lustre/obdclass/scrub.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) 2017, Intel Corporation.
  24  */
  25 /*
  26  * lustre/obdclass/scrub.c
  27  *
  28  * The OI scrub is used for checking and (re)building Object Index files
  29  * that are usually backend special. Here are some general scrub related
  30  * functions that can be shared by different backends for OI scrub.
  31  *
  32  * Author: Fan Yong <fan.yong@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_LFSCK
  36
  37 #include <linux/kthread.h>
  38 #include <lustre_scrub.h>
  39 #include <lustre_lib.h>
  40 #include <lustre_fid.h>
  41
  42 static inline struct dt_device *scrub_obj2dev(struct dt_object *obj)
  43 {
  44         return container_of_safe(obj->do_lu.lo_dev, struct dt_device,
  45                                  dd_lu_dev);
  46 }
  47
  48 static void scrub_file_to_cpu(struct scrub_file *des, struct scrub_file *src)
  49 {
  50         uuid_copy(&des->sf_uuid, &src->sf_uuid);
  51         des->sf_flags   = le64_to_cpu(src->sf_flags);
  52         des->sf_magic   = le32_to_cpu(src->sf_magic);
  53         des->sf_status  = le16_to_cpu(src->sf_status);
  54         des->sf_param   = le16_to_cpu(src->sf_param);
  55         des->sf_time_last_complete      =
  56                                 le64_to_cpu(src->sf_time_last_complete);
  57         des->sf_time_latest_start       =
  58                                 le64_to_cpu(src->sf_time_latest_start);
  59         des->sf_time_last_checkpoint    =
  60                                 le64_to_cpu(src->sf_time_last_checkpoint);
  61         des->sf_pos_latest_start        =
  62                                 le64_to_cpu(src->sf_pos_latest_start);
  63         des->sf_pos_last_checkpoint     =
  64                                 le64_to_cpu(src->sf_pos_last_checkpoint);
  65         des->sf_pos_first_inconsistent  =
  66                                 le64_to_cpu(src->sf_pos_first_inconsistent);
  67         des->sf_items_checked           =
  68                                 le64_to_cpu(src->sf_items_checked);
  69         des->sf_items_updated           =
  70                                 le64_to_cpu(src->sf_items_updated);
  71         des->sf_items_failed            =
  72                                 le64_to_cpu(src->sf_items_failed);
  73         des->sf_items_updated_prior     =
  74                                 le64_to_cpu(src->sf_items_updated_prior);
  75         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
  76         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
  77         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
  78         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
  79         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
  80 }
  81
  82 static void scrub_file_to_le(struct scrub_file *des, struct scrub_file *src)
  83 {
  84         uuid_copy(&des->sf_uuid, &src->sf_uuid);
  85         des->sf_flags   = cpu_to_le64(src->sf_flags);
  86         des->sf_magic   = cpu_to_le32(src->sf_magic);
  87         des->sf_status  = cpu_to_le16(src->sf_status);
  88         des->sf_param   = cpu_to_le16(src->sf_param);
  89         des->sf_time_last_complete      =
  90                                 cpu_to_le64(src->sf_time_last_complete);
  91         des->sf_time_latest_start       =
  92                                 cpu_to_le64(src->sf_time_latest_start);
  93         des->sf_time_last_checkpoint    =
  94                                 cpu_to_le64(src->sf_time_last_checkpoint);
  95         des->sf_pos_latest_start        =
  96                                 cpu_to_le64(src->sf_pos_latest_start);
  97         des->sf_pos_last_checkpoint     =
  98                                 cpu_to_le64(src->sf_pos_last_checkpoint);
  99         des->sf_pos_first_inconsistent  =
 100                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 101         des->sf_items_checked           =
 102                                 cpu_to_le64(src->sf_items_checked);
 103         des->sf_items_updated           =
 104                                 cpu_to_le64(src->sf_items_updated);
 105         des->sf_items_failed            =
 106                                 cpu_to_le64(src->sf_items_failed);
 107         des->sf_items_updated_prior     =
 108                                 cpu_to_le64(src->sf_items_updated_prior);
 109         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 110         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 111         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 112         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 113         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 114 }
 115
 116 void scrub_file_init(struct lustre_scrub *scrub, uuid_t uuid)
 117 {
 118         struct scrub_file *sf = &scrub->os_file;
 119
 120         memset(sf, 0, sizeof(*sf));
 121         uuid_copy(&sf->sf_uuid, &uuid);
 122         sf->sf_magic = SCRUB_MAGIC_V1;
 123         sf->sf_status = SS_INIT;
 124 }
 125 EXPORT_SYMBOL(scrub_file_init);
 126
 127 void scrub_file_reset(struct lustre_scrub *scrub, uuid_t uuid, u64 flags)
 128 {
 129         struct scrub_file *sf = &scrub->os_file;
 130
 131         CDEBUG(D_LFSCK, "%s: reset OI scrub file, old flags = "
 132                "%#llx, add flags = %#llx\n",
 133                scrub->os_name, sf->sf_flags, flags);
 134
 135         uuid_copy(&sf->sf_uuid, &uuid);
 136         sf->sf_status = SS_INIT;
 137         sf->sf_flags |= flags;
 138         sf->sf_flags &= ~SF_AUTO;
 139         sf->sf_run_time = 0;
 140         sf->sf_time_latest_start = 0;
 141         sf->sf_time_last_checkpoint = 0;
 142         sf->sf_pos_latest_start = 0;
 143         sf->sf_pos_last_checkpoint = 0;
 144         sf->sf_pos_first_inconsistent = 0;
 145         sf->sf_items_checked = 0;
 146         sf->sf_items_updated = 0;
 147         sf->sf_items_failed = 0;
 148         sf->sf_items_noscrub = 0;
 149         sf->sf_items_igif = 0;
 150         if (!scrub->os_in_join)
 151                 sf->sf_items_updated_prior = 0;
 152 }
 153 EXPORT_SYMBOL(scrub_file_reset);
 154
 155 int scrub_file_load(const struct lu_env *env, struct lustre_scrub *scrub)
 156 {
 157         struct scrub_file *sf = &scrub->os_file;
 158         struct lu_buf buf = {
 159                 .lb_buf = &scrub->os_file_disk,
 160                 .lb_len = sizeof(scrub->os_file_disk)
 161         };
 162         loff_t pos = 0;
 163         int rc;
 164
 165         rc = dt_read(env, scrub->os_obj, &buf, &pos);
 166         /* failure */
 167         if (rc < 0) {
 168                 CERROR("%s: fail to load scrub file: rc = %d\n",
 169                        scrub->os_name, rc);
 170                 return rc;
 171         }
 172
 173         /* empty */
 174         if (!rc)
 175                 return -ENOENT;
 176
 177         /* corrupted */
 178         if (rc < buf.lb_len) {
 179                 CDEBUG(D_LFSCK, "%s: fail to load scrub file, "
 180                        "expected = %d: rc = %d\n",
 181                        scrub->os_name, (int)buf.lb_len, rc);
 182                 return -EFAULT;
 183         }
 184
 185         scrub_file_to_cpu(sf, &scrub->os_file_disk);
 186         if (sf->sf_magic != SCRUB_MAGIC_V1) {
 187                 CDEBUG(D_LFSCK, "%s: invalid scrub magic 0x%x != 0x%x\n",
 188                        scrub->os_name, sf->sf_magic, SCRUB_MAGIC_V1);
 189                 return -EFAULT;
 190         }
 191
 192         return 0;
 193 }
 194 EXPORT_SYMBOL(scrub_file_load);
 195
 196 int scrub_file_store(const struct lu_env *env, struct lustre_scrub *scrub)
 197 {
 198         struct scrub_file *sf = &scrub->os_file_disk;
 199         struct dt_object *obj = scrub->os_obj;
 200         struct dt_device *dev = scrub_obj2dev(obj);
 201         struct lu_buf buf = {
 202                 .lb_buf = sf,
 203                 .lb_len = sizeof(*sf)
 204         };
 205         struct thandle *th;
 206         loff_t pos = 0;
 207         int rc;
 208         ENTRY;
 209
 210         /* Skip store under rdonly mode. */
 211         if (dev->dd_rdonly)
 212                 RETURN(0);
 213
 214         scrub_file_to_le(sf, &scrub->os_file);
 215         th = dt_trans_create(env, dev);
 216         if (IS_ERR(th))
 217                 GOTO(log, rc = PTR_ERR(th));
 218
 219         rc = dt_declare_record_write(env, obj, &buf, pos, th);
 220         if (rc)
 221                 GOTO(stop, rc);
 222
 223         rc = dt_trans_start_local(env, dev, th);
 224         if (rc)
 225                 GOTO(stop, rc);
 226
 227         rc = dt_record_write(env, obj, &buf, &pos, th);
 228
 229         GOTO(stop, rc);
 230
 231 stop:
 232         dt_trans_stop(env, dev, th);
 233
 234 log:
 235         if (rc)
 236                 CERROR("%s: store scrub file: rc = %d\n",
 237                        scrub->os_name, rc);
 238         else
 239                 CDEBUG(D_LFSCK, "%s: store scrub file: rc = %d\n",
 240                        scrub->os_name, rc);
 241
 242         scrub->os_time_last_checkpoint = ktime_get_seconds();
 243         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 244                                          SCRUB_CHECKPOINT_INTERVAL;
 245         return rc;
 246 }
 247 EXPORT_SYMBOL(scrub_file_store);
 248
 249 int scrub_checkpoint(const struct lu_env *env, struct lustre_scrub *scrub)
 250 {
 251         struct scrub_file *sf = &scrub->os_file;
 252         time64_t now = ktime_get_seconds();
 253         int rc;
 254
 255         if (likely(now < scrub->os_time_next_checkpoint ||
 256                    scrub->os_new_checked == 0))
 257                 return 0;
 258
 259         CDEBUG(D_LFSCK, "%s: OI scrub checkpoint at pos %llu\n",
 260                scrub->os_name, scrub->os_pos_current);
 261
 262         down_write(&scrub->os_rwsem);
 263         sf->sf_items_checked += scrub->os_new_checked;
 264         scrub->os_new_checked = 0;
 265         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 266         sf->sf_time_last_checkpoint = ktime_get_real_seconds();
 267         sf->sf_run_time += now - scrub->os_time_last_checkpoint;
 268         rc = scrub_file_store(env, scrub);
 269         up_write(&scrub->os_rwsem);
 270
 271         return rc;
 272 }
 273 EXPORT_SYMBOL(scrub_checkpoint);
 274
 275 int scrub_start(int (*threadfn)(void *data), struct lustre_scrub *scrub,
 276                 void *data, __u32 flags)
 277 {
 278         struct task_struct *task;
 279         int rc;
 280         ENTRY;
 281
 282         if (scrub->os_task)
 283                 RETURN(-EALREADY);
 284
 285         if (scrub->os_file.sf_status == SS_COMPLETED) {
 286                 if (!(flags & SS_SET_FAILOUT))
 287                         flags |= SS_CLEAR_FAILOUT;
 288
 289                 if (!(flags & SS_SET_DRYRUN))
 290                         flags |= SS_CLEAR_DRYRUN;
 291
 292                 flags |= SS_RESET;
 293         }
 294
 295         task = kthread_create(threadfn, data, "OI_scrub");
 296         if (IS_ERR(task)) {
 297                 rc = PTR_ERR(task);
 298                 CERROR("%s: cannot start iteration thread: rc = %d\n",
 299                        scrub->os_name, rc);
 300                 RETURN(rc);
 301         }
 302         spin_lock(&scrub->os_lock);
 303         if (scrub->os_task) {
 304                 /* Lost a race */
 305                 spin_unlock(&scrub->os_lock);
 306                 kthread_stop(task);
 307                 RETURN(-EALREADY);
 308         }
 309         scrub->os_start_flags = flags;
 310         scrub->os_task = task;
 311         wake_up_process(task);
 312         spin_unlock(&scrub->os_lock);
 313         wait_var_event(scrub, scrub->os_running || !scrub->os_task);
 314
 315         RETURN(0);
 316 }
 317 EXPORT_SYMBOL(scrub_start);
 318
 319 void scrub_stop(struct lustre_scrub *scrub)
 320 {
 321         struct task_struct *task;
 322
 323         spin_lock(&scrub->os_lock);
 324         scrub->os_running = 0;
 325         spin_unlock(&scrub->os_lock);
 326         task = xchg(&scrub->os_task, NULL);
 327         if (task)
 328                 kthread_stop(task);
 329 }
 330 EXPORT_SYMBOL(scrub_stop);
 331
 332 const char *scrub_status_names[] = {
 333         "init",
 334         "scanning",
 335         "completed",
 336         "failed",
 337         "stopped",
 338         "paused",
 339         "crashed",
 340         NULL
 341 };
 342
 343 const char *scrub_flags_names[] = {
 344         "recreated",
 345         "inconsistent",
 346         "auto",
 347         "upgrade",
 348         NULL
 349 };
 350
 351 const char *scrub_param_names[] = {
 352         "failout",
 353         "dryrun",
 354         NULL
 355 };
 356
 357 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
 358                             const char *prefix)
 359 {
 360         int flag;
 361         int i;
 362
 363         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
 364
 365         for (i = 0, flag = 1; bits != 0; i++, flag = BIT(i)) {
 366                 if (flag & bits) {
 367                         bits &= ~flag;
 368                         seq_printf(m, "%s%c", names[i],
 369                                    bits != 0 ? ',' : '\n');
 370                 }
 371         }
 372 }
 373
 374 static void scrub_time_dump(struct seq_file *m, time64_t time,
 375                             const char *prefix)
 376 {
 377         if (time != 0)
 378                 seq_printf(m, "%s: %llu seconds\n", prefix,
 379                            ktime_get_real_seconds() - time);
 380         else
 381                 seq_printf(m, "%s: N/A\n", prefix);
 382 }
 383
 384 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
 385 {
 386         if (pos != 0)
 387                 seq_printf(m, "%s: %llu\n", prefix, pos);
 388         else
 389                 seq_printf(m, "%s: N/A\n", prefix);
 390 }
 391
 392 void scrub_dump(struct seq_file *m, struct lustre_scrub *scrub)
 393 {
 394         struct scrub_file *sf = &scrub->os_file;
 395         u64 checked;
 396         s64 speed;
 397
 398         down_read(&scrub->os_rwsem);
 399         seq_printf(m, "name: OI_scrub\n"
 400                    "magic: 0x%x\n"
 401                    "oi_files: %d\n"
 402                    "status: %s\n",
 403                    sf->sf_magic, (int)sf->sf_oi_count,
 404                    scrub_status_names[sf->sf_status]);
 405
 406         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
 407
 408         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
 409
 410         scrub_time_dump(m, sf->sf_time_last_complete,
 411                         "time_since_last_completed");
 412
 413         scrub_time_dump(m, sf->sf_time_latest_start,
 414                         "time_since_latest_start");
 415
 416         scrub_time_dump(m, sf->sf_time_last_checkpoint,
 417                         "time_since_last_checkpoint");
 418
 419         scrub_pos_dump(m, sf->sf_pos_latest_start,
 420                         "latest_start_position");
 421
 422         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
 423                         "last_checkpoint_position");
 424
 425         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
 426                         "first_failure_position");
 427
 428         checked = sf->sf_items_checked + scrub->os_new_checked;
 429         seq_printf(m, "checked: %llu\n"
 430                    "%s: %llu\n"
 431                    "failed: %llu\n"
 432                    "prior_%s: %llu\n"
 433                    "noscrub: %llu\n"
 434                    "igif: %llu\n"
 435                    "success_count: %u\n",
 436                    checked,
 437                    sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
 438                    sf->sf_items_updated, sf->sf_items_failed,
 439                    sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
 440                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
 441                    sf->sf_items_igif, sf->sf_success_count);
 442
 443         speed = checked;
 444         if (scrub->os_running) {
 445                 s64 new_checked = scrub->os_new_checked;
 446                 time64_t duration;
 447                 time64_t rtime;
 448
 449                 /* Since the time resolution is in seconds for new system
 450                  * or small devices it ismore likely that duration will be
 451                  * zero which will lead to inaccurate results.
 452                  */
 453                 duration = ktime_get_seconds() -
 454                            scrub->os_time_last_checkpoint;
 455                 if (duration != 0)
 456                         new_checked = div_s64(new_checked, duration);
 457
 458                 rtime = sf->sf_run_time + duration;
 459                 if (rtime != 0)
 460                         speed = div_s64(speed, rtime);
 461
 462                 seq_printf(m, "run_time: %lld seconds\n"
 463                            "average_speed: %lld objects/sec\n"
 464                            "real_time_speed: %lld objects/sec\n"
 465                            "current_position: %llu\n"
 466                            "scrub_in_prior: %s\n"
 467                            "scrub_full_speed: %s\n"
 468                            "partial_scan: %s\n",
 469                            rtime, speed, new_checked,
 470                            scrub->os_pos_current,
 471                            scrub->os_in_prior ? "yes" : "no",
 472                            scrub->os_full_speed ? "yes" : "no",
 473                            scrub->os_partial_scan ? "yes" : "no");
 474         } else {
 475                 if (sf->sf_run_time != 0)
 476                         speed = div_s64(speed, sf->sf_run_time);
 477                 seq_printf(m, "run_time: %d seconds\n"
 478                            "average_speed: %lld objects/sec\n"
 479                            "real_time_speed: N/A\n"
 480                            "current_position: N/A\n",
 481                            sf->sf_run_time, speed);
 482         }
 483
 484         up_read(&scrub->os_rwsem);
 485 }
 486 EXPORT_SYMBOL(scrub_dump);
 487
 488 int lustre_liru_new(struct list_head *head, const struct lu_fid *pfid,
 489                     const struct lu_fid *cfid, __u64 child,
 490                     const char *name, int namelen)
 491 {
 492         struct lustre_index_restore_unit *liru;
 493         int len = sizeof(*liru) + namelen + 1;
 494
 495         OBD_ALLOC(liru, len);
 496         if (!liru)
 497                 return -ENOMEM;
 498
 499         INIT_LIST_HEAD(&liru->liru_link);
 500         liru->liru_pfid = *pfid;
 501         liru->liru_cfid = *cfid;
 502         liru->liru_clid = child;
 503         liru->liru_len = len;
 504         memcpy(liru->liru_name, name, namelen);
 505         liru->liru_name[namelen] = 0;
 506         list_add_tail(&liru->liru_link, head);
 507
 508         return 0;
 509 }
 510 EXPORT_SYMBOL(lustre_liru_new);
 511
 512 int lustre_index_register(struct dt_device *dev, const char *devname,
 513                           struct list_head *head, spinlock_t *lock, int *guard,
 514                           const struct lu_fid *fid,
 515                           __u32 keysize, __u32 recsize)
 516 {
 517         struct lustre_index_backup_unit *libu, *pos;
 518         int rc = 0;
 519         ENTRY;
 520
 521         if (dev->dd_rdonly || *guard)
 522                 RETURN(1);
 523
 524         OBD_ALLOC_PTR(libu);
 525         if (!libu)
 526                 RETURN(-ENOMEM);
 527
 528         INIT_LIST_HEAD(&libu->libu_link);
 529         libu->libu_keysize = keysize;
 530         libu->libu_recsize = recsize;
 531         libu->libu_fid = *fid;
 532
 533         spin_lock(lock);
 534         if (unlikely(*guard)) {
 535                 spin_unlock(lock);
 536                 OBD_FREE_PTR(libu);
 537
 538                 RETURN(1);
 539         }
 540
 541         list_for_each_entry_reverse(pos, head, libu_link) {
 542                 rc = lu_fid_cmp(&pos->libu_fid, fid);
 543                 if (rc < 0) {
 544                         list_add(&libu->libu_link, &pos->libu_link);
 545                         spin_unlock(lock);
 546
 547                         RETURN(0);
 548                 }
 549
 550                 if (!rc) {
 551                         /* Registered already. But the former registered one
 552                          * has different keysize/recsize. It may because that
 553                          * the former values are from disk and corrupted, then
 554                          * replace it with new values. */
 555                         if (unlikely(keysize != pos->libu_keysize ||
 556                                      recsize != pos->libu_recsize)) {
 557                                 CWARN("%s: the index "DFID" has registered "
 558                                       "with %u/%u, may be invalid, replace "
 559                                       "with %u/%u\n",
 560                                       devname, PFID(fid), pos->libu_keysize,
 561                                       pos->libu_recsize, keysize, recsize);
 562
 563                                 pos->libu_keysize = keysize;
 564                                 pos->libu_recsize = recsize;
 565                         } else {
 566                                 rc = 1;
 567                         }
 568
 569                         spin_unlock(lock);
 570                         OBD_FREE_PTR(libu);
 571
 572                         RETURN(rc);
 573                 }
 574         }
 575
 576         list_add(&libu->libu_link, head);
 577         spin_unlock(lock);
 578
 579         RETURN(0);
 580 }
 581 EXPORT_SYMBOL(lustre_index_register);
 582
 583 static void lustre_index_degister(struct list_head *head, spinlock_t *lock,
 584                                   const struct lu_fid *fid)
 585 {
 586         struct lustre_index_backup_unit *libu;
 587         int rc = -ENOENT;
 588
 589         spin_lock(lock);
 590         list_for_each_entry_reverse(libu, head, libu_link) {
 591                 rc = lu_fid_cmp(&libu->libu_fid, fid);
 592                 /* NOT registered. */
 593                 if (rc < 0)
 594                         break;
 595
 596                 if (!rc) {
 597                         list_del(&libu->libu_link);
 598                         break;
 599                 }
 600         }
 601         spin_unlock(lock);
 602
 603         if (!rc)
 604                 OBD_FREE_PTR(libu);
 605 }
 606
 607 static void
 608 lustre_index_backup_make_header(struct lustre_index_backup_header *header,
 609                                 __u32 keysize, __u32 recsize,
 610                                 const struct lu_fid *fid, __u32 count)
 611 {
 612         memset(header, 0, sizeof(*header));
 613         header->libh_magic = cpu_to_le32(INDEX_BACKUP_MAGIC_V1);
 614         header->libh_count = cpu_to_le32(count);
 615         header->libh_keysize = cpu_to_le32(keysize);
 616         header->libh_recsize = cpu_to_le32(recsize);
 617         fid_cpu_to_le(&header->libh_owner, fid);
 618 }
 619
 620 static int lustre_index_backup_body(const struct lu_env *env,
 621                                     struct dt_object *obj, loff_t *pos,
 622                                     void *buf, int bufsize)
 623 {
 624         struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
 625         struct thandle *th;
 626         struct lu_buf lbuf = {
 627                 .lb_buf = buf,
 628                 .lb_len = bufsize
 629         };
 630         int rc;
 631         ENTRY;
 632
 633         th = dt_trans_create(env, dev);
 634         if (IS_ERR(th))
 635                 RETURN(PTR_ERR(th));
 636
 637         rc = dt_declare_record_write(env, obj, &lbuf, *pos, th);
 638         if (rc)
 639                 GOTO(stop, rc);
 640
 641         rc = dt_trans_start_local(env, dev, th);
 642         if (rc)
 643                 GOTO(stop, rc);
 644
 645         rc = dt_record_write(env, obj, &lbuf, pos, th);
 646
 647         GOTO(stop, rc);
 648
 649 stop:
 650         dt_trans_stop(env, dev, th);
 651         return rc;
 652 }
 653
 654 static int lustre_index_backup_header(const struct lu_env *env,
 655                                       struct dt_object *obj,
 656                                       const struct lu_fid *tgt_fid,
 657                                       __u32 keysize, __u32 recsize,
 658                                       void *buf, int bufsize, int count)
 659 {
 660         struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
 661         struct lustre_index_backup_header *header = buf;
 662         struct lu_attr *la = buf;
 663         struct thandle *th;
 664         struct lu_buf lbuf = {
 665                 .lb_buf = header,
 666                 .lb_len = sizeof(*header)
 667         };
 668         loff_t size = sizeof(*header) + (keysize + recsize) * count;
 669         loff_t pos = 0;
 670         int rc;
 671         bool punch = false;
 672         ENTRY;
 673
 674         LASSERT(sizeof(*la) <= bufsize);
 675         LASSERT(sizeof(*header) <= bufsize);
 676
 677         rc = dt_attr_get(env, obj, la);
 678         if (rc)
 679                 RETURN(rc);
 680
 681         if (la->la_size > size)
 682                 punch = true;
 683
 684         lustre_index_backup_make_header(header, keysize, recsize,
 685                                         tgt_fid, count);
 686         th = dt_trans_create(env, dev);
 687         if (IS_ERR(th))
 688                 RETURN(PTR_ERR(th));
 689
 690         rc = dt_declare_record_write(env, obj, &lbuf, pos, th);
 691         if (rc)
 692                 GOTO(stop, rc);
 693
 694         if (punch) {
 695                 rc = dt_declare_punch(env, obj, size, OBD_OBJECT_EOF, th);
 696                 if (rc)
 697                         GOTO(stop, rc);
 698         }
 699
 700         rc = dt_trans_start_local(env, dev, th);
 701         if (rc)
 702                 GOTO(stop, rc);
 703
 704         rc = dt_record_write(env, obj, &lbuf, &pos, th);
 705         if (!rc && punch)
 706                 rc = dt_punch(env, obj, size, OBD_OBJECT_EOF, th);
 707
 708         GOTO(stop, rc);
 709
 710 stop:
 711         dt_trans_stop(env, dev, th);
 712         return rc;
 713 }
 714
 715 static int lustre_index_update_lma(const struct lu_env *env,
 716                                    struct dt_object *obj,
 717                                    void *buf, int bufsize)
 718 {
 719         struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
 720         struct lustre_mdt_attrs *lma = buf;
 721         struct lu_buf lbuf = {
 722                 .lb_buf = lma,
 723                 .lb_len = sizeof(struct lustre_ost_attrs)
 724         };
 725         struct thandle *th;
 726         int fl = LU_XATTR_REPLACE;
 727         int rc;
 728         ENTRY;
 729
 730         LASSERT(bufsize >= lbuf.lb_len);
 731
 732         rc = dt_xattr_get(env, obj, &lbuf, XATTR_NAME_LMA);
 733         if (unlikely(rc == -ENODATA)) {
 734                 fl = LU_XATTR_CREATE;
 735                 lustre_lma_init(lma, lu_object_fid(&obj->do_lu),
 736                                 LMAC_IDX_BACKUP, 0);
 737                 rc = sizeof(*lma);
 738         } else if (rc < sizeof(*lma)) {
 739                 RETURN(rc < 0 ? rc : -EFAULT);
 740         } else {
 741                 lustre_lma_swab(lma);
 742                 if (lma->lma_compat & LMAC_IDX_BACKUP)
 743                         RETURN(0);
 744
 745                 lma->lma_compat |= LMAC_IDX_BACKUP;
 746         }
 747
 748         lustre_lma_swab(lma);
 749         lbuf.lb_len = rc;
 750         th = dt_trans_create(env, dev);
 751         if (IS_ERR(th))
 752                 RETURN(rc);
 753
 754         rc = dt_declare_xattr_set(env, obj, &lbuf, XATTR_NAME_LMA, fl, th);
 755         if (rc)
 756                 GOTO(stop, rc);
 757
 758         rc = dt_trans_start_local(env, dev, th);
 759         if (rc)
 760                 GOTO(stop, rc);
 761
 762         rc = dt_xattr_set(env, obj, &lbuf, XATTR_NAME_LMA, fl, th);
 763
 764         GOTO(stop, rc);
 765
 766 stop:
 767         dt_trans_stop(env, dev, th);
 768         return rc;
 769 }
 770
 771 static int lustre_index_backup_one(const struct lu_env *env,
 772                                    struct local_oid_storage *los,
 773                                    struct dt_object *parent,
 774                                    struct lustre_index_backup_unit *libu,
 775                                    char *buf, int bufsize)
 776 {
 777         struct dt_device *dev = scrub_obj2dev(parent);
 778         struct dt_object *tgt_obj = NULL;
 779         struct dt_object *bak_obj = NULL;
 780         const struct dt_it_ops *iops;
 781         struct dt_it *di;
 782         loff_t pos = sizeof(struct lustre_index_backup_header);
 783         int count = 0;
 784         int size = 0;
 785         int rc;
 786         ENTRY;
 787
 788         tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
 789                                              &libu->libu_fid, NULL));
 790         if (IS_ERR_OR_NULL(tgt_obj))
 791                 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
 792
 793         if (!dt_object_exists(tgt_obj))
 794                 GOTO(out, rc = 0);
 795
 796         if (!tgt_obj->do_index_ops) {
 797                 struct dt_index_features feat;
 798
 799                 feat.dif_flags = DT_IND_UPDATE;
 800                 feat.dif_keysize_min = libu->libu_keysize;
 801                 feat.dif_keysize_max = libu->libu_keysize;
 802                 feat.dif_recsize_min = libu->libu_recsize;
 803                 feat.dif_recsize_max = libu->libu_recsize;
 804                 feat.dif_ptrsize = 4;
 805                 rc = tgt_obj->do_ops->do_index_try(env, tgt_obj, &feat);
 806                 if (rc)
 807                         GOTO(out, rc);
 808         }
 809
 810         lustre_fid2lbx(buf, &libu->libu_fid, bufsize);
 811         bak_obj = local_file_find_or_create(env, los, parent, buf,
 812                                             S_IFREG | S_IRUGO | S_IWUSR);
 813         if (IS_ERR_OR_NULL(bak_obj))
 814                 GOTO(out, rc = bak_obj ? PTR_ERR(bak_obj) : -ENOENT);
 815
 816         iops = &tgt_obj->do_index_ops->dio_it;
 817         di = iops->init(env, tgt_obj, 0);
 818         if (IS_ERR(di))
 819                 GOTO(out, rc = PTR_ERR(di));
 820
 821         rc = iops->load(env, di, 0);
 822         if (!rc)
 823                 rc = iops->next(env, di);
 824         else if (rc > 0)
 825                 rc = 0;
 826
 827         while (!rc) {
 828                 void *key;
 829                 void *rec;
 830
 831                 key = iops->key(env, di);
 832                 memcpy(&buf[size], key, libu->libu_keysize);
 833                 size += libu->libu_keysize;
 834                 rec = &buf[size];
 835                 rc = iops->rec(env, di, rec, 0);
 836                 if (rc)
 837                         GOTO(fini, rc);
 838
 839                 size += libu->libu_recsize;
 840                 count++;
 841                 if (size + libu->libu_keysize + libu->libu_recsize > bufsize) {
 842                         rc = lustre_index_backup_body(env, bak_obj, &pos,
 843                                                       buf, size);
 844                         if (rc)
 845                                 GOTO(fini, rc);
 846
 847                         size = 0;
 848                 }
 849
 850                 rc = iops->next(env, di);
 851         }
 852
 853         if (rc >= 0 && size > 0)
 854                 rc = lustre_index_backup_body(env, bak_obj, &pos, buf, size);
 855
 856         if (rc < 0)
 857                 GOTO(fini, rc);
 858
 859         rc = lustre_index_backup_header(env, bak_obj, &libu->libu_fid,
 860                                         libu->libu_keysize, libu->libu_recsize,
 861                                         buf, bufsize, count);
 862         if (!rc)
 863                 rc = lustre_index_update_lma(env, tgt_obj, buf, bufsize);
 864
 865         if (!rc && OBD_FAIL_CHECK(OBD_FAIL_OSD_INDEX_CRASH)) {
 866                 LASSERT(bufsize >= 512);
 867
 868                 pos = 0;
 869                 memset(buf, 0, 512);
 870                 lustre_index_backup_body(env, tgt_obj, &pos, buf, 512);
 871         }
 872
 873         GOTO(fini, rc);
 874
 875 fini:
 876         iops->fini(env, di);
 877 out:
 878         if (!IS_ERR_OR_NULL(tgt_obj))
 879                 dt_object_put_nocache(env, tgt_obj);
 880         if (!IS_ERR_OR_NULL(bak_obj))
 881                 dt_object_put_nocache(env, bak_obj);
 882         return rc;
 883 }
 884
 885 void lustre_index_backup(const struct lu_env *env, struct dt_device *dev,
 886                          const char *devname, struct list_head *head,
 887                          spinlock_t *lock, int *guard, bool backup)
 888 {
 889         struct lustre_index_backup_unit *libu;
 890         struct local_oid_storage *los = NULL;
 891         struct dt_object *parent = NULL;
 892         char *buf = NULL;
 893         struct lu_fid fid;
 894         int rc;
 895         ENTRY;
 896
 897         if (dev->dd_rdonly || *guard)
 898                 RETURN_EXIT;
 899
 900         spin_lock(lock);
 901         *guard = 1;
 902         spin_unlock(lock);
 903
 904         if (list_empty(head))
 905                 RETURN_EXIT;
 906
 907         /* Handle kinds of failures during mount process. */
 908         if (!dev->dd_lu_dev.ld_site || !dev->dd_lu_dev.ld_site->ls_top_dev)
 909                 backup = false;
 910
 911         if (backup) {
 912                 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
 913                 if (!buf) {
 914                         backup = false;
 915                         goto scan;
 916                 }
 917
 918                 lu_local_obj_fid(&fid, INDEX_BACKUP_OID);
 919                 parent = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
 920                                                     &fid, NULL));
 921                 if (IS_ERR_OR_NULL(parent)) {
 922                         CERROR("%s: failed to locate backup dir: rc = %ld\n",
 923                                devname, parent ? PTR_ERR(parent) : -ENOENT);
 924                         backup = false;
 925                         goto scan;
 926                 }
 927
 928                 lu_local_name_obj_fid(&fid, 1);
 929                 rc = local_oid_storage_init(env, dev, &fid, &los);
 930                 if (rc) {
 931                         CERROR("%s: failed to init local storage: rc = %d\n",
 932                                devname, rc);
 933                         backup = false;
 934                 }
 935         }
 936
 937 scan:
 938         spin_lock(lock);
 939         while (!list_empty(head)) {
 940                 libu = list_entry(head->next,
 941                                   struct lustre_index_backup_unit, libu_link);
 942                 list_del_init(&libu->libu_link);
 943                 spin_unlock(lock);
 944
 945                 if (backup) {
 946                         rc = lustre_index_backup_one(env, los, parent, libu,
 947                                                      buf, INDEX_BACKUP_BUFSIZE);
 948                         CDEBUG(D_WARNING, "%s: backup index "DFID": rc = %d\n",
 949                                devname, PFID(&libu->libu_fid), rc);
 950                 }
 951
 952                 OBD_FREE_PTR(libu);
 953                 spin_lock(lock);
 954         }
 955         spin_unlock(lock);
 956
 957         if (los)
 958                 local_oid_storage_fini(env, los);
 959         if (parent)
 960                 dt_object_put_nocache(env, parent);
 961         if (buf)
 962                 OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
 963
 964         EXIT;
 965 }
 966 EXPORT_SYMBOL(lustre_index_backup);
 967
 968 int lustre_index_restore(const struct lu_env *env, struct dt_device *dev,
 969                          const struct lu_fid *parent_fid,
 970                          const struct lu_fid *tgt_fid,
 971                          const struct lu_fid *bak_fid, const char *name,
 972                          struct list_head *head, spinlock_t *lock,
 973                          char *buf, int bufsize)
 974 {
 975         struct dt_object *parent_obj = NULL;
 976         struct dt_object *tgt_obj = NULL;
 977         struct dt_object *bak_obj = NULL;
 978         struct lustre_index_backup_header *header;
 979         struct dt_index_features *feat;
 980         struct dt_object_format *dof;
 981         struct lu_attr *la;
 982         struct thandle *th;
 983         struct lu_object_conf conf;
 984         struct dt_insert_rec ent;
 985         struct lu_buf lbuf;
 986         struct lu_fid tfid;
 987         loff_t pos = 0;
 988         __u32 keysize;
 989         __u32 recsize;
 990         __u32 pairsize;
 991         int count;
 992         int rc;
 993         bool registered = false;
 994         ENTRY;
 995
 996         LASSERT(bufsize >= sizeof(*la) + sizeof(*dof) +
 997                 sizeof(*feat) + sizeof(*header));
 998
 999         memset(buf, 0, bufsize);
1000         la = (struct lu_attr *)buf;
1001         dof = (void *)la + sizeof(*la);
1002         feat = (void *)dof + sizeof(*dof);
1003         header = (void *)feat + sizeof(*feat);
1004         lbuf.lb_buf = header;
1005         lbuf.lb_len = sizeof(*header);
1006
1007         tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1008                                              tgt_fid, NULL));
1009         if (IS_ERR_OR_NULL(tgt_obj))
1010                 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
1011
1012         bak_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1013                                              bak_fid, NULL));
1014         if (IS_ERR_OR_NULL(bak_obj))
1015                 GOTO(out, rc = bak_obj ? PTR_ERR(bak_obj) : -ENOENT);
1016
1017         if (!dt_object_exists(bak_obj))
1018                 GOTO(out, rc = -ENOENT);
1019
1020         parent_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1021                                                 parent_fid, NULL));
1022         if (IS_ERR_OR_NULL(parent_obj))
1023                 GOTO(out, rc = parent_obj ? PTR_ERR(parent_obj) : -ENOENT);
1024
1025         LASSERT(dt_object_exists(parent_obj));
1026
1027         if (unlikely(!dt_try_as_dir(env, parent_obj)))
1028                 GOTO(out, rc = -ENOTDIR);
1029
1030         rc = dt_attr_get(env, tgt_obj, la);
1031         if (rc)
1032                 GOTO(out, rc);
1033
1034         rc = dt_record_read(env, bak_obj, &lbuf, &pos);
1035         if (rc)
1036                 GOTO(out, rc);
1037
1038         if (le32_to_cpu(header->libh_magic) != INDEX_BACKUP_MAGIC_V1)
1039                 GOTO(out, rc = -EINVAL);
1040
1041         fid_le_to_cpu(&tfid, &header->libh_owner);
1042         if (unlikely(!lu_fid_eq(tgt_fid, &tfid)))
1043                 GOTO(out, rc = -EINVAL);
1044
1045         keysize = le32_to_cpu(header->libh_keysize);
1046         recsize = le32_to_cpu(header->libh_recsize);
1047         pairsize = keysize + recsize;
1048
1049         memset(feat, 0, sizeof(*feat));
1050         feat->dif_flags = DT_IND_UPDATE;
1051         feat->dif_keysize_min = feat->dif_keysize_max = keysize;
1052         feat->dif_recsize_min = feat->dif_recsize_max = recsize;
1053         feat->dif_ptrsize = 4;
1054
1055         /* T1: remove old name entry and destroy old index. */
1056         th = dt_trans_create(env, dev);
1057         if (IS_ERR(th))
1058                 GOTO(out, rc = PTR_ERR(th));
1059
1060         rc = dt_declare_delete(env, parent_obj,
1061                                (const struct dt_key *)name, th);
1062         if (rc)
1063                 GOTO(stop, rc);
1064
1065         rc = dt_declare_destroy(env, tgt_obj, th);
1066         if (rc)
1067                 GOTO(stop, rc);
1068
1069         rc = dt_trans_start_local(env, dev, th);
1070         if (rc)
1071                 GOTO(stop, rc);
1072
1073         rc = dt_delete(env, parent_obj, (const struct dt_key *)name, th);
1074         if (rc)
1075                 GOTO(stop, rc);
1076
1077         dt_write_lock(env, tgt_obj, 0);
1078         rc = dt_destroy(env, tgt_obj, th);
1079         dt_write_unlock(env, tgt_obj);
1080         dt_trans_stop(env, dev, th);
1081         if (rc)
1082                 GOTO(out, rc);
1083
1084         la->la_valid = LA_MODE | LA_UID | LA_GID;
1085         conf.loc_flags = LOC_F_NEW;
1086         dof->u.dof_idx.di_feat = feat;
1087         dof->dof_type = DFT_INDEX;
1088         ent.rec_type = S_IFREG;
1089         ent.rec_fid = tgt_fid;
1090
1091         /* Drop cache before re-create it. */
1092         dt_object_put_nocache(env, tgt_obj);
1093         tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1094                                              tgt_fid, &conf));
1095         if (IS_ERR_OR_NULL(tgt_obj))
1096                 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
1097
1098         LASSERT(!dt_object_exists(tgt_obj));
1099
1100         /* T2: create new index and insert new name entry. */
1101         th = dt_trans_create(env, dev);
1102         if (IS_ERR(th))
1103                 GOTO(out, rc = PTR_ERR(th));
1104
1105         rc = dt_declare_create(env, tgt_obj, la, NULL, dof, th);
1106         if (rc)
1107                 GOTO(stop, rc);
1108
1109         rc = dt_declare_insert(env, parent_obj, (const struct dt_rec *)&ent,
1110                                (const struct dt_key *)name, th);
1111         if (rc)
1112                 GOTO(stop, rc);
1113
1114         rc = dt_trans_start_local(env, dev, th);
1115         if (rc)
1116                 GOTO(stop, rc);
1117
1118         dt_write_lock(env, tgt_obj, 0);
1119         rc = dt_create(env, tgt_obj, la, NULL, dof, th);
1120         dt_write_unlock(env, tgt_obj);
1121         if (rc)
1122                 GOTO(stop, rc);
1123
1124         rc = dt_insert(env, parent_obj, (const struct dt_rec *)&ent,
1125                        (const struct dt_key *)name, th);
1126         dt_trans_stop(env, dev, th);
1127         /* Some index name may has been inserted by OSD
1128          * automatically when create the index object. */
1129         if (unlikely(rc == -EEXIST))
1130                 rc = 0;
1131         if (rc)
1132                 GOTO(out, rc);
1133
1134         /* The new index will register via index_try. */
1135         rc = tgt_obj->do_ops->do_index_try(env, tgt_obj, feat);
1136         if (rc)
1137                 GOTO(out, rc);
1138
1139         registered = true;
1140         count = le32_to_cpu(header->libh_count);
1141         while (!rc && count > 0) {
1142                 int size = pairsize * count;
1143                 int items = count;
1144                 int i;
1145
1146                 if (size > bufsize) {
1147                         items = bufsize / pairsize;
1148                         size = pairsize * items;
1149                 }
1150
1151                 lbuf.lb_buf = buf;
1152                 lbuf.lb_len = size;
1153                 rc = dt_record_read(env, bak_obj, &lbuf, &pos);
1154                 for (i = 0; i < items && !rc; i++) {
1155                         void *key = &buf[i * pairsize];
1156                         void *rec = &buf[i * pairsize + keysize];
1157
1158                         /* Tn: restore the records. */
1159                         th = dt_trans_create(env, dev);
1160                         if (!th)
1161                                 GOTO(out, rc = -ENOMEM);
1162
1163                         rc = dt_declare_insert(env, tgt_obj, rec, key, th);
1164                         if (rc)
1165                                 GOTO(stop, rc);
1166
1167                         rc = dt_trans_start_local(env, dev, th);
1168                         if (rc)
1169                                 GOTO(stop, rc);
1170
1171                         rc = dt_insert(env, tgt_obj, rec, key, th);
1172                         if (unlikely(rc == -EEXIST))
1173                                 rc = 0;
1174
1175                         dt_trans_stop(env, dev, th);
1176                 }
1177
1178                 count -= items;
1179         }
1180
1181         GOTO(out, rc);
1182
1183 stop:
1184         dt_trans_stop(env, dev, th);
1185         if (rc && registered)
1186                 /* Degister the index to avoid overwriting the backup. */
1187                 lustre_index_degister(head, lock, tgt_fid);
1188
1189 out:
1190         if (!IS_ERR_OR_NULL(tgt_obj))
1191                 dt_object_put_nocache(env, tgt_obj);
1192         if (!IS_ERR_OR_NULL(bak_obj))
1193                 dt_object_put_nocache(env, bak_obj);
1194         if (!IS_ERR_OR_NULL(parent_obj))
1195                 dt_object_put_nocache(env, parent_obj);
1196         return rc;
1197 }
1198 EXPORT_SYMBOL(lustre_index_restore);