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         if (IS_ERR(inode)) {
 131                 ldiskfs_journal_stop(jh);
 132                 return PTR_ERR(inode);
 133         }
 134
 135         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 136         unlock_new_inode(inode);
 137
 138         if (feat->dif_flags & DT_IND_VARKEY)
 139                 rc = iam_lvar_create(inode, feat->dif_keysize_max,
 140                                      feat->dif_ptrsize, feat->dif_recsize_max,
 141                                      jh);
 142         else
 143                 rc = iam_lfix_create(inode, feat->dif_keysize_max,
 144                                      feat->dif_ptrsize, feat->dif_recsize_max,
 145                                      jh);
 146         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 147         rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
 148         ldiskfs_journal_stop(jh);
 149         iput(inode);
 150         return rc;
 151 }
 152
 153 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 154                                        struct osd_device *osd,
 155                                        const char *name,
 156                                        struct dt_index_features *f,
 157                                        bool create)
 158 {
 159         struct dentry *dentry;
 160         struct inode  *inode;
 161         int            rc;
 162
 163         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 164         if (IS_ERR(dentry))
 165                 return (void *) dentry;
 166
 167         if (dentry->d_inode) {
 168                 LASSERT(!is_bad_inode(dentry->d_inode));
 169                 inode = dentry->d_inode;
 170                 atomic_inc(&inode->i_count);
 171                 dput(dentry);
 172                 return inode;
 173         }
 174
 175         /* create */
 176         dput(dentry);
 177         shrink_dcache_parent(osd_sb(osd)->s_root);
 178         if (!create)
 179                 return ERR_PTR(-ENOENT);
 180
 181         rc = osd_oi_index_create_one(info, osd, name, f);
 182         if (rc)
 183                 return ERR_PTR(rc);
 184
 185         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 186         if (IS_ERR(dentry))
 187                 return (void *) dentry;
 188
 189         if (dentry->d_inode) {
 190                 LASSERT(!is_bad_inode(dentry->d_inode));
 191                 inode = dentry->d_inode;
 192                 atomic_inc(&inode->i_count);
 193                 dput(dentry);
 194                 return inode;
 195         }
 196
 197         return ERR_PTR(-ENOENT);
 198 }
 199
 200 /**
 201  * Open an OI(Ojbect Index) container.
 202  *
 203  * \param       name    Name of OI container
 204  * \param       objp    Pointer of returned OI
 205  *
 206  * \retval      0       success
 207  * \retval      -ve     failure
 208  */
 209 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
 210                        char *name, struct osd_oi **oi_slot, bool create)
 211 {
 212         struct osd_directory *dir;
 213         struct iam_container *bag;
 214         struct inode         *inode;
 215         struct osd_oi        *oi;
 216         int                   rc;
 217
 218         ENTRY;
 219
 220         oi_feat.dif_keysize_min = sizeof(struct lu_fid);
 221         oi_feat.dif_keysize_max = sizeof(struct lu_fid);
 222
 223         inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
 224         if (IS_ERR(inode))
 225                 RETURN(PTR_ERR(inode));
 226
 227         if (!osd->od_dt_dev.dd_rdonly) {
 228                 /* 'What the @fid is' is not imporatant, because these objects
 229                  * have no OI mappings, and only are visible inside the OSD.*/
 230                 lu_igif_build(&info->oti_fid, inode->i_ino,
 231                               inode->i_generation);
 232                 rc = osd_ea_fid_set(info, inode, &info->oti_fid,
 233                                     LMAC_NOT_IN_OI, 0);
 234                 if (rc)
 235                         GOTO(out_inode, rc);
 236         }
 237
 238         OBD_ALLOC_PTR(oi);
 239         if (oi == NULL)
 240                 GOTO(out_inode, rc = -ENOMEM);
 241
 242         oi->oi_inode = inode;
 243         dir = &oi->oi_dir;
 244
 245         bag = &dir->od_container;
 246         rc = iam_container_init(bag, &dir->od_descr, inode);
 247         if (rc < 0)
 248                 GOTO(out_free, rc);
 249
 250         rc = iam_container_setup(bag);
 251         if (rc < 0)
 252                 GOTO(out_container, rc);
 253
 254         *oi_slot = oi;
 255         RETURN(0);
 256
 257 out_container:
 258         iam_container_fini(bag);
 259 out_free:
 260         OBD_FREE_PTR(oi);
 261 out_inode:
 262         iput(inode);
 263         return rc;
 264 }
 265
 266 /**
 267  * Open OI(Object Index) table.
 268  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 269  * will be, this function can either return 0 for new filesystem, or number
 270  * of OIs on existed filesystem.
 271  *
 272  * If \a oi_count is non-zero, which means caller does know number of OIs on
 273  * filesystem, this function should return the exactly same number on
 274  * success, or error code in failure.
 275  *
 276  * \param     oi_count  Number of expected OI containers
 277  * \param     create    Create OIs if doesn't exist
 278  *
 279  * \retval    +ve       number of opened OI containers
 280  * \retval      0       no OI containers found
 281  * \retval    -ve       failure
 282  */
 283 static int
 284 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 285                   struct osd_oi **oi_table, unsigned oi_count, bool create)
 286 {
 287         struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
 288         int                count = 0;
 289         int                rc = 0;
 290         int                i;
 291         ENTRY;
 292
 293         /* NB: oi_count != 0 means that we have already created/known all OIs
 294          * and have known exact number of OIs. */
 295         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 296
 297         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 298                 char name[12];
 299
 300                 if (oi_table[i] != NULL) {
 301                         count++;
 302                         continue;
 303                 }
 304
 305                 sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
 306                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 307                 if (rc == 0) {
 308                         count++;
 309                         continue;
 310                 }
 311
 312                 if (rc == -ENOENT && create == false) {
 313                         if (oi_count == 0)
 314                                 return count;
 315
 316                         rc = 0;
 317                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 318                         continue;
 319                 }
 320
 321                 CERROR("%s: can't open %s: rc = %d\n",
 322                        osd_dev2name(osd), name, rc);
 323                 if (oi_count > 0)
 324                         CERROR("%s: expect to open total %d OI files.\n",
 325                                osd_dev2name(osd), oi_count);
 326                 break;
 327         }
 328
 329         if (rc < 0) {
 330                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 331                 count = rc;
 332         }
 333
 334         RETURN(count);
 335 }
 336
 337 static int osd_remove_oi_one(struct dentry *parent, const char *name,
 338                              int namelen)
 339 {
 340         struct dentry *child;
 341         int rc;
 342
 343         child = ll_lookup_one_len(name, parent, namelen);
 344         if (IS_ERR(child)) {
 345                 rc = PTR_ERR(child);
 346         } else {
 347                 rc = ll_vfs_unlink(parent->d_inode, child);
 348                 dput(child);
 349         }
 350
 351         return rc == -ENOENT ? 0 : rc;
 352 }
 353
 354 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 355 {
 356         char name[16];
 357         int namelen;
 358         int rc;
 359         int i;
 360
 361         if (osd->od_dt_dev.dd_rdonly)
 362                 RETURN(-EROFS);
 363
 364         for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
 365                 namelen = snprintf(name, sizeof(name), "%s.%d",
 366                                    OSD_OI_NAME_BASE, i);
 367                 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
 368                 if (rc != 0) {
 369                         CERROR("%s: fail to remove the stale OI file %s: "
 370                                "rc = %d\n", osd_dev2name(osd), name, rc);
 371                         return rc;
 372                 }
 373         }
 374
 375         namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
 376         rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
 377         if (rc != 0)
 378                 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
 379                        osd_dev2name(osd), name, rc);
 380
 381         return rc;
 382 }
 383
 384 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
 385                 bool restored)
 386 {
 387         struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
 388         struct scrub_file *sf = &scrub->os_file;
 389         struct osd_oi **oi;
 390         int count;
 391         int rc;
 392         ENTRY;
 393
 394         if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
 395                 LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
 396                               osd_dev2name(osd), sf->sf_oi_count);
 397                 sf->sf_oi_count = 0;
 398         }
 399
 400         if (restored) {
 401                 rc = osd_remove_ois(info, osd);
 402                 if (rc)
 403                         RETURN(rc);
 404         }
 405
 406         OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 407         if (oi == NULL)
 408                 RETURN(-ENOMEM);
 409
 410         /* try to open existing multiple OIs first */
 411         count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 412         if (count < 0)
 413                 GOTO(out, rc = count);
 414
 415         if (count > 0) {
 416                 if (count == sf->sf_oi_count)
 417                         GOTO(out, rc = count);
 418
 419                 if (sf->sf_oi_count == 0) {
 420                         if (likely((count & (count - 1)) == 0))
 421                                 GOTO(out, rc = count);
 422
 423                         LCONSOLE_WARN("%s: invalid oi count %d, remove them, "
 424                                       "then set it to %d\n", osd_dev2name(osd),
 425                                       count, osd_oi_count);
 426                         osd_oi_table_put(info, oi, count);
 427                         rc = osd_remove_ois(info, osd);
 428                         if (rc)
 429                                 GOTO(out, rc);
 430
 431                         sf->sf_oi_count = osd_oi_count;
 432                 }
 433
 434                 scrub_file_reset(scrub, LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 435                                  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,
 457                                          LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 458                                          SF_UPGRADE);
 459                 GOTO(out, rc = 1);
 460         } else if (rc != -ENOENT) {
 461                 CERROR("%s: can't open %s: rc = %d\n",
 462                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 463                 GOTO(out, rc);
 464         }
 465
 466         if (sf->sf_oi_count > 0) {
 467                 int i;
 468
 469                 count = sf->sf_oi_count;
 470                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 471                 for (i = 0; i < count; i++)
 472                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 473                 scrub_file_reset(scrub, LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 474                                  SF_RECREATED);
 475         } else {
 476                 count = sf->sf_oi_count = osd_oi_count;
 477         }
 478
 479 create:
 480         rc = scrub_file_store(info->oti_env, scrub);
 481         if (rc < 0) {
 482                 osd_oi_table_put(info, oi, count);
 483                 GOTO(out, rc);
 484         }
 485
 486         /* No OIs exist, new filesystem, create OI objects */
 487         rc = osd_oi_table_open(info, osd, oi, count, true);
 488         LASSERT(ergo(rc >= 0, rc == count));
 489
 490         GOTO(out, rc);
 491
 492 out:
 493         if (rc < 0) {
 494                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 495         } else {
 496                 LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
 497
 498                 osd->od_oi_table = oi;
 499                 osd->od_oi_count = rc;
 500                 if (sf->sf_oi_count != rc) {
 501                         sf->sf_oi_count = rc;
 502                         rc = scrub_file_store(info->oti_env, scrub);
 503                         if (rc < 0) {
 504                                 osd_oi_table_put(info, oi, count);
 505                                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 506                         }
 507                 } else {
 508                         rc = 0;
 509                 }
 510         }
 511
 512         return rc;
 513 }
 514
 515 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 516 {
 517         if (unlikely(!osd->od_oi_table))
 518                 return;
 519
 520         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 521
 522         OBD_FREE(osd->od_oi_table,
 523                  sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
 524         osd->od_oi_table = NULL;
 525 }
 526
 527 static inline int fid_is_fs_root(const struct lu_fid *fid)
 528 {
 529         /* Map root inode to special local object FID */
 530         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 531                          fid_oid(fid) == OSD_FS_ROOT_OID));
 532 }
 533
 534 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 535                              struct osd_oi *oi, struct dt_rec *rec,
 536                              const struct dt_key *key)
 537 {
 538         struct iam_container  *bag;
 539         struct iam_iterator   *it = &oti->oti_idx_it;
 540         struct iam_path_descr *ipd;
 541         int                    rc;
 542         ENTRY;
 543
 544         LASSERT(oi);
 545         LASSERT(oi->oi_inode);
 546
 547         bag = &oi->oi_dir.od_container;
 548         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 549         if (IS_ERR(ipd))
 550                 RETURN(-ENOMEM);
 551
 552         /* got ipd now we can start iterator. */
 553         iam_it_init(it, bag, 0, ipd);
 554
 555         rc = iam_it_get(it, (struct iam_key *)key);
 556         if (rc > 0)
 557                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 558         iam_it_put(it);
 559         iam_it_fini(it);
 560         osd_ipd_put(oti->oti_env, bag, ipd);
 561
 562         LINVRNT(osd_invariant(obj));
 563
 564         RETURN(rc);
 565 }
 566
 567 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 568                   const struct lu_fid *fid, enum oi_check_flags flags)
 569 {
 570         struct lu_seq_range     *range = &info->oti_seq_range;
 571         int                     rc;
 572         ENTRY;
 573
 574         if (flags & OI_KNOWN_ON_OST)
 575                 RETURN(1);
 576
 577         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 578                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 579                      fid_is_quota(fid))
 580                 RETURN(0);
 581
 582         if (fid_is_idif(fid) || fid_is_last_id(fid))
 583                 RETURN(1);
 584
 585         if (!(flags & OI_CHECK_FLD))
 586                 RETURN(0);
 587
 588         if (osd_seq_site(osd)->ss_server_fld == NULL)
 589                 RETURN(0);
 590
 591         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 592         if (rc != 0) {
 593                 /* During upgrade, OST FLDB might not be loaded because
 594                  * OST FLDB is not created until 2.6, so if some DNE
 595                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 596                  * not be able to find the sequence from local FLDB
 597                  * cache see fld_index_init(). */
 598                 if (rc == -ENOENT && osd->od_is_ost)
 599                         RETURN(1);
 600
 601                 if (rc != -ENOENT)
 602                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 603                                osd_name(osd), PFID(fid), rc);
 604                 RETURN(0);
 605         }
 606
 607         if (fld_range_is_ost(range))
 608                 RETURN(1);
 609
 610         RETURN(0);
 611 }
 612
 613 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 614                            const struct lu_fid *fid, struct osd_inode_id *id)
 615 {
 616         struct lu_fid *oi_fid = &info->oti_fid2;
 617         int            rc;
 618
 619         fid_cpu_to_be(oi_fid, fid);
 620         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 621                                (const struct dt_key *)oi_fid);
 622         if (rc > 0) {
 623                 osd_id_unpack(id, id);
 624                 rc = 0;
 625         } else if (rc == 0) {
 626                 rc = -ENOENT;
 627         }
 628         return rc;
 629 }
 630
 631 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 632                   const struct lu_fid *fid, struct osd_inode_id *id,
 633                   enum oi_check_flags flags)
 634 {
 635         if (unlikely(fid_is_last_id(fid)))
 636                 return osd_obj_spec_lookup(info, osd, fid, id);
 637
 638         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 639                 return osd_obj_map_lookup(info, osd, fid, id);
 640
 641
 642         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 643                 int rc;
 644                 if (fid_is_fs_root(fid)) {
 645                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 646                                    osd_sb(osd)->s_root->d_inode->i_generation);
 647                         return 0;
 648                 }
 649                 if (unlikely(fid_is_acct(fid)))
 650                         return osd_acct_obj_lookup(info, osd, fid, id);
 651
 652                 /* For other special FIDs, try OI first, then do spec lookup */
 653                 rc = __osd_oi_lookup(info, osd, fid, id);
 654                 if (rc == -ENOENT)
 655                         return osd_obj_spec_lookup(info, osd, fid, id);
 656                 return rc;
 657         }
 658
 659         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 660                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 661                 return 0;
 662         }
 663
 664         return __osd_oi_lookup(info, osd, fid, id);
 665 }
 666
 667 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 668                              const struct dt_rec *rec, const struct dt_key *key,
 669                              handle_t *th, bool insert)
 670 {
 671         struct iam_container    *bag;
 672         struct iam_path_descr   *ipd;
 673         int                     rc;
 674         ENTRY;
 675
 676         LASSERT(oi);
 677         LASSERT(oi->oi_inode);
 678         ll_vfs_dq_init(oi->oi_inode);
 679
 680         bag = &oi->oi_dir.od_container;
 681         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 682         if (unlikely(ipd == NULL))
 683                 RETURN(-ENOMEM);
 684
 685         LASSERT(th != NULL);
 686         LASSERT(th->h_transaction != NULL);
 687         if (insert)
 688                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 689                                 (const struct iam_rec *)rec, ipd);
 690         else
 691                 rc = iam_update(th, bag, (const struct iam_key *)key,
 692                                 (const struct iam_rec *)rec, ipd);
 693         osd_ipd_put(oti->oti_env, bag, ipd);
 694         LINVRNT(osd_invariant(obj));
 695         RETURN(rc);
 696 }
 697
 698 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 699                   const struct lu_fid *fid, const struct osd_inode_id *id,
 700                   handle_t *th, enum oi_check_flags flags, bool *exist)
 701 {
 702         struct lu_fid       *oi_fid = &info->oti_fid2;
 703         struct osd_inode_id *oi_id  = &info->oti_id2;
 704         int                  rc     = 0;
 705
 706         if (unlikely(fid_is_last_id(fid)))
 707                 return osd_obj_spec_insert(info, osd, fid, id, th);
 708
 709         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 710                 return osd_obj_map_insert(info, osd, fid, id, th);
 711
 712         fid_cpu_to_be(oi_fid, fid);
 713         osd_id_pack(oi_id, id);
 714         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 715                                (const struct dt_rec *)oi_id,
 716                                (const struct dt_key *)oi_fid, th, true);
 717         if (rc != 0) {
 718                 struct inode *inode;
 719                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 720
 721                 if (rc != -EEXIST)
 722                         return rc;
 723
 724                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 725                 if (rc != 0)
 726                         return rc;
 727
 728                 if (unlikely(osd_id_eq(id, oi_id)))
 729                         return 1;
 730
 731                 /* Check whether the mapping for oi_id is valid or not. */
 732                 inode = osd_iget(info, osd, oi_id);
 733                 if (IS_ERR(inode)) {
 734                         rc = PTR_ERR(inode);
 735                         if (rc == -ENOENT || rc == -ESTALE)
 736                                 goto update;
 737                         return rc;
 738                 }
 739
 740                 /* The EA inode should NOT be in OI, old OI scrub may added
 741                  * such OI mapping by wrong, replace it. */
 742                 if (unlikely(osd_is_ea_inode(inode))) {
 743                         iput(inode);
 744                         goto update;
 745                 }
 746
 747                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 748                                  &info->oti_ost_attrs);
 749                 iput(inode);
 750                 if (rc == -ENODATA)
 751                         goto update;
 752
 753                 if (rc != 0)
 754                         return rc;
 755
 756                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 757                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 758                         CERROR("%s: the FID "DFID" is used by two objects: "
 759                                "%u/%u %u/%u\n", osd_dev2name(osd),
 760                                PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
 761                                id->oii_ino, id->oii_gen);
 762                         return -EEXIST;
 763                 }
 764
 765 update:
 766                 osd_id_pack(oi_id, id);
 767                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 768                                         (const struct dt_rec *)oi_id,
 769                                         (const struct dt_key *)oi_fid, th, false);
 770                 if (rc != 0)
 771                         return rc;
 772
 773                 if (exist != NULL)
 774                         *exist = true;
 775         }
 776
 777         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 778                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 779         return rc;
 780 }
 781
 782 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 783                              const struct dt_key *key, handle_t *th)
 784 {
 785         struct iam_container    *bag;
 786         struct iam_path_descr   *ipd;
 787         int                      rc;
 788         ENTRY;
 789
 790         LASSERT(oi);
 791         LASSERT(oi->oi_inode);
 792         ll_vfs_dq_init(oi->oi_inode);
 793
 794         bag = &oi->oi_dir.od_container;
 795         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 796         if (unlikely(ipd == NULL))
 797                 RETURN(-ENOMEM);
 798
 799         LASSERT(th != NULL);
 800         LASSERT(th->h_transaction != NULL);
 801
 802         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 803         osd_ipd_put(oti->oti_env, bag, ipd);
 804         LINVRNT(osd_invariant(obj));
 805         RETURN(rc);
 806 }
 807
 808 int osd_oi_delete(struct osd_thread_info *info,
 809                   struct osd_device *osd, const struct lu_fid *fid,
 810                   handle_t *th, enum oi_check_flags flags)
 811 {
 812         struct lu_fid *oi_fid = &info->oti_fid2;
 813
 814         /* clear idmap cache */
 815         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 816                 fid_zero(&info->oti_cache.oic_fid);
 817
 818         if (fid_is_last_id(fid))
 819                 return 0;
 820
 821         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 822                 return osd_obj_map_delete(info, osd, fid, th);
 823
 824         fid_cpu_to_be(oi_fid, fid);
 825         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 826                                  (const struct dt_key *)oi_fid, th);
 827 }
 828
 829 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 830                   const struct lu_fid *fid, const struct osd_inode_id *id,
 831                   handle_t *th, enum oi_check_flags flags)
 832 {
 833         struct lu_fid       *oi_fid = &info->oti_fid2;
 834         struct osd_inode_id *oi_id  = &info->oti_id2;
 835         int                  rc     = 0;
 836
 837         if (unlikely(fid_is_last_id(fid)))
 838                 return osd_obj_spec_update(info, osd, fid, id, th);
 839
 840         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 841                 return osd_obj_map_update(info, osd, fid, id, th);
 842
 843         fid_cpu_to_be(oi_fid, fid);
 844         osd_id_pack(oi_id, id);
 845         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 846                                (const struct dt_rec *)oi_id,
 847                                (const struct dt_key *)oi_fid, th, false);
 848         if (rc != 0)
 849                 return rc;
 850
 851         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 852                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 853         return rc;
 854 }
 855
 856 int osd_oi_mod_init(void)
 857 {
 858         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 859                 osd_oi_count = OSD_OI_FID_NR;
 860
 861         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 862                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 863                               osd_oi_count, size_roundup_power2(osd_oi_count));
 864                 osd_oi_count = size_roundup_power2(osd_oi_count);
 865         }
 866
 867         return 0;
 868 }