fs/ocfs2/buffer_head_io.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * io.c
   5  *
   6  * Buffer cache handling
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/fs.h>
  27 #include <linux/types.h>
  28 #include <linux/highmem.h>
  29 #include <linux/bio.h>
  30
  31 #include <cluster/masklog.h>
  32
  33 #include "ocfs2.h"
  34
  35 #include "alloc.h"
  36 #include "inode.h"
  37 #include "journal.h"
  38 #include "uptodate.h"
  39 #include "buffer_head_io.h"
  40 #include "ocfs2_trace.h"
  41
  42 /*
  43  * Bits on bh->b_state used by ocfs2.
  44  *
  45  * These MUST be after the JBD2 bits.  Hence, we use BH_JBDPrivateStart.
  46  */
  47 enum ocfs2_state_bits {
  48         BH_NeedsValidate = BH_JBDPrivateStart,
  49 };
  50
  51 /* Expand the magic b_state functions */
  52 BUFFER_FNS(NeedsValidate, needs_validate);
  53
  54 int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
  55                       struct ocfs2_caching_info *ci)
  56 {
  57         int ret = 0;
  58
  59         trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci);
  60
  61         BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
  62         BUG_ON(buffer_jbd(bh));
  63
  64         /* No need to check for a soft readonly file system here. non
  65          * journalled writes are only ever done on system files which
  66          * can get modified during recovery even if read-only. */
  67         if (ocfs2_is_hard_readonly(osb)) {
  68                 ret = -EROFS;
  69                 mlog_errno(ret);
  70                 goto out;
  71         }
  72
  73         ocfs2_metadata_cache_io_lock(ci);
  74
  75         lock_buffer(bh);
  76         set_buffer_uptodate(bh);
  77
  78         /* remove from dirty list before I/O. */
  79         clear_buffer_dirty(bh);
  80
  81         get_bh(bh); /* for end_buffer_write_sync() */
  82         bh->b_end_io = end_buffer_write_sync;
  83         submit_bh(REQ_OP_WRITE, 0, bh);
  84
  85         wait_on_buffer(bh);
  86
  87         if (buffer_uptodate(bh)) {
  88                 ocfs2_set_buffer_uptodate(ci, bh);
  89         } else {
  90                 /* We don't need to remove the clustered uptodate
  91                  * information for this bh as it's not marked locally
  92                  * uptodate. */
  93                 ret = -EIO;
  94                 mlog_errno(ret);
  95         }
  96
  97         ocfs2_metadata_cache_io_unlock(ci);
  98 out:
  99         return ret;
 100 }
 101
 102 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
 103                            unsigned int nr, struct buffer_head *bhs[])
 104 {
 105         int status = 0;
 106         unsigned int i;
 107         struct buffer_head *bh;
 108
 109         trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
 110
 111         if (!nr)
 112                 goto bail;
 113
 114         for (i = 0 ; i < nr ; i++) {
 115                 if (bhs[i] == NULL) {
 116                         bhs[i] = sb_getblk(osb->sb, block++);
 117                         if (bhs[i] == NULL) {
 118                                 status = -ENOMEM;
 119                                 mlog_errno(status);
 120                                 goto bail;
 121                         }
 122                 }
 123                 bh = bhs[i];
 124
 125                 if (buffer_jbd(bh)) {
 126                         trace_ocfs2_read_blocks_sync_jbd(
 127                                         (unsigned long long)bh->b_blocknr);
 128                         continue;
 129                 }
 130
 131                 if (buffer_dirty(bh)) {
 132                         /* This should probably be a BUG, or
 133                          * at least return an error. */
 134                         mlog(ML_ERROR,
 135                              "trying to sync read a dirty "
 136                              "buffer! (blocknr = %llu), skipping\n",
 137                              (unsigned long long)bh->b_blocknr);
 138                         continue;
 139                 }
 140
 141                 lock_buffer(bh);
 142                 if (buffer_jbd(bh)) {
 143 #ifdef CATCH_BH_JBD_RACES
 144                         mlog(ML_ERROR,
 145                              "block %llu had the JBD bit set "
 146                              "while I was in lock_buffer!",
 147                              (unsigned long long)bh->b_blocknr);
 148                         BUG();
 149 #else
 150                         unlock_buffer(bh);
 151                         continue;
 152 #endif
 153                 }
 154
 155                 clear_buffer_uptodate(bh);
 156                 get_bh(bh); /* for end_buffer_read_sync() */
 157                 bh->b_end_io = end_buffer_read_sync;
 158                 submit_bh(REQ_OP_READ, 0, bh);
 159         }
 160
 161         for (i = nr; i > 0; i--) {
 162                 bh = bhs[i - 1];
 163
 164                 /* No need to wait on the buffer if it's managed by JBD. */
 165                 if (!buffer_jbd(bh))
 166                         wait_on_buffer(bh);
 167
 168                 if (!buffer_uptodate(bh)) {
 169                         /* Status won't be cleared from here on out,
 170                          * so we can safely record this and loop back
 171                          * to cleanup the other buffers. */
 172                         status = -EIO;
 173                         put_bh(bh);
 174                         bhs[i - 1] = NULL;
 175                 }
 176         }
 177
 178 bail:
 179         return status;
 180 }
 181
 182 int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
 183                       struct buffer_head *bhs[], int flags,
 184                       int (*validate)(struct super_block *sb,
 185                                       struct buffer_head *bh))
 186 {
 187         int status = 0;
 188         int i, ignore_cache = 0;
 189         struct buffer_head *bh;
 190         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 191
 192         trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
 193
 194         BUG_ON(!ci);
 195         BUG_ON((flags & OCFS2_BH_READAHEAD) &&
 196                (flags & OCFS2_BH_IGNORE_CACHE));
 197
 198         if (bhs == NULL) {
 199                 status = -EINVAL;
 200                 mlog_errno(status);
 201                 goto bail;
 202         }
 203
 204         if (nr < 0) {
 205                 mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
 206                 status = -EINVAL;
 207                 mlog_errno(status);
 208                 goto bail;
 209         }
 210
 211         if (nr == 0) {
 212                 status = 0;
 213                 goto bail;
 214         }
 215
 216         ocfs2_metadata_cache_io_lock(ci);
 217         for (i = 0 ; i < nr ; i++) {
 218                 if (bhs[i] == NULL) {
 219                         bhs[i] = sb_getblk(sb, block++);
 220                         if (bhs[i] == NULL) {
 221                                 ocfs2_metadata_cache_io_unlock(ci);
 222                                 status = -ENOMEM;
 223                                 mlog_errno(status);
 224                                 goto bail;
 225                         }
 226                 }
 227                 bh = bhs[i];
 228                 ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
 229
 230                 /* There are three read-ahead cases here which we need to
 231                  * be concerned with. All three assume a buffer has
 232                  * previously been submitted with OCFS2_BH_READAHEAD
 233                  * and it hasn't yet completed I/O.
 234                  *
 235                  * 1) The current request is sync to disk. This rarely
 236                  *    happens these days, and never when performance
 237                  *    matters - the code can just wait on the buffer
 238                  *    lock and re-submit.
 239                  *
 240                  * 2) The current request is cached, but not
 241                  *    readahead. ocfs2_buffer_uptodate() will return
 242                  *    false anyway, so we'll wind up waiting on the
 243                  *    buffer lock to do I/O. We re-check the request
 244                  *    with after getting the lock to avoid a re-submit.
 245                  *
 246                  * 3) The current request is readahead (and so must
 247                  *    also be a caching one). We short circuit if the
 248                  *    buffer is locked (under I/O) and if it's in the
 249                  *    uptodate cache. The re-check from #2 catches the
 250                  *    case that the previous read-ahead completes just
 251                  *    before our is-it-in-flight check.
 252                  */
 253
 254                 if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
 255                         trace_ocfs2_read_blocks_from_disk(
 256                              (unsigned long long)bh->b_blocknr,
 257                              (unsigned long long)ocfs2_metadata_cache_owner(ci));
 258                         /* We're using ignore_cache here to say
 259                          * "go to disk" */
 260                         ignore_cache = 1;
 261                 }
 262
 263                 trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
 264                         ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
 265
 266                 if (buffer_jbd(bh)) {
 267                         continue;
 268                 }
 269
 270                 if (ignore_cache) {
 271                         if (buffer_dirty(bh)) {
 272                                 /* This should probably be a BUG, or
 273                                  * at least return an error. */
 274                                 continue;
 275                         }
 276
 277                         /* A read-ahead request was made - if the
 278                          * buffer is already under read-ahead from a
 279                          * previously submitted request than we are
 280                          * done here. */
 281                         if ((flags & OCFS2_BH_READAHEAD)
 282                             && ocfs2_buffer_read_ahead(ci, bh))
 283                                 continue;
 284
 285                         lock_buffer(bh);
 286                         if (buffer_jbd(bh)) {
 287 #ifdef CATCH_BH_JBD_RACES
 288                                 mlog(ML_ERROR, "block %llu had the JBD bit set "
 289                                                "while I was in lock_buffer!",
 290                                      (unsigned long long)bh->b_blocknr);
 291                                 BUG();
 292 #else
 293                                 unlock_buffer(bh);
 294                                 continue;
 295 #endif
 296                         }
 297
 298                         /* Re-check ocfs2_buffer_uptodate() as a
 299                          * previously read-ahead buffer may have
 300                          * completed I/O while we were waiting for the
 301                          * buffer lock. */
 302                         if (!(flags & OCFS2_BH_IGNORE_CACHE)
 303                             && !(flags & OCFS2_BH_READAHEAD)
 304                             && ocfs2_buffer_uptodate(ci, bh)) {
 305                                 unlock_buffer(bh);
 306                                 continue;
 307                         }
 308
 309                         clear_buffer_uptodate(bh);
 310                         get_bh(bh); /* for end_buffer_read_sync() */
 311                         if (validate)
 312                                 set_buffer_needs_validate(bh);
 313                         bh->b_end_io = end_buffer_read_sync;
 314                         submit_bh(REQ_OP_READ, 0, bh);
 315                         continue;
 316                 }
 317         }
 318
 319         status = 0;
 320
 321         for (i = (nr - 1); i >= 0; i--) {
 322                 bh = bhs[i];
 323
 324                 if (!(flags & OCFS2_BH_READAHEAD)) {
 325                         if (status) {
 326                                 /* Clear the rest of the buffers on error */
 327                                 put_bh(bh);
 328                                 bhs[i] = NULL;
 329                                 continue;
 330                         }
 331                         /* We know this can't have changed as we hold the
 332                          * owner sem. Avoid doing any work on the bh if the
 333                          * journal has it. */
 334                         if (!buffer_jbd(bh))
 335                                 wait_on_buffer(bh);
 336
 337                         if (!buffer_uptodate(bh)) {
 338                                 /* Status won't be cleared from here on out,
 339                                  * so we can safely record this and loop back
 340                                  * to cleanup the other buffers. Don't need to
 341                                  * remove the clustered uptodate information
 342                                  * for this bh as it's not marked locally
 343                                  * uptodate. */
 344                                 status = -EIO;
 345                                 put_bh(bh);
 346                                 bhs[i] = NULL;
 347                                 continue;
 348                         }
 349
 350                         if (buffer_needs_validate(bh)) {
 351                                 /* We never set NeedsValidate if the
 352                                  * buffer was held by the journal, so
 353                                  * that better not have changed */
 354                                 BUG_ON(buffer_jbd(bh));
 355                                 clear_buffer_needs_validate(bh);
 356                                 status = validate(sb, bh);
 357                                 if (status) {
 358                                         put_bh(bh);
 359                                         bhs[i] = NULL;
 360                                         continue;
 361                                 }
 362                         }
 363                 }
 364
 365                 /* Always set the buffer in the cache, even if it was
 366                  * a forced read, or read-ahead which hasn't yet
 367                  * completed. */
 368                 ocfs2_set_buffer_uptodate(ci, bh);
 369         }
 370         ocfs2_metadata_cache_io_unlock(ci);
 371
 372         trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
 373                                     flags, ignore_cache);
 374
 375 bail:
 376
 377         return status;
 378 }
 379
 380 /* Check whether the blkno is the super block or one of the backups. */
 381 static void ocfs2_check_super_or_backup(struct super_block *sb,
 382                                         sector_t blkno)
 383 {
 384         int i;
 385         u64 backup_blkno;
 386
 387         if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
 388                 return;
 389
 390         for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
 391                 backup_blkno = ocfs2_backup_super_blkno(sb, i);
 392                 if (backup_blkno == blkno)
 393                         return;
 394         }
 395
 396         BUG();
 397 }
 398
 399 /*
 400  * Write super block and backups doesn't need to collaborate with journal,
 401  * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
 402  * into this function.
 403  */
 404 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
 405                                 struct buffer_head *bh)
 406 {
 407         int ret = 0;
 408         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
 409
 410         BUG_ON(buffer_jbd(bh));
 411         ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
 412
 413         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
 414                 ret = -EROFS;
 415                 mlog_errno(ret);
 416                 goto out;
 417         }
 418
 419         lock_buffer(bh);
 420         set_buffer_uptodate(bh);
 421
 422         /* remove from dirty list before I/O. */
 423         clear_buffer_dirty(bh);
 424
 425         get_bh(bh); /* for end_buffer_write_sync() */
 426         bh->b_end_io = end_buffer_write_sync;
 427         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
 428         submit_bh(REQ_OP_WRITE, 0, bh);
 429
 430         wait_on_buffer(bh);
 431
 432         if (!buffer_uptodate(bh)) {
 433                 ret = -EIO;
 434                 mlog_errno(ret);
 435         }
 436
 437 out:
 438         return ret;
 439 }