lustre/osd-ldiskfs/osd_oi.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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/osd/osd_oi.c
  32  *
  33  * Object Index.
  34  *
  35  * Author: Nikita Danilov <nikita@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_OSD
  39
  40 #include <linux/module.h>
  41
  42 /*
  43  * struct OBD_{ALLOC,FREE}*()
  44  * OBD_FAIL_CHECK
  45  */
  46 #include <obd.h>
  47 #include <obd_support.h>
  48
  49 /* fid_cpu_to_be() */
  50 #include <lustre_fid.h>
  51 #include <dt_object.h>
  52 #include <lustre_scrub.h>
  53
  54 #include "osd_oi.h"
  55 /* osd_lookup(), struct osd_thread_info */
  56 #include "osd_internal.h"
  57
  58 unsigned int osd_oi_count = OSD_OI_FID_NR;
  59 module_param(osd_oi_count, int, 0444);
  60 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
  61
  62 static struct dt_index_features oi_feat = {
  63         .dif_flags       = DT_IND_UPDATE,
  64         .dif_recsize_min = sizeof(struct osd_inode_id),
  65         .dif_recsize_max = sizeof(struct osd_inode_id),
  66         .dif_ptrsize     = 4
  67 };
  68
  69 #define OSD_OI_NAME_BASE        "oi.16"
  70
  71 static void osd_oi_table_put(struct osd_thread_info *info,
  72                              struct osd_oi **oi_table, unsigned int oi_count)
  73 {
  74         struct iam_container *bag;
  75         int i;
  76
  77         for (i = 0; i < oi_count; i++) {
  78                 if (oi_table[i] == NULL)
  79                         continue;
  80
  81                 LASSERT(oi_table[i]->oi_inode != NULL);
  82
  83                 bag = &(oi_table[i]->oi_dir.od_container);
  84                 if (bag->ic_object == oi_table[i]->oi_inode)
  85                         iam_container_fini(bag);
  86                 iput(oi_table[i]->oi_inode);
  87                 oi_table[i]->oi_inode = NULL;
  88                 OBD_FREE_PTR(oi_table[i]);
  89                 oi_table[i] = NULL;
  90         }
  91 }
  92
  93 static int osd_oi_index_create_one(struct osd_thread_info *info,
  94                                    struct osd_device *osd, const char *name,
  95                                    struct dt_index_features *feat)
  96 {
  97         const struct lu_env             *env = info->oti_env;
  98         struct osd_inode_id             *id  = &info->oti_id;
  99         struct buffer_head              *bh;
 100         struct inode                    *inode;
 101         struct ldiskfs_dir_entry_2      *de;
 102         struct dentry                   *dentry;
 103         struct super_block              *sb  = osd_sb(osd);
 104         struct inode                    *dir = sb->s_root->d_inode;
 105         handle_t                        *jh;
 106         int                              rc;
 107
 108         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 109         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
 110         if (!IS_ERR(bh)) {
 111                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 112                 brelse(bh);
 113                 inode = osd_iget(info, osd, id);
 114                 if (!IS_ERR(inode)) {
 115                         iput(inode);
 116                         inode = ERR_PTR(-EEXIST);
 117                 }
 118                 return PTR_ERR(inode);
 119         }
 120
 121         if (osd->od_dt_dev.dd_rdonly)
 122                 RETURN(-EROFS);
 123
 124         jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
 125         if (IS_ERR(jh))
 126                 return PTR_ERR(jh);
 127
 128         inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR),
 129                                      NULL);
 130         if (IS_ERR(inode)) {
 131                 ldiskfs_journal_stop(jh);
 132                 return PTR_ERR(inode);
 133         }
 134
 135         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 136         unlock_new_inode(inode);
 137
 138         if (feat->dif_flags & DT_IND_VARKEY)
 139                 rc = iam_lvar_create(inode, feat->dif_keysize_max,
 140                                      feat->dif_ptrsize, feat->dif_recsize_max,
 141                                      jh);
 142         else
 143                 rc = iam_lfix_create(inode, feat->dif_keysize_max,
 144                                      feat->dif_ptrsize, feat->dif_recsize_max,
 145                                      jh);
 146         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 147         rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
 148         ldiskfs_journal_stop(jh);
 149         iput(inode);
 150         return rc;
 151 }
 152
 153 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 154                                        struct osd_device *osd,
 155                                        const char *name,
 156                                        struct dt_index_features *f,
 157                                        bool create)
 158 {
 159         struct dentry *dentry;
 160         struct inode  *inode;
 161         int rc;
 162
 163         dentry = osd_lookup_one_len_unlocked(osd, name, osd_sb(osd)->s_root,
 164                                              strlen(name));
 165         if (IS_ERR(dentry))
 166                 return ERR_CAST(dentry);
 167
 168         if (dentry->d_inode) {
 169                 LASSERT(!is_bad_inode(dentry->d_inode));
 170                 inode = dentry->d_inode;
 171                 atomic_inc(&inode->i_count);
 172                 dput(dentry);
 173                 return inode;
 174         }
 175
 176         /* create */
 177         dput(dentry);
 178         shrink_dcache_parent(osd_sb(osd)->s_root);
 179         if (!create)
 180                 return ERR_PTR(-ENOENT);
 181
 182         rc = osd_oi_index_create_one(info, osd, name, f);
 183         if (rc)
 184                 return ERR_PTR(rc);
 185
 186         dentry = osd_lookup_one_len_unlocked(osd, name, osd_sb(osd)->s_root,
 187                                              strlen(name));
 188         if (IS_ERR(dentry))
 189                 return ERR_CAST(dentry);
 190
 191         if (dentry->d_inode) {
 192                 LASSERT(!is_bad_inode(dentry->d_inode));
 193                 inode = dentry->d_inode;
 194                 atomic_inc(&inode->i_count);
 195                 dput(dentry);
 196                 return inode;
 197         }
 198
 199         return ERR_PTR(-ENOENT);
 200 }
 201
 202 /**
 203  * Open an OI(Ojbect Index) container.
 204  *
 205  * \param       name    Name of OI container
 206  * \param       objp    Pointer of returned OI
 207  *
 208  * \retval      0       success
 209  * \retval      -ve     failure
 210  */
 211 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
 212                        char *name, struct osd_oi **oi_slot, bool create)
 213 {
 214         struct osd_directory *dir;
 215         struct iam_container *bag;
 216         struct inode *inode;
 217         struct osd_oi *oi;
 218         int rc;
 219
 220         ENTRY;
 221
 222         oi_feat.dif_keysize_min = sizeof(struct lu_fid);
 223         oi_feat.dif_keysize_max = sizeof(struct lu_fid);
 224
 225         inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
 226         if (IS_ERR(inode))
 227                 RETURN(PTR_ERR(inode));
 228         ldiskfs_set_inode_state(inode, LDISKFS_STATE_IAM);
 229
 230         if (!osd->od_dt_dev.dd_rdonly) {
 231                 /* 'What the @fid is' is not imporatant, because these objects
 232                  * have no OI mappings, and only are visible inside the OSD.
 233                  */
 234                 lu_igif_build(&info->oti_fid, inode->i_ino,
 235                               inode->i_generation);
 236                 rc = osd_ea_fid_set(info, inode, &info->oti_fid,
 237                                     LMAC_NOT_IN_OI, 0);
 238                 if (rc)
 239                         GOTO(out_inode, rc);
 240         }
 241
 242         OBD_ALLOC_PTR(oi);
 243         if (oi == NULL)
 244                 GOTO(out_inode, rc = -ENOMEM);
 245
 246         oi->oi_inode = inode;
 247         dir = &oi->oi_dir;
 248
 249         bag = &dir->od_container;
 250         rc = iam_container_init(bag, &dir->od_descr, inode);
 251         if (rc < 0)
 252                 GOTO(out_free, rc);
 253
 254         rc = iam_container_setup(bag);
 255         if (rc < 0)
 256                 GOTO(out_container, rc);
 257
 258         *oi_slot = oi;
 259         RETURN(0);
 260
 261 out_container:
 262         iam_container_fini(bag);
 263 out_free:
 264         OBD_FREE_PTR(oi);
 265 out_inode:
 266         iput(inode);
 267         return rc;
 268 }
 269
 270 /**
 271  * Open OI(Object Index) table.
 272  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 273  * will be, this function can either return 0 for new filesystem, or number
 274  * of OIs on existed filesystem.
 275  *
 276  * If \a oi_count is non-zero, which means caller does know number of OIs on
 277  * filesystem, this function should return the exactly same number on
 278  * success, or error code in failure.
 279  *
 280  * \param     oi_count  Number of expected OI containers
 281  * \param     create    Create OIs if doesn't exist
 282  *
 283  * \retval    +ve       number of opened OI containers
 284  * \retval      0       no OI containers found
 285  * \retval    -ve       failure
 286  */
 287 static int
 288 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 289                   struct osd_oi **oi_table, unsigned int oi_count, bool create)
 290 {
 291         struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
 292         int count = 0;
 293         int rc = 0;
 294         int i;
 295         ENTRY;
 296
 297         /* NB: oi_count != 0 means that we have already created/known all OIs
 298          * and have known exact number of OIs.
 299          */
 300         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 301
 302         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 303                 char name[sizeof(OSD_OI_NAME_BASE) + 3 * sizeof(i) + 1];
 304
 305                 if (oi_table[i] != NULL) {
 306                         count++;
 307                         continue;
 308                 }
 309
 310                 snprintf(name, sizeof(name), "%s.%d", OSD_OI_NAME_BASE, i);
 311                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 312                 if (rc == 0) {
 313                         count++;
 314                         continue;
 315                 }
 316
 317                 if (rc == -ENOENT && create == false) {
 318                         if (oi_count == 0)
 319                                 return count;
 320
 321                         rc = 0;
 322                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 323                         continue;
 324                 }
 325
 326                 CERROR("%s: can't open %s: rc = %d\n",
 327                        osd_dev2name(osd), name, rc);
 328                 if (oi_count > 0)
 329                         CERROR("%s: expect to open total %d OI files.\n",
 330                                osd_dev2name(osd), oi_count);
 331                 break;
 332         }
 333
 334         if (rc < 0) {
 335                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 336                 count = rc;
 337         }
 338
 339         RETURN(count);
 340 }
 341
 342 static int osd_remove_oi_one(struct osd_device *osd, struct dentry *parent,
 343                              const char *name, int namelen)
 344 {
 345         struct dentry *child;
 346         int rc;
 347
 348         child = osd_lookup_one_len_unlocked(osd, name, parent, namelen);
 349         if (IS_ERR(child)) {
 350                 rc = PTR_ERR(child);
 351         } else {
 352                 rc = vfs_unlink(&init_user_ns, parent->d_inode, child);
 353                 dput(child);
 354         }
 355
 356         return rc == -ENOENT ? 0 : rc;
 357 }
 358
 359 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 360 {
 361         char name[16];
 362         int namelen;
 363         int rc;
 364         int i;
 365
 366         if (osd->od_dt_dev.dd_rdonly)
 367                 RETURN(-EROFS);
 368
 369         for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
 370                 namelen = scnprintf(name, sizeof(name), "%s.%d",
 371                                     OSD_OI_NAME_BASE, i);
 372                 rc = osd_remove_oi_one(osd, osd_sb(osd)->s_root, name, namelen);
 373                 if (rc != 0) {
 374                         CERROR(
 375                                "%s: fail to remove the stale OI file %s: rc = %d\n",
 376                                osd_dev2name(osd), name, rc);
 377                         return rc;
 378                 }
 379         }
 380
 381         namelen = scnprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
 382         rc = osd_remove_oi_one(osd, osd_sb(osd)->s_root, name, namelen);
 383         if (rc != 0)
 384                 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
 385                        osd_dev2name(osd), name, rc);
 386
 387         return rc;
 388 }
 389
 390 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
 391                 bool restored)
 392 {
 393         struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
 394         struct scrub_file *sf = &scrub->os_file;
 395         struct osd_oi **oi;
 396         int count;
 397         int rc;
 398         ENTRY;
 399
 400         if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
 401                 LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
 402                               osd_dev2name(osd), sf->sf_oi_count);
 403                 sf->sf_oi_count = 0;
 404         }
 405
 406         if (restored) {
 407                 LCONSOLE_WARN("%s: reset Object Index mappings\n",
 408                               osd_dev2name(osd));
 409                 rc = osd_remove_ois(info, osd);
 410                 if (rc)
 411                         RETURN(rc);
 412         }
 413
 414         OBD_ALLOC_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 415         if (oi == NULL)
 416                 RETURN(-ENOMEM);
 417
 418         /* try to open existing multiple OIs first */
 419         count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 420         if (count < 0)
 421                 GOTO(out, rc = count);
 422
 423         if (count > 0) {
 424                 if (count == sf->sf_oi_count)
 425                         GOTO(out, rc = count);
 426
 427                 if (sf->sf_oi_count == 0) {
 428                         if (likely((count & (count - 1)) == 0))
 429                                 GOTO(out, rc = count);
 430
 431                         LCONSOLE_WARN(
 432                                       "%s: invalid oi count %d, remove them, then set it to %d\n",
 433                                       osd_dev2name(osd), count, osd_oi_count);
 434                         osd_oi_table_put(info, oi, count);
 435                         rc = osd_remove_ois(info, osd);
 436                         if (rc)
 437                                 GOTO(out, rc);
 438
 439                         sf->sf_oi_count = osd_oi_count;
 440                 }
 441
 442                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 443                 count = sf->sf_oi_count;
 444                 goto create;
 445         }
 446
 447         /* if previous failed then try found single OI from old filesystem */
 448         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 449         if (rc == 0) { /* found single OI from old filesystem */
 450                 count = 1;
 451                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 452                 if (sf->sf_success_count == 0)
 453                         /* XXX: There is one corner case that if the OI_scrub
 454                          *      file crashed or lost and we regard it upgrade,
 455                          *      then we allow IGIF lookup to bypass OI files.
 456                          *
 457                          *      The risk is that osd_fid_lookup() may found
 458                          *      a wrong inode with the given IGIF especially
 459                          *      when the MDT has performed file-level backup
 460                          *      and restored after former upgrading from 1.8
 461                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 462                          *      verify the inode to decrease the risk.
 463                          */
 464                         scrub_file_reset(scrub, osd->od_uuid, SF_UPGRADE);
 465                 GOTO(out, rc = 1);
 466         } else if (rc != -ENOENT) {
 467                 CERROR("%s: can't open %s: rc = %d\n",
 468                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 469                 GOTO(out, rc);
 470         }
 471
 472         if (sf->sf_oi_count > 0) {
 473                 int i;
 474
 475                 count = sf->sf_oi_count;
 476                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 477                 for (i = 0; i < count; i++)
 478                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 479                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 480         } else {
 481                 count = sf->sf_oi_count = osd_oi_count;
 482         }
 483
 484 create:
 485         rc = scrub_file_store(info->oti_env, scrub);
 486         if (rc < 0) {
 487                 osd_oi_table_put(info, oi, count);
 488                 GOTO(out, rc);
 489         }
 490
 491         /* No OIs exist, new filesystem, create OI objects */
 492         rc = osd_oi_table_open(info, osd, oi, count, true);
 493         LASSERT(ergo(rc >= 0, rc == count));
 494
 495         GOTO(out, rc);
 496
 497 out:
 498         if (rc < 0) {
 499                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 500         } else {
 501                 LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
 502
 503                 osd->od_oi_table = oi;
 504                 osd->od_oi_count = rc;
 505                 if (sf->sf_oi_count != rc) {
 506                         sf->sf_oi_count = rc;
 507                         rc = scrub_file_store(info->oti_env, scrub);
 508                         if (rc < 0) {
 509                                 osd_oi_table_put(info, oi, count);
 510                                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 511                         }
 512                 } else {
 513                         rc = 0;
 514                 }
 515         }
 516
 517         return rc;
 518 }
 519
 520 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 521 {
 522         if (unlikely(!osd->od_oi_table))
 523                 return;
 524
 525         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 526
 527         OBD_FREE_PTR_ARRAY(osd->od_oi_table, OSD_OI_FID_NR_MAX);
 528         osd->od_oi_table = NULL;
 529 }
 530
 531 static inline int fid_is_fs_root(const struct lu_fid *fid)
 532 {
 533         /* Map root inode to special local object FID */
 534         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 535                          fid_oid(fid) == OSD_FS_ROOT_OID));
 536 }
 537
 538 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 539                              struct osd_oi *oi, struct dt_rec *rec,
 540                              const struct dt_key *key)
 541 {
 542         struct iam_container  *bag;
 543         struct iam_iterator   *it = &oti->oti_idx_it;
 544         struct iam_path_descr *ipd;
 545         int rc;
 546         ENTRY;
 547
 548         LASSERT(oi);
 549         LASSERT(oi->oi_inode);
 550
 551         bag = &oi->oi_dir.od_container;
 552         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 553         if (IS_ERR(ipd))
 554                 RETURN(-ENOMEM);
 555
 556         /* got ipd now we can start iterator. */
 557         iam_it_init(it, bag, 0, ipd);
 558
 559         rc = iam_it_get(it, (struct iam_key *)key);
 560         if (rc > 0)
 561                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 562         iam_it_put(it);
 563         iam_it_fini(it);
 564         osd_ipd_put(oti->oti_env, bag, ipd);
 565
 566         LINVRNT(osd_invariant(obj));
 567
 568         RETURN(rc);
 569 }
 570
 571 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 572                   const struct lu_fid *fid, enum oi_check_flags flags)
 573 {
 574         struct lu_seq_range *range = &info->oti_seq_range;
 575         int rc;
 576         ENTRY;
 577
 578         if (flags & OI_KNOWN_ON_OST)
 579                 RETURN(1);
 580
 581         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 582                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 583                      fid_is_quota(fid))
 584                 RETURN(0);
 585
 586         if (fid_is_idif(fid) || fid_is_last_id(fid))
 587                 RETURN(1);
 588
 589         if (!(flags & OI_CHECK_FLD))
 590                 RETURN(0);
 591
 592         if (osd_seq_site(osd)->ss_server_fld == NULL)
 593                 RETURN(0);
 594
 595         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 596         if (rc != 0) {
 597                 /* During upgrade, OST FLDB might not be loaded because
 598                  * OST FLDB is not created until 2.6, so if some DNE
 599                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 600                  * not be able to find the sequence from local FLDB
 601                  * cache see fld_index_init().
 602                  */
 603                 if (rc == -ENOENT && osd->od_is_ost)
 604                         RETURN(1);
 605
 606                 if (rc != -ENOENT)
 607                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 608                                osd_name(osd), PFID(fid), rc);
 609                 RETURN(0);
 610         }
 611
 612         if (fld_range_is_ost(range))
 613                 RETURN(1);
 614
 615         RETURN(0);
 616 }
 617
 618 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 619                            const struct lu_fid *fid, struct osd_inode_id *id)
 620 {
 621         struct lu_fid *oi_fid = &info->oti_fid2;
 622         int rc;
 623
 624         fid_cpu_to_be(oi_fid, fid);
 625         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 626                                (const struct dt_key *)oi_fid);
 627         if (rc > 0) {
 628                 osd_id_unpack(id, id);
 629                 rc = 0;
 630         } else if (rc == 0) {
 631                 rc = -ENOENT;
 632         }
 633         return rc;
 634 }
 635
 636 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 637                   const struct lu_fid *fid, struct osd_inode_id *id,
 638                   enum oi_check_flags flags)
 639 {
 640         if (unlikely(fid_is_last_id(fid)))
 641                 return osd_obj_spec_lookup(info, osd, fid, id, flags);
 642
 643         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 644                 return osd_obj_map_lookup(info, osd, fid, id);
 645
 646         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 647                 int rc;
 648
 649                 if (fid_is_fs_root(fid)) {
 650                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 651                                    osd_sb(osd)->s_root->d_inode->i_generation);
 652                         return 0;
 653                 }
 654                 if (unlikely(fid_is_acct(fid)))
 655                         return osd_acct_obj_lookup(info, osd, fid, id);
 656
 657                 /* For other special FIDs, try OI first, then do spec lookup */
 658                 rc = __osd_oi_lookup(info, osd, fid, id);
 659                 if (rc == -ENOENT)
 660                         return osd_obj_spec_lookup(info, osd, fid, id, flags);
 661                 return rc;
 662         }
 663
 664         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 665                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 666                 return 0;
 667         }
 668
 669         return __osd_oi_lookup(info, osd, fid, id);
 670 }
 671
 672 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 673                              const struct dt_rec *rec, const struct dt_key *key,
 674                              handle_t *th, bool insert)
 675 {
 676         struct iam_container *bag;
 677         struct iam_path_descr *ipd;
 678         int rc;
 679         ENTRY;
 680
 681         LASSERT(oi);
 682         LASSERT(oi->oi_inode);
 683         dquot_initialize(oi->oi_inode);
 684
 685         bag = &oi->oi_dir.od_container;
 686         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 687         if (unlikely(ipd == NULL))
 688                 RETURN(-ENOMEM);
 689
 690         LASSERT(th != NULL);
 691         LASSERT(th->h_transaction != NULL);
 692         if (insert)
 693                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 694                                 (const struct iam_rec *)rec, ipd);
 695         else
 696                 rc = iam_update(th, bag, (const struct iam_key *)key,
 697                                 (const struct iam_rec *)rec, ipd);
 698         osd_ipd_put(oti->oti_env, bag, ipd);
 699         LINVRNT(osd_invariant(obj));
 700         RETURN(rc);
 701 }
 702
 703 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 704                   const struct lu_fid *fid, const struct osd_inode_id *id,
 705                   handle_t *th, enum oi_check_flags flags, bool *exist)
 706 {
 707         struct lu_fid       *oi_fid = &info->oti_fid2;
 708         struct osd_inode_id *oi_id  = &info->oti_id2;
 709         int                  rc     = 0;
 710
 711         if (unlikely(fid_is_last_id(fid)))
 712                 return osd_obj_spec_insert(info, osd, fid, id, th);
 713
 714         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 715                 return osd_obj_map_insert(info, osd, fid, id, th);
 716
 717         fid_cpu_to_be(oi_fid, fid);
 718         osd_id_pack(oi_id, id);
 719         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 720                                (const struct dt_rec *)oi_id,
 721                                (const struct dt_key *)oi_fid, th, true);
 722         if (rc != 0) {
 723                 struct inode *inode;
 724                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 725
 726                 if (rc != -EEXIST)
 727                         return rc;
 728
 729                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 730                 if (rc != 0)
 731                         return rc;
 732
 733                 if (unlikely(osd_id_eq(id, oi_id)))
 734                         return 1;
 735
 736                 /* Check whether the mapping for oi_id is valid or not. */
 737                 inode = osd_iget(info, osd, oi_id);
 738                 if (IS_ERR(inode)) {
 739                         rc = PTR_ERR(inode);
 740                         if (rc == -ENOENT || rc == -ESTALE)
 741                                 goto update;
 742                         return rc;
 743                 }
 744
 745                 /* The EA inode should NOT be in OI, old OI scrub may added
 746                  * such OI mapping by wrong, replace it.
 747                  */
 748                 if (unlikely(osd_is_ea_inode(inode))) {
 749                         iput(inode);
 750                         goto update;
 751                 }
 752
 753                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 754                                  &info->oti_ost_attrs);
 755                 iput(inode);
 756                 if (rc == -ENODATA)
 757                         goto update;
 758
 759                 if (rc != 0)
 760                         return rc;
 761
 762                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 763                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 764                         CERROR(
 765                                "%s: the FID "DFID" is used by two objects: %u/%u %u/%u\n",
 766                                osd_dev2name(osd), PFID(fid), oi_id->oii_ino,
 767                                oi_id->oii_gen, id->oii_ino, id->oii_gen);
 768                         return -EEXIST;
 769                 }
 770
 771 update:
 772                 osd_id_pack(oi_id, id);
 773                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 774                                         (const struct dt_rec *)oi_id,
 775                                         (const struct dt_key *)oi_fid, th,
 776                                         false);
 777                 if (rc != 0)
 778                         return rc;
 779
 780                 if (exist != NULL)
 781                         *exist = true;
 782         }
 783
 784         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 785                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 786         return rc;
 787 }
 788
 789 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 790                              const struct dt_key *key, handle_t *th)
 791 {
 792         struct iam_container    *bag;
 793         struct iam_path_descr   *ipd;
 794         int                      rc;
 795         ENTRY;
 796
 797         LASSERT(oi);
 798         LASSERT(oi->oi_inode);
 799         dquot_initialize(oi->oi_inode);
 800
 801         bag = &oi->oi_dir.od_container;
 802         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 803         if (unlikely(ipd == NULL))
 804                 RETURN(-ENOMEM);
 805
 806         LASSERT(th != NULL);
 807         LASSERT(th->h_transaction != NULL);
 808
 809         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 810         osd_ipd_put(oti->oti_env, bag, ipd);
 811         LINVRNT(osd_invariant(obj));
 812         RETURN(rc);
 813 }
 814
 815 int osd_oi_delete(struct osd_thread_info *info,
 816                   struct osd_device *osd, const struct lu_fid *fid,
 817                   handle_t *th, enum oi_check_flags flags)
 818 {
 819         struct lu_fid *oi_fid = &info->oti_fid2;
 820
 821         /* clear idmap cache */
 822         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 823                 fid_zero(&info->oti_cache.oic_fid);
 824
 825         if (fid_is_last_id(fid))
 826                 return 0;
 827
 828         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 829                 return osd_obj_map_delete(info, osd, fid, th);
 830
 831         fid_cpu_to_be(oi_fid, fid);
 832         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 833                                  (const struct dt_key *)oi_fid, th);
 834 }
 835
 836 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 837                   const struct lu_fid *fid, const struct osd_inode_id *id,
 838                   handle_t *th, enum oi_check_flags flags)
 839 {
 840         struct lu_fid       *oi_fid = &info->oti_fid2;
 841         struct osd_inode_id *oi_id  = &info->oti_id2;
 842         int                  rc     = 0;
 843
 844         if (unlikely(fid_is_last_id(fid)))
 845                 return osd_obj_spec_update(info, osd, fid, id, th);
 846
 847         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 848                 return osd_obj_map_update(info, osd, fid, id, th);
 849
 850         fid_cpu_to_be(oi_fid, fid);
 851         osd_id_pack(oi_id, id);
 852         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 853                                (const struct dt_rec *)oi_id,
 854                                (const struct dt_key *)oi_fid, th, false);
 855         if (rc != 0)
 856                 return rc;
 857
 858         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 859                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 860         return rc;
 861 }
 862
 863 int osd_oi_mod_init(void)
 864 {
 865         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 866                 osd_oi_count = OSD_OI_FID_NR;
 867
 868         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 869                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 870                               osd_oi_count, size_roundup_power2(osd_oi_count));
 871                 osd_oi_count = size_roundup_power2(osd_oi_count);
 872         }
 873
 874         return 0;
 875 }