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 30#include <cluster/masklog.h> 31 32#include "ocfs2.h" 33 34#include "alloc.h" 35#include "inode.h" 36#include "journal.h" 37#include "uptodate.h" 38 39#include "buffer_head_io.h" 40 41/* 42 * Bits on bh->b_state used by ocfs2. 43 * 44 * These MUST be after the JBD2 bits. Hence, we use BH_JBDPrivateStart. 45 */ 46enum ocfs2_state_bits { 47 BH_NeedsValidate = BH_JBDPrivateStart, 48}; 49 50/* Expand the magic b_state functions */ 51BUFFER_FNS(NeedsValidate, needs_validate); 52 53int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh, 54 struct ocfs2_caching_info *ci) 55{ 56 int ret = 0; 57 58 mlog_entry("(bh->b_blocknr = %llu, ci=%p)\n", 59 (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 goto out; 70 } 71 72 ocfs2_metadata_cache_io_lock(ci); 73 74 lock_buffer(bh); 75 set_buffer_uptodate(bh); 76 77 /* remove from dirty list before I/O. */ 78 clear_buffer_dirty(bh); 79 80 get_bh(bh); /* for end_buffer_write_sync() */ 81 bh->b_end_io = end_buffer_write_sync; 82 submit_bh(WRITE, bh); 83 84 wait_on_buffer(bh); 85 86 if (buffer_uptodate(bh)) { 87 ocfs2_set_buffer_uptodate(ci, bh); 88 } else { 89 /* We don't need to remove the clustered uptodate 90 * information for this bh as it's not marked locally 91 * uptodate. */ 92 ret = -EIO; 93 put_bh(bh); 94 } 95 96 ocfs2_metadata_cache_io_unlock(ci); 97out: 98 mlog_exit(ret); 99 return ret; 100} 101 102int 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 if (!nr) { 110 mlog(ML_BH_IO, "No buffers will be read!\n"); 111 goto bail; 112 } 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 = -EIO; 119 mlog_errno(status); 120 goto bail; 121 } 122 } 123 bh = bhs[i]; 124 125 if (buffer_jbd(bh)) { 126 mlog(ML_BH_IO, 127 "trying to sync read a jbd " 128 "managed bh (blocknr = %llu), skipping\n", 129 (unsigned long long)bh->b_blocknr); 130 continue; 131 } 132 133 if (buffer_dirty(bh)) { 134 /* This should probably be a BUG, or 135 * at least return an error. */ 136 mlog(ML_ERROR, 137 "trying to sync read a dirty " 138 "buffer! (blocknr = %llu), skipping\n", 139 (unsigned long long)bh->b_blocknr); 140 continue; 141 } 142 143 lock_buffer(bh); 144 if (buffer_jbd(bh)) { 145 mlog(ML_ERROR, 146 "block %llu had the JBD bit set " 147 "while I was in lock_buffer!", 148 (unsigned long long)bh->b_blocknr); 149 BUG(); 150 } 151 152 clear_buffer_uptodate(bh); 153 get_bh(bh); /* for end_buffer_read_sync() */ 154 bh->b_end_io = end_buffer_read_sync; 155 submit_bh(READ, bh); 156 } 157 158 for (i = nr; i > 0; i--) { 159 bh = bhs[i - 1]; 160 161 /* No need to wait on the buffer if it's managed by JBD. */ 162 if (!buffer_jbd(bh)) 163 wait_on_buffer(bh); 164 165 if (!buffer_uptodate(bh)) { 166 /* Status won't be cleared from here on out, 167 * so we can safely record this and loop back 168 * to cleanup the other buffers. */ 169 status = -EIO; 170 put_bh(bh); 171 bhs[i - 1] = NULL; 172 } 173 } 174 175bail: 176 return status; 177} 178 179int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr, 180 struct buffer_head *bhs[], int flags, 181 int (*validate)(struct super_block *sb, 182 struct buffer_head *bh)) 183{ 184 int status = 0; 185 int i, ignore_cache = 0; 186 struct buffer_head *bh; 187 struct super_block *sb = ocfs2_metadata_cache_get_super(ci); 188 189 mlog_entry("(ci=%p, block=(%llu), nr=(%d), flags=%d)\n", 190 ci, (unsigned long long)block, nr, flags); 191 192 BUG_ON(!ci); 193 BUG_ON((flags & OCFS2_BH_READAHEAD) && 194 (flags & OCFS2_BH_IGNORE_CACHE)); 195 196 if (bhs == NULL) { 197 status = -EINVAL; 198 mlog_errno(status); 199 goto bail; 200 } 201 202 if (nr < 0) { 203 mlog(ML_ERROR, "asked to read %d blocks!\n", nr); 204 status = -EINVAL; 205 mlog_errno(status); 206 goto bail; 207 } 208 209 if (nr == 0) { 210 mlog(ML_BH_IO, "No buffers will be read!\n"); 211 status = 0; 212 goto bail; 213 } 214 215 ocfs2_metadata_cache_io_lock(ci); 216 for (i = 0 ; i < nr ; i++) { 217 if (bhs[i] == NULL) { 218 bhs[i] = sb_getblk(sb, block++); 219 if (bhs[i] == NULL) { 220 ocfs2_metadata_cache_io_unlock(ci); 221 status = -EIO; 222 mlog_errno(status); 223 goto bail; 224 } 225 } 226 bh = bhs[i]; 227 ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE); 228 229 /* There are three read-ahead cases here which we need to 230 * be concerned with. All three assume a buffer has 231 * previously been submitted with OCFS2_BH_READAHEAD 232 * and it hasn't yet completed I/O. 233 * 234 * 1) The current request is sync to disk. This rarely 235 * happens these days, and never when performance 236 * matters - the code can just wait on the buffer 237 * lock and re-submit. 238 * 239 * 2) The current request is cached, but not 240 * readahead. ocfs2_buffer_uptodate() will return 241 * false anyway, so we'll wind up waiting on the 242 * buffer lock to do I/O. We re-check the request 243 * with after getting the lock to avoid a re-submit. 244 * 245 * 3) The current request is readahead (and so must 246 * also be a caching one). We short circuit if the 247 * buffer is locked (under I/O) and if it's in the 248 * uptodate cache. The re-check from #2 catches the 249 * case that the previous read-ahead completes just 250 * before our is-it-in-flight check. 251 */ 252 253 if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) { 254 mlog(ML_UPTODATE, 255 "bh (%llu), owner %llu not uptodate\n", 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 if (buffer_jbd(bh)) { 264 if (ignore_cache) 265 mlog(ML_BH_IO, "trying to sync read a jbd " 266 "managed bh (blocknr = %llu)\n", 267 (unsigned long long)bh->b_blocknr); 268 continue; 269 } 270 271 if (ignore_cache) { 272 if (buffer_dirty(bh)) { 273 /* This should probably be a BUG, or 274 * at least return an error. */ 275 mlog(ML_BH_IO, "asking me to sync read a dirty " 276 "buffer! (blocknr = %llu)\n", 277 (unsigned long long)bh->b_blocknr); 278 continue; 279 } 280 281 /* A read-ahead request was made - if the 282 * buffer is already under read-ahead from a 283 * previously submitted request than we are 284 * done here. */ 285 if ((flags & OCFS2_BH_READAHEAD) 286 && ocfs2_buffer_read_ahead(ci, bh)) 287 continue; 288 289 lock_buffer(bh); 290 if (buffer_jbd(bh)) { 291#ifdef CATCH_BH_JBD_RACES 292 mlog(ML_ERROR, "block %llu had the JBD bit set " 293 "while I was in lock_buffer!", 294 (unsigned long long)bh->b_blocknr); 295 BUG(); 296#else 297 unlock_buffer(bh); 298 continue; 299#endif 300 } 301 302 /* Re-check ocfs2_buffer_uptodate() as a 303 * previously read-ahead buffer may have 304 * completed I/O while we were waiting for the 305 * buffer lock. */ 306 if (!(flags & OCFS2_BH_IGNORE_CACHE) 307 && !(flags & OCFS2_BH_READAHEAD) 308 && ocfs2_buffer_uptodate(ci, bh)) { 309 unlock_buffer(bh); 310 continue; 311 } 312 313 clear_buffer_uptodate(bh); 314 get_bh(bh); /* for end_buffer_read_sync() */ 315 if (validate) 316 set_buffer_needs_validate(bh); 317 bh->b_end_io = end_buffer_read_sync; 318 submit_bh(READ, bh); 319 continue; 320 } 321 } 322 323 status = 0; 324 325 for (i = (nr - 1); i >= 0; i--) { 326 bh = bhs[i]; 327 328 if (!(flags & OCFS2_BH_READAHEAD)) { 329 /* We know this can't have changed as we hold the 330 * owner sem. Avoid doing any work on the bh if the 331 * journal has it. */ 332 if (!buffer_jbd(bh)) 333 wait_on_buffer(bh); 334 335 if (!buffer_uptodate(bh)) { 336 /* Status won't be cleared from here on out, 337 * so we can safely record this and loop back 338 * to cleanup the other buffers. Don't need to 339 * remove the clustered uptodate information 340 * for this bh as it's not marked locally 341 * uptodate. */ 342 status = -EIO; 343 put_bh(bh); 344 bhs[i] = NULL; 345 continue; 346 } 347 348 if (buffer_needs_validate(bh)) { 349 /* We never set NeedsValidate if the 350 * buffer was held by the journal, so 351 * that better not have changed */ 352 BUG_ON(buffer_jbd(bh)); 353 clear_buffer_needs_validate(bh); 354 status = validate(sb, bh); 355 if (status) { 356 put_bh(bh); 357 bhs[i] = NULL; 358 continue; 359 } 360 } 361 } 362 363 /* Always set the buffer in the cache, even if it was 364 * a forced read, or read-ahead which hasn't yet 365 * completed. */ 366 ocfs2_set_buffer_uptodate(ci, bh); 367 } 368 ocfs2_metadata_cache_io_unlock(ci); 369 370 mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n", 371 (unsigned long long)block, nr, 372 ((flags & OCFS2_BH_IGNORE_CACHE) || ignore_cache) ? "no" : "yes", 373 flags); 374 375bail: 376 377 mlog_exit(status); 378 return status; 379} 380 381/* Check whether the blkno is the super block or one of the backups. */ 382static void ocfs2_check_super_or_backup(struct super_block *sb, 383 sector_t blkno) 384{ 385 int i; 386 u64 backup_blkno; 387 388 if (blkno == OCFS2_SUPER_BLOCK_BLKNO) 389 return; 390 391 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { 392 backup_blkno = ocfs2_backup_super_blkno(sb, i); 393 if (backup_blkno == blkno) 394 return; 395 } 396 397 BUG(); 398} 399 400/* 401 * Write super block and backups doesn't need to collaborate with journal, 402 * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed 403 * into this function. 404 */ 405int ocfs2_write_super_or_backup(struct ocfs2_super *osb, 406 struct buffer_head *bh) 407{ 408 int ret = 0; 409 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; 410 411 mlog_entry_void(); 412 413 BUG_ON(buffer_jbd(bh)); 414 ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr); 415 416 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) { 417 ret = -EROFS; 418 goto out; 419 } 420 421 lock_buffer(bh); 422 set_buffer_uptodate(bh); 423 424 /* remove from dirty list before I/O. */ 425 clear_buffer_dirty(bh); 426 427 get_bh(bh); /* for end_buffer_write_sync() */ 428 bh->b_end_io = end_buffer_write_sync; 429 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check); 430 submit_bh(WRITE, bh); 431 432 wait_on_buffer(bh); 433 434 if (!buffer_uptodate(bh)) { 435 ret = -EIO; 436 put_bh(bh); 437 } 438 439out: 440 mlog_exit(ret); 441 return ret; 442} 443