David Teigland | b3b94fa | 2006-01-16 16:50:04 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. |
| 3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. |
| 4 | * |
| 5 | * This copyrighted material is made available to anyone wishing to use, |
| 6 | * modify, copy, or redistribute it subject to the terms and conditions |
| 7 | * of the GNU General Public License v.2. |
| 8 | */ |
| 9 | |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/slab.h> |
| 12 | #include <linux/spinlock.h> |
| 13 | #include <linux/completion.h> |
| 14 | #include <linux/buffer_head.h> |
| 15 | #include <linux/pagemap.h> |
| 16 | #include <asm/semaphore.h> |
| 17 | |
| 18 | #include "gfs2.h" |
| 19 | #include "bmap.h" |
| 20 | #include "glock.h" |
| 21 | #include "inode.h" |
| 22 | #include "jdata.h" |
| 23 | #include "log.h" |
| 24 | #include "meta_io.h" |
| 25 | #include "ops_address.h" |
| 26 | #include "page.h" |
| 27 | #include "quota.h" |
| 28 | #include "trans.h" |
| 29 | |
| 30 | /** |
| 31 | * get_block - Fills in a buffer head with details about a block |
| 32 | * @inode: The inode |
| 33 | * @lblock: The block number to look up |
| 34 | * @bh_result: The buffer head to return the result in |
| 35 | * @create: Non-zero if we may add block to the file |
| 36 | * |
| 37 | * Returns: errno |
| 38 | */ |
| 39 | |
| 40 | static int get_block(struct inode *inode, sector_t lblock, |
| 41 | struct buffer_head *bh_result, int create) |
| 42 | { |
| 43 | struct gfs2_inode *ip = get_v2ip(inode); |
| 44 | int new = create; |
| 45 | uint64_t dblock; |
| 46 | int error; |
| 47 | |
| 48 | error = gfs2_block_map(ip, lblock, &new, &dblock, NULL); |
| 49 | if (error) |
| 50 | return error; |
| 51 | |
| 52 | if (!dblock) |
| 53 | return 0; |
| 54 | |
| 55 | map_bh(bh_result, inode->i_sb, dblock); |
| 56 | if (new) |
| 57 | set_buffer_new(bh_result); |
| 58 | |
| 59 | return 0; |
| 60 | } |
| 61 | |
| 62 | /** |
| 63 | * get_block_noalloc - Fills in a buffer head with details about a block |
| 64 | * @inode: The inode |
| 65 | * @lblock: The block number to look up |
| 66 | * @bh_result: The buffer head to return the result in |
| 67 | * @create: Non-zero if we may add block to the file |
| 68 | * |
| 69 | * Returns: errno |
| 70 | */ |
| 71 | |
| 72 | static int get_block_noalloc(struct inode *inode, sector_t lblock, |
| 73 | struct buffer_head *bh_result, int create) |
| 74 | { |
| 75 | struct gfs2_inode *ip = get_v2ip(inode); |
| 76 | int new = 0; |
| 77 | uint64_t dblock; |
| 78 | int error; |
| 79 | |
| 80 | error = gfs2_block_map(ip, lblock, &new, &dblock, NULL); |
| 81 | if (error) |
| 82 | return error; |
| 83 | |
| 84 | if (dblock) |
| 85 | map_bh(bh_result, inode->i_sb, dblock); |
| 86 | else if (gfs2_assert_withdraw(ip->i_sbd, !create)) |
| 87 | error = -EIO; |
| 88 | |
| 89 | return error; |
| 90 | } |
| 91 | |
| 92 | static int get_blocks(struct inode *inode, sector_t lblock, |
| 93 | unsigned long max_blocks, struct buffer_head *bh_result, |
| 94 | int create) |
| 95 | { |
| 96 | struct gfs2_inode *ip = get_v2ip(inode); |
| 97 | int new = create; |
| 98 | uint64_t dblock; |
| 99 | uint32_t extlen; |
| 100 | int error; |
| 101 | |
| 102 | error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen); |
| 103 | if (error) |
| 104 | return error; |
| 105 | |
| 106 | if (!dblock) |
| 107 | return 0; |
| 108 | |
| 109 | map_bh(bh_result, inode->i_sb, dblock); |
| 110 | if (new) |
| 111 | set_buffer_new(bh_result); |
| 112 | |
| 113 | if (extlen > max_blocks) |
| 114 | extlen = max_blocks; |
| 115 | bh_result->b_size = extlen << inode->i_blkbits; |
| 116 | |
| 117 | return 0; |
| 118 | } |
| 119 | |
| 120 | static int get_blocks_noalloc(struct inode *inode, sector_t lblock, |
| 121 | unsigned long max_blocks, |
| 122 | struct buffer_head *bh_result, int create) |
| 123 | { |
| 124 | struct gfs2_inode *ip = get_v2ip(inode); |
| 125 | int new = 0; |
| 126 | uint64_t dblock; |
| 127 | uint32_t extlen; |
| 128 | int error; |
| 129 | |
| 130 | error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen); |
| 131 | if (error) |
| 132 | return error; |
| 133 | |
| 134 | if (dblock) { |
| 135 | map_bh(bh_result, inode->i_sb, dblock); |
| 136 | if (extlen > max_blocks) |
| 137 | extlen = max_blocks; |
| 138 | bh_result->b_size = extlen << inode->i_blkbits; |
| 139 | } else if (gfs2_assert_withdraw(ip->i_sbd, !create)) |
| 140 | error = -EIO; |
| 141 | |
| 142 | return error; |
| 143 | } |
| 144 | |
| 145 | /** |
| 146 | * gfs2_writepage - Write complete page |
| 147 | * @page: Page to write |
| 148 | * |
| 149 | * Returns: errno |
| 150 | * |
| 151 | * Use Linux VFS block_write_full_page() to write one page, |
| 152 | * using GFS2's get_block_noalloc to find which blocks to write. |
| 153 | */ |
| 154 | |
| 155 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) |
| 156 | { |
| 157 | struct gfs2_inode *ip = get_v2ip(page->mapping->host); |
| 158 | struct gfs2_sbd *sdp = ip->i_sbd; |
| 159 | int error; |
| 160 | |
| 161 | atomic_inc(&sdp->sd_ops_address); |
| 162 | |
| 163 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { |
| 164 | unlock_page(page); |
| 165 | return -EIO; |
| 166 | } |
| 167 | if (get_transaction) { |
| 168 | redirty_page_for_writepage(wbc, page); |
| 169 | unlock_page(page); |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | error = block_write_full_page(page, get_block_noalloc, wbc); |
| 174 | |
| 175 | gfs2_meta_cache_flush(ip); |
| 176 | |
| 177 | return error; |
| 178 | } |
| 179 | |
| 180 | /** |
| 181 | * stuffed_readpage - Fill in a Linux page with stuffed file data |
| 182 | * @ip: the inode |
| 183 | * @page: the page |
| 184 | * |
| 185 | * Returns: errno |
| 186 | */ |
| 187 | |
| 188 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) |
| 189 | { |
| 190 | struct buffer_head *dibh; |
| 191 | void *kaddr; |
| 192 | int error; |
| 193 | |
| 194 | error = gfs2_meta_inode_buffer(ip, &dibh); |
| 195 | if (error) |
| 196 | return error; |
| 197 | |
| 198 | kaddr = kmap(page); |
| 199 | memcpy((char *)kaddr, |
| 200 | dibh->b_data + sizeof(struct gfs2_dinode), |
| 201 | ip->i_di.di_size); |
| 202 | memset((char *)kaddr + ip->i_di.di_size, |
| 203 | 0, |
| 204 | PAGE_CACHE_SIZE - ip->i_di.di_size); |
| 205 | kunmap(page); |
| 206 | |
| 207 | brelse(dibh); |
| 208 | |
| 209 | SetPageUptodate(page); |
| 210 | |
| 211 | return 0; |
| 212 | } |
| 213 | |
| 214 | static int zero_readpage(struct page *page) |
| 215 | { |
| 216 | void *kaddr; |
| 217 | |
| 218 | kaddr = kmap(page); |
| 219 | memset(kaddr, 0, PAGE_CACHE_SIZE); |
| 220 | kunmap(page); |
| 221 | |
| 222 | SetPageUptodate(page); |
| 223 | unlock_page(page); |
| 224 | |
| 225 | return 0; |
| 226 | } |
| 227 | |
| 228 | /** |
| 229 | * jdata_readpage - readpage that goes through gfs2_jdata_read_mem() |
| 230 | * @ip: |
| 231 | * @page: The page to read |
| 232 | * |
| 233 | * Returns: errno |
| 234 | */ |
| 235 | |
| 236 | static int jdata_readpage(struct gfs2_inode *ip, struct page *page) |
| 237 | { |
| 238 | void *kaddr; |
| 239 | int ret; |
| 240 | |
| 241 | kaddr = kmap(page); |
| 242 | |
| 243 | ret = gfs2_jdata_read_mem(ip, kaddr, |
| 244 | (uint64_t)page->index << PAGE_CACHE_SHIFT, |
| 245 | PAGE_CACHE_SIZE); |
| 246 | if (ret >= 0) { |
| 247 | if (ret < PAGE_CACHE_SIZE) |
| 248 | memset(kaddr + ret, 0, PAGE_CACHE_SIZE - ret); |
| 249 | SetPageUptodate(page); |
| 250 | ret = 0; |
| 251 | } |
| 252 | |
| 253 | kunmap(page); |
| 254 | |
| 255 | unlock_page(page); |
| 256 | |
| 257 | return ret; |
| 258 | } |
| 259 | |
| 260 | /** |
| 261 | * gfs2_readpage - readpage with locking |
| 262 | * @file: The file to read a page for |
| 263 | * @page: The page to read |
| 264 | * |
| 265 | * Returns: errno |
| 266 | */ |
| 267 | |
| 268 | static int gfs2_readpage(struct file *file, struct page *page) |
| 269 | { |
| 270 | struct gfs2_inode *ip = get_v2ip(page->mapping->host); |
| 271 | struct gfs2_sbd *sdp = ip->i_sbd; |
| 272 | int error; |
| 273 | |
| 274 | atomic_inc(&sdp->sd_ops_address); |
| 275 | |
| 276 | if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) { |
| 277 | unlock_page(page); |
| 278 | return -EOPNOTSUPP; |
| 279 | } |
| 280 | |
| 281 | if (!gfs2_is_jdata(ip)) { |
| 282 | if (gfs2_is_stuffed(ip)) { |
| 283 | if (!page->index) { |
| 284 | error = stuffed_readpage(ip, page); |
| 285 | unlock_page(page); |
| 286 | } else |
| 287 | error = zero_readpage(page); |
| 288 | } else |
| 289 | error = block_read_full_page(page, get_block); |
| 290 | } else |
| 291 | error = jdata_readpage(ip, page); |
| 292 | |
| 293 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
| 294 | error = -EIO; |
| 295 | |
| 296 | return error; |
| 297 | } |
| 298 | |
| 299 | /** |
| 300 | * gfs2_prepare_write - Prepare to write a page to a file |
| 301 | * @file: The file to write to |
| 302 | * @page: The page which is to be prepared for writing |
| 303 | * @from: From (byte range within page) |
| 304 | * @to: To (byte range within page) |
| 305 | * |
| 306 | * Returns: errno |
| 307 | */ |
| 308 | |
| 309 | static int gfs2_prepare_write(struct file *file, struct page *page, |
| 310 | unsigned from, unsigned to) |
| 311 | { |
| 312 | struct gfs2_inode *ip = get_v2ip(page->mapping->host); |
| 313 | struct gfs2_sbd *sdp = ip->i_sbd; |
| 314 | int error = 0; |
| 315 | |
| 316 | atomic_inc(&sdp->sd_ops_address); |
| 317 | |
| 318 | if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) |
| 319 | return -EOPNOTSUPP; |
| 320 | |
| 321 | if (gfs2_is_stuffed(ip)) { |
| 322 | uint64_t file_size; |
| 323 | file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to; |
| 324 | |
| 325 | if (file_size > sdp->sd_sb.sb_bsize - |
| 326 | sizeof(struct gfs2_dinode)) { |
| 327 | error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_page, |
| 328 | page); |
| 329 | if (!error) |
| 330 | error = block_prepare_write(page, from, to, |
| 331 | get_block); |
| 332 | } else if (!PageUptodate(page)) |
| 333 | error = stuffed_readpage(ip, page); |
| 334 | } else |
| 335 | error = block_prepare_write(page, from, to, get_block); |
| 336 | |
| 337 | return error; |
| 338 | } |
| 339 | |
| 340 | /** |
| 341 | * gfs2_commit_write - Commit write to a file |
| 342 | * @file: The file to write to |
| 343 | * @page: The page containing the data |
| 344 | * @from: From (byte range within page) |
| 345 | * @to: To (byte range within page) |
| 346 | * |
| 347 | * Returns: errno |
| 348 | */ |
| 349 | |
| 350 | static int gfs2_commit_write(struct file *file, struct page *page, |
| 351 | unsigned from, unsigned to) |
| 352 | { |
| 353 | struct inode *inode = page->mapping->host; |
| 354 | struct gfs2_inode *ip = get_v2ip(inode); |
| 355 | struct gfs2_sbd *sdp = ip->i_sbd; |
| 356 | int error; |
| 357 | |
| 358 | atomic_inc(&sdp->sd_ops_address); |
| 359 | |
| 360 | if (gfs2_is_stuffed(ip)) { |
| 361 | struct buffer_head *dibh; |
| 362 | uint64_t file_size; |
| 363 | void *kaddr; |
| 364 | |
| 365 | file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to; |
| 366 | |
| 367 | error = gfs2_meta_inode_buffer(ip, &dibh); |
| 368 | if (error) |
| 369 | goto fail; |
| 370 | |
Steven Whitehouse | d4e9c4c | 2006-01-18 11:19:28 +0000 | [diff] [blame] | 371 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
David Teigland | b3b94fa | 2006-01-16 16:50:04 +0000 | [diff] [blame] | 372 | |
| 373 | kaddr = kmap(page); |
| 374 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, |
| 375 | (char *)kaddr + from, |
| 376 | to - from); |
| 377 | kunmap(page); |
| 378 | |
| 379 | brelse(dibh); |
| 380 | |
| 381 | SetPageUptodate(page); |
| 382 | |
| 383 | if (inode->i_size < file_size) |
| 384 | i_size_write(inode, file_size); |
| 385 | } else { |
| 386 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED) |
| 387 | gfs2_page_add_databufs(sdp, page, from, to); |
| 388 | error = generic_commit_write(file, page, from, to); |
| 389 | if (error) |
| 390 | goto fail; |
| 391 | } |
| 392 | |
| 393 | return 0; |
| 394 | |
| 395 | fail: |
| 396 | ClearPageUptodate(page); |
| 397 | |
| 398 | return error; |
| 399 | } |
| 400 | |
| 401 | /** |
| 402 | * gfs2_bmap - Block map function |
| 403 | * @mapping: Address space info |
| 404 | * @lblock: The block to map |
| 405 | * |
| 406 | * Returns: The disk address for the block or 0 on hole or error |
| 407 | */ |
| 408 | |
| 409 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) |
| 410 | { |
| 411 | struct gfs2_inode *ip = get_v2ip(mapping->host); |
| 412 | struct gfs2_holder i_gh; |
| 413 | sector_t dblock = 0; |
| 414 | int error; |
| 415 | |
| 416 | atomic_inc(&ip->i_sbd->sd_ops_address); |
| 417 | |
| 418 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
| 419 | if (error) |
| 420 | return 0; |
| 421 | |
| 422 | if (!gfs2_is_stuffed(ip)) |
| 423 | dblock = generic_block_bmap(mapping, lblock, get_block); |
| 424 | |
| 425 | gfs2_glock_dq_uninit(&i_gh); |
| 426 | |
| 427 | return dblock; |
| 428 | } |
| 429 | |
| 430 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) |
| 431 | { |
Steven Whitehouse | 64fb4eb | 2006-01-18 13:14:40 +0000 | [diff] [blame^] | 432 | struct gfs2_bufdata *bd; |
David Teigland | b3b94fa | 2006-01-16 16:50:04 +0000 | [diff] [blame] | 433 | |
| 434 | gfs2_log_lock(sdp); |
Steven Whitehouse | 64fb4eb | 2006-01-18 13:14:40 +0000 | [diff] [blame^] | 435 | bd = get_v2bd(bh); |
| 436 | if (bd) { |
| 437 | bd->bd_bh = NULL; |
| 438 | set_v2bd(bh, NULL); |
David Teigland | b3b94fa | 2006-01-16 16:50:04 +0000 | [diff] [blame] | 439 | gfs2_log_unlock(sdp); |
| 440 | brelse(bh); |
| 441 | } else |
| 442 | gfs2_log_unlock(sdp); |
| 443 | |
| 444 | lock_buffer(bh); |
| 445 | clear_buffer_dirty(bh); |
| 446 | bh->b_bdev = NULL; |
| 447 | clear_buffer_mapped(bh); |
| 448 | clear_buffer_req(bh); |
| 449 | clear_buffer_new(bh); |
| 450 | clear_buffer_delay(bh); |
| 451 | unlock_buffer(bh); |
| 452 | } |
| 453 | |
| 454 | static int gfs2_invalidatepage(struct page *page, unsigned long offset) |
| 455 | { |
| 456 | struct gfs2_sbd *sdp = get_v2sdp(page->mapping->host->i_sb); |
| 457 | struct buffer_head *head, *bh, *next; |
| 458 | unsigned int curr_off = 0; |
| 459 | int ret = 1; |
| 460 | |
| 461 | BUG_ON(!PageLocked(page)); |
| 462 | if (!page_has_buffers(page)) |
| 463 | return 1; |
| 464 | |
| 465 | bh = head = page_buffers(page); |
| 466 | do { |
| 467 | unsigned int next_off = curr_off + bh->b_size; |
| 468 | next = bh->b_this_page; |
| 469 | |
| 470 | if (offset <= curr_off) |
| 471 | discard_buffer(sdp, bh); |
| 472 | |
| 473 | curr_off = next_off; |
| 474 | bh = next; |
| 475 | } while (bh != head); |
| 476 | |
| 477 | if (!offset) |
| 478 | ret = try_to_release_page(page, 0); |
| 479 | |
| 480 | return ret; |
| 481 | } |
| 482 | |
| 483 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, |
| 484 | loff_t offset, unsigned long nr_segs) |
| 485 | { |
| 486 | struct file *file = iocb->ki_filp; |
| 487 | struct inode *inode = file->f_mapping->host; |
| 488 | struct gfs2_inode *ip = get_v2ip(inode); |
| 489 | struct gfs2_sbd *sdp = ip->i_sbd; |
| 490 | get_blocks_t *gb = get_blocks; |
| 491 | |
| 492 | atomic_inc(&sdp->sd_ops_address); |
| 493 | |
| 494 | if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)) || |
| 495 | gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip))) |
| 496 | return -EINVAL; |
| 497 | |
| 498 | if (rw == WRITE && !get_transaction) |
| 499 | gb = get_blocks_noalloc; |
| 500 | |
| 501 | return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, |
| 502 | offset, nr_segs, gb, NULL); |
| 503 | } |
| 504 | |
| 505 | struct address_space_operations gfs2_file_aops = { |
| 506 | .writepage = gfs2_writepage, |
| 507 | .readpage = gfs2_readpage, |
| 508 | .sync_page = block_sync_page, |
| 509 | .prepare_write = gfs2_prepare_write, |
| 510 | .commit_write = gfs2_commit_write, |
| 511 | .bmap = gfs2_bmap, |
| 512 | .invalidatepage = gfs2_invalidatepage, |
| 513 | .direct_IO = gfs2_direct_IO, |
| 514 | }; |
| 515 | |