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, 2016, 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 /* LUSTRE_VERSION_CODE */
  44 #include <lustre_ver.h>
  45 /*
  46  * struct OBD_{ALLOC,FREE}*()
  47  * OBD_FAIL_CHECK
  48  */
  49 #include <obd.h>
  50 #include <obd_support.h>
  51
  52 /* fid_cpu_to_be() */
  53 #include <lustre_fid.h>
  54 #include <dt_object.h>
  55
  56 #include "osd_oi.h"
  57 /* osd_lookup(), struct osd_thread_info */
  58 #include "osd_internal.h"
  59 #include "osd_scrub.h"
  60
  61 static unsigned int osd_oi_count = OSD_OI_FID_NR;
  62 module_param(osd_oi_count, int, 0444);
  63 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
  64
  65 /** to serialize concurrent OI index initialization */
  66 static struct mutex oi_init_lock;
  67
  68 static struct dt_index_features oi_feat = {
  69         .dif_flags       = DT_IND_UPDATE,
  70         .dif_recsize_min = sizeof(struct osd_inode_id),
  71         .dif_recsize_max = sizeof(struct osd_inode_id),
  72         .dif_ptrsize     = 4
  73 };
  74
  75 #define OSD_OI_NAME_BASE        "oi.16"
  76
  77 static void osd_oi_table_put(struct osd_thread_info *info,
  78                              struct osd_oi **oi_table, unsigned oi_count)
  79 {
  80         struct iam_container *bag;
  81         int                   i;
  82
  83         for (i = 0; i < oi_count; i++) {
  84                 if (oi_table[i] == NULL)
  85                         continue;
  86
  87                 LASSERT(oi_table[i]->oi_inode != NULL);
  88
  89                 bag = &(oi_table[i]->oi_dir.od_container);
  90                 if (bag->ic_object == oi_table[i]->oi_inode)
  91                         iam_container_fini(bag);
  92                 iput(oi_table[i]->oi_inode);
  93                 oi_table[i]->oi_inode = NULL;
  94                 OBD_FREE_PTR(oi_table[i]);
  95                 oi_table[i] = NULL;
  96         }
  97 }
  98
  99 static int osd_oi_index_create_one(struct osd_thread_info *info,
 100                                    struct osd_device *osd, const char *name,
 101                                    struct dt_index_features *feat)
 102 {
 103         const struct lu_env             *env = info->oti_env;
 104         struct osd_inode_id             *id  = &info->oti_id;
 105         struct buffer_head              *bh;
 106         struct inode                    *inode;
 107         struct ldiskfs_dir_entry_2      *de;
 108         struct dentry                   *dentry;
 109         struct super_block              *sb  = osd_sb(osd);
 110         struct inode                    *dir = sb->s_root->d_inode;
 111         handle_t                        *jh;
 112         int                              rc;
 113
 114         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 115         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
 116         if (!IS_ERR(bh)) {
 117                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 118                 brelse(bh);
 119                 inode = osd_iget(info, osd, id);
 120                 if (!IS_ERR(inode)) {
 121                         iput(inode);
 122                         inode = ERR_PTR(-EEXIST);
 123                 }
 124                 return PTR_ERR(inode);
 125         }
 126
 127         jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
 128         if (IS_ERR(jh))
 129                 return PTR_ERR(jh);
 130
 131         inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR));
 132         if (IS_ERR(inode)) {
 133                 ldiskfs_journal_stop(jh);
 134                 return PTR_ERR(inode);
 135         }
 136
 137         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 138         unlock_new_inode(inode);
 139
 140         if (feat->dif_flags & DT_IND_VARKEY)
 141                 rc = iam_lvar_create(inode, feat->dif_keysize_max,
 142                                      feat->dif_ptrsize, feat->dif_recsize_max,
 143                                      jh);
 144         else
 145                 rc = iam_lfix_create(inode, feat->dif_keysize_max,
 146                                      feat->dif_ptrsize, feat->dif_recsize_max,
 147                                      jh);
 148         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 149         rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
 150         ldiskfs_journal_stop(jh);
 151         iput(inode);
 152         return rc;
 153 }
 154
 155 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 156                                        struct osd_device *osd,
 157                                        const char *name,
 158                                        struct dt_index_features *f,
 159                                        bool create)
 160 {
 161         struct dentry *dentry;
 162         struct inode  *inode;
 163         int            rc;
 164
 165         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 166         if (IS_ERR(dentry))
 167                 return (void *) 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 = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 188         if (IS_ERR(dentry))
 189                 return (void *) 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
 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, inode->i_generation);
 232         rc = osd_ea_fid_set(info, inode, &info->oti_fid, LMAC_NOT_IN_OI, 0);
 233         if (rc != 0)
 234                 GOTO(out_inode, rc);
 235
 236         OBD_ALLOC_PTR(oi);
 237         if (oi == NULL)
 238                 GOTO(out_inode, rc = -ENOMEM);
 239
 240         oi->oi_inode = inode;
 241         dir = &oi->oi_dir;
 242
 243         bag = &dir->od_container;
 244         rc = iam_container_init(bag, &dir->od_descr, inode);
 245         if (rc < 0)
 246                 GOTO(out_free, rc);
 247
 248         rc = iam_container_setup(bag);
 249         if (rc < 0)
 250                 GOTO(out_container, rc);
 251
 252         *oi_slot = oi;
 253         RETURN(0);
 254
 255 out_container:
 256         iam_container_fini(bag);
 257 out_free:
 258         OBD_FREE_PTR(oi);
 259 out_inode:
 260         iput(inode);
 261         return rc;
 262 }
 263
 264 /**
 265  * Open OI(Object Index) table.
 266  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 267  * will be, this function can either return 0 for new filesystem, or number
 268  * of OIs on existed filesystem.
 269  *
 270  * If \a oi_count is non-zero, which means caller does know number of OIs on
 271  * filesystem, this function should return the exactly same number on
 272  * success, or error code in failure.
 273  *
 274  * \param     oi_count  Number of expected OI containers
 275  * \param     create    Create OIs if doesn't exist
 276  *
 277  * \retval    +ve       number of opened OI containers
 278  * \retval      0       no OI containers found
 279  * \retval    -ve       failure
 280  */
 281 static int
 282 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 283                   struct osd_oi **oi_table, unsigned oi_count, bool create)
 284 {
 285         struct scrub_file *sf = &osd->od_scrub.os_file;
 286         int                count = 0;
 287         int                rc = 0;
 288         int                i;
 289         ENTRY;
 290
 291         /* NB: oi_count != 0 means that we have already created/known all OIs
 292          * and have known exact number of OIs. */
 293         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 294
 295         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 296                 char name[12];
 297
 298                 if (oi_table[i] != NULL) {
 299                         count++;
 300                         continue;
 301                 }
 302
 303                 sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
 304                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 305                 if (rc == 0) {
 306                         count++;
 307                         continue;
 308                 }
 309
 310                 if (rc == -ENOENT && create == false) {
 311                         if (oi_count == 0)
 312                                 return count;
 313
 314                         rc = 0;
 315                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 316                         continue;
 317                 }
 318
 319                 CERROR("%s: can't open %s: rc = %d\n",
 320                        osd_dev2name(osd), name, rc);
 321                 if (oi_count > 0)
 322                         CERROR("%s: expect to open total %d OI files.\n",
 323                                osd_dev2name(osd), oi_count);
 324                 break;
 325         }
 326
 327         if (rc < 0) {
 328                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 329                 count = rc;
 330         }
 331
 332         RETURN(count);
 333 }
 334
 335 static int osd_remove_oi_one(struct dentry *parent, const char *name,
 336                              int namelen)
 337 {
 338         struct dentry *child;
 339         int rc;
 340
 341         child = ll_lookup_one_len(name, parent, namelen);
 342         if (IS_ERR(child)) {
 343                 rc = PTR_ERR(child);
 344         } else {
 345                 rc = ll_vfs_unlink(parent->d_inode, child);
 346                 dput(child);
 347         }
 348
 349         return rc == -ENOENT ? 0 : rc;
 350 }
 351
 352 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 353 {
 354         char name[16];
 355         int namelen;
 356         int rc;
 357         int i;
 358
 359         for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; i++) {
 360                 namelen = snprintf(name, sizeof(name), "%s.%d",
 361                                    OSD_OI_NAME_BASE, i);
 362                 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
 363                 if (rc != 0) {
 364                         CERROR("%s: fail to remove the stale OI file %s: "
 365                                "rc = %d\n", osd_dev2name(osd), name, rc);
 366                         return rc;
 367                 }
 368         }
 369
 370         namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
 371         rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
 372         if (rc != 0)
 373                 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
 374                        osd_dev2name(osd), name, rc);
 375
 376         return rc;
 377 }
 378
 379 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
 380                 bool restored)
 381 {
 382         struct osd_scrub  *scrub = &osd->od_scrub;
 383         struct scrub_file *sf = &scrub->os_file;
 384         struct osd_oi    **oi;
 385         int                rc;
 386         ENTRY;
 387
 388         if (restored) {
 389                 rc = osd_remove_ois(info, osd);
 390                 if (rc != 0)
 391                         return rc;
 392         }
 393
 394         OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 395         if (oi == NULL)
 396                 RETURN(-ENOMEM);
 397
 398         mutex_lock(&oi_init_lock);
 399         /* try to open existing multiple OIs first */
 400         rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 401         if (rc < 0)
 402                 GOTO(out, rc);
 403
 404         if (rc > 0) {
 405                 if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
 406                         GOTO(out, rc);
 407
 408                 osd_scrub_file_reset(scrub,
 409                                      LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 410                                      SF_RECREATED);
 411                 osd_oi_count = sf->sf_oi_count;
 412                 goto create;
 413         }
 414
 415         /* if previous failed then try found single OI from old filesystem */
 416         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 417         if (rc == 0) { /* found single OI from old filesystem */
 418                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 419                 if (sf->sf_success_count == 0)
 420                         /* XXX: There is one corner case that if the OI_scrub
 421                          *      file crashed or lost and we regard it upgrade,
 422                          *      then we allow IGIF lookup to bypass OI files.
 423                          *
 424                          *      The risk is that osd_fid_lookup() may found
 425                          *      a wrong inode with the given IGIF especially
 426                          *      when the MDT has performed file-level backup
 427                          *      and restored after former upgrading from 1.8
 428                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 429                          *      verify the inode to decrease the risk. */
 430                         osd_scrub_file_reset(scrub,
 431                                         LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 432                                         SF_UPGRADE);
 433                 GOTO(out, rc = 1);
 434         } else if (rc != -ENOENT) {
 435                 CERROR("%s: can't open %s: rc = %d\n",
 436                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 437                 GOTO(out, rc);
 438         }
 439
 440         if (sf->sf_oi_count > 0) {
 441                 int i;
 442
 443                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 444                 for (i = 0; i < osd_oi_count; i++)
 445                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 446                 osd_scrub_file_reset(scrub,
 447                                      LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 448                                      SF_RECREATED);
 449         }
 450         sf->sf_oi_count = osd_oi_count;
 451
 452 create:
 453         rc = osd_scrub_file_store(scrub);
 454         if (rc < 0) {
 455                 osd_oi_table_put(info, oi, sf->sf_oi_count);
 456                 GOTO(out, rc);
 457         }
 458
 459         /* No OIs exist, new filesystem, create OI objects */
 460         rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
 461         LASSERT(ergo(rc >= 0, rc == osd_oi_count));
 462
 463         GOTO(out, rc);
 464
 465 out:
 466         if (rc < 0) {
 467                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 468         } else {
 469                 LASSERT((rc & (rc - 1)) == 0);
 470                 osd->od_oi_table = oi;
 471                 osd->od_oi_count = rc;
 472                 if (sf->sf_oi_count != rc) {
 473                         sf->sf_oi_count = rc;
 474                         rc = osd_scrub_file_store(scrub);
 475                         if (rc < 0) {
 476                                 osd_oi_table_put(info, oi, sf->sf_oi_count);
 477                                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 478                         }
 479                 } else {
 480                         rc = 0;
 481                 }
 482         }
 483
 484         mutex_unlock(&oi_init_lock);
 485         return rc;
 486 }
 487
 488 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 489 {
 490         if (unlikely(osd->od_oi_table == NULL))
 491                 return;
 492
 493         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 494
 495         OBD_FREE(osd->od_oi_table,
 496                  sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
 497         osd->od_oi_table = NULL;
 498 }
 499
 500 static inline int fid_is_fs_root(const struct lu_fid *fid)
 501 {
 502         /* Map root inode to special local object FID */
 503         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 504                          fid_oid(fid) == OSD_FS_ROOT_OID));
 505 }
 506
 507 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 508                              struct osd_oi *oi, struct dt_rec *rec,
 509                              const struct dt_key *key)
 510 {
 511         struct iam_container  *bag;
 512         struct iam_iterator   *it = &oti->oti_idx_it;
 513         struct iam_path_descr *ipd;
 514         int                    rc;
 515         ENTRY;
 516
 517         LASSERT(oi);
 518         LASSERT(oi->oi_inode);
 519
 520         bag = &oi->oi_dir.od_container;
 521         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 522         if (IS_ERR(ipd))
 523                 RETURN(-ENOMEM);
 524
 525         /* got ipd now we can start iterator. */
 526         iam_it_init(it, bag, 0, ipd);
 527
 528         rc = iam_it_get(it, (struct iam_key *)key);
 529         if (rc > 0)
 530                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 531         iam_it_put(it);
 532         iam_it_fini(it);
 533         osd_ipd_put(oti->oti_env, bag, ipd);
 534
 535         LINVRNT(osd_invariant(obj));
 536
 537         RETURN(rc);
 538 }
 539
 540 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 541                   const struct lu_fid *fid, enum oi_check_flags flags)
 542 {
 543         struct lu_seq_range     *range = &info->oti_seq_range;
 544         int                     rc;
 545         ENTRY;
 546
 547         if (flags & OI_KNOWN_ON_OST)
 548                 RETURN(1);
 549
 550         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 551                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 552                      fid_is_quota(fid))
 553                 RETURN(0);
 554
 555         if (fid_is_idif(fid) || fid_is_last_id(fid))
 556                 RETURN(1);
 557
 558         if (!(flags & OI_CHECK_FLD))
 559                 RETURN(0);
 560
 561         if (osd_seq_site(osd)->ss_server_fld == NULL)
 562                 RETURN(0);
 563
 564         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 565         if (rc != 0) {
 566                 /* During upgrade, OST FLDB might not be loaded because
 567                  * OST FLDB is not created until 2.6, so if some DNE
 568                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 569                  * not be able to find the sequence from local FLDB
 570                  * cache see fld_index_init(). */
 571                 if (rc == -ENOENT && osd->od_is_ost)
 572                         RETURN(1);
 573
 574                 if (rc != -ENOENT)
 575                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 576                                osd_name(osd), PFID(fid), rc);
 577                 RETURN(0);
 578         }
 579
 580         if (fld_range_is_ost(range))
 581                 RETURN(1);
 582
 583         RETURN(0);
 584 }
 585
 586 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 587                            const struct lu_fid *fid, struct osd_inode_id *id)
 588 {
 589         struct lu_fid *oi_fid = &info->oti_fid2;
 590         int            rc;
 591
 592         fid_cpu_to_be(oi_fid, fid);
 593         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 594                                (const struct dt_key *)oi_fid);
 595         if (rc > 0) {
 596                 osd_id_unpack(id, id);
 597                 rc = 0;
 598         } else if (rc == 0) {
 599                 rc = -ENOENT;
 600         }
 601         return rc;
 602 }
 603
 604 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 605                   const struct lu_fid *fid, struct osd_inode_id *id,
 606                   enum oi_check_flags flags)
 607 {
 608         if (unlikely(fid_is_last_id(fid)))
 609                 return osd_obj_spec_lookup(info, osd, fid, id);
 610
 611         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 612                 return osd_obj_map_lookup(info, osd, fid, id);
 613
 614
 615         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 616                 int rc;
 617                 if (fid_is_fs_root(fid)) {
 618                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 619                                    osd_sb(osd)->s_root->d_inode->i_generation);
 620                         return 0;
 621                 }
 622                 if (unlikely(fid_is_acct(fid)))
 623                         return osd_acct_obj_lookup(info, osd, fid, id);
 624
 625                 /* For other special FIDs, try OI first, then do spec lookup */
 626                 rc = __osd_oi_lookup(info, osd, fid, id);
 627                 if (rc == -ENOENT)
 628                         return osd_obj_spec_lookup(info, osd, fid, id);
 629                 return rc;
 630         }
 631
 632         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 633                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 634                 return 0;
 635         }
 636
 637         return __osd_oi_lookup(info, osd, fid, id);
 638 }
 639
 640 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 641                              const struct dt_rec *rec, const struct dt_key *key,
 642                              handle_t *th, bool insert)
 643 {
 644         struct iam_container    *bag;
 645         struct iam_path_descr   *ipd;
 646         int                     rc;
 647         ENTRY;
 648
 649         LASSERT(oi);
 650         LASSERT(oi->oi_inode);
 651         ll_vfs_dq_init(oi->oi_inode);
 652
 653         bag = &oi->oi_dir.od_container;
 654         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 655         if (unlikely(ipd == NULL))
 656                 RETURN(-ENOMEM);
 657
 658         LASSERT(th != NULL);
 659         LASSERT(th->h_transaction != NULL);
 660         if (insert)
 661                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 662                                 (const struct iam_rec *)rec, ipd);
 663         else
 664                 rc = iam_update(th, bag, (const struct iam_key *)key,
 665                                 (const struct iam_rec *)rec, ipd);
 666         osd_ipd_put(oti->oti_env, bag, ipd);
 667         LINVRNT(osd_invariant(obj));
 668         RETURN(rc);
 669 }
 670
 671 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 672                   const struct lu_fid *fid, const struct osd_inode_id *id,
 673                   handle_t *th, enum oi_check_flags flags, bool *exist)
 674 {
 675         struct lu_fid       *oi_fid = &info->oti_fid2;
 676         struct osd_inode_id *oi_id  = &info->oti_id2;
 677         int                  rc     = 0;
 678
 679         if (unlikely(fid_is_last_id(fid)))
 680                 return osd_obj_spec_insert(info, osd, fid, id, th);
 681
 682         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 683                 return osd_obj_map_insert(info, osd, fid, id, th);
 684
 685         fid_cpu_to_be(oi_fid, fid);
 686         osd_id_pack(oi_id, id);
 687         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 688                                (const struct dt_rec *)oi_id,
 689                                (const struct dt_key *)oi_fid, th, true);
 690         if (rc != 0) {
 691                 struct inode *inode;
 692                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 693
 694                 if (rc != -EEXIST)
 695                         return rc;
 696
 697                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 698                 if (rc != 0)
 699                         return rc;
 700
 701                 if (unlikely(osd_id_eq(id, oi_id)))
 702                         return 1;
 703
 704                 /* Check whether the mapping for oi_id is valid or not. */
 705                 inode = osd_iget(info, osd, oi_id);
 706                 if (IS_ERR(inode)) {
 707                         rc = PTR_ERR(inode);
 708                         if (rc == -ENOENT || rc == -ESTALE)
 709                                 goto update;
 710                         return rc;
 711                 }
 712
 713                 /* The EA inode should NOT be in OI, old OI scrub may added
 714                  * such OI mapping by wrong, replace it. */
 715                 if (unlikely(osd_is_ea_inode(inode))) {
 716                         iput(inode);
 717                         goto update;
 718                 }
 719
 720                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 721                                  &info->oti_ost_attrs);
 722                 iput(inode);
 723                 if (rc == -ENODATA)
 724                         goto update;
 725
 726                 if (rc != 0)
 727                         return rc;
 728
 729                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 730                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 731                         CERROR("%s: the FID "DFID" is used by two objects: "
 732                                "%u/%u %u/%u\n", osd_dev2name(osd),
 733                                PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
 734                                id->oii_ino, id->oii_gen);
 735                         return -EEXIST;
 736                 }
 737
 738 update:
 739                 osd_id_pack(oi_id, id);
 740                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 741                                         (const struct dt_rec *)oi_id,
 742                                         (const struct dt_key *)oi_fid, th, false);
 743                 if (rc != 0)
 744                         return rc;
 745
 746                 if (exist != NULL)
 747                         *exist = true;
 748         }
 749
 750         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 751                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 752         return rc;
 753 }
 754
 755 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 756                              const struct dt_key *key, handle_t *th)
 757 {
 758         struct iam_container    *bag;
 759         struct iam_path_descr   *ipd;
 760         int                      rc;
 761         ENTRY;
 762
 763         LASSERT(oi);
 764         LASSERT(oi->oi_inode);
 765         ll_vfs_dq_init(oi->oi_inode);
 766
 767         bag = &oi->oi_dir.od_container;
 768         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 769         if (unlikely(ipd == NULL))
 770                 RETURN(-ENOMEM);
 771
 772         LASSERT(th != NULL);
 773         LASSERT(th->h_transaction != NULL);
 774
 775         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 776         osd_ipd_put(oti->oti_env, bag, ipd);
 777         LINVRNT(osd_invariant(obj));
 778         RETURN(rc);
 779 }
 780
 781 int osd_oi_delete(struct osd_thread_info *info,
 782                   struct osd_device *osd, const struct lu_fid *fid,
 783                   handle_t *th, enum oi_check_flags flags)
 784 {
 785         struct lu_fid *oi_fid = &info->oti_fid2;
 786
 787         /* clear idmap cache */
 788         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 789                 fid_zero(&info->oti_cache.oic_fid);
 790
 791         if (fid_is_last_id(fid))
 792                 return 0;
 793
 794         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 795                 return osd_obj_map_delete(info, osd, fid, th);
 796
 797         fid_cpu_to_be(oi_fid, fid);
 798         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 799                                  (const struct dt_key *)oi_fid, th);
 800 }
 801
 802 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 803                   const struct lu_fid *fid, const struct osd_inode_id *id,
 804                   handle_t *th, enum oi_check_flags flags)
 805 {
 806         struct lu_fid       *oi_fid = &info->oti_fid2;
 807         struct osd_inode_id *oi_id  = &info->oti_id2;
 808         int                  rc     = 0;
 809
 810         if (unlikely(fid_is_last_id(fid)))
 811                 return osd_obj_spec_update(info, osd, fid, id, th);
 812
 813         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 814                 return osd_obj_map_update(info, osd, fid, id, th);
 815
 816         fid_cpu_to_be(oi_fid, fid);
 817         osd_id_pack(oi_id, id);
 818         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 819                                (const struct dt_rec *)oi_id,
 820                                (const struct dt_key *)oi_fid, th, false);
 821         if (rc != 0)
 822                 return rc;
 823
 824         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 825                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 826         return rc;
 827 }
 828
 829 int osd_oi_mod_init(void)
 830 {
 831         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 832                 osd_oi_count = OSD_OI_FID_NR;
 833
 834         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 835                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 836                               osd_oi_count, size_roundup_power2(osd_oi_count));
 837                 osd_oi_count = size_roundup_power2(osd_oi_count);
 838         }
 839
 840         mutex_init(&oi_init_lock);
 841         return 0;
 842 }