lustre/obdclass/llog_cat.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *   Author: Andreas Dilger <adilger@clusterfs.com>
   6  *
   7  *   This file is part of the Lustre file system, http://www.lustre.org
   8  *   Lustre is a trademark of Cluster File Systems, Inc.
   9  *
  10  *   You may have signed or agreed to another license before downloading
  11  *   this software.  If so, you are bound by the terms and conditions
  12  *   of that agreement, and the following does not apply to you.  See the
  13  *   LICENSE file included with this distribution for more information.
  14  *
  15  *   If you did not agree to a different license, then this copy of Lustre
  16  *   is open source software; you can redistribute it and/or modify it
  17  *   under the terms of version 2 of the GNU General Public License as
  18  *   published by the Free Software Foundation.
  19  *
  20  *   In either case, Lustre is distributed in the hope that it will be
  21  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  22  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  23  *   license text for more details.
  24  *
  25  * OST<->MDS recovery logging infrastructure.
  26  *
  27  * Invariants in implementation:
  28  * - we do not share logs among different OST<->MDS connections, so that
  29  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  30  */
  31
  32 #define DEBUG_SUBSYSTEM S_LOG
  33
  34 #ifndef EXPORT_SYMTAB
  35 #define EXPORT_SYMTAB
  36 #endif
  37
  38 #ifndef __KERNEL__
  39 #include <liblustre.h>
  40 #endif
  41
  42 #include <obd_class.h>
  43 #include <lustre_log.h>
  44 #include <libcfs/list.h>
  45
  46 /* Create a new log handle and add it to the open list.
  47  * This log handle will be closed when all of the records in it are removed.
  48  *
  49  * Assumes caller has already pushed us into the kernel context and is locking.
  50  */
  51 static struct llog_handle *llog_cat_new_log(struct llog_handle *cathandle)
  52 {
  53         struct llog_handle *loghandle;
  54         struct llog_log_hdr *llh;
  55         struct llog_logid_rec rec = { { 0 }, };
  56         int rc, index, bitmap_size;
  57         ENTRY;
  58
  59         llh = cathandle->lgh_hdr;
  60         bitmap_size = LLOG_BITMAP_SIZE(llh);
  61
  62         index = (cathandle->lgh_last_idx + 1) % bitmap_size;
  63
  64         /* maximum number of available slots in catlog is bitmap_size - 2 */
  65         if (llh->llh_cat_idx == index) {
  66                 CERROR("no free catalog slots for log...\n");
  67                 RETURN(ERR_PTR(-ENOSPC));
  68         }
  69
  70         if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
  71                 RETURN(ERR_PTR(-ENOSPC));
  72
  73         rc = llog_create(cathandle->lgh_ctxt, &loghandle, NULL, NULL);
  74         if (rc)
  75                 RETURN(ERR_PTR(rc));
  76
  77         rc = llog_init_handle(loghandle,
  78                               LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
  79                               &cathandle->lgh_hdr->llh_tgtuuid);
  80         if (rc)
  81                 GOTO(out_destroy, rc);
  82         if (index == 0)
  83                 index = 1;
  84         if (ext2_set_bit(index, llh->llh_bitmap)) {
  85                 CERROR("argh, index %u already set in log bitmap?\n",
  86                        index);
  87                 LBUG(); /* should never happen */
  88         }
  89         cathandle->lgh_last_idx = index;
  90         llh->llh_count++;
  91         llh->llh_tail.lrt_index = index;
  92
  93         CDEBUG(D_RPCTRACE,"new recovery log "LPX64":%x for index %u of catalog "
  94                LPX64"\n", loghandle->lgh_id.lgl_oid, loghandle->lgh_id.lgl_ogen,
  95                index, cathandle->lgh_id.lgl_oid);
  96         /* build the record for this log in the catalog */
  97         rec.lid_hdr.lrh_len = sizeof(rec);
  98         rec.lid_hdr.lrh_index = index;
  99         rec.lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
 100         rec.lid_id = loghandle->lgh_id;
 101         rec.lid_tail.lrt_len = sizeof(rec);
 102         rec.lid_tail.lrt_index = index;
 103
 104         /* update the catalog: header and record */
 105         rc = llog_write_rec(cathandle, &rec.lid_hdr,
 106                             &loghandle->u.phd.phd_cookie, 1, NULL, index);
 107         if (rc < 0) {
 108                 GOTO(out_destroy, rc);
 109         }
 110
 111         loghandle->lgh_hdr->llh_cat_idx = index;
 112         cathandle->u.chd.chd_current_log = loghandle;
 113         LASSERT(list_empty(&loghandle->u.phd.phd_entry));
 114         list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
 115
 116  out_destroy:
 117         if (rc < 0)
 118                 llog_destroy(loghandle);
 119
 120         RETURN(loghandle);
 121 }
 122 EXPORT_SYMBOL(llog_cat_new_log);
 123
 124 /* Open an existent log handle and add it to the open list.
 125  * This log handle will be closed when all of the records in it are removed.
 126  *
 127  * Assumes caller has already pushed us into the kernel context and is locking.
 128  * We return a lock on the handle to ensure nobody yanks it from us.
 129  */
 130 int llog_cat_id2handle(struct llog_handle *cathandle, struct llog_handle **res,
 131                        struct llog_logid *logid)
 132 {
 133         struct llog_handle *loghandle;
 134         int rc = 0;
 135         ENTRY;
 136
 137         if (cathandle == NULL)
 138                 RETURN(-EBADF);
 139
 140         list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
 141                             u.phd.phd_entry) {
 142                 struct llog_logid *cgl = &loghandle->lgh_id;
 143                 if (cgl->lgl_oid == logid->lgl_oid) {
 144                         if (cgl->lgl_ogen != logid->lgl_ogen) {
 145                                 CERROR("log "LPX64" generation %x != %x\n",
 146                                        logid->lgl_oid, cgl->lgl_ogen,
 147                                        logid->lgl_ogen);
 148                                 continue;
 149                         }
 150                         loghandle->u.phd.phd_cat_handle = cathandle;
 151                         GOTO(out, rc = 0);
 152                 }
 153         }
 154
 155         rc = llog_create(cathandle->lgh_ctxt, &loghandle, logid, NULL);
 156         if (rc) {
 157                 CERROR("error opening log id "LPX64":%x: rc %d\n",
 158                        logid->lgl_oid, logid->lgl_ogen, rc);
 159         } else {
 160                 rc = llog_init_handle(loghandle, LLOG_F_IS_PLAIN, NULL);
 161                 if (!rc) {
 162                         list_add(&loghandle->u.phd.phd_entry,
 163                                  &cathandle->u.chd.chd_head);
 164                 }
 165         }
 166         if (!rc) {
 167                 loghandle->u.phd.phd_cat_handle = cathandle;
 168                 loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
 169                 loghandle->u.phd.phd_cookie.lgc_index =
 170                         loghandle->lgh_hdr->llh_cat_idx;
 171         }
 172
 173 out:
 174         *res = loghandle;
 175         RETURN(rc);
 176 }
 177
 178 int llog_cat_put(struct llog_handle *cathandle)
 179 {
 180         struct llog_handle *loghandle, *n;
 181         int rc;
 182         ENTRY;
 183
 184         list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
 185                                  u.phd.phd_entry) {
 186                 int err = llog_close(loghandle);
 187                 if (err)
 188                         CERROR("error closing loghandle\n");
 189         }
 190         rc = llog_close(cathandle);
 191         RETURN(rc);
 192 }
 193 EXPORT_SYMBOL(llog_cat_put);
 194
 195 /* Return the currently active log handle.  If the current log handle doesn't
 196  * have enough space left for the current record, start a new one.
 197  *
 198  * If reclen is 0, we only want to know what the currently active log is,
 199  * otherwise we get a lock on this log so nobody can steal our space.
 200  *
 201  * Assumes caller has already pushed us into the kernel context and is locking.
 202  *
 203  * NOTE: loghandle is write-locked upon successful return
 204  */
 205 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
 206                                                 int create)
 207 {
 208         struct llog_handle *loghandle = NULL;
 209         ENTRY;
 210
 211         down_read(&cathandle->lgh_lock);
 212         loghandle = cathandle->u.chd.chd_current_log;
 213         if (loghandle) {
 214                 struct llog_log_hdr *llh = loghandle->lgh_hdr;
 215                 down_write(&loghandle->lgh_lock);
 216                 if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) {
 217                         up_read(&cathandle->lgh_lock);
 218                         RETURN(loghandle);
 219                 } else {
 220                         up_write(&loghandle->lgh_lock);
 221                 }
 222         }
 223         if (!create) {
 224                 if (loghandle)
 225                         down_write(&loghandle->lgh_lock);
 226                 up_read(&cathandle->lgh_lock);
 227                 RETURN(loghandle);
 228         }
 229         up_read(&cathandle->lgh_lock);
 230
 231         /* time to create new log */
 232
 233         /* first, we have to make sure the state hasn't changed */
 234         down_write(&cathandle->lgh_lock);
 235         loghandle = cathandle->u.chd.chd_current_log;
 236         if (loghandle) {
 237                 struct llog_log_hdr *llh = loghandle->lgh_hdr;
 238                 down_write(&loghandle->lgh_lock);
 239                 if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) {
 240                         up_write(&cathandle->lgh_lock);
 241                         RETURN(loghandle);
 242                 } else {
 243                         up_write(&loghandle->lgh_lock);
 244                 }
 245         }
 246
 247         CDEBUG(D_INODE, "creating new log\n");
 248         loghandle = llog_cat_new_log(cathandle);
 249         if (!IS_ERR(loghandle))
 250                 down_write(&loghandle->lgh_lock);
 251         up_write(&cathandle->lgh_lock);
 252         RETURN(loghandle);
 253 }
 254
 255 /* Add a single record to the recovery log(s) using a catalog
 256  * Returns as llog_write_record
 257  *
 258  * Assumes caller has already pushed us into the kernel context.
 259  */
 260 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
 261                      struct llog_cookie *reccookie, void *buf)
 262 {
 263         struct llog_handle *loghandle;
 264         int rc;
 265         ENTRY;
 266
 267         LASSERT(rec->lrh_len <= LLOG_CHUNK_SIZE);
 268         loghandle = llog_cat_current_log(cathandle, 1);
 269         if (IS_ERR(loghandle))
 270                 RETURN(PTR_ERR(loghandle));
 271         /* loghandle is already locked by llog_cat_current_log() for us */
 272         rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
 273         up_write(&loghandle->lgh_lock);
 274         if (rc == -ENOSPC) {
 275                 /* to create a new plain log */
 276                 loghandle = llog_cat_current_log(cathandle, 1);
 277                 if (IS_ERR(loghandle))
 278                         RETURN(PTR_ERR(loghandle));
 279                 rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
 280                 up_write(&loghandle->lgh_lock);
 281         }
 282
 283         RETURN(rc);
 284 }
 285 EXPORT_SYMBOL(llog_cat_add_rec);
 286
 287 /* For each cookie in the cookie array, we clear the log in-use bit and either:
 288  * - the log is empty, so mark it free in the catalog header and delete it
 289  * - the log is not empty, just write out the log header
 290  *
 291  * The cookies may be in different log files, so we need to get new logs
 292  * each time.
 293  *
 294  * Assumes caller has already pushed us into the kernel context.
 295  */
 296 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
 297                         struct llog_cookie *cookies)
 298 {
 299         int i, index, rc = 0;
 300         ENTRY;
 301
 302         down_write(&cathandle->lgh_lock);
 303         for (i = 0; i < count; i++, cookies++) {
 304                 struct llog_handle *loghandle;
 305                 struct llog_logid *lgl = &cookies->lgc_lgl;
 306
 307                 rc = llog_cat_id2handle(cathandle, &loghandle, lgl);
 308                 if (rc) {
 309                         CERROR("Cannot find log "LPX64"\n", lgl->lgl_oid);
 310                         break;
 311                 }
 312
 313                 down_write(&loghandle->lgh_lock);
 314                 rc = llog_cancel_rec(loghandle, cookies->lgc_index);
 315                 up_write(&loghandle->lgh_lock);
 316
 317                 if (rc == 1) {          /* log has been destroyed */
 318                         index = loghandle->u.phd.phd_cookie.lgc_index;
 319                         if (cathandle->u.chd.chd_current_log == loghandle)
 320                                 cathandle->u.chd.chd_current_log = NULL;
 321                         llog_free_handle(loghandle);
 322
 323                         LASSERT(index);
 324                         llog_cat_set_first_idx(cathandle, index);
 325                         rc = llog_cancel_rec(cathandle, index);
 326                         if (rc == 0)
 327                                 CDEBUG(D_RPCTRACE,"cancel plain log at index %u"
 328                                        " of catalog "LPX64"\n",
 329                                        index, cathandle->lgh_id.lgl_oid);
 330                 }
 331         }
 332         up_write(&cathandle->lgh_lock);
 333
 334         RETURN(rc);
 335 }
 336 EXPORT_SYMBOL(llog_cat_cancel_records);
 337
 338 int llog_cat_process_cb(struct llog_handle *cat_llh, struct llog_rec_hdr *rec,
 339                         void *data)
 340 {
 341         struct llog_process_data *d = data;
 342         struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
 343         struct llog_handle *llh;
 344         int rc;
 345
 346         ENTRY;
 347         if (rec->lrh_type != LLOG_LOGID_MAGIC) {
 348                 CERROR("invalid record in catalog\n");
 349                 RETURN(-EINVAL);
 350         }
 351         CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
 352                LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
 353                rec->lrh_index, cat_llh->lgh_id.lgl_oid);
 354
 355         rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
 356         if (rc) {
 357                 CERROR("Cannot find handle for log "LPX64"\n",
 358                        lir->lid_id.lgl_oid);
 359                 RETURN(rc);
 360         }
 361
 362         rc = llog_process(llh, d->lpd_cb, d->lpd_data, NULL);
 363         RETURN(rc);
 364 }
 365
 366 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data)
 367 {
 368         struct llog_process_data d;
 369         struct llog_process_cat_data cd;
 370         struct llog_log_hdr *llh = cat_llh->lgh_hdr;
 371         int rc;
 372         ENTRY;
 373
 374         LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
 375         d.lpd_data = data;
 376         d.lpd_cb = cb;
 377
 378         if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
 379                 CWARN("catlog "LPX64" crosses index zero\n",
 380                       cat_llh->lgh_id.lgl_oid);
 381
 382                 cd.first_idx = llh->llh_cat_idx;
 383                 cd.last_idx = 0;
 384                 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
 385                 if (rc != 0)
 386                         RETURN(rc);
 387
 388                 cd.first_idx = 0;
 389                 cd.last_idx = cat_llh->lgh_last_idx;
 390                 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
 391         } else {
 392                 rc = llog_process(cat_llh, llog_cat_process_cb, &d, NULL);
 393         }
 394
 395         RETURN(rc);
 396 }
 397 EXPORT_SYMBOL(llog_cat_process);
 398
 399 static int llog_cat_reverse_process_cb(struct llog_handle *cat_llh,
 400                                        struct llog_rec_hdr *rec, void *data)
 401 {
 402         struct llog_process_data *d = data;
 403         struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
 404         struct llog_handle *llh;
 405         int rc;
 406
 407         if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
 408                 CERROR("invalid record in catalog\n");
 409                 RETURN(-EINVAL);
 410         }
 411         CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
 412                LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
 413                le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid);
 414
 415         rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
 416         if (rc) {
 417                 CERROR("Cannot find handle for log "LPX64"\n",
 418                        lir->lid_id.lgl_oid);
 419                 RETURN(rc);
 420         }
 421
 422         rc = llog_reverse_process(llh, d->lpd_cb, d->lpd_data, NULL);
 423         RETURN(rc);
 424 }
 425
 426 int llog_cat_reverse_process(struct llog_handle *cat_llh,
 427                              llog_cb_t cb, void *data)
 428 {
 429         struct llog_process_data d;
 430         struct llog_process_cat_data cd;
 431         struct llog_log_hdr *llh = cat_llh->lgh_hdr;
 432         int rc;
 433         ENTRY;
 434
 435         LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
 436         d.lpd_data = data;
 437         d.lpd_cb = cb;
 438
 439         if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
 440                 CWARN("catalog "LPX64" crosses index zero\n",
 441                       cat_llh->lgh_id.lgl_oid);
 442
 443                 cd.first_idx = 0;
 444                 cd.last_idx = cat_llh->lgh_last_idx;
 445                 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
 446                                           &d, &cd);
 447                 if (rc != 0)
 448                         RETURN(rc);
 449
 450                 cd.first_idx = le32_to_cpu(llh->llh_cat_idx);
 451                 cd.last_idx = 0;
 452                 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
 453                                           &d, &cd);
 454         } else {
 455                 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
 456                                           &d, NULL);
 457         }
 458
 459         RETURN(rc);
 460 }
 461 EXPORT_SYMBOL(llog_cat_reverse_process);
 462
 463 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
 464 {
 465         struct llog_log_hdr *llh = cathandle->lgh_hdr;
 466         int i, bitmap_size, idx;
 467         ENTRY;
 468
 469         bitmap_size = LLOG_BITMAP_SIZE(llh);
 470         if (llh->llh_cat_idx == (index - 1)) {
 471                 idx = llh->llh_cat_idx + 1;
 472                 llh->llh_cat_idx = idx;
 473                 if (idx == cathandle->lgh_last_idx)
 474                         goto out;
 475                 for (i = (index + 1) % bitmap_size;
 476                      i != cathandle->lgh_last_idx;
 477                      i = (i + 1) % bitmap_size) {
 478                         if (!ext2_test_bit(i, llh->llh_bitmap)) {
 479                                 idx = llh->llh_cat_idx + 1;
 480                                 llh->llh_cat_idx = idx;
 481                         } else if (i == 0) {
 482                                 llh->llh_cat_idx = 0;
 483                         } else {
 484                                 break;
 485                         }
 486                 }
 487 out:
 488                 CDEBUG(D_RPCTRACE, "set catlog "LPX64" first idx %u\n",
 489                        cathandle->lgh_id.lgl_oid, llh->llh_cat_idx);
 490         }
 491
 492         RETURN(0);
 493 }
 494
 495 #if 0
 496 /* Assumes caller has already pushed us into the kernel context. */
 497 int llog_cat_init(struct llog_handle *cathandle, struct obd_uuid *tgtuuid)
 498 {
 499         struct llog_log_hdr *llh;
 500         loff_t offset = 0;
 501         int rc = 0;
 502         ENTRY;
 503
 504         LASSERT(sizeof(*llh) == LLOG_CHUNK_SIZE);
 505
 506         down(&cathandle->lgh_lock);
 507         llh = cathandle->lgh_hdr;
 508
 509         if (i_size_read(cathandle->lgh_file->f_dentry->d_inode) == 0) {
 510                 llog_write_rec(cathandle, &llh->llh_hdr, NULL, 0, NULL, 0);
 511
 512 write_hdr:
 513                 rc = lustre_fwrite(cathandle->lgh_file, llh, LLOG_CHUNK_SIZE,
 514                                    &offset);
 515                 if (rc != LLOG_CHUNK_SIZE) {
 516                         CERROR("error writing catalog header: rc %d\n", rc);
 517                         OBD_FREE(llh, sizeof(*llh));
 518                         if (rc >= 0)
 519                                 rc = -ENOSPC;
 520                 } else
 521                         rc = 0;
 522         } else {
 523                 rc = lustre_fread(cathandle->lgh_file, llh, LLOG_CHUNK_SIZE,
 524                                   &offset);
 525                 if (rc != LLOG_CHUNK_SIZE) {
 526                         CERROR("error reading catalog header: rc %d\n", rc);
 527                         /* Can we do much else if the header is bad? */
 528                         goto write_hdr;
 529                 } else
 530                         rc = 0;
 531         }
 532
 533         cathandle->lgh_tgtuuid = &llh->llh_tgtuuid;
 534         up(&cathandle->lgh_lock);
 535         RETURN(rc);
 536 }
 537 EXPORT_SYMBOL(llog_cat_init);
 538
 539 #endif