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 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 = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 165         if (IS_ERR(dentry))
 166                 return (void *) 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 = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 187         if (IS_ERR(dentry))
 188                 return (void *) dentry;
 189
 190         if (dentry->d_inode) {
 191                 LASSERT(!is_bad_inode(dentry->d_inode));
 192                 inode = dentry->d_inode;
 193                 atomic_inc(&inode->i_count);
 194                 dput(dentry);
 195                 return inode;
 196         }
 197
 198         return ERR_PTR(-ENOENT);
 199 }
 200
 201 /**
 202  * Open an OI(Ojbect Index) container.
 203  *
 204  * \param       name    Name of OI container
 205  * \param       objp    Pointer of returned OI
 206  *
 207  * \retval      0       success
 208  * \retval      -ve     failure
 209  */
 210 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
 211                        char *name, struct osd_oi **oi_slot, bool create)
 212 {
 213         struct osd_directory *dir;
 214         struct iam_container *bag;
 215         struct inode         *inode;
 216         struct osd_oi        *oi;
 217         int                   rc;
 218
 219         ENTRY;
 220
 221         oi_feat.dif_keysize_min = sizeof(struct lu_fid);
 222         oi_feat.dif_keysize_max = sizeof(struct lu_fid);
 223
 224         inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
 225         if (IS_ERR(inode))
 226                 RETURN(PTR_ERR(inode));
 227
 228         if (!osd->od_dt_dev.dd_rdonly) {
 229                 /* 'What the @fid is' is not imporatant, because these objects
 230                  * have no OI mappings, and only are visible inside the OSD.*/
 231                 lu_igif_build(&info->oti_fid, inode->i_ino,
 232                               inode->i_generation);
 233                 rc = osd_ea_fid_set(info, inode, &info->oti_fid,
 234                                     LMAC_NOT_IN_OI, 0);
 235                 if (rc)
 236                         GOTO(out_inode, rc);
 237         }
 238
 239         OBD_ALLOC_PTR(oi);
 240         if (oi == NULL)
 241                 GOTO(out_inode, rc = -ENOMEM);
 242
 243         oi->oi_inode = inode;
 244         dir = &oi->oi_dir;
 245
 246         bag = &dir->od_container;
 247         rc = iam_container_init(bag, &dir->od_descr, inode);
 248         if (rc < 0)
 249                 GOTO(out_free, rc);
 250
 251         rc = iam_container_setup(bag);
 252         if (rc < 0)
 253                 GOTO(out_container, rc);
 254
 255         *oi_slot = oi;
 256         RETURN(0);
 257
 258 out_container:
 259         iam_container_fini(bag);
 260 out_free:
 261         OBD_FREE_PTR(oi);
 262 out_inode:
 263         iput(inode);
 264         return rc;
 265 }
 266
 267 /**
 268  * Open OI(Object Index) table.
 269  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 270  * will be, this function can either return 0 for new filesystem, or number
 271  * of OIs on existed filesystem.
 272  *
 273  * If \a oi_count is non-zero, which means caller does know number of OIs on
 274  * filesystem, this function should return the exactly same number on
 275  * success, or error code in failure.
 276  *
 277  * \param     oi_count  Number of expected OI containers
 278  * \param     create    Create OIs if doesn't exist
 279  *
 280  * \retval    +ve       number of opened OI containers
 281  * \retval      0       no OI containers found
 282  * \retval    -ve       failure
 283  */
 284 static int
 285 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 286                   struct osd_oi **oi_table, unsigned oi_count, bool create)
 287 {
 288         struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
 289         int count = 0;
 290         int rc = 0;
 291         int i;
 292         ENTRY;
 293
 294         /* NB: oi_count != 0 means that we have already created/known all OIs
 295          * and have known exact number of OIs. */
 296         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 297
 298         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 299                 char name[sizeof(OSD_OI_NAME_BASE) + 3 * sizeof(i) + 1];
 300
 301                 if (oi_table[i] != NULL) {
 302                         count++;
 303                         continue;
 304                 }
 305
 306                 snprintf(name, sizeof(name), "%s.%d", OSD_OI_NAME_BASE, i);
 307                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 308                 if (rc == 0) {
 309                         count++;
 310                         continue;
 311                 }
 312
 313                 if (rc == -ENOENT && create == false) {
 314                         if (oi_count == 0)
 315                                 return count;
 316
 317                         rc = 0;
 318                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 319                         continue;
 320                 }
 321
 322                 CERROR("%s: can't open %s: rc = %d\n",
 323                        osd_dev2name(osd), name, rc);
 324                 if (oi_count > 0)
 325                         CERROR("%s: expect to open total %d OI files.\n",
 326                                osd_dev2name(osd), oi_count);
 327                 break;
 328         }
 329
 330         if (rc < 0) {
 331                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 332                 count = rc;
 333         }
 334
 335         RETURN(count);
 336 }
 337
 338 static int osd_remove_oi_one(struct dentry *parent, const char *name,
 339                              int namelen)
 340 {
 341         struct dentry *child;
 342         int rc;
 343
 344         child = ll_lookup_one_len(name, parent, namelen);
 345         if (IS_ERR(child)) {
 346                 rc = PTR_ERR(child);
 347         } else {
 348                 rc = ll_vfs_unlink(parent->d_inode, child);
 349                 dput(child);
 350         }
 351
 352         return rc == -ENOENT ? 0 : rc;
 353 }
 354
 355 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 356 {
 357         char name[16];
 358         int namelen;
 359         int rc;
 360         int i;
 361
 362         if (osd->od_dt_dev.dd_rdonly)
 363                 RETURN(-EROFS);
 364
 365         for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
 366                 namelen = snprintf(name, sizeof(name), "%s.%d",
 367                                    OSD_OI_NAME_BASE, i);
 368                 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
 369                 if (rc != 0) {
 370                         CERROR("%s: fail to remove the stale OI file %s: "
 371                                "rc = %d\n", osd_dev2name(osd), name, rc);
 372                         return rc;
 373                 }
 374         }
 375
 376         namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
 377         rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
 378         if (rc != 0)
 379                 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
 380                        osd_dev2name(osd), name, rc);
 381
 382         return rc;
 383 }
 384
 385 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
 386                 bool restored)
 387 {
 388         struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
 389         struct scrub_file *sf = &scrub->os_file;
 390         struct osd_oi **oi;
 391         int count;
 392         int rc;
 393         ENTRY;
 394
 395         if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
 396                 LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
 397                               osd_dev2name(osd), sf->sf_oi_count);
 398                 sf->sf_oi_count = 0;
 399         }
 400
 401         if (restored) {
 402                 rc = osd_remove_ois(info, osd);
 403                 if (rc)
 404                         RETURN(rc);
 405         }
 406
 407         OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 408         if (oi == NULL)
 409                 RETURN(-ENOMEM);
 410
 411         /* try to open existing multiple OIs first */
 412         count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 413         if (count < 0)
 414                 GOTO(out, rc = count);
 415
 416         if (count > 0) {
 417                 if (count == sf->sf_oi_count)
 418                         GOTO(out, rc = count);
 419
 420                 if (sf->sf_oi_count == 0) {
 421                         if (likely((count & (count - 1)) == 0))
 422                                 GOTO(out, rc = count);
 423
 424                         LCONSOLE_WARN("%s: invalid oi count %d, remove them, "
 425                                       "then set it to %d\n", osd_dev2name(osd),
 426                                       count, osd_oi_count);
 427                         osd_oi_table_put(info, oi, count);
 428                         rc = osd_remove_ois(info, osd);
 429                         if (rc)
 430                                 GOTO(out, rc);
 431
 432                         sf->sf_oi_count = osd_oi_count;
 433                 }
 434
 435                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 436                 count = sf->sf_oi_count;
 437                 goto create;
 438         }
 439
 440         /* if previous failed then try found single OI from old filesystem */
 441         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 442         if (rc == 0) { /* found single OI from old filesystem */
 443                 count = 1;
 444                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 445                 if (sf->sf_success_count == 0)
 446                         /* XXX: There is one corner case that if the OI_scrub
 447                          *      file crashed or lost and we regard it upgrade,
 448                          *      then we allow IGIF lookup to bypass OI files.
 449                          *
 450                          *      The risk is that osd_fid_lookup() may found
 451                          *      a wrong inode with the given IGIF especially
 452                          *      when the MDT has performed file-level backup
 453                          *      and restored after former upgrading from 1.8
 454                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 455                          *      verify the inode to decrease the risk. */
 456                         scrub_file_reset(scrub, osd->od_uuid, SF_UPGRADE);
 457                 GOTO(out, rc = 1);
 458         } else if (rc != -ENOENT) {
 459                 CERROR("%s: can't open %s: rc = %d\n",
 460                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 461                 GOTO(out, rc);
 462         }
 463
 464         if (sf->sf_oi_count > 0) {
 465                 int i;
 466
 467                 count = sf->sf_oi_count;
 468                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 469                 for (i = 0; i < count; i++)
 470                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 471                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 472         } else {
 473                 count = sf->sf_oi_count = osd_oi_count;
 474         }
 475
 476 create:
 477         rc = scrub_file_store(info->oti_env, scrub);
 478         if (rc < 0) {
 479                 osd_oi_table_put(info, oi, count);
 480                 GOTO(out, rc);
 481         }
 482
 483         /* No OIs exist, new filesystem, create OI objects */
 484         rc = osd_oi_table_open(info, osd, oi, count, true);
 485         LASSERT(ergo(rc >= 0, rc == count));
 486
 487         GOTO(out, rc);
 488
 489 out:
 490         if (rc < 0) {
 491                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 492         } else {
 493                 LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
 494
 495                 osd->od_oi_table = oi;
 496                 osd->od_oi_count = rc;
 497                 if (sf->sf_oi_count != rc) {
 498                         sf->sf_oi_count = rc;
 499                         rc = scrub_file_store(info->oti_env, scrub);
 500                         if (rc < 0) {
 501                                 osd_oi_table_put(info, oi, count);
 502                                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 503                         }
 504                 } else {
 505                         rc = 0;
 506                 }
 507         }
 508
 509         return rc;
 510 }
 511
 512 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 513 {
 514         if (unlikely(!osd->od_oi_table))
 515                 return;
 516
 517         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 518
 519         OBD_FREE(osd->od_oi_table,
 520                  sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
 521         osd->od_oi_table = NULL;
 522 }
 523
 524 static inline int fid_is_fs_root(const struct lu_fid *fid)
 525 {
 526         /* Map root inode to special local object FID */
 527         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 528                          fid_oid(fid) == OSD_FS_ROOT_OID));
 529 }
 530
 531 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 532                              struct osd_oi *oi, struct dt_rec *rec,
 533                              const struct dt_key *key)
 534 {
 535         struct iam_container  *bag;
 536         struct iam_iterator   *it = &oti->oti_idx_it;
 537         struct iam_path_descr *ipd;
 538         int                    rc;
 539         ENTRY;
 540
 541         LASSERT(oi);
 542         LASSERT(oi->oi_inode);
 543
 544         bag = &oi->oi_dir.od_container;
 545         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 546         if (IS_ERR(ipd))
 547                 RETURN(-ENOMEM);
 548
 549         /* got ipd now we can start iterator. */
 550         iam_it_init(it, bag, 0, ipd);
 551
 552         rc = iam_it_get(it, (struct iam_key *)key);
 553         if (rc > 0)
 554                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 555         iam_it_put(it);
 556         iam_it_fini(it);
 557         osd_ipd_put(oti->oti_env, bag, ipd);
 558
 559         LINVRNT(osd_invariant(obj));
 560
 561         RETURN(rc);
 562 }
 563
 564 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 565                   const struct lu_fid *fid, enum oi_check_flags flags)
 566 {
 567         struct lu_seq_range     *range = &info->oti_seq_range;
 568         int                     rc;
 569         ENTRY;
 570
 571         if (flags & OI_KNOWN_ON_OST)
 572                 RETURN(1);
 573
 574         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 575                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 576                      fid_is_quota(fid))
 577                 RETURN(0);
 578
 579         if (fid_is_idif(fid) || fid_is_last_id(fid))
 580                 RETURN(1);
 581
 582         if (!(flags & OI_CHECK_FLD))
 583                 RETURN(0);
 584
 585         if (osd_seq_site(osd)->ss_server_fld == NULL)
 586                 RETURN(0);
 587
 588         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 589         if (rc != 0) {
 590                 /* During upgrade, OST FLDB might not be loaded because
 591                  * OST FLDB is not created until 2.6, so if some DNE
 592                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 593                  * not be able to find the sequence from local FLDB
 594                  * cache see fld_index_init(). */
 595                 if (rc == -ENOENT && osd->od_is_ost)
 596                         RETURN(1);
 597
 598                 if (rc != -ENOENT)
 599                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 600                                osd_name(osd), PFID(fid), rc);
 601                 RETURN(0);
 602         }
 603
 604         if (fld_range_is_ost(range))
 605                 RETURN(1);
 606
 607         RETURN(0);
 608 }
 609
 610 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 611                            const struct lu_fid *fid, struct osd_inode_id *id)
 612 {
 613         struct lu_fid *oi_fid = &info->oti_fid2;
 614         int            rc;
 615
 616         fid_cpu_to_be(oi_fid, fid);
 617         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 618                                (const struct dt_key *)oi_fid);
 619         if (rc > 0) {
 620                 osd_id_unpack(id, id);
 621                 rc = 0;
 622         } else if (rc == 0) {
 623                 rc = -ENOENT;
 624         }
 625         return rc;
 626 }
 627
 628 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 629                   const struct lu_fid *fid, struct osd_inode_id *id,
 630                   enum oi_check_flags flags)
 631 {
 632         if (unlikely(fid_is_last_id(fid)))
 633                 return osd_obj_spec_lookup(info, osd, fid, id);
 634
 635         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 636                 return osd_obj_map_lookup(info, osd, fid, id);
 637
 638
 639         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 640                 int rc;
 641                 if (fid_is_fs_root(fid)) {
 642                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 643                                    osd_sb(osd)->s_root->d_inode->i_generation);
 644                         return 0;
 645                 }
 646                 if (unlikely(fid_is_acct(fid)))
 647                         return osd_acct_obj_lookup(info, osd, fid, id);
 648
 649                 /* For other special FIDs, try OI first, then do spec lookup */
 650                 rc = __osd_oi_lookup(info, osd, fid, id);
 651                 if (rc == -ENOENT)
 652                         return osd_obj_spec_lookup(info, osd, fid, id);
 653                 return rc;
 654         }
 655
 656         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 657                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 658                 return 0;
 659         }
 660
 661         return __osd_oi_lookup(info, osd, fid, id);
 662 }
 663
 664 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 665                              const struct dt_rec *rec, const struct dt_key *key,
 666                              handle_t *th, bool insert)
 667 {
 668         struct iam_container    *bag;
 669         struct iam_path_descr   *ipd;
 670         int                     rc;
 671         ENTRY;
 672
 673         LASSERT(oi);
 674         LASSERT(oi->oi_inode);
 675         ll_vfs_dq_init(oi->oi_inode);
 676
 677         bag = &oi->oi_dir.od_container;
 678         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 679         if (unlikely(ipd == NULL))
 680                 RETURN(-ENOMEM);
 681
 682         LASSERT(th != NULL);
 683         LASSERT(th->h_transaction != NULL);
 684         if (insert)
 685                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 686                                 (const struct iam_rec *)rec, ipd);
 687         else
 688                 rc = iam_update(th, bag, (const struct iam_key *)key,
 689                                 (const struct iam_rec *)rec, ipd);
 690         osd_ipd_put(oti->oti_env, bag, ipd);
 691         LINVRNT(osd_invariant(obj));
 692         RETURN(rc);
 693 }
 694
 695 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 696                   const struct lu_fid *fid, const struct osd_inode_id *id,
 697                   handle_t *th, enum oi_check_flags flags, bool *exist)
 698 {
 699         struct lu_fid       *oi_fid = &info->oti_fid2;
 700         struct osd_inode_id *oi_id  = &info->oti_id2;
 701         int                  rc     = 0;
 702
 703         if (unlikely(fid_is_last_id(fid)))
 704                 return osd_obj_spec_insert(info, osd, fid, id, th);
 705
 706         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 707                 return osd_obj_map_insert(info, osd, fid, id, th);
 708
 709         fid_cpu_to_be(oi_fid, fid);
 710         osd_id_pack(oi_id, id);
 711         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 712                                (const struct dt_rec *)oi_id,
 713                                (const struct dt_key *)oi_fid, th, true);
 714         if (rc != 0) {
 715                 struct inode *inode;
 716                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 717
 718                 if (rc != -EEXIST)
 719                         return rc;
 720
 721                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 722                 if (rc != 0)
 723                         return rc;
 724
 725                 if (unlikely(osd_id_eq(id, oi_id)))
 726                         return 1;
 727
 728                 /* Check whether the mapping for oi_id is valid or not. */
 729                 inode = osd_iget(info, osd, oi_id);
 730                 if (IS_ERR(inode)) {
 731                         rc = PTR_ERR(inode);
 732                         if (rc == -ENOENT || rc == -ESTALE)
 733                                 goto update;
 734                         return rc;
 735                 }
 736
 737                 /* The EA inode should NOT be in OI, old OI scrub may added
 738                  * such OI mapping by wrong, replace it. */
 739                 if (unlikely(osd_is_ea_inode(inode))) {
 740                         iput(inode);
 741                         goto update;
 742                 }
 743
 744                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 745                                  &info->oti_ost_attrs);
 746                 iput(inode);
 747                 if (rc == -ENODATA)
 748                         goto update;
 749
 750                 if (rc != 0)
 751                         return rc;
 752
 753                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 754                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 755                         CERROR("%s: the FID "DFID" is used by two objects: "
 756                                "%u/%u %u/%u\n", osd_dev2name(osd),
 757                                PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
 758                                id->oii_ino, id->oii_gen);
 759                         return -EEXIST;
 760                 }
 761
 762 update:
 763                 osd_id_pack(oi_id, id);
 764                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 765                                         (const struct dt_rec *)oi_id,
 766                                         (const struct dt_key *)oi_fid, th, false);
 767                 if (rc != 0)
 768                         return rc;
 769
 770                 if (exist != NULL)
 771                         *exist = true;
 772         }
 773
 774         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 775                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 776         return rc;
 777 }
 778
 779 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 780                              const struct dt_key *key, handle_t *th)
 781 {
 782         struct iam_container    *bag;
 783         struct iam_path_descr   *ipd;
 784         int                      rc;
 785         ENTRY;
 786
 787         LASSERT(oi);
 788         LASSERT(oi->oi_inode);
 789         ll_vfs_dq_init(oi->oi_inode);
 790
 791         bag = &oi->oi_dir.od_container;
 792         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 793         if (unlikely(ipd == NULL))
 794                 RETURN(-ENOMEM);
 795
 796         LASSERT(th != NULL);
 797         LASSERT(th->h_transaction != NULL);
 798
 799         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 800         osd_ipd_put(oti->oti_env, bag, ipd);
 801         LINVRNT(osd_invariant(obj));
 802         RETURN(rc);
 803 }
 804
 805 int osd_oi_delete(struct osd_thread_info *info,
 806                   struct osd_device *osd, const struct lu_fid *fid,
 807                   handle_t *th, enum oi_check_flags flags)
 808 {
 809         struct lu_fid *oi_fid = &info->oti_fid2;
 810
 811         /* clear idmap cache */
 812         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 813                 fid_zero(&info->oti_cache.oic_fid);
 814
 815         if (fid_is_last_id(fid))
 816                 return 0;
 817
 818         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 819                 return osd_obj_map_delete(info, osd, fid, th);
 820
 821         fid_cpu_to_be(oi_fid, fid);
 822         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 823                                  (const struct dt_key *)oi_fid, th);
 824 }
 825
 826 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 827                   const struct lu_fid *fid, const struct osd_inode_id *id,
 828                   handle_t *th, enum oi_check_flags flags)
 829 {
 830         struct lu_fid       *oi_fid = &info->oti_fid2;
 831         struct osd_inode_id *oi_id  = &info->oti_id2;
 832         int                  rc     = 0;
 833
 834         if (unlikely(fid_is_last_id(fid)))
 835                 return osd_obj_spec_update(info, osd, fid, id, th);
 836
 837         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 838                 return osd_obj_map_update(info, osd, fid, id, th);
 839
 840         fid_cpu_to_be(oi_fid, fid);
 841         osd_id_pack(oi_id, id);
 842         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 843                                (const struct dt_rec *)oi_id,
 844                                (const struct dt_key *)oi_fid, th, false);
 845         if (rc != 0)
 846                 return rc;
 847
 848         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 849                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 850         return rc;
 851 }
 852
 853 int osd_oi_mod_init(void)
 854 {
 855         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 856                 osd_oi_count = OSD_OI_FID_NR;
 857
 858         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 859                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 860                               osd_oi_count, size_roundup_power2(osd_oi_count));
 861                 osd_oi_count = size_roundup_power2(osd_oi_count);
 862         }
 863
 864         return 0;
 865 }