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  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_oi.c
  33  *
  34  * Object Index.
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_OSD
  40
  41 #include <linux/module.h>
  42
  43 /*
  44  * struct OBD_{ALLOC,FREE}*()
  45  * OBD_FAIL_CHECK
  46  */
  47 #include <obd.h>
  48 #include <obd_support.h>
  49
  50 /* fid_cpu_to_be() */
  51 #include <lustre_fid.h>
  52 #include <dt_object.h>
  53 #include <lustre_scrub.h>
  54
  55 #include "osd_oi.h"
  56 /* osd_lookup(), struct osd_thread_info */
  57 #include "osd_internal.h"
  58
  59 unsigned int osd_oi_count = OSD_OI_FID_NR;
  60 module_param(osd_oi_count, int, 0444);
  61 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
  62
  63 static struct dt_index_features oi_feat = {
  64         .dif_flags       = DT_IND_UPDATE,
  65         .dif_recsize_min = sizeof(struct osd_inode_id),
  66         .dif_recsize_max = sizeof(struct osd_inode_id),
  67         .dif_ptrsize     = 4
  68 };
  69
  70 #define OSD_OI_NAME_BASE        "oi.16"
  71
  72 static void osd_oi_table_put(struct osd_thread_info *info,
  73                              struct osd_oi **oi_table, unsigned int oi_count)
  74 {
  75         struct iam_container *bag;
  76         int i;
  77
  78         for (i = 0; i < oi_count; i++) {
  79                 if (oi_table[i] == NULL)
  80                         continue;
  81
  82                 LASSERT(oi_table[i]->oi_inode != NULL);
  83
  84                 bag = &(oi_table[i]->oi_dir.od_container);
  85                 if (bag->ic_object == oi_table[i]->oi_inode)
  86                         iam_container_fini(bag);
  87                 iput(oi_table[i]->oi_inode);
  88                 oi_table[i]->oi_inode = NULL;
  89                 OBD_FREE_PTR(oi_table[i]);
  90                 oi_table[i] = NULL;
  91         }
  92 }
  93
  94 static int osd_oi_index_create_one(struct osd_thread_info *info,
  95                                    struct osd_device *osd, const char *name,
  96                                    struct dt_index_features *feat)
  97 {
  98         const struct lu_env             *env = info->oti_env;
  99         struct osd_inode_id             *id  = &info->oti_id;
 100         struct buffer_head              *bh;
 101         struct inode                    *inode;
 102         struct ldiskfs_dir_entry_2      *de;
 103         struct dentry                   *dentry;
 104         struct super_block              *sb  = osd_sb(osd);
 105         struct inode                    *dir = sb->s_root->d_inode;
 106         handle_t                        *jh;
 107         int                              rc;
 108
 109         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 110         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
 111         if (!IS_ERR(bh)) {
 112                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 113                 brelse(bh);
 114                 inode = osd_iget(info, osd, id);
 115                 if (!IS_ERR(inode)) {
 116                         iput(inode);
 117                         inode = ERR_PTR(-EEXIST);
 118                 }
 119                 return PTR_ERR(inode);
 120         }
 121
 122         if (osd->od_dt_dev.dd_rdonly)
 123                 RETURN(-EROFS);
 124
 125         jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
 126         if (IS_ERR(jh))
 127                 return PTR_ERR(jh);
 128
 129         inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR),
 130                                      NULL);
 131         if (IS_ERR(inode)) {
 132                 ldiskfs_journal_stop(jh);
 133                 return PTR_ERR(inode);
 134         }
 135
 136         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 137         unlock_new_inode(inode);
 138
 139         if (feat->dif_flags & DT_IND_VARKEY)
 140                 rc = iam_lvar_create(inode, feat->dif_keysize_max,
 141                                      feat->dif_ptrsize, feat->dif_recsize_max,
 142                                      jh);
 143         else
 144                 rc = iam_lfix_create(inode, feat->dif_keysize_max,
 145                                      feat->dif_ptrsize, feat->dif_recsize_max,
 146                                      jh);
 147         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 148         rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
 149         ldiskfs_journal_stop(jh);
 150         iput(inode);
 151         return rc;
 152 }
 153
 154 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 155                                        struct osd_device *osd,
 156                                        const char *name,
 157                                        struct dt_index_features *f,
 158                                        bool create)
 159 {
 160         struct dentry *dentry;
 161         struct inode  *inode;
 162         int rc;
 163
 164         dentry = osd_lookup_one_len_unlocked(osd, name, osd_sb(osd)->s_root,
 165                                              strlen(name));
 166         if (IS_ERR(dentry))
 167                 return ERR_CAST(dentry);
 168
 169         if (dentry->d_inode) {
 170                 LASSERT(!is_bad_inode(dentry->d_inode));
 171                 inode = dentry->d_inode;
 172                 atomic_inc(&inode->i_count);
 173                 dput(dentry);
 174                 return inode;
 175         }
 176
 177         /* create */
 178         dput(dentry);
 179         shrink_dcache_parent(osd_sb(osd)->s_root);
 180         if (!create)
 181                 return ERR_PTR(-ENOENT);
 182
 183         rc = osd_oi_index_create_one(info, osd, name, f);
 184         if (rc)
 185                 return ERR_PTR(rc);
 186
 187         dentry = osd_lookup_one_len_unlocked(osd, name, osd_sb(osd)->s_root,
 188                                              strlen(name));
 189         if (IS_ERR(dentry))
 190                 return ERR_CAST(dentry);
 191
 192         if (dentry->d_inode) {
 193                 LASSERT(!is_bad_inode(dentry->d_inode));
 194                 inode = dentry->d_inode;
 195                 atomic_inc(&inode->i_count);
 196                 dput(dentry);
 197                 return inode;
 198         }
 199
 200         return ERR_PTR(-ENOENT);
 201 }
 202
 203 /**
 204  * Open an OI(Ojbect Index) container.
 205  *
 206  * \param       name    Name of OI container
 207  * \param       objp    Pointer of returned OI
 208  *
 209  * \retval      0       success
 210  * \retval      -ve     failure
 211  */
 212 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
 213                        char *name, struct osd_oi **oi_slot, bool create)
 214 {
 215         struct osd_directory *dir;
 216         struct iam_container *bag;
 217         struct inode *inode;
 218         struct osd_oi *oi;
 219         int rc;
 220
 221         ENTRY;
 222
 223         oi_feat.dif_keysize_min = sizeof(struct lu_fid);
 224         oi_feat.dif_keysize_max = sizeof(struct lu_fid);
 225
 226         inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
 227         if (IS_ERR(inode))
 228                 RETURN(PTR_ERR(inode));
 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 = ll_vfs_unlink(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 = snprintf(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 = snprintf(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                 rc = osd_remove_ois(info, osd);
 408                 if (rc)
 409                         RETURN(rc);
 410         }
 411
 412         OBD_ALLOC_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 413         if (oi == NULL)
 414                 RETURN(-ENOMEM);
 415
 416         /* try to open existing multiple OIs first */
 417         count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 418         if (count < 0)
 419                 GOTO(out, rc = count);
 420
 421         if (count > 0) {
 422                 if (count == sf->sf_oi_count)
 423                         GOTO(out, rc = count);
 424
 425                 if (sf->sf_oi_count == 0) {
 426                         if (likely((count & (count - 1)) == 0))
 427                                 GOTO(out, rc = count);
 428
 429                         LCONSOLE_WARN(
 430                                       "%s: invalid oi count %d, remove them, then set it to %d\n",
 431                                       osd_dev2name(osd), count, osd_oi_count);
 432                         osd_oi_table_put(info, oi, count);
 433                         rc = osd_remove_ois(info, osd);
 434                         if (rc)
 435                                 GOTO(out, rc);
 436
 437                         sf->sf_oi_count = osd_oi_count;
 438                 }
 439
 440                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 441                 count = sf->sf_oi_count;
 442                 goto create;
 443         }
 444
 445         /* if previous failed then try found single OI from old filesystem */
 446         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 447         if (rc == 0) { /* found single OI from old filesystem */
 448                 count = 1;
 449                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 450                 if (sf->sf_success_count == 0)
 451                         /* XXX: There is one corner case that if the OI_scrub
 452                          *      file crashed or lost and we regard it upgrade,
 453                          *      then we allow IGIF lookup to bypass OI files.
 454                          *
 455                          *      The risk is that osd_fid_lookup() may found
 456                          *      a wrong inode with the given IGIF especially
 457                          *      when the MDT has performed file-level backup
 458                          *      and restored after former upgrading from 1.8
 459                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 460                          *      verify the inode to decrease the risk.
 461                          */
 462                         scrub_file_reset(scrub, osd->od_uuid, SF_UPGRADE);
 463                 GOTO(out, rc = 1);
 464         } else if (rc != -ENOENT) {
 465                 CERROR("%s: can't open %s: rc = %d\n",
 466                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 467                 GOTO(out, rc);
 468         }
 469
 470         if (sf->sf_oi_count > 0) {
 471                 int i;
 472
 473                 count = sf->sf_oi_count;
 474                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 475                 for (i = 0; i < count; i++)
 476                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 477                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 478         } else {
 479                 count = sf->sf_oi_count = osd_oi_count;
 480         }
 481
 482 create:
 483         rc = scrub_file_store(info->oti_env, scrub);
 484         if (rc < 0) {
 485                 osd_oi_table_put(info, oi, count);
 486                 GOTO(out, rc);
 487         }
 488
 489         /* No OIs exist, new filesystem, create OI objects */
 490         rc = osd_oi_table_open(info, osd, oi, count, true);
 491         LASSERT(ergo(rc >= 0, rc == count));
 492
 493         GOTO(out, rc);
 494
 495 out:
 496         if (rc < 0) {
 497                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 498         } else {
 499                 LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
 500
 501                 osd->od_oi_table = oi;
 502                 osd->od_oi_count = rc;
 503                 if (sf->sf_oi_count != rc) {
 504                         sf->sf_oi_count = rc;
 505                         rc = scrub_file_store(info->oti_env, scrub);
 506                         if (rc < 0) {
 507                                 osd_oi_table_put(info, oi, count);
 508                                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 509                         }
 510                 } else {
 511                         rc = 0;
 512                 }
 513         }
 514
 515         return rc;
 516 }
 517
 518 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 519 {
 520         if (unlikely(!osd->od_oi_table))
 521                 return;
 522
 523         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 524
 525         OBD_FREE_PTR_ARRAY(osd->od_oi_table, OSD_OI_FID_NR_MAX);
 526         osd->od_oi_table = NULL;
 527 }
 528
 529 static inline int fid_is_fs_root(const struct lu_fid *fid)
 530 {
 531         /* Map root inode to special local object FID */
 532         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 533                          fid_oid(fid) == OSD_FS_ROOT_OID));
 534 }
 535
 536 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 537                              struct osd_oi *oi, struct dt_rec *rec,
 538                              const struct dt_key *key)
 539 {
 540         struct iam_container  *bag;
 541         struct iam_iterator   *it = &oti->oti_idx_it;
 542         struct iam_path_descr *ipd;
 543         int rc;
 544         ENTRY;
 545
 546         LASSERT(oi);
 547         LASSERT(oi->oi_inode);
 548
 549         bag = &oi->oi_dir.od_container;
 550         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 551         if (IS_ERR(ipd))
 552                 RETURN(-ENOMEM);
 553
 554         /* got ipd now we can start iterator. */
 555         iam_it_init(it, bag, 0, ipd);
 556
 557         rc = iam_it_get(it, (struct iam_key *)key);
 558         if (rc > 0)
 559                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 560         iam_it_put(it);
 561         iam_it_fini(it);
 562         osd_ipd_put(oti->oti_env, bag, ipd);
 563
 564         LINVRNT(osd_invariant(obj));
 565
 566         RETURN(rc);
 567 }
 568
 569 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 570                   const struct lu_fid *fid, enum oi_check_flags flags)
 571 {
 572         struct lu_seq_range *range = &info->oti_seq_range;
 573         int rc;
 574         ENTRY;
 575
 576         if (flags & OI_KNOWN_ON_OST)
 577                 RETURN(1);
 578
 579         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 580                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 581                      fid_is_quota(fid))
 582                 RETURN(0);
 583
 584         if (fid_is_idif(fid) || fid_is_last_id(fid))
 585                 RETURN(1);
 586
 587         if (!(flags & OI_CHECK_FLD))
 588                 RETURN(0);
 589
 590         if (osd_seq_site(osd)->ss_server_fld == NULL)
 591                 RETURN(0);
 592
 593         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 594         if (rc != 0) {
 595                 /* During upgrade, OST FLDB might not be loaded because
 596                  * OST FLDB is not created until 2.6, so if some DNE
 597                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 598                  * not be able to find the sequence from local FLDB
 599                  * cache see fld_index_init().
 600                  */
 601                 if (rc == -ENOENT && osd->od_is_ost)
 602                         RETURN(1);
 603
 604                 if (rc != -ENOENT)
 605                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 606                                osd_name(osd), PFID(fid), rc);
 607                 RETURN(0);
 608         }
 609
 610         if (fld_range_is_ost(range))
 611                 RETURN(1);
 612
 613         RETURN(0);
 614 }
 615
 616 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 617                            const struct lu_fid *fid, struct osd_inode_id *id)
 618 {
 619         struct lu_fid *oi_fid = &info->oti_fid2;
 620         int rc;
 621
 622         fid_cpu_to_be(oi_fid, fid);
 623         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 624                                (const struct dt_key *)oi_fid);
 625         if (rc > 0) {
 626                 osd_id_unpack(id, id);
 627                 rc = 0;
 628         } else if (rc == 0) {
 629                 rc = -ENOENT;
 630         }
 631         return rc;
 632 }
 633
 634 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 635                   const struct lu_fid *fid, struct osd_inode_id *id,
 636                   enum oi_check_flags flags)
 637 {
 638         if (unlikely(fid_is_last_id(fid)))
 639                 return osd_obj_spec_lookup(info, osd, fid, id, flags);
 640
 641         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 642                 return osd_obj_map_lookup(info, osd, fid, id);
 643
 644         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 645                 int rc;
 646
 647                 if (fid_is_fs_root(fid)) {
 648                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 649                                    osd_sb(osd)->s_root->d_inode->i_generation);
 650                         return 0;
 651                 }
 652                 if (unlikely(fid_is_acct(fid)))
 653                         return osd_acct_obj_lookup(info, osd, fid, id);
 654
 655                 /* For other special FIDs, try OI first, then do spec lookup */
 656                 rc = __osd_oi_lookup(info, osd, fid, id);
 657                 if (rc == -ENOENT)
 658                         return osd_obj_spec_lookup(info, osd, fid, id, flags);
 659                 return rc;
 660         }
 661
 662         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 663                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 664                 return 0;
 665         }
 666
 667         return __osd_oi_lookup(info, osd, fid, id);
 668 }
 669
 670 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 671                              const struct dt_rec *rec, const struct dt_key *key,
 672                              handle_t *th, bool insert)
 673 {
 674         struct iam_container *bag;
 675         struct iam_path_descr *ipd;
 676         int rc;
 677         ENTRY;
 678
 679         LASSERT(oi);
 680         LASSERT(oi->oi_inode);
 681         dquot_initialize(oi->oi_inode);
 682
 683         bag = &oi->oi_dir.od_container;
 684         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 685         if (unlikely(ipd == NULL))
 686                 RETURN(-ENOMEM);
 687
 688         LASSERT(th != NULL);
 689         LASSERT(th->h_transaction != NULL);
 690         if (insert)
 691                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 692                                 (const struct iam_rec *)rec, ipd);
 693         else
 694                 rc = iam_update(th, bag, (const struct iam_key *)key,
 695                                 (const struct iam_rec *)rec, ipd);
 696         osd_ipd_put(oti->oti_env, bag, ipd);
 697         LINVRNT(osd_invariant(obj));
 698         RETURN(rc);
 699 }
 700
 701 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 702                   const struct lu_fid *fid, const struct osd_inode_id *id,
 703                   handle_t *th, enum oi_check_flags flags, bool *exist)
 704 {
 705         struct lu_fid       *oi_fid = &info->oti_fid2;
 706         struct osd_inode_id *oi_id  = &info->oti_id2;
 707         int                  rc     = 0;
 708
 709         if (unlikely(fid_is_last_id(fid)))
 710                 return osd_obj_spec_insert(info, osd, fid, id, th);
 711
 712         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 713                 return osd_obj_map_insert(info, osd, fid, id, th);
 714
 715         fid_cpu_to_be(oi_fid, fid);
 716         osd_id_pack(oi_id, id);
 717         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 718                                (const struct dt_rec *)oi_id,
 719                                (const struct dt_key *)oi_fid, th, true);
 720         if (rc != 0) {
 721                 struct inode *inode;
 722                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 723
 724                 if (rc != -EEXIST)
 725                         return rc;
 726
 727                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 728                 if (rc != 0)
 729                         return rc;
 730
 731                 if (unlikely(osd_id_eq(id, oi_id)))
 732                         return 1;
 733
 734                 /* Check whether the mapping for oi_id is valid or not. */
 735                 inode = osd_iget(info, osd, oi_id);
 736                 if (IS_ERR(inode)) {
 737                         rc = PTR_ERR(inode);
 738                         if (rc == -ENOENT || rc == -ESTALE)
 739                                 goto update;
 740                         return rc;
 741                 }
 742
 743                 /* The EA inode should NOT be in OI, old OI scrub may added
 744                  * such OI mapping by wrong, replace it.
 745                  */
 746                 if (unlikely(osd_is_ea_inode(inode))) {
 747                         iput(inode);
 748                         goto update;
 749                 }
 750
 751                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 752                                  &info->oti_ost_attrs);
 753                 iput(inode);
 754                 if (rc == -ENODATA)
 755                         goto update;
 756
 757                 if (rc != 0)
 758                         return rc;
 759
 760                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 761                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 762                         CERROR(
 763                                "%s: the FID "DFID" is used by two objects: %u/%u %u/%u\n",
 764                                osd_dev2name(osd), PFID(fid), oi_id->oii_ino,
 765                                oi_id->oii_gen, id->oii_ino, id->oii_gen);
 766                         return -EEXIST;
 767                 }
 768
 769 update:
 770                 osd_id_pack(oi_id, id);
 771                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 772                                         (const struct dt_rec *)oi_id,
 773                                         (const struct dt_key *)oi_fid, th,
 774                                         false);
 775                 if (rc != 0)
 776                         return rc;
 777
 778                 if (exist != NULL)
 779                         *exist = true;
 780         }
 781
 782         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 783                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 784         return rc;
 785 }
 786
 787 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 788                              const struct dt_key *key, handle_t *th)
 789 {
 790         struct iam_container    *bag;
 791         struct iam_path_descr   *ipd;
 792         int                      rc;
 793         ENTRY;
 794
 795         LASSERT(oi);
 796         LASSERT(oi->oi_inode);
 797         dquot_initialize(oi->oi_inode);
 798
 799         bag = &oi->oi_dir.od_container;
 800         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 801         if (unlikely(ipd == NULL))
 802                 RETURN(-ENOMEM);
 803
 804         LASSERT(th != NULL);
 805         LASSERT(th->h_transaction != NULL);
 806
 807         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 808         osd_ipd_put(oti->oti_env, bag, ipd);
 809         LINVRNT(osd_invariant(obj));
 810         RETURN(rc);
 811 }
 812
 813 int osd_oi_delete(struct osd_thread_info *info,
 814                   struct osd_device *osd, const struct lu_fid *fid,
 815                   handle_t *th, enum oi_check_flags flags)
 816 {
 817         struct lu_fid *oi_fid = &info->oti_fid2;
 818
 819         /* clear idmap cache */
 820         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 821                 fid_zero(&info->oti_cache.oic_fid);
 822
 823         if (fid_is_last_id(fid))
 824                 return 0;
 825
 826         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 827                 return osd_obj_map_delete(info, osd, fid, th);
 828
 829         fid_cpu_to_be(oi_fid, fid);
 830         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 831                                  (const struct dt_key *)oi_fid, th);
 832 }
 833
 834 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 835                   const struct lu_fid *fid, const struct osd_inode_id *id,
 836                   handle_t *th, enum oi_check_flags flags)
 837 {
 838         struct lu_fid       *oi_fid = &info->oti_fid2;
 839         struct osd_inode_id *oi_id  = &info->oti_id2;
 840         int                  rc     = 0;
 841
 842         if (unlikely(fid_is_last_id(fid)))
 843                 return osd_obj_spec_update(info, osd, fid, id, th);
 844
 845         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 846                 return osd_obj_map_update(info, osd, fid, id, th);
 847
 848         fid_cpu_to_be(oi_fid, fid);
 849         osd_id_pack(oi_id, id);
 850         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 851                                (const struct dt_rec *)oi_id,
 852                                (const struct dt_key *)oi_fid, th, false);
 853         if (rc != 0)
 854                 return rc;
 855
 856         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 857                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 858         return rc;
 859 }
 860
 861 int osd_oi_mod_init(void)
 862 {
 863         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 864                 osd_oi_count = OSD_OI_FID_NR;
 865
 866         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 867                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 868                               osd_oi_count, size_roundup_power2(osd_oi_count));
 869                 osd_oi_count = size_roundup_power2(osd_oi_count);
 870         }
 871
 872         return 0;
 873 }