610         if (bits & LA_CTIME)
1611                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1612         if (bits & LA_MTIME)
1613                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1614         if (bits & LA_SIZE) {
1615                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1616                 i_size_write(inode, attr->la_size);
1617         }
1618
1619 #if 0
1620         /* OSD should not change "i_blocks" which is used by quota.
1621          * "i_blocks" should be changed by ldiskfs only. */
1622         if (bits & LA_BLOCKS)
1623                 inode->i_blocks = attr->la_blocks;
1624 #endif
1625         if (bits & LA_MODE)
1626                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1627                         (attr->la_mode & ~S_IFMT);
1628         if (bits & LA_UID)
1629                 inode->i_uid    = attr->la_uid;
1630         if (bits & LA_GID)
1631                 inode->i_gid    = attr->la_gid;
1632         if (bits & LA_NLINK)
1633                 set_nlink(inode, attr->la_nlink);
1634         if (bits & LA_RDEV)
1635                 inode->i_rdev   = attr->la_rdev;
1636
1637         if (bits & LA_FLAGS) {
1638                 /* always keep S_NOCMTIME */
1639                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1640                                  S_NOCMTIME;
1641         }
1642         return 0;
1643 }
1644
1645 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1646 {
1647         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1648             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1649                 struct iattr    iattr;
1650                 int             rc;
1651
1652                 iattr.ia_valid = 0;
1653                 if (attr->la_valid & LA_UID)
1654                         iattr.ia_valid |= ATTR_UID;
1655                 if (attr->la_valid & LA_GID)
1656                         iattr.ia_valid |= ATTR_GID;
1657                 iattr.ia_uid = attr->la_uid;
1658                 iattr.ia_gid = attr->la_gid;
1659
1660                 rc = ll_vfs_dq_transfer(inode, &iattr);
1661                 if (rc) {
1662                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1663                                "enforcement enabled on the ldiskfs filesystem?",
1664                                inode->i_sb->s_id, rc);
1665                         return rc;
1666                 }
1667         }
1668         return 0;
1669 }
1670
1671 static int osd_attr_set(const struct lu_env *env,
1672                         struct dt_object *dt,
1673                         const struct lu_attr *attr,
1674                         struct thandle *handle,
1675                         struct lustre_capa *capa)
1676 {
1677         struct osd_object *obj = osd_dt_obj(dt);
1678         struct inode      *inode;
1679         int rc;
1680
1681         LASSERT(handle != NULL);
1682         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1683         LASSERT(osd_invariant(obj));
1684
1685         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1686                 return -EACCES;
1687
1688         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1689
1690         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1691                 struct osd_thread_info  *oti  = osd_oti_get(env);
1692                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1693                 struct lu_fid           *fid1 = &oti->oti_fid;
1694                 struct osd_inode_id     *id   = &oti->oti_id;
1695                 struct iam_path_descr   *ipd;
1696                 struct iam_container    *bag;
1697                 struct osd_thandle      *oh;
1698                 int                      rc;
1699
1700                 fid_cpu_to_be(fid1, fid0);
1701                 memset(id, 1, sizeof(*id));
1702                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1703                                   fid0)->oi_dir.od_container;
1704                 ipd = osd_idx_ipd_get(env, bag);
1705                 if (unlikely(ipd == NULL))
1706                         RETURN(-ENOMEM);
1707
1708                 oh = container_of0(handle, struct osd_thandle, ot_super);
1709                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1710                                 (const struct iam_rec *)id, ipd);
1711                 osd_ipd_put(env, bag, ipd);
1712                 return(rc > 0 ? 0 : rc);
1713         }
1714
1715         inode = obj->oo_inode;
1716         ll_vfs_dq_init(inode);
1717
1718         rc = osd_quota_transfer(inode, attr);
1719         if (rc)
1720                 return rc;
1721
1722         spin_lock(&obj->oo_guard);
1723         rc = osd_inode_setattr(env, inode, attr);
1724         spin_unlock(&obj->oo_guard);
1725
1726         if (!rc)
1727                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1728         return rc;
1729 }
1730
1731 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1732                                     struct osd_object *obj,
1733                                     const char *name, const int namelen)
1734 {
1735         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1736 }
1737
1738 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1739                       cfs_umode_t mode,
1740                       struct dt_allocation_hint *hint,
1741                       struct thandle *th)
1742 {
1743         int result;
1744         struct osd_device  *osd = osd_obj2dev(obj);
1745         struct osd_thandle *oth;
1746         struct dt_object   *parent = NULL;
1747         struct inode       *inode;
1748
1749         LINVRNT(osd_invariant(obj));
1750         LASSERT(obj->oo_inode == NULL);
1751         LASSERT(obj->oo_hl_head == NULL);
1752
1753         if (S_ISDIR(mode) && ldiskfs_pdo) {
1754                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1755                 if (obj->oo_hl_head == NULL)
1756                         return -ENOMEM;
1757         }
1758
1759         oth = container_of(th, struct osd_thandle, ot_super);
1760         LASSERT(oth->ot_handle->h_transaction != NULL);
1761
1762         if (hint && hint->dah_parent)
1763                 parent = hint->dah_parent;
1764
1765         inode = ldiskfs_create_inode(oth->ot_handle,
1766                                      parent ? osd_dt_obj(parent)->oo_inode :
1767                                               osd_sb(osd)->s_root->d_inode,
1768                                      mode);
1769         if (!IS_ERR(inode)) {
1770                 /* Do not update file c/mtime in ldiskfs.
1771                  * NB: don't need any lock because no contention at this
1772                  * early stage */
1773                 inode->i_flags |= S_NOCMTIME;
1774
1775                 /* For new created object, it must be consistent,
1776                  * and it is unnecessary to scrub against it. */
1777                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1778                 obj->oo_inode = inode;
1779                 result = 0;
1780         } else {
1781                 if (obj->oo_hl_head != NULL) {
1782                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1783                         obj->oo_hl_head = NULL;
1784                 }
1785                 result = PTR_ERR(inode);
1786         }
1787         LINVRNT(osd_invariant(obj));
1788         return result;
1789 }
1790
1791 enum {
1792         OSD_NAME_LEN = 255
1793 };
1794
1795 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1796                      struct lu_attr *attr,
1797                      struct dt_allocation_hint *hint,
1798                      struct dt_object_format *dof,
1799                      struct thandle *th)
1800 {
1801         int result;
1802         struct osd_thandle *oth;
1803         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1804
1805         LASSERT(S_ISDIR(attr->la_mode));
1806
1807         oth = container_of(th, struct osd_thandle, ot_super);
1808         LASSERT(oth->ot_handle->h_transaction != NULL);
1809         result = osd_mkfile(info, obj, mode, hint, th);
1810
1811         return result;
1812 }
1813
1814 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1815                         struct lu_attr *attr,
1816                         struct dt_allocation_hint *hint,
1817                         struct dt_object_format *dof,
1818                         struct thandle *th)
1819 {
1820         int result;
1821         struct osd_thandle *oth;
1822         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1823
1824         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1825
1826         LASSERT(S_ISREG(attr->la_mode));
1827
1828         oth = container_of(th, struct osd_thandle, ot_super);
1829         LASSERT(oth->ot_handle->h_transaction != NULL);
1830
1831         result = osd_mkfile(info, obj, mode, hint, th);
1832         if (result == 0) {
1833                 LASSERT(obj->oo_inode != NULL);
1834                 if (feat->dif_flags & DT_IND_VARKEY)
1835                         result = iam_lvar_create(obj->oo_inode,
1836                                                  feat->dif_keysize_max,
1837                                                  feat->dif_ptrsize,
1838                                                  feat->dif_recsize_max,
1839                                                  oth->ot_handle);
1840                 else
1841                         result = iam_lfix_create(obj->oo_inode,
1842                                                  feat->dif_keysize_max,
1843                                                  feat->dif_ptrsize,
1844                                                  feat->dif_recsize_max,
1845                                                  oth->ot_handle);
1846
1847         }
1848         return result;
1849 }
1850
1851 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1852                      struct lu_attr *attr,
1853                      struct dt_allocation_hint *hint,
1854                      struct dt_object_format *dof,
1855                      struct thandle *th)
1856 {
1857         LASSERT(S_ISREG(attr->la_mode));
1858         return osd_mkfile(info, obj, (attr->la_mode &
1859                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1860 }
1861
1862 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1863                      struct lu_attr *attr,
1864                      struct dt_allocation_hint *hint,
1865                      struct dt_object_format *dof,
1866                      struct thandle *th)
1867 {
1868         LASSERT(S_ISLNK(attr->la_mode));
1869         return osd_mkfile(info, obj, (attr->la_mode &
1870                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1871 }
1872
1873 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1874                      struct lu_attr *attr,
1875                      struct dt_allocation_hint *hint,
1876                      struct dt_object_format *dof,
1877                      struct thandle *th)
1878 {
1879         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1880         int result;
1881
1882         LINVRNT(osd_invariant(obj));
1883         LASSERT(obj->oo_inode == NULL);
1884         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1885                 S_ISFIFO(mode) || S_ISSOCK(mode));
1886
1887         result = osd_mkfile(info, obj, mode, hint, th);
1888         if (result == 0) {
1889                 LASSERT(obj->oo_inode != NULL);
1890                 /*
1891                  * This inode should be marked dirty for i_rdev.  Currently
1892                  * that is done in the osd_attr_init().
1893                  */
1894                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1895                                    attr->la_rdev);
1896         }
1897         LINVRNT(osd_invariant(obj));
1898         return result;
1899 }
1900
1901 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1902                               struct lu_attr *,
1903                               struct dt_allocation_hint *hint,
1904                               struct dt_object_format *dof,
1905                               struct thandle *);
1906
1907 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1908 {
1909         osd_obj_type_f result;
1910
1911         switch (type) {
1912         case DFT_DIR:
1913                 result = osd_mkdir;
1914                 break;
1915         case DFT_REGULAR:
1916                 result = osd_mkreg;
1917                 break;
1918         case DFT_SYM:
1919                 result = osd_mksym;
1920                 break;
1921         case DFT_NODE:
1922                 result = osd_mknod;
1923                 break;
1924         case DFT_INDEX:
1925                 result = osd_mk_index;
1926                 break;
1927
1928         default:
1929                 LBUG();
1930                 break;
1931         }
1932         return result;
1933 }
1934
1935
1936 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1937                         struct dt_object *parent, struct dt_object *child,
1938                         cfs_umode_t child_mode)
1939 {
1940         LASSERT(ah);
1941
1942         memset(ah, 0, sizeof(*ah));
1943         ah->dah_parent = parent;
1944         ah->dah_mode = child_mode;
1945 }
1946
1947 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1948                           struct lu_attr *attr, struct dt_object_format *dof)
1949 {
1950         struct inode   *inode = obj->oo_inode;
1951         __u64           valid = attr->la_valid;
1952         int             result;
1953
1954         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1955
1956         if (dof->dof_type != DFT_NODE)
1957                 attr->la_valid &= ~LA_RDEV;
1958         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1959                 attr->la_valid &= ~LA_ATIME;
1960         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1961                 attr->la_valid &= ~LA_CTIME;
1962         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1963                 attr->la_valid &= ~LA_MTIME;
1964
1965         result = osd_quota_transfer(inode, attr);
1966         if (result)
1967                 return;
1968
1969         if (attr->la_valid != 0) {
1970                 result = osd_inode_setattr(info->oti_env, inode, attr);
1971                 /*
1972                  * The osd_inode_setattr() should always succeed here.  The
1973                  * only error that could be returned is EDQUOT when we are
1974                  * trying to change the UID or GID of the inode. However, this
1975                  * should not happen since quota enforcement is no longer
1976                  * enabled on ldiskfs (lquota takes care of it).
1977                  */
1978                 LASSERTF(result == 0, "%d", result);
1979                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1980         }
1981
1982         attr->la_valid = valid;
1983 }
1984
1985 /**
1986  * Helper function for osd_object_create()
1987  *
1988  * \retval 0, on success
1989  */
1990 static int __osd_object_create(struct osd_thread_info *info,
1991                                struct osd_object *obj, struct lu_attr *attr,
1992                                struct dt_allocation_hint *hint,
1993                                struct dt_object_format *dof,
1994                                struct thandle *th)
1995 {
1996         int     result;
1997         __u32   umask;
1998
1999         /* we drop umask so that permissions we pass are not affected */
2000         umask = current->fs->umask;
2001         current->fs->umask = 0;
2002
2003         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2004                                                   th);
2005         if (result == 0) {
2006                 osd_attr_init(info, obj, attr, dof);
2007                 osd_object_init0(obj);
2008                 /* bz 24037 */
2009                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2010                         unlock_new_inode(obj->oo_inode);
2011         }
2012
2013         /* restore previous umask value */
2014         current->fs->umask = umask;
2015
2016         return result;
2017 }
2018
2019 /**
2020  * Helper function for osd_object_create()
2021  *
2022  * \retval 0, on success
2023  */
2024 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2025                            const struct lu_fid *fid, struct thandle *th)
2026 {
2027         struct osd_thread_info *info = osd_oti_get(env);
2028         struct osd_inode_id    *id   = &info->oti_id;
2029         struct osd_device      *osd  = osd_obj2dev(obj);
2030
2031         LASSERT(obj->oo_inode != NULL);
2032
2033         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2034         return osd_oi_insert(info, osd, fid, id, th);
2035 }
2036
2037 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2038                    const struct lu_fid *fid, struct lu_seq_range *range)
2039 {
2040         struct seq_server_site  *ss = osd_seq_site(osd);
2041         int                     rc;
2042
2043         if (fid_is_idif(fid)) {
2044                 fld_range_set_ost(range);
2045                 range->lsr_index = fid_idif_ost_idx(fid);
2046                 return 0;
2047         }
2048
2049         if (!fid_seq_in_fldb(fid_seq(fid))) {
2050                 fld_range_set_mdt(range);
2051                 if (ss != NULL)
2052                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2053                          * at all */
2054                         range->lsr_index = ss->ss_node_id;
2055                 return 0;
2056         }
2057
2058         LASSERT(ss != NULL);
2059         fld_range_set_any(range);
2060         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2061         if (rc != 0) {
2062                 CERROR("%s: cannot find FLD range for "DFID": rc = %d\n",
2063                        osd_name(osd), PFID(fid), rc);
2064         }
2065         return rc;
2066 }
2067
2068 /*
2069  * Concurrency: no external locking is necessary.
2070  */
2071 static int osd_declare_object_create(const struct lu_env *env,
2072                                      struct dt_object *dt,
2073                                      struct lu_attr *attr,
2074                                      struct dt_allocation_hint *hint,
2075                                      struct dt_object_format *dof,
2076                                      struct thandle *handle)
2077 {
2078         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
2079         struct osd_thandle      *oh;
2080         int                      rc;
2081         ENTRY;
2082
2083         LASSERT(handle != NULL);
2084
2085         oh = container_of0(handle, struct osd_thandle, ot_super);
2086         LASSERT(oh->ot_handle == NULL);
2087
2088         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2089                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2090         if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2091                            lu_object_fid(&dt->do_lu)))
2092                 /* Reuse idle OI block may cause additional one OI block
2093                  * to be changed. */
2094                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2095                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2096
2097         /* If this is directory, then we expect . and .. to be inserted as
2098          * well. The one directory block always needs to be created for the
2099          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2100          * block), there is no danger of needing a tree for the first block.
2101          */
2102         if (attr && S_ISDIR(attr->la_mode)) {
2103                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2104                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2105                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2106         }
2107
2108         if (!attr)
2109                 RETURN(0);
2110
2111         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2112                                    false, false, NULL, false);
2113         if (rc != 0)
2114                 RETURN(rc);
2115
2116         /* It does fld look up inside declare, and the result will be
2117          * added to fld cache, so the following fld lookup inside insert
2118          * does not need send RPC anymore, so avoid send rpc with holding
2119          * transaction */
2120         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2121                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2122                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2123                                lu_object_fid(&dt->do_lu), range);
2124
2125
2126         RETURN(rc);
2127 }
2128
2129 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2130                              struct lu_attr *attr,
2131                              struct dt_allocation_hint *hint,
2132                              struct dt_object_format *dof,
2133                              struct thandle *th)
2134 {
2135         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2136         struct osd_object      *obj    = osd_dt_obj(dt);
2137         struct osd_thread_info *info   = osd_oti_get(env);
2138         int result;
2139
2140         ENTRY;
2141
2142         LINVRNT(osd_invariant(obj));
2143         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2144         LASSERT(osd_write_locked(env, obj));
2145         LASSERT(th != NULL);
2146
2147         if (unlikely(fid_is_acct(fid)))
2148                 /* Quota files can't be created from the kernel any more,
2149                  * 'tune2fs -O quota' will take care of creating them */
2150                 RETURN(-EPERM);
2151
2152         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2153         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2154
2155         result = __osd_object_create(info, obj, attr, hint, dof, th);
2156         if (result == 0)
2157                 result = __osd_oi_insert(env, obj, fid, th);
2158
2159         LASSERT(ergo(result == 0,
2160                      dt_object_exists(dt) && !dt_object_remote(dt)));
2161
2162         LASSERT(osd_invariant(obj));
2163         RETURN(result);
2164 }
2165
2166 /**
2167  * Called to destroy on-disk representation of the object
2168  *
2169  * Concurrency: must be locked
2170  */
2171 static int osd_declare_object_destroy(const struct lu_env *env,
2172                                       struct dt_object *dt,
2173                                       struct thandle *th)
2174 {
2175         struct osd_object  *obj = osd_dt_obj(dt);
2176         struct inode       *inode = obj->oo_inode;
2177         struct osd_thandle *oh;
2178         int                 rc;
2179         ENTRY;
2180
2181         oh = container_of0(th, struct osd_thandle, ot_super);
2182         LASSERT(oh->ot_handle == NULL);
2183         LASSERT(inode);
2184
2185         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2186                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2187         /* Recycle idle OI leaf may cause additional three OI blocks
2188          * to be changed. */
2189         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2190                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2191         /* one less inode */
2192         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2193                                    false, true, NULL, false);
2194         if (rc)
2195                 RETURN(rc);
2196         /* data to be truncated */
2197         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2198                                    true, true, NULL, false);
2199         RETURN(rc);
2200 }
2201
2202 static int osd_object_destroy(const struct lu_env *env,
2203                               struct dt_object *dt,
2204                               struct thandle *th)
2205 {
2206         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2207         struct osd_object      *obj = osd_dt_obj(dt);
2208         struct inode           *inode = obj->oo_inode;
2209         struct osd_device      *osd = osd_obj2dev(obj);
2210         struct osd_thandle     *oh;
2211         int                     result;
2212         ENTRY;
2213
2214         oh = container_of0(th, struct osd_thandle, ot_super);
2215         LASSERT(oh->ot_handle);
2216         LASSERT(inode);
2217         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2218
2219         if (unlikely(fid_is_acct(fid)))
2220                 RETURN(-EPERM);
2221
2222         if (S_ISDIR(inode->i_mode)) {
2223                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2224                 /* it will check/delete the inode from remote parent,
2225                  * how to optimize it? unlink performance impaction XXX */
2226                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2227                 if (result != 0 && result != -ENOENT) {
2228                         CERROR("%s: delete inode "DFID": rc = %d\n",
2229                                osd_name(osd), PFID(fid), result);
2230                 }
2231                 spin_lock(&obj->oo_guard);
2232                 clear_nlink(inode);
2233                 spin_unlock(&obj->oo_guard);
2234                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2235         }
2236
2237         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2238
2239         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2240
2241         /* XXX: add to ext3 orphan list */
2242         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2243
2244         /* not needed in the cache anymore */
2245         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2246
2247         RETURN(0);
2248 }
2249
2250 /**
2251  * Put the fid into lustre_mdt_attrs, and then place the structure
2252  * inode's ea. This fid should not be altered during the life time
2253  * of the inode.
2254  *
2255  * \retval +ve, on success
2256  * \retval -ve, on error
2257  *
2258  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2259  */
2260 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2261                    const struct lu_fid *fid, __u64 flags)
2262 {
2263         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2264         int                      rc;
2265
2266         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2267                 return 0;
2268
2269         lustre_lma_init(lma, fid, flags);
2270         lustre_lma_swab(lma);
2271
2272         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2273                              XATTR_CREATE);
2274         /* Someone may created the EA by race. */
2275         if (unlikely(rc == -EEXIST))
2276                 rc = 0;
2277         return rc;
2278 }
2279
2280 /**
2281  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2282  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2283  * To have compatilibility with 1.8 ldiskfs driver we need to have
2284  * magic number at start of fid data.
2285  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2286  * its inmemory API.
2287  */
2288 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2289                                   const struct dt_rec *fid)
2290 {
2291         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2292             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2293                 param->edp_magic = 0;
2294                 return;
2295         }
2296
2297         param->edp_magic = LDISKFS_LUFID_MAGIC;
2298         param->edp_len =  sizeof(struct lu_fid) + 1;
2299         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2300 }
2301
2302 /**
2303  * Try to read the fid from inode ea into dt_rec.
2304  *
2305  * \param fid object fid.
2306  *
2307  * \retval 0 on success
2308  */
2309 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2310                           __u32 ino, struct lu_fid *fid,
2311                           struct osd_inode_id *id)
2312 {
2313         struct osd_thread_info *info  = osd_oti_get(env);
2314         struct inode           *inode;
2315         ENTRY;
2316
2317         osd_id_gen(id, ino, OSD_OII_NOGEN);
2318         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2319         if (IS_ERR(inode))
2320                 RETURN(PTR_ERR(inode));
2321
2322         iput(inode);
2323         RETURN(0);
2324 }
2325
2326 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2327                                         struct inode *dir,
2328                                         struct inode  *parent_dir,
2329                                         const struct dt_rec *dot_fid,
2330                                         const struct dt_rec *dot_dot_fid,
2331                                         struct osd_thandle *oth)
2332 {
2333         struct ldiskfs_dentry_param *dot_ldp;
2334         struct ldiskfs_dentry_param *dot_dot_ldp;
2335
2336         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2337         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2338
2339         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2340         dot_ldp->edp_magic = 0;
2341         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2342                                         dir, dot_ldp, dot_dot_ldp);
2343 }
2344
2345 /**
2346  * Create an local agent inode for remote entry
2347  */
2348 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2349                                                   struct osd_device *osd,
2350                                                   struct osd_object *pobj,
2351                                                   const struct lu_fid *fid,
2352                                                   struct thandle *th)
2353 {
2354         struct osd_thread_info  *info = osd_oti_get(env);
2355         struct inode            *local;
2356         struct osd_thandle      *oh;
2357         int                     rc;
2358         ENTRY;
2359
2360         LASSERT(th);
2361         oh = container_of(th, struct osd_thandle, ot_super);
2362         LASSERT(oh->ot_handle->h_transaction != NULL);
2363
2364         /* FIXME: Insert index api needs to know the mode of
2365          * the remote object. Just use S_IFDIR for now */
2366         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2367         if (IS_ERR(local)) {
2368                 CERROR("%s: create local error %d\n", osd_name(osd),
2369                        (int)PTR_ERR(local));
2370                 RETURN(local);
2371         }
2372
2373         /* Set special LMA flag for local agent inode */
2374         rc = osd_ea_fid_set(info, local, fid, LMAI_AGENT);
2375         if (rc != 0) {
2376                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2377                        osd_name(osd), PFID(fid), rc);
2378                 RETURN(ERR_PTR(rc));
2379         }
2380
2381         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2382                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2383                 (const struct dt_rec *)fid, oh);
2384         if (rc != 0) {
2385                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2386                         osd_name(osd), PFID(fid), rc);
2387                 RETURN(ERR_PTR(rc));
2388         }
2389
2390         RETURN(local);
2391 }
2392
2393 /**
2394  * Delete local agent inode for remote entry
2395  */
2396 static int osd_delete_local_agent_inode(const struct lu_env *env,
2397                                         struct osd_device *osd,
2398                                         const struct lu_fid *fid,
2399                                         __u32 ino, struct osd_thandle *oh)
2400 {
2401         struct osd_thread_info  *oti = osd_oti_get(env);
2402         struct osd_inode_id     *id = &oti->oti_id;
2403         struct inode            *inode;
2404         ENTRY;
2405
2406         id->oii_ino = le32_to_cpu(ino);
2407         id->oii_gen = OSD_OII_NOGEN;
2408         inode = osd_iget(oti, osd, id);
2409         if (IS_ERR(inode)) {
2410                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2411                        PFID(fid), id->oii_ino, id->oii_gen);
2412                 RETURN(PTR_ERR(inode));
2413         }
2414
2415         clear_nlink(inode);
2416         mark_inode_dirty(inode);
2417         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2418                 osd_name(osd), PFID(fid), inode->i_ino);
2419         iput(inode);
2420         RETURN(0);
2421 }
2422
2423 /**
2424  * OSD layer object create function for interoperability mode (b11826).
2425  * This is mostly similar to osd_object_create(). Only difference being, fid is
2426  * inserted into inode ea here.
2427  *
2428  * \retval   0, on success
2429  * \retval -ve, on error
2430  */
2431 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2432                                 struct lu_attr *attr,
2433                                 struct dt_allocation_hint *hint,
2434                                 struct dt_object_format *dof,
2435                                 struct thandle *th)
2436 {
2437         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2438         struct osd_object      *obj    = osd_dt_obj(dt);
2439         struct osd_thread_info *info   = osd_oti_get(env);
2440         int                     result;
2441
2442         ENTRY;
2443
2444         LASSERT(osd_invariant(obj));
2445         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2446         LASSERT(osd_write_locked(env, obj));
2447         LASSERT(th != NULL);
2448
2449         if (unlikely(fid_is_acct(fid)))
2450                 /* Quota files can't be created from the kernel any more,
2451                  * 'tune2fs -O quota' will take care of creating them */
2452                 RETURN(-EPERM);
2453
2454         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2455         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2456
2457         result = __osd_object_create(info, obj, attr, hint, dof, th);
2458         if (result == 0)
2459                 result = osd_ea_fid_set(info, obj->oo_inode, fid, 0);
2460
2461         if (result == 0)
2462                 result = __osd_oi_insert(env, obj, fid, th);
2463
2464         LASSERT(ergo(result == 0,
2465                      dt_object_exists(dt) && !dt_object_remote(dt)));
2466         LINVRNT(osd_invariant(obj));
2467         RETURN(result);
2468 }
2469
2470 static int osd_declare_object_ref_add(const struct lu_env *env,
2471                                       struct dt_object *dt,
2472                                       struct thandle *handle)
2473 {
2474         struct osd_thandle       *oh;
2475
2476         /* it's possible that object doesn't exist yet */
2477         LASSERT(handle != NULL);
2478
2479         oh = container_of0(handle, struct osd_thandle, ot_super);
2480         LASSERT(oh->ot_handle == NULL);
2481
2482         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2483                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2484
2485         return 0;
2486 }
2487
2488 /*
2489  * Concurrency: @dt is write locked.
2490  */
2491 static int osd_object_ref_add(const struct lu_env *env,
2492                               struct dt_object *dt, struct thandle *th)
2493 {
2494         struct osd_object *obj = osd_dt_obj(dt);
2495         struct inode      *inode = obj->oo_inode;
2496         bool               need_dirty = false;
2497         int                rc = 0;
2498
2499         LINVRNT(osd_invariant(obj));
2500         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2501         LASSERT(osd_write_locked(env, obj));
2502         LASSERT(th != NULL);
2503
2504         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2505
2506         /* This based on ldiskfs_inc_count(), which is not exported.
2507          *
2508          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2509          * (65000) subdirectories by storing "1" in i_nlink if the link count
2510          * would otherwise overflow. Directory tranversal tools understand
2511          * that (st_nlink == 1) indicates that the filesystem dose not track
2512          * hard links count on the directory, and will not abort subdirectory
2513          * scanning early once (st_nlink - 2) subdirs have been found.
2514          *
2515          * This also has to properly handle the case of inodes with nlink == 0
2516          * in case they are being linked into the PENDING directory
2517          */
2518         spin_lock(&obj->oo_guard);
2519         if (unlikely(!S_ISDIR(inode->i_mode) &&
2520                      inode->i_nlink >= LDISKFS_LINK_MAX)) {
2521                 /* MDD should have checked this, but good to be safe */
2522                 rc = -EMLINK;
2523         } else if (unlikely(inode->i_nlink == 0 ||
2524                             (S_ISDIR(inode->i_mode) &&
2525                              inode->i_nlink >= LDISKFS_LINK_MAX))) {
2526                 /* inc_nlink from 0 may cause WARN_ON */
2527                 set_nlink(inode, 1);
2528                 need_dirty = true;
2529         } else if (!S_ISDIR(inode->i_mode) ||
2530                    (S_ISDIR(inode->i_mode) && inode->i_nlink >= 2)) {
2531                 inc_nlink(inode);
2532                 need_dirty = true;
2533         } /* else (S_ISDIR(inode->i_mode) && inode->i_nlink == 1) { ; } */
2534
2535         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2536         spin_unlock(&obj->oo_guard);
2537
2538         if (need_dirty)
2539                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2540
2541         LINVRNT(osd_invariant(obj));
2542
2543         return rc;
2544 }
2545
2546 static int osd_declare_object_ref_del(const struct lu_env *env,
2547                                       struct dt_object *dt,
2548                                       struct thandle *handle)
2549 {
2550         struct osd_thandle *oh;
2551
2552         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2553         LASSERT(handle != NULL);
2554
2555         oh = container_of0(handle, struct osd_thandle, ot_super);
2556         LASSERT(oh->ot_handle == NULL);
2557
2558         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2559                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2560
2561         return 0;
2562 }
2563
2564 /*
2565  * Concurrency: @dt is write locked.
2566  */
2567 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2568                               struct thandle *th)
2569 {
2570         struct osd_object *obj = osd_dt_obj(dt);
2571         struct inode      *inode = obj->oo_inode;
2572
2573         LINVRNT(osd_invariant(obj));
2574         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2575         LASSERT(osd_write_locked(env, obj));
2576         LASSERT(th != NULL);
2577
2578         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2579
2580         spin_lock(&obj->oo_guard);
2581         LASSERT(inode->i_nlink > 0);
2582
2583         /* This based on ldiskfs_dec_count(), which is not exported.
2584          *
2585          * If a directory already has nlink == 1, then do not drop the nlink
2586          * count to 0, even temporarily, to avoid race conditions with other
2587          * threads not holding oo_guard seeing i_nlink == 0 in rare cases.
2588          *
2589          * nlink == 1 means the directory has/had > EXT4_LINK_MAX subdirs.
2590          * */
2591         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 1) {
2592                 drop_nlink(inode);
2593
2594                 spin_unlock(&obj->oo_guard);
2595                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2596                 LINVRNT(osd_invariant(obj));
2597         } else {
2598                 spin_unlock(&obj->oo_guard);
2599         }
2600
2601         return 0;
2602 }
2603
2604 /*
2605  * Get the 64-bit version for an inode.
2606  */
2607 static int osd_object_version_get(const struct lu_env *env,
2608                                   struct dt_object *dt, dt_obj_version_t *ver)
2609 {
2610         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2611
2612         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2613                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2614         *ver = LDISKFS_I(inode)->i_fs_version;
2615         return 0;
2616 }
2617
2618 /*
2619  * Concurrency: @dt is read locked.
2620  */
2621 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2622                          struct lu_buf *buf, const char *name,
2623                          struct lustre_capa *capa)
2624 {
2625         struct osd_object      *obj    = osd_dt_obj(dt);
2626         struct inode           *inode  = obj->oo_inode;
2627         struct osd_thread_info *info   = osd_oti_get(env);
2628         struct dentry          *dentry = &info->oti_obj_dentry;
2629
2630         /* version get is not real XATTR but uses xattr API */
2631         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2632                 /* for version we are just using xattr API but change inode
2633                  * field instead */
2634                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2635                 osd_object_version_get(env, dt, buf->lb_buf);
2636                 return sizeof(dt_obj_version_t);
2637         }
2638
2639         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2640         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2641
2642         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2643                 return -EACCES;
2644
2645         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2646 }
2647
2648
2649 static int osd_declare_xattr_set(const struct lu_env *env,
2650                                  struct dt_object *dt,
2651                                  const struct lu_buf *buf, const char *name,
2652                                  int fl, struct thandle *handle)
2653 {
2654         struct osd_thandle *oh;
2655
2656         LASSERT(handle != NULL);
2657
2658         oh = container_of0(handle, struct osd_thandle, ot_super);
2659         LASSERT(oh->ot_handle == NULL);
2660
2661         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2662                              strcmp(name, XATTR_NAME_VERSION) == 0 ?
2663                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2664                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2665
2666         return 0;
2667 }
2668
2669 /*
2670  * Set the 64-bit version for object
2671  */
2672 static void osd_object_version_set(const struct lu_env *env,
2673                                    struct dt_object *dt,
2674                                    dt_obj_version_t *new_version)
2675 {
2676         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2677
2678         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2679                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2680
2681         LDISKFS_I(inode)->i_fs_version = *new_version;
2682         /** Version is set after all inode operations are finished,
2683          *  so we should mark it dirty here */
2684         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2685 }
2686
2687 /*
2688  * Concurrency: @dt is write locked.
2689  */
2690 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2691                          const struct lu_buf *buf, const char *name, int fl,
2692                          struct thandle *handle, struct lustre_capa *capa)
2693 {
2694         struct osd_object      *obj      = osd_dt_obj(dt);
2695         struct inode           *inode    = obj->oo_inode;
2696         struct osd_thread_info *info     = osd_oti_get(env);
2697         int                     fs_flags = 0;
2698         ENTRY;
2699
2700         LASSERT(handle != NULL);
2701
2702         /* version set is not real XATTR */
2703         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2704                 /* for version we are just using xattr API but change inode
2705                  * field instead */
2706                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2707                 osd_object_version_set(env, dt, buf->lb_buf);
2708                 return sizeof(dt_obj_version_t);
2709         }
2710
2711         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2712                 return -EACCES;
2713
2714         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2715         if (fl & LU_XATTR_REPLACE)
2716                 fs_flags |= XATTR_REPLACE;
2717
2718         if (fl & LU_XATTR_CREATE)
2719                 fs_flags |= XATTR_CREATE;
2720
2721         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2722                                fs_flags);
2723 }
2724
2725 /*
2726  * Concurrency: @dt is read locked.
2727  */
2728 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2729                           struct lu_buf *buf, struct lustre_capa *capa)
2730 {
2731         struct osd_object      *obj    = osd_dt_obj(dt);
2732         struct inode           *inode  = obj->oo_inode;
2733         struct osd_thread_info *info   = osd_oti_get(env);
2734         struct dentry          *dentry = &info->oti_obj_dentry;
2735
2736         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2737         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2738         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2739
2740         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2741                 return -EACCES;
2742
2743         dentry->d_inode = inode;
2744         dentry->d_sb = inode->i_sb;
2745         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2746 }
2747
2748 static int osd_declare_xattr_del(const struct lu_env *env,
2749                                  struct dt_object *dt, const char *name,
2750                                  struct thandle *handle)
2751 {
2752         struct osd_thandle *oh;
2753
2754         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2755         LASSERT(handle != NULL);
2756
2757         oh = container_of0(handle, struct osd_thandle, ot_super);
2758         LASSERT(oh->ot_handle == NULL);
2759
2760         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2761                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2762
2763         return 0;
2764 }
2765
2766 /*
2767  * Concurrency: @dt is write locked.
2768  */
2769 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2770                          const char *name, struct thandle *handle,
2771                          struct lustre_capa *capa)
2772 {
2773         struct osd_object      *obj    = osd_dt_obj(dt);
2774         struct inode           *inode  = obj->oo_inode;
2775         struct osd_thread_info *info   = osd_oti_get(env);
2776         struct dentry          *dentry = &info->oti_obj_dentry;
2777         int                     rc;
2778
2779         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2780         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2781         LASSERT(handle != NULL);
2782
2783         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2784                 return -EACCES;
2785
2786         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2787
2788         ll_vfs_dq_init(inode);
2789         dentry->d_inode = inode;
2790         dentry->d_sb = inode->i_sb;
2791         rc = inode->i_op->removexattr(dentry, name);
2792         return rc;
2793 }
2794
2795 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2796                                      struct dt_object *dt,
2797                                      struct lustre_capa *old,
2798                                      __u64 opc)
2799 {
2800         struct osd_thread_info *info = osd_oti_get(env);
2801         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2802         struct osd_object *obj = osd_dt_obj(dt);
2803         struct osd_device *dev = osd_obj2dev(obj);
2804         struct lustre_capa_key *key = &info->oti_capa_key;
2805         struct lustre_capa *capa = &info->oti_capa;
2806         struct obd_capa *oc;
2807         struct md_capainfo *ci;
2808         int rc;
2809         ENTRY;
2810
2811         if (!dev->od_fl_capa)
2812                 RETURN(ERR_PTR(-ENOENT));
2813
2814         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2815         LINVRNT(osd_invariant(obj));
2816
2817         /* renewal sanity check */
2818         if (old && osd_object_auth(env, dt, old, opc))
2819                 RETURN(ERR_PTR(-EACCES));
2820
2821         ci = md_capainfo(env);
2822         if (unlikely(!ci))
2823                 RETURN(ERR_PTR(-ENOENT));
2824
2825         switch (ci->mc_auth) {
2826         case LC_ID_NONE:
2827                 RETURN(NULL);
2828         case LC_ID_PLAIN:
2829                 capa->lc_uid = obj->oo_inode->i_uid;
2830                 capa->lc_gid = obj->oo_inode->i_gid;
2831                 capa->lc_flags = LC_ID_PLAIN;
2832                 break;
2833         case LC_ID_CONVERT: {
2834                 __u32 d[4], s[4];
2835
2836                 s[0] = obj->oo_inode->i_uid;
2837                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2838                 s[2] = obj->oo_inode->i_gid;
2839                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2840                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2841                 if (unlikely(rc))
2842                         RETURN(ERR_PTR(rc));
2843
2844                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2845                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2846                 capa->lc_flags = LC_ID_CONVERT;
2847                 break;
2848         }
2849         default:
2850                 RETURN(ERR_PTR(-EINVAL));
2851         }
2852
2853         capa->lc_fid = *fid;
2854         capa->lc_opc = opc;
2855         capa->lc_flags |= dev->od_capa_alg << 24;
2856         capa->lc_timeout = dev->od_capa_timeout;
2857         capa->lc_expiry = 0;
2858
2859         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2860         if (oc) {
2861                 LASSERT(!capa_is_expired(oc));
2862                 RETURN(oc);
2863         }
2864
2865         spin_lock(&capa_lock);
2866         *key = dev->od_capa_keys[1];
2867         spin_unlock(&capa_lock);
2868
2869         capa->lc_keyid = key->lk_keyid;
2870         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2871
2872         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2873         if (rc) {
2874                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2875                 RETURN(ERR_PTR(rc));
2876         }
2877
2878         oc = capa_add(dev->od_capa_hash, capa);
2879         RETURN(oc);
2880 }
2881
2882 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2883 {
2884         struct osd_object       *obj    = osd_dt_obj(dt);
2885         struct inode            *inode  = obj->oo_inode;
2886         struct osd_thread_info  *info   = osd_oti_get(env);
2887         struct dentry           *dentry = &info->oti_obj_dentry;
2888         struct file             *file   = &info->oti_file;
2889         int                     rc;
2890
2891         ENTRY;
2892
2893         dentry->d_inode = inode;
2894         dentry->d_sb = inode->i_sb;
2895         file->f_dentry = dentry;
2896         file->f_mapping = inode->i_mapping;
2897         file->f_op = inode->i_fop;
2898 #ifndef HAVE_FILE_FSYNC_4ARGS
2899         mutex_lock(&inode->i_mutex);
2900 #endif
2901         rc = do_fsync(file, 0);
2902 #ifndef HAVE_FILE_FSYNC_4ARGS
2903         mutex_unlock(&inode->i_mutex);
2904 #endif
2905         RETURN(rc);
2906 }
2907
2908 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2909                         void **data)
2910 {
2911         struct osd_object *obj = osd_dt_obj(dt);
2912         ENTRY;
2913
2914         *data = (void *)obj->oo_inode;
2915         RETURN(0);
2916 }
2917
2918 /*
2919  * Index operations.
2920  */
2921
2922 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2923                            const struct dt_index_features *feat)
2924 {
2925         struct iam_descr *descr;
2926
2927         if (osd_object_is_root(o))
2928                 return feat == &dt_directory_features;
2929
2930         LASSERT(o->oo_dir != NULL);
2931
2932         descr = o->oo_dir->od_container.ic_descr;
2933         if (feat == &dt_directory_features) {
2934                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2935                         return 1;
2936                 else
2937                         return 0;
2938         } else {
2939                 return
2940                         feat->dif_keysize_min <= descr->id_key_size &&
2941                         descr->id_key_size <= feat->dif_keysize_max &&
2942                         feat->dif_recsize_min <= descr->id_rec_size &&
2943                         descr->id_rec_size <= feat->dif_recsize_max &&
2944                         !(feat->dif_flags & (DT_IND_VARKEY |
2945                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2946                         ergo(feat->dif_flags & DT_IND_UPDATE,
2947                              1 /* XXX check that object (and file system) is
2948                                 * writable */);
2949         }
2950 }
2951
2952 static int osd_iam_container_init(const struct lu_env *env,
2953                                   struct osd_object *obj,
2954                                   struct osd_directory *dir)
2955 {
2956         struct iam_container *bag = &dir->od_container;
2957         int result;
2958
2959         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2960         if (result != 0)
2961                 return result;
2962
2963         result = iam_container_setup(bag);
2964         if (result == 0)
2965                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2966         else
2967                 iam_container_fini(bag);
2968
2969         return result;
2970 }
2971
2972
2973 /*
2974  * Concurrency: no external locking is necessary.
2975  */
2976 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2977                          const struct dt_index_features *feat)
2978 {
2979         int                      result;
2980         int                      skip_iam = 0;
2981         struct osd_object       *obj = osd_dt_obj(dt);
2982
2983         LINVRNT(osd_invariant(obj));
2984
2985         if (osd_object_is_root(obj)) {
2986                 dt->do_index_ops = &osd_index_ea_ops;
2987                 result = 0;
2988         } else if (feat == &dt_directory_features) {
2989                 dt->do_index_ops = &osd_index_ea_ops;
2990                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2991                         result = 0;
2992                 else
2993                         result = -ENOTDIR;
2994                 skip_iam = 1;
2995         } else if (unlikely(feat == &dt_otable_features)) {
2996                 dt->do_index_ops = &osd_otable_ops;
2997                 return 0;
2998         } else if (unlikely(feat == &dt_acct_features)) {
2999                 dt->do_index_ops = &osd_acct_index_ops;
3000                 result = 0;
3001                 skip_iam = 1;
3002         } else if (!osd_has_index(obj)) {
3003                 struct osd_directory *dir;
3004
3005                 OBD_ALLOC_PTR(dir);
3006                 if (dir != NULL) {
3007
3008                         spin_lock(&obj->oo_guard);
3009                         if (obj->oo_dir == NULL)
3010                                 obj->oo_dir = dir;
3011                         else
3012                                 /*
3013                                  * Concurrent thread allocated container data.
3014                                  */
3015                                 OBD_FREE_PTR(dir);
3016                         spin_unlock(&obj->oo_guard);
3017                         /*
3018                          * Now, that we have container data, serialize its
3019                          * initialization.
3020                          */
3021                         down_write(&obj->oo_ext_idx_sem);
3022                         /*
3023                          * recheck under lock.
3024                          */
3025                         if (!osd_has_index(obj))
3026                                 result = osd_iam_container_init(env, obj, dir);
3027                         else
3028                                 result = 0;
3029                         up_write(&obj->oo_ext_idx_sem);
3030                 } else {
3031                         result = -ENOMEM;
3032                 }
3033         } else {
3034                 result = 0;
3035         }
3036
3037         if (result == 0 && skip_iam == 0) {
3038                 if (!osd_iam_index_probe(env, obj, feat))
3039                         result = -ENOTDIR;
3040         }
3041         LINVRNT(osd_invariant(obj));
3042
3043         if (result == 0 && is_quota_glb_feat(feat) &&
3044             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3045                 result = osd_quota_migration(env, dt, feat);
3046
3047         return result;
3048 }
3049
3050 static int osd_otable_it_attr_get(const struct lu_env *env,
3051                                  struct dt_object *dt,
3052                                  struct lu_attr *attr,
3053                                  struct lustre_capa *capa)
3054 {
3055         attr->la_valid = 0;
3056         return 0;
3057 }
3058
3059 static const struct dt_object_operations osd_obj_ops = {
3060         .do_read_lock         = osd_object_read_lock,
3061         .do_write_lock        = osd_object_write_lock,
3062         .do_read_unlock       = osd_object_read_unlock,
3063         .do_write_unlock      = osd_object_write_unlock,
3064         .do_write_locked      = osd_object_write_locked,
3065         .do_attr_get          = osd_attr_get,
3066         .do_declare_attr_set  = osd_declare_attr_set,
3067         .do_attr_set          = osd_attr_set,
3068         .do_ah_init           = osd_ah_init,
3069         .do_declare_create    = osd_declare_object_create,
3070         .do_create            = osd_object_create,
3071         .do_declare_destroy   = osd_declare_object_destroy,
3072         .do_destroy           = osd_object_destroy,
3073         .do_index_try         = osd_index_try,
3074         .do_declare_ref_add   = osd_declare_object_ref_add,
3075         .do_ref_add           = osd_object_ref_add,
3076         .do_declare_ref_del   = osd_declare_object_ref_del,
3077         .do_ref_del           = osd_object_ref_del,
3078         .do_xattr_get         = osd_xattr_get,
3079         .do_declare_xattr_set = osd_declare_xattr_set,
3080         .do_xattr_set         = osd_xattr_set,
3081         .do_declare_xattr_del = osd_declare_xattr_del,
3082         .do_xattr_del         = osd_xattr_del,
3083         .do_xattr_list        = osd_xattr_list,
3084         .do_capa_get          = osd_capa_get,
3085         .do_object_sync       = osd_object_sync,
3086         .do_data_get          = osd_data_get,
3087 };
3088
3089 /**
3090  * dt_object_operations for interoperability mode
3091  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3092  */
3093 static const struct dt_object_operations osd_obj_ea_ops = {
3094         .do_read_lock         = osd_object_read_lock,
3095         .do_write_lock        = osd_object_write_lock,
3096         .do_read_unlock       = osd_object_read_unlock,
3097         .do_write_unlock      = osd_object_write_unlock,
3098         .do_write_locked      = osd_object_write_locked,
3099         .do_attr_get          = osd_attr_get,
3100         .do_declare_attr_set  = osd_declare_attr_set,
3101         .do_attr_set          = osd_attr_set,
3102         .do_ah_init           = osd_ah_init,
3103         .do_declare_create    = osd_declare_object_create,
3104         .do_create            = osd_object_ea_create,
3105         .do_declare_destroy   = osd_declare_object_destroy,
3106         .do_destroy           = osd_object_destroy,
3107         .do_index_try         = osd_index_try,
3108         .do_declare_ref_add   = osd_declare_object_ref_add,
3109         .do_ref_add           = osd_object_ref_add,
3110         .do_declare_ref_del   = osd_declare_object_ref_del,
3111         .do_ref_del           = osd_object_ref_del,
3112         .do_xattr_get         = osd_xattr_get,
3113         .do_declare_xattr_set = osd_declare_xattr_set,
3114         .do_xattr_set         = osd_xattr_set,
3115         .do_declare_xattr_del = osd_declare_xattr_del,
3116         .do_xattr_del         = osd_xattr_del,
3117         .do_xattr_list        = osd_xattr_list,
3118         .do_capa_get          = osd_capa_get,
3119         .do_object_sync       = osd_object_sync,
3120         .do_data_get          = osd_data_get,
3121 };
3122
3123 static const struct dt_object_operations osd_obj_otable_it_ops = {
3124         .do_attr_get    = osd_otable_it_attr_get,
3125         .do_index_try   = osd_index_try,
3126 };
3127
3128 static int osd_index_declare_iam_delete(const struct lu_env *env,
3129                                         struct dt_object *dt,
3130                                         const struct dt_key *key,
3131                                         struct thandle *handle)
3132 {
3133         struct osd_thandle    *oh;
3134
3135         oh = container_of0(handle, struct osd_thandle, ot_super);
3136         LASSERT(oh->ot_handle == NULL);
3137
3138         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3139                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3140
3141         return 0;
3142 }
3143
3144 /**
3145  *      delete a (key, value) pair from index \a dt specified by \a key
3146  *
3147  *      \param  dt      osd index object
3148  *      \param  key     key for index
3149  *      \param  rec     record reference
3150  *      \param  handle  transaction handler
3151  *
3152  *      \retval  0  success
3153  *      \retval -ve   failure
3154  */
3155
3156 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3157                                 const struct dt_key *key,
3158                                 struct thandle *handle,
3159                                 struct lustre_capa *capa)
3160 {
3161         struct osd_thread_info *oti = osd_oti_get(env);
3162         struct osd_object      *obj = osd_dt_obj(dt);
3163         struct osd_thandle     *oh;
3164         struct iam_path_descr  *ipd;
3165         struct iam_container   *bag = &obj->oo_dir->od_container;
3166         int                     rc;
3167
3168         ENTRY;
3169
3170         LINVRNT(osd_invariant(obj));
3171         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3172         LASSERT(bag->ic_object == obj->oo_inode);
3173         LASSERT(handle != NULL);
3174
3175         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3176                 RETURN(-EACCES);
3177
3178         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3179
3180         ipd = osd_idx_ipd_get(env, bag);
3181         if (unlikely(ipd == NULL))
3182                 RETURN(-ENOMEM);
3183
3184         oh = container_of0(handle, struct osd_thandle, ot_super);
3185         LASSERT(oh->ot_handle != NULL);
3186         LASSERT(oh->ot_handle->h_transaction != NULL);
3187
3188         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3189                 /* swab quota uid/gid provided by caller */
3190                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3191                 key = (const struct dt_key *)&oti->oti_quota_id;
3192         }
3193
3194         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3195         osd_ipd_put(env, bag, ipd);
3196         LINVRNT(osd_invariant(obj));
3197         RETURN(rc);
3198 }
3199
3200 static int osd_index_declare_ea_delete(const struct lu_env *env,
3201                                        struct dt_object *dt,
3202                                        const struct dt_key *key,
3203                                        struct thandle *handle)
3204 {
3205         struct osd_thandle *oh;
3206         struct inode       *inode;
3207         int                 rc;
3208         ENTRY;
3209
3210         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3211         LASSERT(handle != NULL);
3212
3213         oh = container_of0(handle, struct osd_thandle, ot_super);
3214         LASSERT(oh->ot_handle == NULL);
3215
3216         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3217                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3218
3219         inode = osd_dt_obj(dt)->oo_inode;
3220         LASSERT(inode);
3221
3222         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3223                                    true, true, NULL, false);
3224         RETURN(rc);
3225 }
3226
3227 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3228                                           struct dt_rec *fid)
3229 {
3230         struct osd_fid_pack *rec;
3231         int                  rc = -ENODATA;
3232
3233         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3234                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3235                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3236         }
3237         return rc;
3238 }
3239
3240 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3241                           struct lu_fid *fid)
3242 {
3243         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
3244         struct seq_server_site  *ss = osd_seq_site(osd);
3245         int                     rc;
3246         ENTRY;
3247
3248         /* Those FID seqs, which are not in FLDB, must be local seq */
3249         if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
3250                 RETURN(0);
3251
3252         rc = osd_fld_lookup(env, osd, fid, range);
3253         if (rc != 0) {
3254                 CERROR("%s: Can not lookup fld for "DFID"\n",
3255                        osd_name(osd), PFID(fid));
3256                 RETURN(rc);
3257         }
3258
3259         RETURN(ss->ss_node_id != range->lsr_index);
3260 }
3261
3262 /**
3263  * Index delete function for interoperability mode (b11826).
3264  * It will remove the directory entry added by osd_index_ea_insert().
3265  * This entry is needed to maintain name->fid mapping.
3266  *
3267  * \param key,  key i.e. file entry to be deleted
3268  *
3269  * \retval   0, on success
3270  * \retval -ve, on error
3271  */
3272 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3273                                const struct dt_key *key,
3274                                struct thandle *handle,
3275                                struct lustre_capa *capa)
3276 {
3277         struct osd_object          *obj    = osd_dt_obj(dt);
3278         struct inode               *dir    = obj->oo_inode;
3279         struct dentry              *dentry;
3280         struct osd_thandle         *oh;
3281         struct ldiskfs_dir_entry_2 *de = NULL;
3282         struct buffer_head         *bh;
3283         struct htree_lock          *hlock = NULL;
3284         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3285         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3286         int                        rc;
3287         ENTRY;
3288
3289         LINVRNT(osd_invariant(obj));
3290         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3291         LASSERT(handle != NULL);
3292
3293         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3294
3295         oh = container_of(handle, struct osd_thandle, ot_super);
3296         LASSERT(oh->ot_handle != NULL);
3297         LASSERT(oh->ot_handle->h_transaction != NULL);
3298
3299         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3300                 RETURN(-EACCES);
3301
3302         ll_vfs_dq_init(dir);
3303         dentry = osd_child_dentry_get(env, obj,
3304                                       (char *)key, strlen((char *)key));
3305
3306         if (obj->oo_hl_head != NULL) {
3307                 hlock = osd_oti_get(env)->oti_hlock;
3308                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3309                                    dir, LDISKFS_HLOCK_DEL);
3310         } else {
3311                 down_write(&obj->oo_ext_idx_sem);
3312         }
3313
3314         bh = ldiskfs_find_entry(dir, &dentry->d_name, &de, hlock);
3315         if (bh) {
3316                 __u32 ino = 0;
3317
3318                 /* If this is not the ".." entry, it might be a remote DNE
3319                  * entry and  we need to check if the FID is for a remote
3320                  * MDT.  If the FID is  not in the directory entry (e.g.
3321                  * upgraded 1.8 filesystem without dirdata enabled) then
3322                  * we need to get the FID from the LMA. For a remote directory
3323                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3324                  * case.
3325                  *
3326                  * Delete the entry before the agent inode in order to
3327                  * simplify error handling.  At worst an error after deleting
3328                  * the entry first might leak the agent inode afterward. The
3329                  * reverse would need filesystem abort in case of error deleting
3330                  * the entry after the agent had been removed, or leave a
3331                  * dangling entry pointing at a random inode. */
3332                 if (strcmp((char *)key, dotdot) != 0) {
3333                         LASSERT(de != NULL);
3334                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3335                         /* If Fid is not in dentry, try to get it from LMA */
3336                         if (rc == -ENODATA) {
3337                                 struct osd_inode_id *id;
3338                                 struct inode *inode;
3339
3340                                 /* Before trying to get fid from the inode,
3341                                  * check whether the inode is valid.
3342                                  *
3343                                  * If the inode has been deleted, do not go
3344                                  * ahead to do osd_ea_fid_get, which will set
3345                                  * the inode to bad inode, which might cause
3346                                  * the inode to be deleted uncorrectly */
3347                                 inode = ldiskfs_iget(osd_sb(osd),
3348                                                      le32_to_cpu(de->inode));
3349                                 if (IS_ERR(inode)) {
3350                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3351                                                "error.\n", osd_name(osd),
3352                                                PFID(fid));
3353                                         rc = PTR_ERR(inode);
3354                                 } else {
3355                                         if (likely(inode->i_nlink != 0)) {
3356                                                 id = &osd_oti_get(env)->oti_id;
3357                                                 rc = osd_ea_fid_get(env, obj,
3358                                                         le32_to_cpu(de->inode),
3359                                                                     fid, id);
3360                                         } else {
3361                                                 CDEBUG(D_INFO, "%s: %u "DFID
3362                                                        "deleted.\n",
3363                                                        osd_name(osd),
3364                                                        le32_to_cpu(de->inode),
3365                                                        PFID(fid));
3366                                                 rc = -ESTALE;
3367                                         }
3368                                         iput(inode);
3369                                 }
3370                         }
3371                         if (rc == 0 &&
3372                             unlikely(osd_remote_fid(env, osd, fid)))
3373                                 /* Need to delete agent inode */
3374                                 ino = le32_to_cpu(de->inode);
3375                 }
3376                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3377                 brelse(bh);
3378                 if (rc == 0 && unlikely(ino != 0)) {
3379                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3380                                                           oh);
3381                         if (rc != 0)
3382                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3383                                        osd_name(osd), PFID(fid), rc);
3384                 }
3385         } else {
3386                 rc = -ENOENT;
3387         }
3388         if (hlock != NULL)
3389                 ldiskfs_htree_unlock(hlock);
3390         else
3391                 up_write(&obj->oo_ext_idx_sem);
3392
3393         if (rc != 0)
3394                 GOTO(out, rc);
3395
3396         /* For inode on the remote MDT, .. will point to
3397          * /Agent directory, Check whether it needs to delete
3398          * from agent directory */
3399         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3400                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3401                                                    oh);
3402                 if (rc != 0 && rc != -ENOENT) {
3403                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3404                                osd_name(osd), PFID(fid), rc);
3405                 }
3406
3407                 if (rc == -ENOENT)
3408                         rc = 0;
3409
3410                 GOTO(out, rc);
3411         }
3412 out:
3413
3414         LASSERT(osd_invariant(obj));
3415         RETURN(rc);
3416 }
3417
3418 /**
3419  *      Lookup index for \a key and copy record to \a rec.
3420  *
3421  *      \param  dt      osd index object
3422  *      \param  key     key for index
3423  *      \param  rec     record reference
3424  *
3425  *      \retval  +ve  success : exact mach
3426  *      \retval  0    return record with key not greater than \a key
3427  *      \retval -ve   failure
3428  */
3429 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3430                                 struct dt_rec *rec, const struct dt_key *key,
3431                                 struct lustre_capa *capa)
3432 {
3433         struct osd_object      *obj = osd_dt_obj(dt);
3434         struct iam_path_descr  *ipd;
3435         struct iam_container   *bag = &obj->oo_dir->od_container;
3436         struct osd_thread_info *oti = osd_oti_get(env);
3437         struct iam_iterator    *it = &oti->oti_idx_it;
3438         struct iam_rec         *iam_rec;
3439         int                     rc;
3440
3441         ENTRY;
3442
3443         LASSERT(osd_invariant(obj));
3444         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3445         LASSERT(bag->ic_object == obj->oo_inode);
3446
3447         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3448                 RETURN(-EACCES);
3449
3450         ipd = osd_idx_ipd_get(env, bag);
3451         if (IS_ERR(ipd))
3452                 RETURN(-ENOMEM);
3453
3454         /* got ipd now we can start iterator. */
3455         iam_it_init(it, bag, 0, ipd);
3456
3457         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3458                 /* swab quota uid/gid provided by caller */
3459                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3460                 key = (const struct dt_key *)&oti->oti_quota_id;
3461         }
3462
3463         rc = iam_it_get(it, (struct iam_key *)key);
3464         if (rc >= 0) {
3465                 if (S_ISDIR(obj->oo_inode->i_mode))
3466                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3467                 else
3468                         iam_rec = (struct iam_rec *) rec;
3469
3470                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3471
3472                 if (S_ISDIR(obj->oo_inode->i_mode))
3473                         osd_fid_unpack((struct lu_fid *) rec,
3474                                        (struct osd_fid_pack *)iam_rec);
3475                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3476                         osd_quota_unpack(obj, rec);
3477         }
3478
3479         iam_it_put(it);
3480         iam_it_fini(it);
3481         osd_ipd_put(env, bag, ipd);
3482
3483         LINVRNT(osd_invariant(obj));
3484
3485         RETURN(rc);
3486 }
3487
3488 static int osd_index_declare_iam_insert(const struct lu_env *env,
3489                                         struct dt_object *dt,
3490                                         const struct dt_rec *rec,
3491                                         const struct dt_key *key,
3492                                         struct thandle *handle)
3493 {
3494         struct osd_thandle *oh;
3495
3496         LASSERT(handle != NULL);
3497
3498         oh = container_of0(handle, struct osd_thandle, ot_super);
3499         LASSERT(oh->ot_handle == NULL);
3500
3501         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3502                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3503
3504         return 0;
3505 }
3506
3507 /**
3508  *      Inserts (key, value) pair in \a dt index object.
3509  *
3510  *      \param  dt      osd index object
3511  *      \param  key     key for index
3512  *      \param  rec     record reference
3513  *      \param  th      transaction handler
3514  *
3515  *      \retval  0  success
3516  *      \retval -ve failure
3517  */
3518 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3519                                 const struct dt_rec *rec,
3520                                 const struct dt_key *key, struct thandle *th,
3521                                 struct lustre_capa *capa, int ignore_quota)
3522 {
3523         struct osd_object     *obj = osd_dt_obj(dt);
3524         struct iam_path_descr *ipd;
3525         struct osd_thandle    *oh;
3526         struct iam_container  *bag = &obj->oo_dir->od_container;
3527         struct osd_thread_info *oti = osd_oti_get(env);
3528         struct iam_rec         *iam_rec;
3529         int                     rc;
3530
3531         ENTRY;
3532
3533         LINVRNT(osd_invariant(obj));
3534         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3535         LASSERT(bag->ic_object == obj->oo_inode);
3536         LASSERT(th != NULL);
3537
3538         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3539                 RETURN(-EACCES);
3540
3541         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3542
3543         ipd = osd_idx_ipd_get(env, bag);
3544         if (unlikely(ipd == NULL))
3545                 RETURN(-ENOMEM);
3546
3547         oh = container_of0(th, struct osd_thandle, ot_super);
3548         LASSERT(oh->ot_handle != NULL);
3549         LASSERT(oh->ot_handle->h_transaction != NULL);
3550         if (S_ISDIR(obj->oo_inode->i_mode)) {
3551                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3552                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3553         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3554                 /* pack quota uid/gid */
3555                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3556                 key = (const struct dt_key *)&oti->oti_quota_id;
3557                 /* pack quota record */
3558                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3559                 iam_rec = (struct iam_rec *)rec;
3560         } else {
3561                 iam_rec = (struct iam_rec *)rec;
3562         }
3563
3564         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3565                         iam_rec, ipd);
3566         osd_ipd_put(env, bag, ipd);
3567         LINVRNT(osd_invariant(obj));
3568         RETURN(rc);
3569 }
3570
3571 /**
3572  * Calls ldiskfs_add_entry() to add directory entry
3573  * into the directory. This is required for
3574  * interoperability mode (b11826)
3575  *
3576  * \retval   0, on success
3577  * \retval -ve, on error
3578  */
3579 static int __osd_ea_add_rec(struct osd_thread_info *info,
3580                             struct osd_object *pobj, struct inode  *cinode,
3581                             const char *name, const struct dt_rec *fid,
3582                             struct htree_lock *hlock, struct thandle *th)
3583 {
3584         struct ldiskfs_dentry_param *ldp;
3585         struct dentry               *child;
3586         struct osd_thandle          *oth;
3587         int                          rc;
3588
3589         oth = container_of(th, struct osd_thandle, ot_super);
3590         LASSERT(oth->ot_handle != NULL);
3591         LASSERT(oth->ot_handle->h_transaction != NULL);
3592         LASSERT(pobj->oo_inode);
3593
3594         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3595         if (unlikely(pobj->oo_inode ==
3596                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3597                 ldp->edp_magic = 0;
3598         else
3599                 osd_get_ldiskfs_dirent_param(ldp, fid);
3600         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3601         child->d_fsdata = (void *)ldp;
3602         ll_vfs_dq_init(pobj->oo_inode);
3603         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3604
3605         RETURN(rc);
3606 }
3607
3608 /**
3609  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3610  * into the directory.Also sets flags into osd object to
3611  * indicate dot and dotdot are created. This is required for
3612  * interoperability mode (b11826)
3613  *
3614  * \param dir   directory for dot and dotdot fixup.
3615  * \param obj   child object for linking
3616  *
3617  * \retval   0, on success
3618  * \retval -ve, on error
3619  */
3620 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3621                               struct osd_object *dir,
3622                               struct inode  *parent_dir, const char *name,
3623                               const struct dt_rec *dot_fid,
3624                               const struct dt_rec *dot_dot_fid,
3625                               struct thandle *th)
3626 {
3627         struct inode                *inode = dir->oo_inode;
3628         struct osd_thandle          *oth;
3629         int result = 0;
3630
3631         oth = container_of(th, struct osd_thandle, ot_super);
3632         LASSERT(oth->ot_handle->h_transaction != NULL);
3633         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3634
3635         if (strcmp(name, dot) == 0) {
3636                 if (dir->oo_compat_dot_created) {
3637                         result = -EEXIST;
3638                 } else {
3639                         LASSERT(inode == parent_dir);
3640                         dir->oo_compat_dot_created = 1;
3641                         result = 0;
3642                 }
3643         } else if (strcmp(name, dotdot) == 0) {
3644                 if (!dir->oo_compat_dot_created)
3645                         return -EINVAL;
3646                 /* in case of rename, dotdot is already created */
3647                 if (dir->oo_compat_dotdot_created) {
3648                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3649                                                 dot_dot_fid, NULL, th);
3650                 }
3651
3652                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3653                                                 parent_dir, dot_fid,
3654                                                 dot_dot_fid, oth);
3655                 if (result == 0)
3656                         dir->oo_compat_dotdot_created = 1;
3657         }
3658
3659         return result;
3660 }
3661
3662
3663 /**
3664  * It will call the appropriate osd_add* function and return the
3665  * value, return by respective functions.
3666  */
3667 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3668                           struct inode *cinode, const char *name,
3669                           const struct dt_rec *fid, struct thandle *th)
3670 {
3671         struct osd_thread_info *info   = osd_oti_get(env);
3672         struct htree_lock      *hlock;
3673         int                     rc;
3674
3675         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3676
3677         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3678                                                    name[2] =='\0'))) {
3679                 if (hlock != NULL) {
3680                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3681                                            pobj->oo_inode, 0);
3682                 } else {
3683                         down_write(&pobj->oo_ext_idx_sem);
3684                 }
3685                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3686                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3687                                         fid, th);
3688         } else {
3689                 if (hlock != NULL) {
3690                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3691                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3692                 } else {
3693                         down_write(&pobj->oo_ext_idx_sem);
3694                 }
3695
3696                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3697                         struct lu_fid *tfid = &info->oti_fid;
3698
3699                         *tfid = *(const struct lu_fid *)fid;
3700                         tfid->f_ver = ~0;
3701                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3702                                               (const struct dt_rec *)tfid,
3703                                               hlock, th);
3704                 } else {
3705                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3706                                               hlock, th);
3707                 }
3708         }
3709         if (hlock != NULL)
3710                 ldiskfs_htree_unlock(hlock);
3711         else
3712                 up_write(&pobj->oo_ext_idx_sem);
3713
3714         return rc;
3715 }
3716
3717 static void
3718 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3719                       struct osd_idmap_cache *oic)
3720 {
3721         struct osd_scrub    *scrub = &dev->od_scrub;
3722         struct lu_fid       *fid   = &oic->oic_fid;
3723         struct osd_inode_id *id    = &oti->oti_id;
3724         int                  once  = 0;
3725         int                  rc;
3726         ENTRY;
3727
3728         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3729                 RETURN_EXIT;
3730
3731 again:
3732         rc = osd_oi_lookup(oti, dev, fid, id, true);
3733         if (rc != 0 && rc != -ENOENT)
3734                 RETURN_EXIT;
3735
3736         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3737                 RETURN_EXIT;
3738
3739         if (thread_is_running(&scrub->os_thread)) {
3740                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3741                 /* There is race condition between osd_oi_lookup and OI scrub.
3742                  * The OI scrub finished just after osd_oi_lookup() failure.
3743                  * Under such case, it is unnecessary to trigger OI scrub again,
3744                  * but try to call osd_oi_lookup() again. */
3745                 if (unlikely(rc == -EAGAIN))
3746                         goto again;
3747
3748                 RETURN_EXIT;
3749         }
3750
3751         if (!dev->od_noscrub && ++once == 1) {
3752                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3753                        PFID(fid));
3754                 rc = osd_scrub_start(dev);
3755                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3756                                ", rc = %d [2]\n",
3757                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3758                                PFID(fid), rc);
3759                 if (rc == 0)
3760                         goto again;
3761         }
3762
3763         EXIT;
3764 }
3765
3766 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3767                                struct osd_device *dev,
3768                                struct osd_idmap_cache *oic,
3769                                struct lu_fid *fid, __u32 ino)
3770 {
3771         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3772         struct inode            *inode;
3773         int                      rc;
3774
3775         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3776         inode = osd_iget(oti, dev, &oic->oic_lid);
3777         if (IS_ERR(inode)) {
3778                 fid_zero(&oic->oic_fid);
3779                 return PTR_ERR(inode);
3780         }
3781
3782         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3783         iput(inode);
3784         if (rc != 0)
3785                 fid_zero(&oic->oic_fid);
3786         else
3787                 *fid = oic->oic_fid = lma->lma_self_fid;
3788         return rc;
3789 }
3790
3791 static int osd_add_oi_cache(struct osd_thread_info *info,
3792                             struct osd_device *osd,
3793                             struct osd_inode_id *id,
3794                             struct lu_fid *fid)
3795 {
3796         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
3797                id->oii_ino, id->oii_gen, info);
3798         info->oti_cache.oic_lid = *id;
3799         info->oti_cache.oic_fid = *fid;
3800
3801         return 0;
3802 }
3803
3804 /**
3805  * Calls ->lookup() to find dentry. From dentry get inode and
3806  * read inode's ea to get fid. This is required for  interoperability
3807  * mode (b11826)
3808  *
3809  * \retval   0, on success
3810  * \retval -ve, on error
3811  */
3812 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3813                              struct dt_rec *rec, const struct dt_key *key)
3814 {
3815         struct inode               *dir    = obj->oo_inode;
3816         struct dentry              *dentry;
3817         struct ldiskfs_dir_entry_2 *de;
3818         struct buffer_head         *bh;
3819         struct lu_fid              *fid = (struct lu_fid *) rec;
3820         struct htree_lock          *hlock = NULL;
3821         int                         ino;
3822         int                         rc;
3823         ENTRY;
3824
3825         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3826
3827         dentry = osd_child_dentry_get(env, obj,
3828                                       (char *)key, strlen((char *)key));
3829
3830         if (obj->oo_hl_head != NULL) {
3831                 hlock = osd_oti_get(env)->oti_hlock;
3832                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3833                                    dir, LDISKFS_HLOCK_LOOKUP);
3834         } else {
3835                 down_read(&obj->oo_ext_idx_sem);
3836         }
3837
3838         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3839         if (bh) {
3840                 struct osd_thread_info *oti = osd_oti_get(env);
3841                 struct osd_inode_id *id = &oti->oti_id;
3842                 struct osd_idmap_cache *oic = &oti->oti_cache;
3843                 struct osd_device *dev = osd_obj2dev(obj);
3844                 struct osd_scrub *scrub = &dev->od_scrub;
3845                 struct scrub_file *sf = &scrub->os_file;
3846
3847                 ino = le32_to_cpu(de->inode);
3848                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
3849                         brelse(bh);
3850                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
3851                         GOTO(out, rc);
3852                 }
3853
3854                 rc = osd_get_fid_from_dentry(de, rec);
3855
3856                 /* done with de, release bh */
3857                 brelse(bh);
3858                 if (rc != 0)
3859                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
3860                 else
3861                         osd_id_gen(id, ino, OSD_OII_NOGEN);
3862                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
3863                         fid_zero(&oic->oic_fid);
3864                         GOTO(out, rc);
3865                 }
3866
3867                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
3868                                       fid);
3869                 if (rc != 0)
3870                         GOTO(out, rc);
3871                 if ((scrub->os_pos_current <= ino) &&
3872                     ((sf->sf_flags & SF_INCONSISTENT) ||
3873                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
3874                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3875                                       sf->sf_oi_bitmap)))
3876                         osd_consistency_check(oti, dev, oic);
3877         } else {
3878                 rc = -ENOENT;
3879         }
3880
3881         GOTO(out, rc);
3882
3883 out:
3884         if (hlock != NULL)
3885                 ldiskfs_htree_unlock(hlock);
3886         else
3887                 up_read(&obj->oo_ext_idx_sem);
3888         return rc;
3889 }
3890
3891 /**
3892  * Find the osd object for given fid.
3893  *
3894  * \param fid need to find the osd object having this fid
3895  *
3896  * \retval osd_object on success
3897  * \retval        -ve on error
3898  */
3899 struct osd_object *osd_object_find(const struct lu_env *env,
3900                                    struct dt_object *dt,
3901                                    const struct lu_fid *fid)
3902 {
3903         struct lu_device  *ludev = dt->do_lu.lo_dev;
3904         struct osd_object *child = NULL;
3905         struct lu_object  *luch;
3906         struct lu_object  *lo;
3907
3908         /*
3909          * at this point topdev might not exist yet
3910          * (i.e. MGS is preparing profiles). so we can
3911          * not rely on topdev and instead lookup with
3912          * our device passed as topdev. this can't work
3913          * if the object isn't cached yet (as osd doesn't
3914          * allocate lu_header). IOW, the object must be
3915          * in the cache, otherwise lu_object_alloc() crashes
3916          * -bzzz
3917          */
3918         luch = lu_object_find_at(env, ludev, fid, NULL);
3919         if (!IS_ERR(luch)) {
3920                 if (lu_object_exists(luch)) {
3921                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3922                         if (lo != NULL)
3923                                 child = osd_obj(lo);
3924                         else
3925                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3926                                                 "lu_object can't be located"
3927                                                 DFID"\n", PFID(fid));
3928
3929                         if (child == NULL) {
3930                                 lu_object_put(env, luch);
3931                                 CERROR("Unable to get osd_object\n");
3932                                 child = ERR_PTR(-ENOENT);
3933                         }
3934                 } else {
3935                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3936                                         "lu_object does not exists "DFID"\n",
3937                                         PFID(fid));
3938                         lu_object_put(env, luch);
3939                         child = ERR_PTR(-ENOENT);
3940                 }
3941         } else
3942                 child = (void *)luch;
3943
3944         return child;
3945 }
3946
3947 /**
3948  * Put the osd object once done with it.
3949  *
3950  * \param obj osd object that needs to be put
3951  */
3952 static inline void osd_object_put(const struct lu_env *env,
3953                                   struct osd_object *obj)
3954 {
3955         lu_object_put(env, &obj->oo_dt.do_lu);
3956 }
3957
3958 static int osd_index_declare_ea_insert(const struct lu_env *env,
3959                                        struct dt_object *dt,
3960                                        const struct dt_rec *rec,
3961                                        const struct dt_key *key,
3962                                        struct thandle *handle)
3963 {
3964         struct osd_thandle      *oh;
3965         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
3966         struct lu_fid           *fid = (struct lu_fid *)rec;
3967         int                     rc;
3968         ENTRY;
3969
3970         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3971         LASSERT(handle != NULL);
3972
3973         oh = container_of0(handle, struct osd_thandle, ot_super);
3974         LASSERT(oh->ot_handle == NULL);
3975
3976         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3977                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3978
3979         if (osd_dt_obj(dt)->oo_inode == NULL) {
3980                 const char *name  = (const char *)key;
3981                 /* Object is not being created yet. Only happens when
3982                  *     1. declare directory create
3983                  *     2. declare insert .
3984                  *     3. declare insert ..
3985                  */
3986                 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
3987         } else {
3988                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3989
3990                 /* We ignore block quota on meta pool (MDTs), so needn't
3991                  * calculate how many blocks will be consumed by this index
3992                  * insert */
3993                 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
3994                                            oh, true, true, NULL, false);
3995         }
3996
3997         if (fid == NULL)
3998                 RETURN(0);
3999
4000         rc = osd_remote_fid(env, osd, fid);
4001         if (rc <= 0)
4002                 RETURN(rc);
4003
4004         rc = 0;
4005
4006         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4007                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4008         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4009                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4010         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4011                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4012
4013         RETURN(rc);
4014 }
4015
4016 /**
4017  * Index add function for interoperability mode (b11826).
4018  * It will add the directory entry.This entry is needed to
4019  * maintain name->fid mapping.
4020  *
4021  * \param key it is key i.e. file entry to be inserted
4022  * \param rec it is value of given key i.e. fid
4023  *
4024  * \retval   0, on success
4025  * \retval -ve, on error
4026  */
4027 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4028                                const struct dt_rec *rec,
4029                                const struct dt_key *key, struct thandle *th,
4030                                struct lustre_capa *capa, int ignore_quota)
4031 {
4032         struct osd_object       *obj = osd_dt_obj(dt);
4033         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4034         struct lu_fid           *fid = (struct lu_fid *) rec;
4035         const char              *name = (const char *)key;
4036         struct osd_thread_info  *oti   = osd_oti_get(env);
4037         struct osd_inode_id     *id    = &oti->oti_id;
4038         struct inode            *child_inode = NULL;
4039         struct osd_object       *child = NULL;
4040         int                     rc;
4041         ENTRY;
4042
4043         LASSERT(osd_invariant(obj));
4044         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4045         LASSERT(th != NULL);
4046
4047         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4048
4049         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4050                 RETURN(-EACCES);
4051
4052         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4053
4054         rc = osd_remote_fid(env, osd, fid);
4055         if (rc < 0) {
4056                 CERROR("%s: Can not find object "DFID" rc %d\n",
4057                        osd_name(osd), PFID(fid), rc);
4058                 RETURN(rc);
4059         }
4060
4061         if (rc == 1) {
4062                 /* Insert remote entry */
4063                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4064                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4065                         struct osd_thandle      *oh;
4066
4067                         /* If parent on remote MDT, we need put this object
4068                          * under AGENT */
4069                         oh = container_of(th, typeof(*oh), ot_super);
4070                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4071                         if (rc != 0) {
4072                                 CERROR("%s: add "DFID" error: rc = %d\n",
4073                                        osd_name(osd),
4074                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4075                                 RETURN(rc);
4076                         }
4077
4078                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4079                 } else {
4080                         child_inode = osd_create_local_agent_inode(env, osd,
4081                                                                    obj, fid,
4082                                                                    th);
4083                         if (IS_ERR(child_inode))
4084                                 RETURN(PTR_ERR(child_inode));
4085                 }
4086         } else {
4087                 /* Insert local entry */
4088                 child = osd_object_find(env, dt, fid);
4089                 if (IS_ERR(child)) {
4090                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4091                                osd_name(osd), PFID(fid),
4092                                id->oii_ino, id->oii_gen,
4093                                (int)PTR_ERR(child_inode));
4094                         RETURN(PTR_ERR(child_inode));
4095                 }
4096                 child_inode = igrab(child->oo_inode);
4097         }
4098
4099         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4100
4101         iput(child_inode);
4102         if (child != NULL)
4103                 osd_object_put(env, child);
4104         LASSERT(osd_invariant(obj));
4105         RETURN(rc);
4106 }
4107
4108 /**
4109  *  Initialize osd Iterator for given osd index object.
4110  *
4111  *  \param  dt      osd index object
4112  */
4113
4114 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4115                                      struct dt_object *dt,
4116                                      __u32 unused,
4117                                      struct lustre_capa *capa)
4118 {
4119         struct osd_it_iam      *it;
4120         struct osd_thread_info *oti = osd_oti_get(env);
4121         struct osd_object      *obj = osd_dt_obj(dt);
4122         struct lu_object       *lo  = &dt->do_lu;
4123         struct iam_path_descr  *ipd;
4124         struct iam_container   *bag = &obj->oo_dir->od_container;
4125
4126         LASSERT(lu_object_exists(lo));
4127
4128         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4129                 return ERR_PTR(-EACCES);
4130
4131         it = &oti->oti_it;
4132         ipd = osd_it_ipd_get(env, bag);
4133         if (likely(ipd != NULL)) {
4134                 it->oi_obj = obj;
4135                 it->oi_ipd = ipd;
4136                 lu_object_get(lo);
4137                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4138                 return (struct dt_it *)it;
4139         }
4140         return ERR_PTR(-ENOMEM);
4141 }
4142
4143 /**
4144  * free given Iterator.
4145  */
4146
4147 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4148 {
4149         struct osd_it_iam *it = (struct osd_it_iam *)di;
4150         struct osd_object *obj = it->oi_obj;
4151
4152         iam_it_fini(&it->oi_it);
4153         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4154         lu_object_put(env, &obj->oo_dt.do_lu);
4155 }
4156
4157 /**
4158  *  Move Iterator to record specified by \a key
4159  *
4160  *  \param  di      osd iterator
4161  *  \param  key     key for index
4162  *
4163  *  \retval +ve  di points to record with least key not larger than key
4164  *  \retval  0   di points to exact matched key
4165  *  \retval -ve  failure
4166  */
4167
4168 static int osd_it_iam_get(const struct lu_env *env,
4169                           struct dt_it *di, const struct dt_key *key)
4170 {
4171         struct osd_thread_info  *oti = osd_oti_get(env);
4172         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4173
4174         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4175                 /* swab quota uid/gid */
4176                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4177                 key = (struct dt_key *)&oti->oti_quota_id;
4178         }
4179
4180         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4181 }
4182
4183 /**
4184  *  Release Iterator
4185  *
4186  *  \param  di      osd iterator
4187  */
4188 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4189 {
4190         struct osd_it_iam *it = (struct osd_it_iam *)di;
4191
4192         iam_it_put(&it->oi_it);
4193 }
4194
4195 /**
4196  *  Move iterator by one record
4197  *
4198  *  \param  di      osd iterator
4199  *
4200  *  \retval +1   end of container reached
4201  *  \retval  0   success
4202  *  \retval -ve  failure
4203  */
4204
4205 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4206 {
4207         struct osd_it_iam *it = (struct osd_it_iam *)di;
4208
4209         return iam_it_next(&it->oi_it);
4210 }
4211
4212 /**
4213  * Return pointer to the key under iterator.
4214  */
4215
4216 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4217                                  const struct dt_it *di)
4218 {
4219         struct osd_thread_info *oti = osd_oti_get(env);
4220         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4221         struct osd_object      *obj = it->oi_obj;
4222         struct dt_key          *key;
4223
4224         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4225
4226         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4227                 /* swab quota uid/gid */
4228                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4229                 key = (struct dt_key *)&oti->oti_quota_id;
4230         }
4231
4232         return key;
4233 }
4234
4235 /**
4236  * Return size of key under iterator (in bytes)
4237  */
4238
4239 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4240 {
4241         struct osd_it_iam *it = (struct osd_it_iam *)di;
4242
4243         return iam_it_key_size(&it->oi_it);
4244 }
4245
4246 static inline void
4247 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4248 {
4249         /* check if file type is required */
4250         if (ent->lde_attrs & LUDA_TYPE) {
4251                 struct luda_type *lt;
4252                 int align = sizeof(*lt) - 1;
4253
4254                 len = (len + align) & ~align;
4255                 lt = (struct luda_type *)(ent->lde_name + len);
4256                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4257         }
4258
4259         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4260 }
4261
4262 /**
4263  * build lu direct from backend fs dirent.
4264  */
4265
4266 static inline void
4267 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4268                    char *name, __u16 namelen, __u16 type, __u32 attr)
4269 {
4270         ent->lde_attrs = attr | LUDA_FID;
4271         fid_cpu_to_le(&ent->lde_fid, fid);
4272
4273         ent->lde_hash = cpu_to_le64(offset);
4274         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4275
4276         strncpy(ent->lde_name, name, namelen);
4277         ent->lde_name[namelen] = '\0';
4278         ent->lde_namelen = cpu_to_le16(namelen);
4279
4280         /* append lustre attributes */
4281         osd_it_append_attrs(ent, namelen, type);
4282 }
4283
4284 /**
4285  * Return pointer to the record under iterator.
4286  */
4287 static int osd_it_iam_rec(const struct lu_env *env,
4288                           const struct dt_it *di,
4289                           struct dt_rec *dtrec, __u32 attr)
4290 {
4291         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4292         struct osd_thread_info *info = osd_oti_get(env);
4293         ENTRY;
4294
4295         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4296                 const struct osd_fid_pack *rec;
4297                 struct lu_fid             *fid = &info->oti_fid;
4298                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4299                 char                      *name;
4300                 int                        namelen;
4301                 __u64                      hash;
4302                 int                        rc;
4303
4304                 name = (char *)iam_it_key_get(&it->oi_it);
4305                 if (IS_ERR(name))
4306                         RETURN(PTR_ERR(name));
4307
4308                 namelen = iam_it_key_size(&it->oi_it);
4309
4310                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4311                 if (IS_ERR(rec))
4312                         RETURN(PTR_ERR(rec));
4313
4314                 rc = osd_fid_unpack(fid, rec);
4315                 if (rc)
4316                         RETURN(rc);
4317
4318                 hash = iam_it_store(&it->oi_it);
4319
4320                 /* IAM does not store object type in IAM index (dir) */
4321                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4322                                    0, LUDA_FID);
4323         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4324                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4325                            (struct iam_rec *)dtrec);
4326                 osd_quota_unpack(it->oi_obj, dtrec);
4327         } else {
4328                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4329                            (struct iam_rec *)dtrec);
4330         }
4331
4332         RETURN(0);
4333 }
4334
4335 /**
4336  * Returns cookie for current Iterator position.
4337  */
4338 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4339 {
4340         struct osd_it_iam *it = (struct osd_it_iam *)di;
4341
4342         return iam_it_store(&it->oi_it);
4343 }
4344
4345 /**
4346  * Restore iterator from cookie.
4347  *
4348  * \param  di      osd iterator
4349  * \param  hash    Iterator location cookie
4350  *
4351  * \retval +ve  di points to record with least key not larger than key.
4352  * \retval  0   di points to exact matched key
4353  * \retval -ve  failure
4354  */
4355
4356 static int osd_it_iam_load(const struct lu_env *env,
4357                            const struct dt_it *di, __u64 hash)
4358 {
4359         struct osd_it_iam *it = (struct osd_it_iam *)di;
4360
4361         return iam_it_load(&it->oi_it, hash);
4362 }
4363
4364 static const struct dt_index_operations osd_index_iam_ops = {
4365         .dio_lookup         = osd_index_iam_lookup,
4366         .dio_declare_insert = osd_index_declare_iam_insert,
4367         .dio_insert         = osd_index_iam_insert,
4368         .dio_declare_delete = osd_index_declare_iam_delete,
4369         .dio_delete         = osd_index_iam_delete,
4370         .dio_it     = {
4371                 .init     = osd_it_iam_init,
4372                 .fini     = osd_it_iam_fini,
4373                 .get      = osd_it_iam_get,
4374                 .put      = osd_it_iam_put,
4375                 .next     = osd_it_iam_next,
4376                 .key      = osd_it_iam_key,
4377                 .key_size = osd_it_iam_key_size,
4378                 .rec      = osd_it_iam_rec,
4379                 .store    = osd_it_iam_store,
4380                 .load     = osd_it_iam_load
4381         }
4382 };
4383
4384
4385 /**
4386  * Creates or initializes iterator context.
4387  *
4388  * \retval struct osd_it_ea, iterator structure on success
4389  *
4390  */
4391 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4392                                     struct dt_object *dt,
4393                                     __u32 attr,
4394                                     struct lustre_capa *capa)
4395 {
4396         struct osd_object       *obj  = osd_dt_obj(dt);
4397         struct osd_thread_info  *info = osd_oti_get(env);
4398         struct osd_it_ea        *it   = &info->oti_it_ea;
4399         struct file             *file = &it->oie_file;
4400         struct lu_object        *lo   = &dt->do_lu;
4401         struct dentry           *obj_dentry = &info->oti_it_dentry;
4402         ENTRY;
4403         LASSERT(lu_object_exists(lo));
4404
4405         obj_dentry->d_inode = obj->oo_inode;
4406         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4407         obj_dentry->d_name.hash = 0;
4408
4409         it->oie_rd_dirent       = 0;
4410         it->oie_it_dirent       = 0;
4411         it->oie_dirent          = NULL;
4412         it->oie_buf             = info->oti_it_ea_buf;
4413         it->oie_obj             = obj;
4414
4415         /* Reset the "file" totally to avoid to reuse any old value from
4416          * former readdir handling, the "file->f_pos" should be zero. */
4417         memset(file, 0, sizeof(*file));
4418         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4419         if (attr & LUDA_64BITHASH)
4420                 file->f_mode    = FMODE_64BITHASH;
4421         else
4422                 file->f_mode    = FMODE_32BITHASH;
4423         file->f_dentry          = obj_dentry;
4424         file->f_mapping         = obj->oo_inode->i_mapping;
4425         file->f_op              = obj->oo_inode->i_fop;
4426         lu_object_get(lo);
4427         RETURN((struct dt_it *) it);
4428 }
4429
4430 /**
4431  * Destroy or finishes iterator context.
4432  *
4433  * \param di iterator structure to be destroyed
4434  */
4435 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4436 {
4437         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4438         struct osd_object    *obj  = it->oie_obj;
4439         struct inode       *inode  = obj->oo_inode;
4440
4441         ENTRY;
4442         it->oie_file.f_op->release(inode, &it->oie_file);
4443         lu_object_put(env, &obj->oo_dt.do_lu);
4444         EXIT;
4445 }
4446
4447 /**
4448  * It position the iterator at given key, so that next lookup continues from
4449  * that key Or it is similar to dio_it->load() but based on a key,
4450  * rather than file position.
4451  *
4452  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4453  * to the beginning.
4454  *
4455  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4456  */
4457 static int osd_it_ea_get(const struct lu_env *env,
4458                          struct dt_it *di, const struct dt_key *key)
4459 {
4460         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4461
4462         ENTRY;
4463         LASSERT(((const char *)key)[0] == '\0');
4464         it->oie_file.f_pos      = 0;
4465         it->oie_rd_dirent       = 0;
4466         it->oie_it_dirent       = 0;
4467         it->oie_dirent          = NULL;
4468
4469         RETURN(+1);
4470 }
4471
4472 /**
4473  * Does nothing
4474  */
4475 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4476 {
4477 }
4478
4479 /**
4480  * It is called internally by ->readdir(). It fills the
4481  * iterator's in-memory data structure with required
4482  * information i.e. name, namelen, rec_size etc.
4483  *
4484  * \param buf in which information to be filled in.
4485  * \param name name of the file in given dir
4486  *
4487  * \retval 0 on success
4488  * \retval 1 on buffer full
4489  */
4490 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4491                                loff_t offset, __u64 ino,
4492                                unsigned d_type)
4493 {
4494         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4495         struct osd_object       *obj  = it->oie_obj;
4496         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4497         struct lu_fid           *fid  = &ent->oied_fid;
4498         struct osd_fid_pack     *rec;
4499         ENTRY;
4500
4501         /* this should never happen */
4502         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4503                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4504                 RETURN(-EIO);
4505         }
4506
4507         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4508             OSD_IT_EA_BUFSIZE)
4509                 RETURN(1);
4510
4511         /* "." is just the object itself. */
4512         if (namelen == 1 && name[0] == '.') {
4513                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4514         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4515                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4516                 if (osd_fid_unpack(fid, rec) != 0)
4517                         fid_zero(fid);
4518         } else {
4519                 fid_zero(fid);
4520         }
4521         d_type &= ~LDISKFS_DIRENT_LUFID;
4522
4523         /* NOT export local root. */
4524         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4525                 ino = obj->oo_inode->i_ino;
4526                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4527         }
4528
4529         ent->oied_ino     = ino;
4530         ent->oied_off     = offset;
4531         ent->oied_namelen = namelen;
4532         ent->oied_type    = d_type;
4533
4534         memcpy(ent->oied_name, name, namelen);
4535
4536         it->oie_rd_dirent++;
4537         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4538         RETURN(0);
4539 }
4540
4541 /**
4542  * Calls ->readdir() to load a directory entry at a time
4543  * and stored it in iterator's in-memory data structure.
4544  *
4545  * \param di iterator's in memory structure
4546  *
4547  * \retval   0 on success
4548  * \retval -ve on error
4549  */
4550 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4551                                const struct dt_it *di)
4552 {
4553         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4554         struct osd_object  *obj   = it->oie_obj;
4555         struct inode       *inode = obj->oo_inode;
4556         struct htree_lock  *hlock = NULL;
4557         int                 result = 0;
4558
4559         ENTRY;
4560         it->oie_dirent = it->oie_buf;
4561         it->oie_rd_dirent = 0;
4562
4563         if (obj->oo_hl_head != NULL) {
4564                 hlock = osd_oti_get(env)->oti_hlock;
4565                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4566                                    inode, LDISKFS_HLOCK_READDIR);
4567         } else {
4568                 down_read(&obj->oo_ext_idx_sem);
4569         }
4570
4571         result = inode->i_fop->readdir(&it->oie_file, it,
4572                                        (filldir_t) osd_ldiskfs_filldir);
4573
4574         if (hlock != NULL)
4575                 ldiskfs_htree_unlock(hlock);
4576         else
4577                 up_read(&obj->oo_ext_idx_sem);
4578
4579         if (it->oie_rd_dirent == 0) {
4580                 result = -EIO;
4581         } else {
4582                 it->oie_dirent = it->oie_buf;
4583                 it->oie_it_dirent = 1;
4584         }
4585
4586         RETURN(result);
4587 }
4588
4589 /**
4590  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4591  * to load a directory entry at a time and stored it in
4592  * iterator's in-memory data structure.
4593  *
4594  * \param di iterator's in memory structure
4595  *
4596  * \retval +ve iterator reached to end
4597  * \retval   0 iterator not reached to end
4598  * \retval -ve on error
4599  */
4600 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4601 {
4602         struct osd_it_ea *it = (struct osd_it_ea *)di;
4603         int rc;
4604
4605         ENTRY;
4606
4607         if (it->oie_it_dirent < it->oie_rd_dirent) {
4608                 it->oie_dirent =
4609                         (void *) it->oie_dirent +
4610                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4611                                        it->oie_dirent->oied_namelen);
4612                 it->oie_it_dirent++;
4613                 RETURN(0);
4614         } else {
4615                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4616                         rc = +1;
4617                 else
4618                         rc = osd_ldiskfs_it_fill(env, di);
4619         }
4620
4621         RETURN(rc);
4622 }
4623
4624 /**
4625  * Returns the key at current position from iterator's in memory structure.
4626  *
4627  * \param di iterator's in memory structure
4628  *
4629  * \retval key i.e. struct dt_key on success
4630  */
4631 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4632                                     const struct dt_it *di)
4633 {
4634         struct osd_it_ea *it = (struct osd_it_ea *)di;
4635
4636         return (struct dt_key *)it->oie_dirent->oied_name;
4637 }
4638
4639 /**
4640  * Returns the key's size at current position from iterator's in memory structure.
4641  *
4642  * \param di iterator's in memory structure
4643  *
4644  * \retval key_size i.e. struct dt_key on success
4645  */
4646 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4647 {
4648         struct osd_it_ea *it = (struct osd_it_ea *)di;
4649
4650         return it->oie_dirent->oied_namelen;
4651 }
4652
4653 static int
4654 osd_dirent_update(handle_t *jh, struct super_block *sb,
4655                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4656                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4657 {
4658         struct osd_fid_pack *rec;
4659         int                  rc;
4660         ENTRY;
4661
4662         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4663         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4664
4665         rc = ldiskfs_journal_get_write_access(jh, bh);
4666         if (rc != 0) {
4667                 CERROR("%.16s: fail to write access for update dirent: "
4668                        "name = %.*s, rc = %d\n",
4669                        LDISKFS_SB(sb)->s_es->s_volume_name,
4670                        ent->oied_namelen, ent->oied_name, rc);
4671                 RETURN(rc);
4672         }
4673
4674         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4675         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4676         rc = ldiskfs_journal_dirty_metadata(jh, bh);
4677         if (rc != 0)
4678                 CERROR("%.16s: fail to dirty metadata for update dirent: "
4679                        "name = %.*s, rc = %d\n",
4680                        LDISKFS_SB(sb)->s_es->s_volume_name,
4681                        ent->oied_namelen, ent->oied_name, rc);
4682
4683         RETURN(rc);
4684 }
4685
4686 static inline int
4687 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4688 {
4689         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4690             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4691                 return 1;
4692         else
4693                 return 0;
4694 }
4695
4696 static inline int
4697 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
4698 {
4699         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
4700                  "dot_dotdot = %d\n", dot_dotdot);
4701
4702         if (LDISKFS_DIR_REC_LEN(de) >=
4703             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
4704                 return 1;
4705         else
4706                 return 0;
4707 }
4708
4709 static int
4710 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
4711                     struct inode *dir, struct inode *inode,
4712                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4713                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
4714                     struct htree_lock *hlock)
4715 {
4716         struct dentry               *dentry;
4717         struct osd_fid_pack         *rec;
4718         struct ldiskfs_dentry_param *ldp;
4719         int                          rc;
4720         ENTRY;
4721
4722         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
4723                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
4724                 RETURN(0);
4725
4726         /* There is enough space to hold the FID-in-dirent. */
4727         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
4728                                  dir->i_sb->s_blocksize)) {
4729                 rc = ldiskfs_journal_get_write_access(jh, bh);
4730                 if (rc != 0) {
4731                         CERROR("%.16s: fail to write access for reinsert "
4732                                "dirent: name = %.*s, rc = %d\n",
4733                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4734                                ent->oied_namelen, ent->oied_name, rc);
4735                         RETURN(rc);
4736                 }
4737
4738                 de->name[de->name_len] = 0;
4739                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4740                 rec->fp_len = sizeof(struct lu_fid) + 1;
4741                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4742                 de->file_type |= LDISKFS_DIRENT_LUFID;
4743
4744                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4745                 if (rc != 0)
4746                         CERROR("%.16s: fail to dirty metadata for reinsert "
4747                                "dirent: name = %.*s, rc = %d\n",
4748                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4749                                ent->oied_namelen, ent->oied_name, rc);
4750
4751                 RETURN(rc);
4752         }
4753
4754         rc = ldiskfs_delete_entry(jh, dir, de, bh);
4755         if (rc != 0) {
4756                 CERROR("%.16s: fail to delete entry for reinsert dirent: "
4757                        "name = %.*s, rc = %d\n",
4758                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4759                        ent->oied_namelen, ent->oied_name, rc);
4760                 RETURN(rc);
4761         }
4762
4763         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
4764                                            ent->oied_namelen);
4765         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
4766         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
4767         dentry->d_fsdata = (void *)ldp;
4768         ll_vfs_dq_init(dir);
4769         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
4770         /* It is too bad, we cannot reinsert the name entry back.
4771          * That means we lose it! */
4772         if (rc != 0)
4773                 CERROR("%.16s: fail to insert entry for reinsert dirent: "
4774                        "name = %.*s, rc = %d\n",
4775                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4776                        ent->oied_namelen, ent->oied_name, rc);
4777
4778         RETURN(rc);
4779 }
4780
4781 static int
4782 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
4783                         struct osd_it_ea *it, struct lu_fid *fid,
4784                         struct osd_inode_id *id, __u32 *attr)
4785 {
4786         struct osd_thread_info     *info        = osd_oti_get(env);
4787         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
4788         struct osd_device          *dev         = osd_obj2dev(obj);
4789         struct super_block         *sb          = osd_sb(dev);
4790         const char                 *devname     =
4791                                         LDISKFS_SB(sb)->s_es->s_volume_name;
4792         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
4793         struct inode               *dir         = obj->oo_inode;
4794         struct htree_lock          *hlock       = NULL;
4795         struct buffer_head         *bh          = NULL;
4796         handle_t                   *jh          = NULL;
4797         struct ldiskfs_dir_entry_2 *de;
4798         struct dentry              *dentry;
4799         struct inode               *inode;
4800         int                         credits;
4801         int                         rc;
4802         int                         dot_dotdot  = 0;
4803         bool                        dirty       = false;
4804         ENTRY;
4805
4806         if (ent->oied_name[0] == '.') {
4807                 if (ent->oied_namelen == 1)
4808                         dot_dotdot = 1;
4809                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
4810                         dot_dotdot = 2;
4811         }
4812
4813         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
4814                                       ent->oied_namelen);
4815
4816         /* We need to ensure that the name entry is still valid.
4817          * Because it may be removed or renamed by other already.
4818          *
4819          * The unlink or rename operation will start journal before PDO lock,
4820          * so to avoid deadlock, here we need to start journal handle before
4821          * related PDO lock also. But because we do not know whether there
4822          * will be something to be repaired before PDO lock, we just start
4823          * journal without conditions.
4824          *
4825          * We may need to remove the name entry firstly, then insert back.
4826          * One credit is for user quota file update.
4827          * One credit is for group quota file update.
4828          * Two credits are for dirty inode. */
4829         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4830                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
4831
4832 again:
4833         if (dev->od_dirent_journal) {
4834                 jh = ldiskfs_journal_start_sb(sb, credits);
4835                 if (IS_ERR(jh)) {
4836                         rc = PTR_ERR(jh);
4837                         CERROR("%.16s: fail to start trans for dirent "
4838                                "check_repair: credits %d, name %.*s, rc %d\n",
4839                                devname, credits, ent->oied_namelen,
4840                                ent->oied_name, rc);
4841                         RETURN(rc);
4842                 }
4843
4844                 if (obj->oo_hl_head != NULL) {
4845                         hlock = osd_oti_get(env)->oti_hlock;
4846                         /* "0" means exclusive lock for the whole directory.
4847                          * We need to prevent others access such name entry
4848                          * during the delete + insert. Neither HLOCK_ADD nor
4849                          * HLOCK_DEL cannot guarantee the atomicity. */
4850                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
4851                 } else {
4852                         down_write(&obj->oo_ext_idx_sem);
4853                 }
4854         } else {
4855                 if (obj->oo_hl_head != NULL) {
4856                         hlock = osd_oti_get(env)->oti_hlock;
4857                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
4858                                            LDISKFS_HLOCK_LOOKUP);
4859                 } else {
4860                         down_read(&obj->oo_ext_idx_sem);
4861                 }
4862         }
4863
4864         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
4865         /* For dot/dotdot entry, if there is not enough space to hold the
4866          * FID-in-dirent, just keep them there. It only happens when the
4867          * device upgraded from 1.8 or restored from MDT file-level backup.
4868          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
4869          * and needs to get FID from LMA when readdir, it will not affect the
4870          * performance much. */
4871         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
4872             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
4873                 *attr |= LUDA_IGNORE;
4874                 GOTO(out_journal, rc = 0);
4875         }
4876
4877         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
4878         inode = osd_iget(info, dev, id);
4879         if (IS_ERR(inode)) {
4880                 rc = PTR_ERR(inode);
4881                 if (rc == -ENOENT || rc == -ESTALE) {
4882                         *attr |= LUDA_IGNORE;
4883                         rc = 0;
4884                 }
4885
4886                 GOTO(out_journal, rc);
4887         }
4888
4889         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
4890         if (rc == 0) {
4891                 if (fid_is_sane(fid)) {
4892                         /* FID-in-dirent is valid. */
4893                         if (lu_fid_eq(fid, &lma->lma_self_fid))
4894                                 GOTO(out_inode, rc = 0);
4895
4896                         /* Do not repair under dryrun mode. */
4897                         if (*attr & LUDA_VERIFY_DRYRUN) {
4898                                 *attr |= LUDA_REPAIR;
4899                                 GOTO(out_inode, rc = 0);
4900                         }
4901
4902                         if (!dev->od_dirent_journal) {
4903                                 iput(inode);
4904                                 brelse(bh);
4905                                 if (hlock != NULL)
4906                                         ldiskfs_htree_unlock(hlock);
4907                                 else
4908                                         up_read(&obj->oo_ext_idx_sem);
4909                                 dev->od_dirent_journal = 1;
4910                                 goto again;
4911                         }
4912
4913                         *fid = lma->lma_self_fid;
4914                         dirty = true;
4915                         /* Update the FID-in-dirent. */
4916                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
4917                         if (rc == 0)
4918                                 *attr |= LUDA_REPAIR;
4919                 } else {
4920                         /* Do not repair under dryrun mode. */
4921                         if (*attr & LUDA_VERIFY_DRYRUN) {
4922                                 *fid = lma->lma_self_fid;
4923                                 *attr |= LUDA_REPAIR;
4924                                 GOTO(out_inode, rc = 0);
4925                         }
4926
4927                         if (!dev->od_dirent_journal) {
4928                                 iput(inode);
4929                                 brelse(bh);
4930                                 if (hlock != NULL)
4931                                         ldiskfs_htree_unlock(hlock);
4932                                 else
4933                                         up_read(&obj->oo_ext_idx_sem);
4934                                 dev->od_dirent_journal = 1;
4935                                 goto again;
4936                         }
4937
4938                         *fid = lma->lma_self_fid;
4939                         dirty = true;
4940                         /* Append the FID-in-dirent. */
4941                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4942                                                  fid, bh, de, hlock);
4943                         if (rc == 0)
4944                                 *attr |= LUDA_REPAIR;
4945                 }
4946         } else if (rc == -ENODATA) {
4947                 /* Do not repair under dryrun mode. */
4948                 if (*attr & LUDA_VERIFY_DRYRUN) {
4949                         if (fid_is_sane(fid)) {
4950                                 *attr |= LUDA_REPAIR;
4951                         } else {
4952                                 lu_igif_build(fid, inode->i_ino,
4953                                               inode->i_generation);
4954                                 *attr |= LUDA_UPGRADE;
4955                         }
4956                         GOTO(out_inode, rc = 0);
4957                 }
4958
4959                 if (!dev->od_dirent_journal) {
4960                         iput(inode);
4961                         brelse(bh);
4962                         if (hlock != NULL)
4963                                 ldiskfs_htree_unlock(hlock);
4964                         else
4965                                 up_read(&obj->oo_ext_idx_sem);
4966                         dev->od_dirent_journal = 1;
4967                         goto again;
4968                 }
4969
4970                 dirty = true;
4971                 if (unlikely(fid_is_sane(fid))) {
4972                         /* FID-in-dirent exists, but FID-in-LMA is lost.
4973                          * Trust the FID-in-dirent, and add FID-in-LMA. */
4974                         rc = osd_ea_fid_set(info, inode, fid, 0);
4975                         if (rc == 0)
4976                                 *attr |= LUDA_REPAIR;
4977                 } else {
4978                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
4979                         /* It is probably IGIF object. Only aappend the
4980                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
4981                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4982                                                  fid, bh, de, hlock);
4983                         if (rc == 0)
4984                                 *attr |= LUDA_UPGRADE;
4985                 }
4986         }
4987
4988         GOTO(out_inode, rc);
4989
4990 out_inode:
4991         iput(inode);
4992
4993 out_journal:
4994         brelse(bh);
4995         if (hlock != NULL) {
4996                 ldiskfs_htree_unlock(hlock);
4997         } else {
4998                 if (dev->od_dirent_journal)
4999                         up_write(&obj->oo_ext_idx_sem);
5000                 else
5001                         up_read(&obj->oo_ext_idx_sem);
5002         }
5003         if (jh != NULL)
5004                 ldiskfs_journal_stop(jh);
5005         if (rc >= 0 && !dirty)
5006                 dev->od_dirent_journal = 0;
5007         return rc;
5008 }
5009
5010 /**
5011  * Returns the value at current position from iterator's in memory structure.
5012  *
5013  * \param di struct osd_it_ea, iterator's in memory structure
5014  * \param attr attr requested for dirent.
5015  * \param lde lustre dirent
5016  *
5017  * \retval   0 no error and \param lde has correct lustre dirent.
5018  * \retval -ve on error
5019  */
5020 static inline int osd_it_ea_rec(const struct lu_env *env,
5021                                 const struct dt_it *di,
5022                                 struct dt_rec *dtrec, __u32 attr)
5023 {
5024         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5025         struct osd_object      *obj   = it->oie_obj;
5026         struct osd_device      *dev   = osd_obj2dev(obj);
5027         struct osd_scrub       *scrub = &dev->od_scrub;
5028         struct scrub_file      *sf    = &scrub->os_file;
5029         struct osd_thread_info *oti   = osd_oti_get(env);
5030         struct osd_inode_id    *id    = &oti->oti_id;
5031         struct osd_idmap_cache *oic   = &oti->oti_cache;
5032         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5033         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5034         __u32                   ino   = it->oie_dirent->oied_ino;
5035         int                     rc    = 0;
5036         ENTRY;
5037
5038         if (attr & LUDA_VERIFY) {
5039                 attr |= LUDA_TYPE;
5040                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5041                         attr |= LUDA_IGNORE;
5042                         rc = 0;
5043                         goto pack;
5044                 }
5045
5046                 rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr);
5047         } else {
5048                 attr &= ~LU_DIRENT_ATTRS_MASK;
5049                 if (!fid_is_sane(fid)) {
5050                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5051                                 RETURN(-ENOENT);
5052
5053                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
5054                 } else {
5055                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5056                 }
5057         }
5058
5059         if (rc < 0)
5060                 RETURN(rc);
5061
5062 pack:
5063         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5064                            it->oie_dirent->oied_name,
5065                            it->oie_dirent->oied_namelen,
5066                            it->oie_dirent->oied_type, attr);
5067
5068         if (osd_remote_fid(env, dev, fid))
5069                 RETURN(0);
5070
5071         if (likely(!(attr & LUDA_IGNORE)))
5072                 rc = osd_add_oi_cache(oti, dev, id, fid);
5073
5074         if (!(attr & LUDA_VERIFY) &&
5075             (scrub->os_pos_current <= ino) &&
5076             ((sf->sf_flags & SF_INCONSISTENT) ||
5077              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5078              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5079                 osd_consistency_check(oti, dev, oic);
5080
5081         RETURN(rc);
5082 }
5083
5084 /**
5085  * Returns a cookie for current position of the iterator head, so that
5086  * user can use this cookie to load/start the iterator next time.
5087  *
5088  * \param di iterator's in memory structure
5089  *
5090  * \retval cookie for current position, on success
5091  */
5092 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5093 {
5094         struct osd_it_ea *it = (struct osd_it_ea *)di;
5095
5096         return it->oie_dirent->oied_off;
5097 }
5098
5099 /**
5100  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5101  * to load a directory entry at a time and stored it i inn,
5102  * in iterator's in-memory data structure.
5103  *
5104  * \param di struct osd_it_ea, iterator's in memory structure
5105  *
5106  * \retval +ve on success
5107  * \retval -ve on error
5108  */
5109 static int osd_it_ea_load(const struct lu_env *env,
5110                           const struct dt_it *di, __u64 hash)
5111 {
5112         struct osd_it_ea *it = (struct osd_it_ea *)di;
5113         int rc;
5114
5115         ENTRY;
5116         it->oie_file.f_pos = hash;
5117
5118         rc =  osd_ldiskfs_it_fill(env, di);
5119         if (rc == 0)
5120                 rc = +1;
5121
5122         RETURN(rc);
5123 }
5124
5125 /**
5126  * Index lookup function for interoperability mode (b11826).
5127  *
5128  * \param key,  key i.e. file name to be searched
5129  *
5130  * \retval +ve, on success
5131  * \retval -ve, on error
5132  */
5133 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5134                                struct dt_rec *rec, const struct dt_key *key,
5135                                struct lustre_capa *capa)
5136 {
5137         struct osd_object *obj = osd_dt_obj(dt);
5138         int rc = 0;
5139
5140         ENTRY;
5141
5142         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5143         LINVRNT(osd_invariant(obj));
5144
5145         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5146                 return -EACCES;
5147
5148         rc = osd_ea_lookup_rec(env, obj, rec, key);
5149         if (rc == 0)
5150                 rc = +1;
5151         RETURN(rc);
5152 }
5153
5154 /**
5155  * Index and Iterator operations for interoperability
5156  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5157  */
5158 static const struct dt_index_operations osd_index_ea_ops = {
5159         .dio_lookup         = osd_index_ea_lookup,
5160         .dio_declare_insert = osd_index_declare_ea_insert,
5161         .dio_insert         = osd_index_ea_insert,
5162         .dio_declare_delete = osd_index_declare_ea_delete,
5163         .dio_delete         = osd_index_ea_delete,
5164         .dio_it     = {
5165                 .init     = osd_it_ea_init,
5166                 .fini     = osd_it_ea_fini,
5167                 .get      = osd_it_ea_get,
5168                 .put      = osd_it_ea_put,
5169                 .next     = osd_it_ea_next,
5170                 .key      = osd_it_ea_key,
5171                 .key_size = osd_it_ea_key_size,
5172                 .rec      = osd_it_ea_rec,
5173                 .store    = osd_it_ea_store,
5174                 .load     = osd_it_ea_load
5175         }
5176 };
5177
5178 static void *osd_key_init(const struct lu_context *ctx,
5179                           struct lu_context_key *key)
5180 {
5181         struct osd_thread_info *info;
5182
5183         OBD_ALLOC_PTR(info);
5184         if (info == NULL)
5185                 return ERR_PTR(-ENOMEM);
5186
5187         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5188         if (info->oti_it_ea_buf == NULL)
5189                 goto out_free_info;
5190
5191         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5192
5193         info->oti_hlock = ldiskfs_htree_lock_alloc();
5194         if (info->oti_hlock == NULL)
5195                 goto out_free_ea;
5196
5197         return info;
5198
5199  out_free_ea:
5200         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5201  out_free_info:
5202         OBD_FREE_PTR(info);
5203         return ERR_PTR(-ENOMEM);
5204 }
5205
5206 static void osd_key_fini(const struct lu_context *ctx,
5207                          struct lu_context_key *key, void* data)
5208 {
5209         struct osd_thread_info *info = data;
5210
5211         if (info->oti_hlock != NULL)
5212                 ldiskfs_htree_lock_free(info->oti_hlock);
5213         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5214         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5215         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5216         OBD_FREE_PTR(info);
5217 }
5218
5219 static void osd_key_exit(const struct lu_context *ctx,
5220                          struct lu_context_key *key, void *data)
5221 {
5222         struct osd_thread_info *info = data;
5223
5224         LASSERT(info->oti_r_locks == 0);
5225         LASSERT(info->oti_w_locks == 0);
5226         LASSERT(info->oti_txns    == 0);
5227 }
5228
5229 /* type constructor/destructor: osd_type_init, osd_type_fini */
5230 LU_TYPE_INIT_FINI(osd, &osd_key);
5231
5232 struct lu_context_key osd_key = {
5233         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5234         .lct_init = osd_key_init,
5235         .lct_fini = osd_key_fini,
5236         .lct_exit = osd_key_exit
5237 };
5238
5239
5240 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5241                            const char *name, struct lu_device *next)
5242 {
5243         struct osd_device *osd = osd_dev(d);
5244
5245         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5246             >= sizeof(osd->od_svname))
5247                 return -E2BIG;
5248         return osd_procfs_init(osd, name);
5249 }
5250
5251 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5252 {
5253         ENTRY;
5254
5255         /* shutdown quota slave instance associated with the device */
5256         if (o->od_quota_slave != NULL) {
5257                 qsd_fini(env, o->od_quota_slave);
5258                 o->od_quota_slave = NULL;
5259         }
5260
5261         RETURN(0);
5262 }
5263
5264 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5265 {
5266         ENTRY;
5267
5268         if (o->od_fsops) {
5269                 fsfilt_put_ops(o->od_fsops);
5270                 o->od_fsops = NULL;
5271         }
5272
5273         if (o->od_mnt != NULL) {
5274                 shrink_dcache_sb(osd_sb(o));
5275                 osd_sync(env, &o->od_dt_dev);
5276
5277                 mntput(o->od_mnt);
5278                 o->od_mnt = NULL;
5279         }
5280
5281         EXIT;
5282 }
5283
5284 static int osd_mount(const struct lu_env *env,
5285                      struct osd_device *o, struct lustre_cfg *cfg)
5286 {
5287         const char              *name  = lustre_cfg_string(cfg, 0);
5288         const char              *dev  = lustre_cfg_string(cfg, 1);
5289         const char              *opts;
5290         unsigned long            page, s_flags, lmd_flags = 0;
5291         struct page             *__page;
5292         struct file_system_type *type;
5293         char                    *options = NULL;
5294         char                    *str;
5295         int                       rc = 0;
5296         ENTRY;
5297
5298         if (o->od_mnt != NULL)
5299                 RETURN(0);
5300
5301         if (strlen(dev) >= sizeof(o->od_mntdev))
5302                 RETURN(-E2BIG);
5303         strcpy(o->od_mntdev, dev);
5304
5305         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
5306         if (o->od_fsops == NULL) {
5307                 CERROR("Can't find fsfilt_ldiskfs\n");
5308                 RETURN(-ENOTSUPP);
5309         }
5310
5311         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
5312         if (__page == NULL)
5313                 GOTO(out, rc = -ENOMEM);
5314
5315         str = lustre_cfg_string(cfg, 2);
5316         s_flags = simple_strtoul(str, NULL, 0);
5317         str = strstr(str, ":");
5318         if (str)
5319                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5320         opts = lustre_cfg_string(cfg, 3);
5321         page = (unsigned long)cfs_page_address(__page);
5322         options = (char *)page;
5323         *options = '\0';
5324         if (opts == NULL)
5325                 strcat(options, "user_xattr,acl");
5326         else
5327                 strcat(options, opts);
5328
5329         /* Glom up mount options */
5330         if (*options != '\0')
5331                 strcat(options, ",");
5332         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
5333
5334         type = get_fs_type("ldiskfs");
5335         if (!type) {
5336                 CERROR("%s: cannot find ldiskfs module\n", name);
5337                 GOTO(out, rc = -ENODEV);
5338         }
5339
5340         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5341         cfs_module_put(type->owner);
5342
5343         if (IS_ERR(o->od_mnt)) {
5344                 rc = PTR_ERR(o->od_mnt);
5345                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5346                 o->od_mnt = NULL;
5347                 GOTO(out, rc);
5348         }
5349
5350 #ifdef HAVE_DEV_SET_RDONLY
5351         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5352                 CERROR("%s: underlying device %s is marked as read-only. "
5353                        "Setup failed\n", name, dev);
5354                 mntput(o->od_mnt);
5355                 o->od_mnt = NULL;
5356                 GOTO(out, rc = -EROFS);
5357         }
5358 #endif
5359
5360         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5361             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5362                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5363                 mntput(o->od_mnt);
5364                 o->od_mnt = NULL;
5365                 GOTO(out, rc = -EINVAL);
5366         }
5367
5368         ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
5369                                 LDISKFS_STATE_LUSTRE_NO_OI);
5370         if (lmd_flags & LMD_FLG_NOSCRUB)
5371                 o->od_noscrub = 1;
5372
5373 out:
5374         if (__page)
5375                 OBD_PAGE_FREE(__page);
5376         if (rc)
5377                 fsfilt_put_ops(o->od_fsops);
5378
5379         RETURN(rc);
5380 }
5381
5382 static struct lu_device *osd_device_fini(const struct lu_env *env,
5383                                          struct lu_device *d)
5384 {
5385         struct osd_device *o = osd_dev(d);
5386         ENTRY;
5387
5388         osd_procfs_fini(o);
5389         osd_shutdown(env, o);
5390         osd_scrub_cleanup(env, o);
5391         osd_obj_map_fini(o);
5392         osd_umount(env, o);
5393
5394         RETURN(NULL);
5395 }
5396
5397 static int osd_device_init0(const struct lu_env *env,
5398                             struct osd_device *o,
5399                             struct lustre_cfg *cfg)
5400 {
5401         struct lu_device        *l = osd2lu_dev(o);
5402         struct osd_thread_info *info;
5403         int                     rc;
5404         int                     cplen = 0;
5405
5406         /* if the module was re-loaded, env can loose its keys */
5407         rc = lu_env_refill((struct lu_env *) env);
5408         if (rc)
5409                 GOTO(out, rc);
5410         info = osd_oti_get(env);
5411         LASSERT(info);
5412
5413         l->ld_ops = &osd_lu_ops;
5414         o->od_dt_dev.dd_ops = &osd_dt_ops;
5415
5416         spin_lock_init(&o->od_osfs_lock);
5417         mutex_init(&o->od_otable_mutex);
5418         o->od_osfs_age = cfs_time_shift_64(-1000);
5419
5420         o->od_capa_hash = init_capa_hash();
5421         if (o->od_capa_hash == NULL)
5422                 GOTO(out, rc = -ENOMEM);
5423
5424         o->od_read_cache = 1;
5425         o->od_writethrough_cache = 1;
5426         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5427
5428         rc = osd_mount(env, o, cfg);
5429         if (rc)
5430                 GOTO(out_capa, rc);
5431
5432         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5433                         sizeof(o->od_svname));
5434         if (cplen >= sizeof(o->od_svname)) {
5435                 rc = -E2BIG;
5436                 GOTO(out_mnt, rc);
5437         }
5438
5439         rc = osd_obj_map_init(env, o);
5440         if (rc != 0)
5441                 GOTO(out_mnt, rc);
5442
5443         rc = lu_site_init(&o->od_site, l);
5444         if (rc != 0)
5445                 GOTO(out_compat, rc);
5446         o->od_site.ls_bottom_dev = l;
5447
5448         rc = lu_site_init_finish(&o->od_site);
5449         if (rc != 0)
5450                 GOTO(out_site, rc);
5451
5452         CFS_INIT_LIST_HEAD(&o->od_ios_list);
5453         /* setup scrub, including OI files initialization */
5454         rc = osd_scrub_setup(env, o);
5455         if (rc < 0)
5456                 GOTO(out_site, rc);
5457
5458         rc = osd_procfs_init(o, o->od_svname);
5459         if (rc != 0) {
5460                 CERROR("%s: can't initialize procfs: rc = %d\n",
5461                        o->od_svname, rc);
5462                 GOTO(out_scrub, rc);
5463         }
5464
5465         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5466
5467         /* initialize quota slave instance */
5468         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5469                                      o->od_proc_entry);
5470         if (IS_ERR(o->od_quota_slave)) {
5471                 rc = PTR_ERR(o->od_quota_slave);
5472                 o->od_quota_slave = NULL;
5473                 GOTO(out_procfs, rc);
5474         }
5475
5476         RETURN(0);
5477
5478 out_procfs:
5479         osd_procfs_fini(o);
5480 out_scrub:
5481         osd_scrub_cleanup(env, o);
5482 out_site:
5483         lu_site_fini(&o->od_site);
5484 out_compat:
5485         osd_obj_map_fini(o);
5486 out_mnt:
5487         osd_umount(env, o);
5488 out_capa:
5489         cleanup_capa_hash(o->od_capa_hash);
5490 out:
5491         return rc;
5492 }
5493
5494 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5495                                           struct lu_device_type *t,
5496                                           struct lustre_cfg *cfg)
5497 {
5498         struct osd_device *o;
5499         int                rc;
5500
5501         OBD_ALLOC_PTR(o);
5502         if (o == NULL)
5503                 return ERR_PTR(-ENOMEM);
5504
5505         rc = dt_device_init(&o->od_dt_dev, t);
5506         if (rc == 0) {
5507                 /* Because the ctx might be revived in dt_device_init,
5508                  * refill the env here */
5509                 lu_env_refill((struct lu_env *)env);
5510                 rc = osd_device_init0(env, o, cfg);
5511                 if (rc)
5512                         dt_device_fini(&o->od_dt_dev);
5513         }
5514
5515         if (unlikely(rc != 0))
5516                 OBD_FREE_PTR(o);
5517
5518         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5519 }
5520
5521 static struct lu_device *osd_device_free(const struct lu_env *env,
5522                                          struct lu_device *d)
5523 {
5524         struct osd_device *o = osd_dev(d);
5525         ENTRY;
5526
5527         cleanup_capa_hash(o->od_capa_hash);
5528         /* XXX: make osd top device in order to release reference */
5529         d->ld_site->ls_top_dev = d;
5530         lu_site_purge(env, d->ld_site, -1);
5531         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5532                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5533                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5534         }
5535         lu_site_fini(&o->od_site);
5536         dt_device_fini(&o->od_dt_dev);
5537         OBD_FREE_PTR(o);
5538         RETURN(NULL);
5539 }
5540
5541 static int osd_process_config(const struct lu_env *env,
5542                               struct lu_device *d, struct lustre_cfg *cfg)
5543 {
5544         struct osd_device *o = osd_dev(d);
5545         int err;
5546         ENTRY;
5547
5548         switch(cfg->lcfg_command) {
5549         case LCFG_SETUP:
5550                 err = osd_mount(env, o, cfg);
5551                 break;
5552         case LCFG_CLEANUP:
5553                 lu_dev_del_linkage(d->ld_site, d);
5554                 err = osd_shutdown(env, o);
5555                 break;
5556         default:
5557                 err = -ENOSYS;
5558         }
5559
5560         RETURN(err);
5561 }
5562
5563 static int osd_recovery_complete(const struct lu_env *env,
5564                                  struct lu_device *d)
5565 {
5566         struct osd_device       *osd = osd_dev(d);
5567         int                      rc = 0;
5568         ENTRY;
5569
5570         if (osd->od_quota_slave == NULL)
5571                 RETURN(0);
5572
5573         /* start qsd instance on recovery completion, this notifies the quota
5574          * slave code that we are about to process new requests now */
5575         rc = qsd_start(env, osd->od_quota_slave);
5576         RETURN(rc);
5577 }
5578
5579 /*
5580  * we use exports to track all osd users
5581  */
5582 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5583                            struct obd_device *obd, struct obd_uuid *cluuid,
5584                            struct obd_connect_data *data, void *localdata)
5585 {
5586         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
5587         struct lustre_handle  conn;
5588         int                   rc;
5589         ENTRY;
5590
5591         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5592
5593         rc = class_connect(&conn, obd, cluuid);
5594         if (rc)
5595                 RETURN(rc);
5596
5597         *exp = class_conn2export(&conn);
5598
5599         spin_lock(&osd->od_osfs_lock);
5600         osd->od_connects++;
5601         spin_unlock(&osd->od_osfs_lock);
5602
5603         RETURN(0);
5604 }
5605
5606 /*
5607  * once last export (we don't count self-export) disappeared
5608  * osd can be released
5609  */
5610 static int osd_obd_disconnect(struct obd_export *exp)
5611 {
5612         struct obd_device *obd = exp->exp_obd;
5613         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5614         int                rc, release = 0;
5615         ENTRY;
5616
5617         /* Only disconnect the underlying layers on the final disconnect. */
5618         spin_lock(&osd->od_osfs_lock);
5619         osd->od_connects--;
5620         if (osd->od_connects == 0)
5621                 release = 1;
5622         spin_unlock(&osd->od_osfs_lock);
5623
5624         rc = class_disconnect(exp); /* bz 9811 */
5625
5626         if (rc == 0 && release)
5627                 class_manual_cleanup(obd);
5628         RETURN(rc);
5629 }
5630
5631 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5632                        struct lu_device *dev)
5633 {
5634         struct osd_device *osd = osd_dev(dev);
5635         int                result = 0;
5636         ENTRY;
5637
5638         if (osd->od_quota_slave != NULL)
5639                 /* set up quota slave objects */
5640                 result = qsd_prepare(env, osd->od_quota_slave);
5641
5642         RETURN(result);
5643 }
5644
5645 static const struct lu_object_operations osd_lu_obj_ops = {
5646         .loo_object_init      = osd_object_init,
5647         .loo_object_delete    = osd_object_delete,
5648         .loo_object_release   = osd_object_release,
5649         .loo_object_free      = osd_object_free,
5650         .loo_object_print     = osd_object_print,
5651         .loo_object_invariant = osd_object_invariant
5652 };
5653
5654 const struct lu_device_operations osd_lu_ops = {
5655         .ldo_object_alloc      = osd_object_alloc,
5656         .ldo_process_config    = osd_process_config,
5657         .ldo_recovery_complete = osd_recovery_complete,
5658         .ldo_prepare           = osd_prepare,
5659 };
5660
5661 static const struct lu_device_type_operations osd_device_type_ops = {
5662         .ldto_init = osd_type_init,
5663         .ldto_fini = osd_type_fini,
5664
5665         .ldto_start = osd_type_start,
5666         .ldto_stop  = osd_type_stop,
5667
5668         .ldto_device_alloc = osd_device_alloc,
5669         .ldto_device_free  = osd_device_free,
5670
5671         .ldto_device_init    = osd_device_init,
5672         .ldto_device_fini    = osd_device_fini
5673 };
5674
5675 struct lu_device_type osd_device_type = {
5676         .ldt_tags     = LU_DEVICE_DT,
5677         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
5678         .ldt_ops      = &osd_device_type_ops,
5679         .ldt_ctx_tags = LCT_LOCAL,
5680 };
5681
5682 /*
5683  * lprocfs legacy support.
5684  */
5685 static struct obd_ops osd_obd_device_ops = {
5686         .o_owner = THIS_MODULE,
5687         .o_connect      = osd_obd_connect,
5688         .o_disconnect   = osd_obd_disconnect
5689 };
5690
5691 static int __init osd_mod_init(void)
5692 {
5693         struct lprocfs_static_vars lvars;
5694
5695         osd_oi_mod_init();
5696         lprocfs_osd_init_vars(&lvars);
5697         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
5698                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5699 }
5700
5701 static void __exit osd_mod_exit(void)
5702 {
5703         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
5704 }
5705
5706 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5707 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
5708 MODULE_LICENSE("GPL");
5709
5710 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);