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         ldiskfs_set_inode_state(inode, LDISKFS_STATE_IAM);
 230
 231         if (!osd->od_dt_dev.dd_rdonly) {
 232                 /* 'What the @fid is' is not imporatant, because these objects
 233                  * have no OI mappings, and only are visible inside the OSD.
 234                  */
 235                 lu_igif_build(&info->oti_fid, inode->i_ino,
 236                               inode->i_generation);
 237                 rc = osd_ea_fid_set(info, inode, &info->oti_fid,
 238                                     LMAC_NOT_IN_OI, 0);
 239                 if (rc)
 240                         GOTO(out_inode, rc);
 241         }
 242
 243         OBD_ALLOC_PTR(oi);
 244         if (oi == NULL)
 245                 GOTO(out_inode, rc = -ENOMEM);
 246
 247         oi->oi_inode = inode;
 248         dir = &oi->oi_dir;
 249
 250         bag = &dir->od_container;
 251         rc = iam_container_init(bag, &dir->od_descr, inode);
 252         if (rc < 0)
 253                 GOTO(out_free, rc);
 254
 255         rc = iam_container_setup(bag);
 256         if (rc < 0)
 257                 GOTO(out_container, rc);
 258
 259         *oi_slot = oi;
 260         RETURN(0);
 261
 262 out_container:
 263         iam_container_fini(bag);
 264 out_free:
 265         OBD_FREE_PTR(oi);
 266 out_inode:
 267         iput(inode);
 268         return rc;
 269 }
 270
 271 /**
 272  * Open OI(Object Index) table.
 273  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 274  * will be, this function can either return 0 for new filesystem, or number
 275  * of OIs on existed filesystem.
 276  *
 277  * If \a oi_count is non-zero, which means caller does know number of OIs on
 278  * filesystem, this function should return the exactly same number on
 279  * success, or error code in failure.
 280  *
 281  * \param     oi_count  Number of expected OI containers
 282  * \param     create    Create OIs if doesn't exist
 283  *
 284  * \retval    +ve       number of opened OI containers
 285  * \retval      0       no OI containers found
 286  * \retval    -ve       failure
 287  */
 288 static int
 289 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 290                   struct osd_oi **oi_table, unsigned int oi_count, bool create)
 291 {
 292         struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
 293         int count = 0;
 294         int rc = 0;
 295         int i;
 296         ENTRY;
 297
 298         /* NB: oi_count != 0 means that we have already created/known all OIs
 299          * and have known exact number of OIs.
 300          */
 301         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 302
 303         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 304                 char name[sizeof(OSD_OI_NAME_BASE) + 3 * sizeof(i) + 1];
 305
 306                 if (oi_table[i] != NULL) {
 307                         count++;
 308                         continue;
 309                 }
 310
 311                 snprintf(name, sizeof(name), "%s.%d", OSD_OI_NAME_BASE, i);
 312                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 313                 if (rc == 0) {
 314                         count++;
 315                         continue;
 316                 }
 317
 318                 if (rc == -ENOENT && create == false) {
 319                         if (oi_count == 0)
 320                                 return count;
 321
 322                         rc = 0;
 323                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 324                         continue;
 325                 }
 326
 327                 CERROR("%s: can't open %s: rc = %d\n",
 328                        osd_dev2name(osd), name, rc);
 329                 if (oi_count > 0)
 330                         CERROR("%s: expect to open total %d OI files.\n",
 331                                osd_dev2name(osd), oi_count);
 332                 break;
 333         }
 334
 335         if (rc < 0) {
 336                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 337                 count = rc;
 338         }
 339
 340         RETURN(count);
 341 }
 342
 343 static int osd_remove_oi_one(struct osd_device *osd, struct dentry *parent,
 344                              const char *name, int namelen)
 345 {
 346         struct dentry *child;
 347         int rc;
 348
 349         child = osd_lookup_one_len_unlocked(osd, name, parent, namelen);
 350         if (IS_ERR(child)) {
 351                 rc = PTR_ERR(child);
 352         } else {
 353                 rc = ll_vfs_unlink(parent->d_inode, child);
 354                 dput(child);
 355         }
 356
 357         return rc == -ENOENT ? 0 : rc;
 358 }
 359
 360 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 361 {
 362         char name[16];
 363         int namelen;
 364         int rc;
 365         int i;
 366
 367         if (osd->od_dt_dev.dd_rdonly)
 368                 RETURN(-EROFS);
 369
 370         for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
 371                 namelen = scnprintf(name, sizeof(name), "%s.%d",
 372                                     OSD_OI_NAME_BASE, i);
 373                 rc = osd_remove_oi_one(osd, osd_sb(osd)->s_root, name, namelen);
 374                 if (rc != 0) {
 375                         CERROR(
 376                                "%s: fail to remove the stale OI file %s: rc = %d\n",
 377                                osd_dev2name(osd), name, rc);
 378                         return rc;
 379                 }
 380         }
 381
 382         namelen = scnprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
 383         rc = osd_remove_oi_one(osd, osd_sb(osd)->s_root, name, namelen);
 384         if (rc != 0)
 385                 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
 386                        osd_dev2name(osd), name, rc);
 387
 388         return rc;
 389 }
 390
 391 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
 392                 bool restored)
 393 {
 394         struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
 395         struct scrub_file *sf = &scrub->os_file;
 396         struct osd_oi **oi;
 397         int count;
 398         int rc;
 399         ENTRY;
 400
 401         if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
 402                 LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
 403                               osd_dev2name(osd), sf->sf_oi_count);
 404                 sf->sf_oi_count = 0;
 405         }
 406
 407         if (restored) {
 408                 LCONSOLE_WARN("%s: reset Object Index mappings\n",
 409                               osd_dev2name(osd));
 410                 rc = osd_remove_ois(info, osd);
 411                 if (rc)
 412                         RETURN(rc);
 413         }
 414
 415         OBD_ALLOC_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 416         if (oi == NULL)
 417                 RETURN(-ENOMEM);
 418
 419         /* try to open existing multiple OIs first */
 420         count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 421         if (count < 0)
 422                 GOTO(out, rc = count);
 423
 424         if (count > 0) {
 425                 if (count == sf->sf_oi_count)
 426                         GOTO(out, rc = count);
 427
 428                 if (sf->sf_oi_count == 0) {
 429                         if (likely((count & (count - 1)) == 0))
 430                                 GOTO(out, rc = count);
 431
 432                         LCONSOLE_WARN(
 433                                       "%s: invalid oi count %d, remove them, then set it to %d\n",
 434                                       osd_dev2name(osd), count, osd_oi_count);
 435                         osd_oi_table_put(info, oi, count);
 436                         rc = osd_remove_ois(info, osd);
 437                         if (rc)
 438                                 GOTO(out, rc);
 439
 440                         sf->sf_oi_count = osd_oi_count;
 441                 }
 442
 443                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 444                 count = sf->sf_oi_count;
 445                 goto create;
 446         }
 447
 448         /* if previous failed then try found single OI from old filesystem */
 449         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 450         if (rc == 0) { /* found single OI from old filesystem */
 451                 count = 1;
 452                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 453                 if (sf->sf_success_count == 0)
 454                         /* XXX: There is one corner case that if the OI_scrub
 455                          *      file crashed or lost and we regard it upgrade,
 456                          *      then we allow IGIF lookup to bypass OI files.
 457                          *
 458                          *      The risk is that osd_fid_lookup() may found
 459                          *      a wrong inode with the given IGIF especially
 460                          *      when the MDT has performed file-level backup
 461                          *      and restored after former upgrading from 1.8
 462                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 463                          *      verify the inode to decrease the risk.
 464                          */
 465                         scrub_file_reset(scrub, osd->od_uuid, SF_UPGRADE);
 466                 GOTO(out, rc = 1);
 467         } else if (rc != -ENOENT) {
 468                 CERROR("%s: can't open %s: rc = %d\n",
 469                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 470                 GOTO(out, rc);
 471         }
 472
 473         if (sf->sf_oi_count > 0) {
 474                 int i;
 475
 476                 count = sf->sf_oi_count;
 477                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 478                 for (i = 0; i < count; i++)
 479                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 480                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 481         } else {
 482                 count = sf->sf_oi_count = osd_oi_count;
 483         }
 484
 485 create:
 486         rc = scrub_file_store(info->oti_env, scrub);
 487         if (rc < 0) {
 488                 osd_oi_table_put(info, oi, count);
 489                 GOTO(out, rc);
 490         }
 491
 492         /* No OIs exist, new filesystem, create OI objects */
 493         rc = osd_oi_table_open(info, osd, oi, count, true);
 494         LASSERT(ergo(rc >= 0, rc == count));
 495
 496         GOTO(out, rc);
 497
 498 out:
 499         if (rc < 0) {
 500                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 501         } else {
 502                 LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
 503
 504                 osd->od_oi_table = oi;
 505                 osd->od_oi_count = rc;
 506                 if (sf->sf_oi_count != rc) {
 507                         sf->sf_oi_count = rc;
 508                         rc = scrub_file_store(info->oti_env, scrub);
 509                         if (rc < 0) {
 510                                 osd_oi_table_put(info, oi, count);
 511                                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 512                         }
 513                 } else {
 514                         rc = 0;
 515                 }
 516         }
 517
 518         return rc;
 519 }
 520
 521 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 522 {
 523         if (unlikely(!osd->od_oi_table))
 524                 return;
 525
 526         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 527
 528         OBD_FREE_PTR_ARRAY(osd->od_oi_table, OSD_OI_FID_NR_MAX);
 529         osd->od_oi_table = NULL;
 530 }
 531
 532 static inline int fid_is_fs_root(const struct lu_fid *fid)
 533 {
 534         /* Map root inode to special local object FID */
 535         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 536                          fid_oid(fid) == OSD_FS_ROOT_OID));
 537 }
 538
 539 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 540                              struct osd_oi *oi, struct dt_rec *rec,
 541                              const struct dt_key *key)
 542 {
 543         struct iam_container  *bag;
 544         struct iam_iterator   *it = &oti->oti_idx_it;
 545         struct iam_path_descr *ipd;
 546         int rc;
 547         ENTRY;
 548
 549         LASSERT(oi);
 550         LASSERT(oi->oi_inode);
 551
 552         bag = &oi->oi_dir.od_container;
 553         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 554         if (IS_ERR(ipd))
 555                 RETURN(-ENOMEM);
 556
 557         /* got ipd now we can start iterator. */
 558         iam_it_init(it, bag, 0, ipd);
 559
 560         rc = iam_it_get(it, (struct iam_key *)key);
 561         if (rc > 0)
 562                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 563         iam_it_put(it);
 564         iam_it_fini(it);
 565         osd_ipd_put(oti->oti_env, bag, ipd);
 566
 567         LINVRNT(osd_invariant(obj));
 568
 569         RETURN(rc);
 570 }
 571
 572 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 573                   const struct lu_fid *fid, enum oi_check_flags flags)
 574 {
 575         struct lu_seq_range *range = &info->oti_seq_range;
 576         int rc;
 577         ENTRY;
 578
 579         if (flags & OI_KNOWN_ON_OST)
 580                 RETURN(1);
 581
 582         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 583                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 584                      fid_is_quota(fid))
 585                 RETURN(0);
 586
 587         if (fid_is_idif(fid) || fid_is_last_id(fid))
 588                 RETURN(1);
 589
 590         if (!(flags & OI_CHECK_FLD))
 591                 RETURN(0);
 592
 593         if (osd_seq_site(osd)->ss_server_fld == NULL)
 594                 RETURN(0);
 595
 596         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 597         if (rc != 0) {
 598                 /* During upgrade, OST FLDB might not be loaded because
 599                  * OST FLDB is not created until 2.6, so if some DNE
 600                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 601                  * not be able to find the sequence from local FLDB
 602                  * cache see fld_index_init().
 603                  */
 604                 if (rc == -ENOENT && osd->od_is_ost)
 605                         RETURN(1);
 606
 607                 if (rc != -ENOENT)
 608                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 609                                osd_name(osd), PFID(fid), rc);
 610                 RETURN(0);
 611         }
 612
 613         if (fld_range_is_ost(range))
 614                 RETURN(1);
 615
 616         RETURN(0);
 617 }
 618
 619 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 620                            const struct lu_fid *fid, struct osd_inode_id *id)
 621 {
 622         struct lu_fid *oi_fid = &info->oti_fid2;
 623         int rc;
 624
 625         fid_cpu_to_be(oi_fid, fid);
 626         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 627                                (const struct dt_key *)oi_fid);
 628         if (rc > 0) {
 629                 osd_id_unpack(id, id);
 630                 rc = 0;
 631         } else if (rc == 0) {
 632                 rc = -ENOENT;
 633         }
 634         return rc;
 635 }
 636
 637 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 638                   const struct lu_fid *fid, struct osd_inode_id *id,
 639                   enum oi_check_flags flags)
 640 {
 641         if (unlikely(fid_is_last_id(fid)))
 642                 return osd_obj_spec_lookup(info, osd, fid, id, flags);
 643
 644         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 645                 return osd_obj_map_lookup(info, osd, fid, id);
 646
 647         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 648                 int rc;
 649
 650                 if (fid_is_fs_root(fid)) {
 651                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 652                                    osd_sb(osd)->s_root->d_inode->i_generation);
 653                         return 0;
 654                 }
 655                 if (unlikely(fid_is_acct(fid)))
 656                         return osd_acct_obj_lookup(info, osd, fid, id);
 657
 658                 /* For other special FIDs, try OI first, then do spec lookup */
 659                 rc = __osd_oi_lookup(info, osd, fid, id);
 660                 if (rc == -ENOENT)
 661                         return osd_obj_spec_lookup(info, osd, fid, id, flags);
 662                 return rc;
 663         }
 664
 665         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 666                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 667                 return 0;
 668         }
 669
 670         return __osd_oi_lookup(info, osd, fid, id);
 671 }
 672
 673 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 674                              const struct dt_rec *rec, const struct dt_key *key,
 675                              handle_t *th, bool insert)
 676 {
 677         struct iam_container *bag;
 678         struct iam_path_descr *ipd;
 679         int rc;
 680         ENTRY;
 681
 682         LASSERT(oi);
 683         LASSERT(oi->oi_inode);
 684         dquot_initialize(oi->oi_inode);
 685
 686         bag = &oi->oi_dir.od_container;
 687         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 688         if (unlikely(ipd == NULL))
 689                 RETURN(-ENOMEM);
 690
 691         LASSERT(th != NULL);
 692         LASSERT(th->h_transaction != NULL);
 693         if (insert)
 694                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 695                                 (const struct iam_rec *)rec, ipd);
 696         else
 697                 rc = iam_update(th, bag, (const struct iam_key *)key,
 698                                 (const struct iam_rec *)rec, ipd);
 699         osd_ipd_put(oti->oti_env, bag, ipd);
 700         LINVRNT(osd_invariant(obj));
 701         RETURN(rc);
 702 }
 703
 704 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 705                   const struct lu_fid *fid, const struct osd_inode_id *id,
 706                   handle_t *th, enum oi_check_flags flags, bool *exist)
 707 {
 708         struct lu_fid       *oi_fid = &info->oti_fid2;
 709         struct osd_inode_id *oi_id  = &info->oti_id2;
 710         int                  rc     = 0;
 711
 712         if (unlikely(fid_is_last_id(fid)))
 713                 return osd_obj_spec_insert(info, osd, fid, id, th);
 714
 715         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 716                 return osd_obj_map_insert(info, osd, fid, id, th);
 717
 718         fid_cpu_to_be(oi_fid, fid);
 719         osd_id_pack(oi_id, id);
 720         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 721                                (const struct dt_rec *)oi_id,
 722                                (const struct dt_key *)oi_fid, th, true);
 723         if (rc != 0) {
 724                 struct inode *inode;
 725                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 726
 727                 if (rc != -EEXIST)
 728                         return rc;
 729
 730                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 731                 if (rc != 0)
 732                         return rc;
 733
 734                 if (unlikely(osd_id_eq(id, oi_id)))
 735                         return 1;
 736
 737                 /* Check whether the mapping for oi_id is valid or not. */
 738                 inode = osd_iget(info, osd, oi_id);
 739                 if (IS_ERR(inode)) {
 740                         rc = PTR_ERR(inode);
 741                         if (rc == -ENOENT || rc == -ESTALE)
 742                                 goto update;
 743                         return rc;
 744                 }
 745
 746                 /* The EA inode should NOT be in OI, old OI scrub may added
 747                  * such OI mapping by wrong, replace it.
 748                  */
 749                 if (unlikely(osd_is_ea_inode(inode))) {
 750                         iput(inode);
 751                         goto update;
 752                 }
 753
 754                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 755                                  &info->oti_ost_attrs);
 756                 iput(inode);
 757                 if (rc == -ENODATA)
 758                         goto update;
 759
 760                 if (rc != 0)
 761                         return rc;
 762
 763                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 764                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 765                         CERROR(
 766                                "%s: the FID "DFID" is used by two objects: %u/%u %u/%u\n",
 767                                osd_dev2name(osd), PFID(fid), oi_id->oii_ino,
 768                                oi_id->oii_gen, id->oii_ino, id->oii_gen);
 769                         return -EEXIST;
 770                 }
 771
 772 update:
 773                 osd_id_pack(oi_id, id);
 774                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 775                                         (const struct dt_rec *)oi_id,
 776                                         (const struct dt_key *)oi_fid, th,
 777                                         false);
 778                 if (rc != 0)
 779                         return rc;
 780
 781                 if (exist != NULL)
 782                         *exist = true;
 783         }
 784
 785         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 786                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 787         return rc;
 788 }
 789
 790 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 791                              const struct dt_key *key, handle_t *th)
 792 {
 793         struct iam_container    *bag;
 794         struct iam_path_descr   *ipd;
 795         int                      rc;
 796         ENTRY;
 797
 798         LASSERT(oi);
 799         LASSERT(oi->oi_inode);
 800         dquot_initialize(oi->oi_inode);
 801
 802         bag = &oi->oi_dir.od_container;
 803         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 804         if (unlikely(ipd == NULL))
 805                 RETURN(-ENOMEM);
 806
 807         LASSERT(th != NULL);
 808         LASSERT(th->h_transaction != NULL);
 809
 810         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 811         osd_ipd_put(oti->oti_env, bag, ipd);
 812         LINVRNT(osd_invariant(obj));
 813         RETURN(rc);
 814 }
 815
 816 int osd_oi_delete(struct osd_thread_info *info,
 817                   struct osd_device *osd, const struct lu_fid *fid,
 818                   handle_t *th, enum oi_check_flags flags)
 819 {
 820         struct lu_fid *oi_fid = &info->oti_fid2;
 821
 822         /* clear idmap cache */
 823         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 824                 fid_zero(&info->oti_cache.oic_fid);
 825
 826         if (fid_is_last_id(fid))
 827                 return 0;
 828
 829         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 830                 return osd_obj_map_delete(info, osd, fid, th);
 831
 832         fid_cpu_to_be(oi_fid, fid);
 833         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 834                                  (const struct dt_key *)oi_fid, th);
 835 }
 836
 837 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 838                   const struct lu_fid *fid, const struct osd_inode_id *id,
 839                   handle_t *th, enum oi_check_flags flags)
 840 {
 841         struct lu_fid       *oi_fid = &info->oti_fid2;
 842         struct osd_inode_id *oi_id  = &info->oti_id2;
 843         int                  rc     = 0;
 844
 845         if (unlikely(fid_is_last_id(fid)))
 846                 return osd_obj_spec_update(info, osd, fid, id, th);
 847
 848         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 849                 return osd_obj_map_update(info, osd, fid, id, th);
 850
 851         fid_cpu_to_be(oi_fid, fid);
 852         osd_id_pack(oi_id, id);
 853         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 854                                (const struct dt_rec *)oi_id,
 855                                (const struct dt_key *)oi_fid, th, false);
 856         if (rc != 0)
 857                 return rc;
 858
 859         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 860                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 861         return rc;
 862 }
 863
 864 int osd_oi_mod_init(void)
 865 {
 866         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 867                 osd_oi_count = OSD_OI_FID_NR;
 868
 869         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 870                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 871                               osd_oi_count, size_roundup_power2(osd_oi_count));
 872                 osd_oi_count = size_roundup_power2(osd_oi_count);
 873         }
 874
 875         return 0;
 876 }