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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_oi.c
  37  *
  38  * Object Index.
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_OSD
  44
  45 #include <linux/module.h>
  46
  47 /* LUSTRE_VERSION_CODE */
  48 #include <lustre_ver.h>
  49 /*
  50  * struct OBD_{ALLOC,FREE}*()
  51  * OBD_FAIL_CHECK
  52  */
  53 #include <obd.h>
  54 #include <obd_support.h>
  55
  56 /* fid_cpu_to_be() */
  57 #include <lustre_fid.h>
  58 #include <dt_object.h>
  59
  60 #include "osd_oi.h"
  61 /* osd_lookup(), struct osd_thread_info */
  62 #include "osd_internal.h"
  63 #include "osd_scrub.h"
  64
  65 static unsigned int osd_oi_count = OSD_OI_FID_NR;
  66 CFS_MODULE_PARM(osd_oi_count, "i", int, 0444,
  67                 "Number of Object Index containers to be created, "
  68                 "it's only valid for new filesystem.");
  69
  70 /** to serialize concurrent OI index initialization */
  71 static struct mutex oi_init_lock;
  72
  73 static struct dt_index_features oi_feat = {
  74         .dif_flags       = DT_IND_UPDATE,
  75         .dif_recsize_min = sizeof(struct osd_inode_id),
  76         .dif_recsize_max = sizeof(struct osd_inode_id),
  77         .dif_ptrsize     = 4
  78 };
  79
  80 #define OSD_OI_NAME_BASE        "oi.16"
  81
  82 static void osd_oi_table_put(struct osd_thread_info *info,
  83                              struct osd_oi **oi_table, unsigned oi_count)
  84 {
  85         struct iam_container *bag;
  86         int                   i;
  87
  88         for (i = 0; i < oi_count; i++) {
  89                 if (oi_table[i] == NULL)
  90                         continue;
  91
  92                 LASSERT(oi_table[i]->oi_inode != NULL);
  93
  94                 bag = &(oi_table[i]->oi_dir.od_container);
  95                 if (bag->ic_object == oi_table[i]->oi_inode)
  96                         iam_container_fini(bag);
  97                 iput(oi_table[i]->oi_inode);
  98                 oi_table[i]->oi_inode = NULL;
  99                 OBD_FREE_PTR(oi_table[i]);
 100                 oi_table[i] = NULL;
 101         }
 102 }
 103
 104 static int osd_oi_index_create_one(struct osd_thread_info *info,
 105                                    struct osd_device *osd, const char *name,
 106                                    struct dt_index_features *feat)
 107 {
 108         const struct lu_env             *env = info->oti_env;
 109         struct osd_inode_id             *id  = &info->oti_id;
 110         struct buffer_head              *bh;
 111         struct inode                    *inode;
 112         struct ldiskfs_dir_entry_2      *de;
 113         struct dentry                   *dentry;
 114         struct super_block              *sb  = osd_sb(osd);
 115         struct inode                    *dir = sb->s_root->d_inode;
 116         handle_t                        *jh;
 117         int                              rc;
 118
 119         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 120         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
 121         if (bh) {
 122                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 123                 brelse(bh);
 124                 inode = osd_iget(info, osd, id);
 125                 if (!IS_ERR(inode)) {
 126                         iput(inode);
 127                         inode = ERR_PTR(-EEXIST);
 128                 }
 129                 return PTR_ERR(inode);
 130         }
 131
 132         jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
 133         if (IS_ERR(jh))
 134                 return PTR_ERR(jh);
 135
 136         inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR));
 137         if (IS_ERR(inode)) {
 138                 ldiskfs_journal_stop(jh);
 139                 return PTR_ERR(inode);
 140         }
 141
 142         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 143         unlock_new_inode(inode);
 144
 145         if (feat->dif_flags & DT_IND_VARKEY)
 146                 rc = iam_lvar_create(inode, feat->dif_keysize_max,
 147                                      feat->dif_ptrsize, feat->dif_recsize_max,
 148                                      jh);
 149         else
 150                 rc = iam_lfix_create(inode, feat->dif_keysize_max,
 151                                      feat->dif_ptrsize, feat->dif_recsize_max,
 152                                      jh);
 153         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 154         rc = osd_ldiskfs_add_entry(jh, dentry, inode, NULL);
 155         ldiskfs_journal_stop(jh);
 156         iput(inode);
 157         return rc;
 158 }
 159
 160 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 161                                        struct osd_device *osd,
 162                                        const char *name,
 163                                        struct dt_index_features *f,
 164                                        bool create)
 165 {
 166         struct dentry *dentry;
 167         struct inode  *inode;
 168         int            rc;
 169
 170         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 171         if (IS_ERR(dentry))
 172                 return (void *) dentry;
 173
 174         if (dentry->d_inode) {
 175                 LASSERT(!is_bad_inode(dentry->d_inode));
 176                 inode = dentry->d_inode;
 177                 atomic_inc(&inode->i_count);
 178                 dput(dentry);
 179                 return inode;
 180         }
 181
 182         /* create */
 183         dput(dentry);
 184         shrink_dcache_parent(osd_sb(osd)->s_root);
 185         if (!create)
 186                 return ERR_PTR(-ENOENT);
 187
 188         rc = osd_oi_index_create_one(info, osd, name, f);
 189         if (rc)
 190                 return ERR_PTR(rc);
 191
 192         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 193         if (IS_ERR(dentry))
 194                 return (void *) dentry;
 195
 196         if (dentry->d_inode) {
 197                 LASSERT(!is_bad_inode(dentry->d_inode));
 198                 inode = dentry->d_inode;
 199                 atomic_inc(&inode->i_count);
 200                 dput(dentry);
 201                 return inode;
 202         }
 203
 204         return ERR_PTR(-ENOENT);
 205 }
 206
 207 /**
 208  * Open an OI(Ojbect Index) container.
 209  *
 210  * \param       name    Name of OI container
 211  * \param       objp    Pointer of returned OI
 212  *
 213  * \retval      0       success
 214  * \retval      -ve     failure
 215  */
 216 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
 217                        char *name, struct osd_oi **oi_slot, bool create)
 218 {
 219         struct osd_directory *dir;
 220         struct iam_container *bag;
 221         struct inode         *inode;
 222         struct osd_oi        *oi;
 223         int                   rc;
 224
 225         ENTRY;
 226
 227         oi_feat.dif_keysize_min = sizeof(struct lu_fid);
 228         oi_feat.dif_keysize_max = sizeof(struct lu_fid);
 229
 230         inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
 231         if (IS_ERR(inode))
 232                 RETURN(PTR_ERR(inode));
 233
 234         /* 'What the @fid is' is not imporatant, because these objects
 235          * have no OI mappings, and only are visible inside the OSD.*/
 236         lu_igif_build(&info->oti_fid, inode->i_ino, inode->i_generation);
 237         rc = osd_ea_fid_set(info, inode, &info->oti_fid, LMAC_NOT_IN_OI, 0);
 238         if (rc != 0)
 239                 GOTO(out_inode, rc);
 240
 241         OBD_ALLOC_PTR(oi);
 242         if (oi == NULL)
 243                 GOTO(out_inode, rc = -ENOMEM);
 244
 245         oi->oi_inode = inode;
 246         dir = &oi->oi_dir;
 247
 248         bag = &dir->od_container;
 249         rc = iam_container_init(bag, &dir->od_descr, inode);
 250         if (rc < 0)
 251                 GOTO(out_free, rc);
 252
 253         rc = iam_container_setup(bag);
 254         if (rc < 0)
 255                 GOTO(out_container, rc);
 256
 257         *oi_slot = oi;
 258         RETURN(0);
 259
 260 out_container:
 261         iam_container_fini(bag);
 262 out_free:
 263         OBD_FREE_PTR(oi);
 264 out_inode:
 265         iput(inode);
 266         return rc;
 267 }
 268
 269 /**
 270  * Open OI(Object Index) table.
 271  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 272  * will be, this function can either return 0 for new filesystem, or number
 273  * of OIs on existed filesystem.
 274  *
 275  * If \a oi_count is non-zero, which means caller does know number of OIs on
 276  * filesystem, this function should return the exactly same number on
 277  * success, or error code in failure.
 278  *
 279  * \param     oi_count  Number of expected OI containers
 280  * \param     create    Create OIs if doesn't exist
 281  *
 282  * \retval    +ve       number of opened OI containers
 283  * \retval      0       no OI containers found
 284  * \retval    -ve       failure
 285  */
 286 static int
 287 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 288                   struct osd_oi **oi_table, unsigned oi_count, bool create)
 289 {
 290         struct scrub_file *sf = &osd->od_scrub.os_file;
 291         int                count = 0;
 292         int                rc = 0;
 293         int                i;
 294         ENTRY;
 295
 296         /* NB: oi_count != 0 means that we have already created/known all OIs
 297          * and have known exact number of OIs. */
 298         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 299
 300         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 301                 char name[12];
 302
 303                 if (oi_table[i] != NULL) {
 304                         count++;
 305                         continue;
 306                 }
 307
 308                 sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
 309                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 310                 if (rc == 0) {
 311                         count++;
 312                         continue;
 313                 }
 314
 315                 if (rc == -ENOENT && create == false) {
 316                         if (oi_count == 0)
 317                                 return count;
 318
 319                         rc = 0;
 320                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 321                         continue;
 322                 }
 323
 324                 CERROR("%.16s: can't open %s: rc = %d\n",
 325                        LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, name, rc);
 326                 if (oi_count > 0)
 327                         CERROR("%.16s: expect to open total %d OI files.\n",
 328                                LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
 329                                oi_count);
 330                 break;
 331         }
 332
 333         if (rc < 0) {
 334                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 335                 count = rc;
 336         }
 337
 338         RETURN(count);
 339 }
 340
 341 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
 342 {
 343         struct osd_scrub  *scrub = &osd->od_scrub;
 344         struct scrub_file *sf = &scrub->os_file;
 345         struct osd_oi    **oi;
 346         int                rc;
 347         ENTRY;
 348
 349         OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 350         if (oi == NULL)
 351                 RETURN(-ENOMEM);
 352
 353         mutex_lock(&oi_init_lock);
 354         /* try to open existing multiple OIs first */
 355         rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 356         if (rc < 0)
 357                 GOTO(out, rc);
 358
 359         if (rc > 0) {
 360                 if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
 361                         GOTO(out, rc);
 362
 363                 osd_scrub_file_reset(scrub,
 364                                      LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 365                                      SF_RECREATED);
 366                 osd_oi_count = sf->sf_oi_count;
 367                 goto create;
 368         }
 369
 370         /* if previous failed then try found single OI from old filesystem */
 371         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 372         if (rc == 0) { /* found single OI from old filesystem */
 373                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 374                 if (sf->sf_success_count == 0)
 375                         /* XXX: There is one corner case that if the OI_scrub
 376                          *      file crashed or lost and we regard it upgrade,
 377                          *      then we allow IGIF lookup to bypass OI files.
 378                          *
 379                          *      The risk is that osd_fid_lookup() may found
 380                          *      a wrong inode with the given IGIF especially
 381                          *      when the MDT has performed file-level backup
 382                          *      and restored after former upgrading from 1.8
 383                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 384                          *      verify the inode to decrease the risk. */
 385                         osd_scrub_file_reset(scrub,
 386                                         LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 387                                         SF_UPGRADE);
 388                 GOTO(out, rc = 1);
 389         } else if (rc != -ENOENT) {
 390                 CERROR("%.16s: can't open %s: rc = %d\n",
 391                        LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
 392                        OSD_OI_NAME_BASE, rc);
 393                 GOTO(out, rc);
 394         }
 395
 396         if (sf->sf_oi_count > 0) {
 397                 int i;
 398
 399                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 400                 for (i = 0; i < osd_oi_count; i++)
 401                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 402                 osd_scrub_file_reset(scrub,
 403                                      LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
 404                                      SF_RECREATED);
 405         }
 406         sf->sf_oi_count = osd_oi_count;
 407
 408 create:
 409         rc = osd_scrub_file_store(scrub);
 410         if (rc < 0) {
 411                 osd_oi_table_put(info, oi, sf->sf_oi_count);
 412                 GOTO(out, rc);
 413         }
 414
 415         /* No OIs exist, new filesystem, create OI objects */
 416         rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
 417         LASSERT(ergo(rc >= 0, rc == osd_oi_count));
 418
 419         GOTO(out, rc);
 420
 421 out:
 422         if (rc < 0) {
 423                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 424         } else {
 425                 LASSERT((rc & (rc - 1)) == 0);
 426                 osd->od_oi_table = oi;
 427                 osd->od_oi_count = rc;
 428                 if (sf->sf_oi_count != rc) {
 429                         sf->sf_oi_count = rc;
 430                         rc = osd_scrub_file_store(scrub);
 431                         if (rc < 0) {
 432                                 osd_oi_table_put(info, oi, sf->sf_oi_count);
 433                                 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
 434                         }
 435                 } else {
 436                         rc = 0;
 437                 }
 438         }
 439
 440         mutex_unlock(&oi_init_lock);
 441         return rc;
 442 }
 443
 444 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 445 {
 446         if (unlikely(osd->od_oi_table == NULL))
 447                 return;
 448
 449         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 450
 451         OBD_FREE(osd->od_oi_table,
 452                  sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
 453         osd->od_oi_table = NULL;
 454 }
 455
 456 static inline int fid_is_fs_root(const struct lu_fid *fid)
 457 {
 458         /* Map root inode to special local object FID */
 459         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 460                          fid_oid(fid) == OSD_FS_ROOT_OID));
 461 }
 462
 463 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 464                              struct osd_oi *oi, struct dt_rec *rec,
 465                              const struct dt_key *key)
 466 {
 467         struct iam_container  *bag;
 468         struct iam_iterator   *it = &oti->oti_idx_it;
 469         struct iam_path_descr *ipd;
 470         int                    rc;
 471         ENTRY;
 472
 473         LASSERT(oi);
 474         LASSERT(oi->oi_inode);
 475
 476         bag = &oi->oi_dir.od_container;
 477         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 478         if (IS_ERR(ipd))
 479                 RETURN(-ENOMEM);
 480
 481         /* got ipd now we can start iterator. */
 482         iam_it_init(it, bag, 0, ipd);
 483
 484         rc = iam_it_get(it, (struct iam_key *)key);
 485         if (rc > 0)
 486                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 487         iam_it_put(it);
 488         iam_it_fini(it);
 489         osd_ipd_put(oti->oti_env, bag, ipd);
 490
 491         LINVRNT(osd_invariant(obj));
 492
 493         RETURN(rc);
 494 }
 495
 496 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 497                   const struct lu_fid *fid, enum oi_check_flags flags)
 498 {
 499         struct lu_seq_range     *range = &info->oti_seq_range;
 500         int                     rc;
 501         ENTRY;
 502
 503         if (flags & OI_KNOWN_ON_OST)
 504                 RETURN(1);
 505
 506         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 507                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 508                      fid_is_quota(fid))
 509                 RETURN(0);
 510
 511         if (fid_is_idif(fid) || fid_is_last_id(fid))
 512                 RETURN(1);
 513
 514         if (!(flags & OI_CHECK_FLD))
 515                 RETURN(0);
 516
 517         if (osd_seq_site(osd)->ss_server_fld == NULL)
 518                 RETURN(0);
 519
 520         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 521         if (rc != 0) {
 522                 if (rc != -ENOENT)
 523                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 524                                osd_name(osd), PFID(fid), rc);
 525                 RETURN(0);
 526         }
 527
 528         if (fld_range_is_ost(range))
 529                 RETURN(1);
 530
 531         RETURN(0);
 532 }
 533
 534 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 535                            const struct lu_fid *fid, struct osd_inode_id *id)
 536 {
 537         struct lu_fid *oi_fid = &info->oti_fid2;
 538         int            rc;
 539
 540         fid_cpu_to_be(oi_fid, fid);
 541         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 542                                (const struct dt_key *)oi_fid);
 543         if (rc > 0) {
 544                 osd_id_unpack(id, id);
 545                 rc = 0;
 546         } else if (rc == 0) {
 547                 rc = -ENOENT;
 548         }
 549         return rc;
 550 }
 551
 552 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 553                   const struct lu_fid *fid, struct osd_inode_id *id,
 554                   enum oi_check_flags flags)
 555 {
 556         if (unlikely(fid_is_last_id(fid)))
 557                 return osd_obj_spec_lookup(info, osd, fid, id);
 558
 559         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 560                 return osd_obj_map_lookup(info, osd, fid, id);
 561
 562
 563         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 564                 int rc;
 565                 if (fid_is_fs_root(fid)) {
 566                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 567                                    osd_sb(osd)->s_root->d_inode->i_generation);
 568                         return 0;
 569                 }
 570                 if (unlikely(fid_is_acct(fid)))
 571                         return osd_acct_obj_lookup(info, osd, fid, id);
 572
 573                 /* For other special FIDs, try OI first, then do spec lookup */
 574                 rc = __osd_oi_lookup(info, osd, fid, id);
 575                 if (rc == -ENOENT)
 576                         return osd_obj_spec_lookup(info, osd, fid, id);
 577                 return rc;
 578         }
 579
 580         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 581                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 582                 return 0;
 583         }
 584
 585         return __osd_oi_lookup(info, osd, fid, id);
 586 }
 587
 588 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 589                              const struct dt_rec *rec, const struct dt_key *key,
 590                              handle_t *th, bool insert)
 591 {
 592         struct iam_container    *bag;
 593         struct iam_path_descr   *ipd;
 594         int                     rc;
 595         ENTRY;
 596
 597         LASSERT(oi);
 598         LASSERT(oi->oi_inode);
 599         ll_vfs_dq_init(oi->oi_inode);
 600
 601         bag = &oi->oi_dir.od_container;
 602         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 603         if (unlikely(ipd == NULL))
 604                 RETURN(-ENOMEM);
 605
 606         LASSERT(th != NULL);
 607         LASSERT(th->h_transaction != NULL);
 608         if (insert)
 609                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 610                                 (const struct iam_rec *)rec, ipd);
 611         else
 612                 rc = iam_update(th, bag, (const struct iam_key *)key,
 613                                 (const struct iam_rec *)rec, ipd);
 614         osd_ipd_put(oti->oti_env, bag, ipd);
 615         LINVRNT(osd_invariant(obj));
 616         RETURN(rc);
 617 }
 618
 619 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 620                   const struct lu_fid *fid, const struct osd_inode_id *id,
 621                   handle_t *th, enum oi_check_flags flags)
 622 {
 623         struct lu_fid       *oi_fid = &info->oti_fid2;
 624         struct osd_inode_id *oi_id  = &info->oti_id2;
 625         int                  rc     = 0;
 626
 627         if (unlikely(fid_is_last_id(fid)))
 628                 return osd_obj_spec_insert(info, osd, fid, id, th);
 629
 630         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 631                 return osd_obj_map_insert(info, osd, fid, id, th);
 632
 633         fid_cpu_to_be(oi_fid, fid);
 634         osd_id_pack(oi_id, id);
 635         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 636                                (const struct dt_rec *)oi_id,
 637                                (const struct dt_key *)oi_fid, th, true);
 638         if (rc != 0) {
 639                 struct inode *inode;
 640                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
 641
 642                 if (rc != -EEXIST)
 643                         return rc;
 644
 645                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 646                 if (rc != 0)
 647                         return rc;
 648
 649                 if (unlikely(osd_id_eq(id, oi_id)))
 650                         return 1;
 651
 652                 /* Check whether the mapping for oi_id is valid or not. */
 653                 inode = osd_iget(info, osd, oi_id);
 654                 if (IS_ERR(inode)) {
 655                         rc = PTR_ERR(inode);
 656                         if (rc == -ENOENT || rc == -ESTALE)
 657                                 goto update;
 658                         return rc;
 659                 }
 660
 661                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 662                 iput(inode);
 663                 if (rc == -ENODATA)
 664                         goto update;
 665
 666                 if (rc != 0)
 667                         return rc;
 668
 669                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 670                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 671                         CERROR("%.16s: the FID "DFID" is used by two objects: "
 672                                "%u/%u %u/%u\n",
 673                                LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
 674                                PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
 675                                id->oii_ino, id->oii_gen);
 676                         return -EEXIST;
 677                 }
 678
 679 update:
 680                 osd_id_pack(oi_id, id);
 681                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 682                                         (const struct dt_rec *)oi_id,
 683                                         (const struct dt_key *)oi_fid, th, false);
 684                 if (rc != 0)
 685                         return rc;
 686         }
 687
 688         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 689                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 690         return rc;
 691 }
 692
 693 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 694                              const struct dt_key *key, handle_t *th)
 695 {
 696         struct iam_container    *bag;
 697         struct iam_path_descr   *ipd;
 698         int                      rc;
 699         ENTRY;
 700
 701         LASSERT(oi);
 702         LASSERT(oi->oi_inode);
 703         ll_vfs_dq_init(oi->oi_inode);
 704
 705         bag = &oi->oi_dir.od_container;
 706         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 707         if (unlikely(ipd == NULL))
 708                 RETURN(-ENOMEM);
 709
 710         LASSERT(th != NULL);
 711         LASSERT(th->h_transaction != NULL);
 712
 713         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 714         osd_ipd_put(oti->oti_env, bag, ipd);
 715         LINVRNT(osd_invariant(obj));
 716         RETURN(rc);
 717 }
 718
 719 int osd_oi_delete(struct osd_thread_info *info,
 720                   struct osd_device *osd, const struct lu_fid *fid,
 721                   handle_t *th, enum oi_check_flags flags)
 722 {
 723         struct lu_fid *oi_fid = &info->oti_fid2;
 724
 725         /* clear idmap cache */
 726         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 727                 fid_zero(&info->oti_cache.oic_fid);
 728
 729         if (fid_is_last_id(fid))
 730                 return 0;
 731
 732         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 733                 return osd_obj_map_delete(info, osd, fid, th);
 734
 735         fid_cpu_to_be(oi_fid, fid);
 736         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 737                                  (const struct dt_key *)oi_fid, th);
 738 }
 739
 740 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 741                   const struct lu_fid *fid, const struct osd_inode_id *id,
 742                   handle_t *th, enum oi_check_flags flags)
 743 {
 744         struct lu_fid       *oi_fid = &info->oti_fid2;
 745         struct osd_inode_id *oi_id  = &info->oti_id2;
 746         int                  rc     = 0;
 747
 748         if (unlikely(fid_is_last_id(fid)))
 749                 return osd_obj_spec_update(info, osd, fid, id, th);
 750
 751         if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
 752                 return osd_obj_map_update(info, osd, fid, id, th);
 753
 754         fid_cpu_to_be(oi_fid, fid);
 755         osd_id_pack(oi_id, id);
 756         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 757                                (const struct dt_rec *)oi_id,
 758                                (const struct dt_key *)oi_fid, th, false);
 759         if (rc != 0)
 760                 return rc;
 761
 762         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 763                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 764         return rc;
 765 }
 766
 767 int osd_oi_mod_init(void)
 768 {
 769         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 770                 osd_oi_count = OSD_OI_FID_NR;
 771
 772         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 773                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 774                               osd_oi_count, size_roundup_power2(osd_oi_count));
 775                 osd_oi_count = size_roundup_power2(osd_oi_count);
 776         }
 777
 778         mutex_init(&oi_init_lock);
 779         return 0;
 780 }