| 1 | /* |
| 2 | * linux/fs/ext4/page-io.c |
| 3 | * |
| 4 | * This contains the new page_io functions for ext4 |
| 5 | * |
| 6 | * Written by Theodore Ts'o, 2010. |
| 7 | */ |
| 8 | |
| 9 | #include <linux/fs.h> |
| 10 | #include <linux/time.h> |
| 11 | #include <linux/jbd2.h> |
| 12 | #include <linux/highuid.h> |
| 13 | #include <linux/pagemap.h> |
| 14 | #include <linux/quotaops.h> |
| 15 | #include <linux/string.h> |
| 16 | #include <linux/buffer_head.h> |
| 17 | #include <linux/writeback.h> |
| 18 | #include <linux/pagevec.h> |
| 19 | #include <linux/mpage.h> |
| 20 | #include <linux/namei.h> |
| 21 | #include <linux/aio.h> |
| 22 | #include <linux/uio.h> |
| 23 | #include <linux/bio.h> |
| 24 | #include <linux/workqueue.h> |
| 25 | #include <linux/kernel.h> |
| 26 | #include <linux/slab.h> |
| 27 | #include <linux/mm.h> |
| 28 | #include <linux/ratelimit.h> |
| 29 | |
| 30 | #include "ext4_jbd2.h" |
| 31 | #include "xattr.h" |
| 32 | #include "acl.h" |
| 33 | |
| 34 | static struct kmem_cache *io_end_cachep; |
| 35 | |
| 36 | int __init ext4_init_pageio(void) |
| 37 | { |
| 38 | io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT); |
| 39 | if (io_end_cachep == NULL) |
| 40 | return -ENOMEM; |
| 41 | return 0; |
| 42 | } |
| 43 | |
| 44 | void ext4_exit_pageio(void) |
| 45 | { |
| 46 | kmem_cache_destroy(io_end_cachep); |
| 47 | } |
| 48 | |
| 49 | /* |
| 50 | * Print an buffer I/O error compatible with the fs/buffer.c. This |
| 51 | * provides compatibility with dmesg scrapers that look for a specific |
| 52 | * buffer I/O error message. We really need a unified error reporting |
| 53 | * structure to userspace ala Digital Unix's uerf system, but it's |
| 54 | * probably not going to happen in my lifetime, due to LKML politics... |
| 55 | */ |
| 56 | static void buffer_io_error(struct buffer_head *bh) |
| 57 | { |
| 58 | char b[BDEVNAME_SIZE]; |
| 59 | printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", |
| 60 | bdevname(bh->b_bdev, b), |
| 61 | (unsigned long long)bh->b_blocknr); |
| 62 | } |
| 63 | |
| 64 | static void ext4_finish_bio(struct bio *bio) |
| 65 | { |
| 66 | int i; |
| 67 | int error = !test_bit(BIO_UPTODATE, &bio->bi_flags); |
| 68 | |
| 69 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 70 | struct bio_vec *bvec = &bio->bi_io_vec[i]; |
| 71 | struct page *page = bvec->bv_page; |
| 72 | struct buffer_head *bh, *head; |
| 73 | unsigned bio_start = bvec->bv_offset; |
| 74 | unsigned bio_end = bio_start + bvec->bv_len; |
| 75 | unsigned under_io = 0; |
| 76 | unsigned long flags; |
| 77 | |
| 78 | if (!page) |
| 79 | continue; |
| 80 | |
| 81 | if (error) { |
| 82 | SetPageError(page); |
| 83 | set_bit(AS_EIO, &page->mapping->flags); |
| 84 | } |
| 85 | bh = head = page_buffers(page); |
| 86 | /* |
| 87 | * We check all buffers in the page under BH_Uptodate_Lock |
| 88 | * to avoid races with other end io clearing async_write flags |
| 89 | */ |
| 90 | local_irq_save(flags); |
| 91 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); |
| 92 | do { |
| 93 | if (bh_offset(bh) < bio_start || |
| 94 | bh_offset(bh) + bh->b_size > bio_end) { |
| 95 | if (buffer_async_write(bh)) |
| 96 | under_io++; |
| 97 | continue; |
| 98 | } |
| 99 | clear_buffer_async_write(bh); |
| 100 | if (error) |
| 101 | buffer_io_error(bh); |
| 102 | } while ((bh = bh->b_this_page) != head); |
| 103 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); |
| 104 | local_irq_restore(flags); |
| 105 | if (!under_io) |
| 106 | end_page_writeback(page); |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | static void ext4_release_io_end(ext4_io_end_t *io_end) |
| 111 | { |
| 112 | struct bio *bio, *next_bio; |
| 113 | |
| 114 | BUG_ON(!list_empty(&io_end->list)); |
| 115 | BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); |
| 116 | WARN_ON(io_end->handle); |
| 117 | |
| 118 | if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count)) |
| 119 | wake_up_all(ext4_ioend_wq(io_end->inode)); |
| 120 | |
| 121 | for (bio = io_end->bio; bio; bio = next_bio) { |
| 122 | next_bio = bio->bi_private; |
| 123 | ext4_finish_bio(bio); |
| 124 | bio_put(bio); |
| 125 | } |
| 126 | kmem_cache_free(io_end_cachep, io_end); |
| 127 | } |
| 128 | |
| 129 | static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end) |
| 130 | { |
| 131 | struct inode *inode = io_end->inode; |
| 132 | |
| 133 | io_end->flag &= ~EXT4_IO_END_UNWRITTEN; |
| 134 | /* Wake up anyone waiting on unwritten extent conversion */ |
| 135 | if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten)) |
| 136 | wake_up_all(ext4_ioend_wq(inode)); |
| 137 | } |
| 138 | |
| 139 | /* |
| 140 | * Check a range of space and convert unwritten extents to written. Note that |
| 141 | * we are protected from truncate touching same part of extent tree by the |
| 142 | * fact that truncate code waits for all DIO to finish (thus exclusion from |
| 143 | * direct IO is achieved) and also waits for PageWriteback bits. Thus we |
| 144 | * cannot get to ext4_ext_truncate() before all IOs overlapping that range are |
| 145 | * completed (happens from ext4_free_ioend()). |
| 146 | */ |
| 147 | static int ext4_end_io(ext4_io_end_t *io) |
| 148 | { |
| 149 | struct inode *inode = io->inode; |
| 150 | loff_t offset = io->offset; |
| 151 | ssize_t size = io->size; |
| 152 | handle_t *handle = io->handle; |
| 153 | int ret = 0; |
| 154 | |
| 155 | ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," |
| 156 | "list->prev 0x%p\n", |
| 157 | io, inode->i_ino, io->list.next, io->list.prev); |
| 158 | |
| 159 | io->handle = NULL; /* Following call will use up the handle */ |
| 160 | ret = ext4_convert_unwritten_extents(handle, inode, offset, size); |
| 161 | if (ret < 0) { |
| 162 | ext4_msg(inode->i_sb, KERN_EMERG, |
| 163 | "failed to convert unwritten extents to written " |
| 164 | "extents -- potential data loss! " |
| 165 | "(inode %lu, offset %llu, size %zd, error %d)", |
| 166 | inode->i_ino, offset, size, ret); |
| 167 | } |
| 168 | ext4_clear_io_unwritten_flag(io); |
| 169 | ext4_release_io_end(io); |
| 170 | return ret; |
| 171 | } |
| 172 | |
| 173 | static void dump_completed_IO(struct inode *inode, struct list_head *head) |
| 174 | { |
| 175 | #ifdef EXT4FS_DEBUG |
| 176 | struct list_head *cur, *before, *after; |
| 177 | ext4_io_end_t *io, *io0, *io1; |
| 178 | |
| 179 | if (list_empty(head)) |
| 180 | return; |
| 181 | |
| 182 | ext4_debug("Dump inode %lu completed io list\n", inode->i_ino); |
| 183 | list_for_each_entry(io, head, list) { |
| 184 | cur = &io->list; |
| 185 | before = cur->prev; |
| 186 | io0 = container_of(before, ext4_io_end_t, list); |
| 187 | after = cur->next; |
| 188 | io1 = container_of(after, ext4_io_end_t, list); |
| 189 | |
| 190 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", |
| 191 | io, inode->i_ino, io0, io1); |
| 192 | } |
| 193 | #endif |
| 194 | } |
| 195 | |
| 196 | /* Add the io_end to per-inode completed end_io list. */ |
| 197 | static void ext4_add_complete_io(ext4_io_end_t *io_end) |
| 198 | { |
| 199 | struct ext4_inode_info *ei = EXT4_I(io_end->inode); |
| 200 | struct workqueue_struct *wq; |
| 201 | unsigned long flags; |
| 202 | |
| 203 | /* Only reserved conversions from writeback should enter here */ |
| 204 | WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN)); |
| 205 | WARN_ON(!io_end->handle); |
| 206 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
| 207 | wq = EXT4_SB(io_end->inode->i_sb)->rsv_conversion_wq; |
| 208 | if (list_empty(&ei->i_rsv_conversion_list)) |
| 209 | queue_work(wq, &ei->i_rsv_conversion_work); |
| 210 | list_add_tail(&io_end->list, &ei->i_rsv_conversion_list); |
| 211 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
| 212 | } |
| 213 | |
| 214 | static int ext4_do_flush_completed_IO(struct inode *inode, |
| 215 | struct list_head *head) |
| 216 | { |
| 217 | ext4_io_end_t *io; |
| 218 | struct list_head unwritten; |
| 219 | unsigned long flags; |
| 220 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 221 | int err, ret = 0; |
| 222 | |
| 223 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
| 224 | dump_completed_IO(inode, head); |
| 225 | list_replace_init(head, &unwritten); |
| 226 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
| 227 | |
| 228 | while (!list_empty(&unwritten)) { |
| 229 | io = list_entry(unwritten.next, ext4_io_end_t, list); |
| 230 | BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN)); |
| 231 | list_del_init(&io->list); |
| 232 | |
| 233 | err = ext4_end_io(io); |
| 234 | if (unlikely(!ret && err)) |
| 235 | ret = err; |
| 236 | } |
| 237 | return ret; |
| 238 | } |
| 239 | |
| 240 | /* |
| 241 | * work on completed IO, to convert unwritten extents to extents |
| 242 | */ |
| 243 | void ext4_end_io_rsv_work(struct work_struct *work) |
| 244 | { |
| 245 | struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info, |
| 246 | i_rsv_conversion_work); |
| 247 | ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list); |
| 248 | } |
| 249 | |
| 250 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) |
| 251 | { |
| 252 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
| 253 | if (io) { |
| 254 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
| 255 | io->inode = inode; |
| 256 | INIT_LIST_HEAD(&io->list); |
| 257 | atomic_set(&io->count, 1); |
| 258 | } |
| 259 | return io; |
| 260 | } |
| 261 | |
| 262 | void ext4_put_io_end_defer(ext4_io_end_t *io_end) |
| 263 | { |
| 264 | if (atomic_dec_and_test(&io_end->count)) { |
| 265 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) { |
| 266 | ext4_release_io_end(io_end); |
| 267 | return; |
| 268 | } |
| 269 | ext4_add_complete_io(io_end); |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | int ext4_put_io_end(ext4_io_end_t *io_end) |
| 274 | { |
| 275 | int err = 0; |
| 276 | |
| 277 | if (atomic_dec_and_test(&io_end->count)) { |
| 278 | if (io_end->flag & EXT4_IO_END_UNWRITTEN) { |
| 279 | err = ext4_convert_unwritten_extents(io_end->handle, |
| 280 | io_end->inode, io_end->offset, |
| 281 | io_end->size); |
| 282 | io_end->handle = NULL; |
| 283 | ext4_clear_io_unwritten_flag(io_end); |
| 284 | } |
| 285 | ext4_release_io_end(io_end); |
| 286 | } |
| 287 | return err; |
| 288 | } |
| 289 | |
| 290 | ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end) |
| 291 | { |
| 292 | atomic_inc(&io_end->count); |
| 293 | return io_end; |
| 294 | } |
| 295 | |
| 296 | /* BIO completion function for page writeback */ |
| 297 | static void ext4_end_bio(struct bio *bio, int error) |
| 298 | { |
| 299 | ext4_io_end_t *io_end = bio->bi_private; |
| 300 | sector_t bi_sector = bio->bi_sector; |
| 301 | |
| 302 | BUG_ON(!io_end); |
| 303 | bio->bi_end_io = NULL; |
| 304 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
| 305 | error = 0; |
| 306 | |
| 307 | if (error) { |
| 308 | struct inode *inode = io_end->inode; |
| 309 | |
| 310 | ext4_warning(inode->i_sb, "I/O error writing to inode %lu " |
| 311 | "(offset %llu size %ld starting block %llu)", |
| 312 | inode->i_ino, |
| 313 | (unsigned long long) io_end->offset, |
| 314 | (long) io_end->size, |
| 315 | (unsigned long long) |
| 316 | bi_sector >> (inode->i_blkbits - 9)); |
| 317 | } |
| 318 | |
| 319 | if (io_end->flag & EXT4_IO_END_UNWRITTEN) { |
| 320 | /* |
| 321 | * Link bio into list hanging from io_end. We have to do it |
| 322 | * atomically as bio completions can be racing against each |
| 323 | * other. |
| 324 | */ |
| 325 | bio->bi_private = xchg(&io_end->bio, bio); |
| 326 | ext4_put_io_end_defer(io_end); |
| 327 | } else { |
| 328 | /* |
| 329 | * Drop io_end reference early. Inode can get freed once |
| 330 | * we finish the bio. |
| 331 | */ |
| 332 | ext4_put_io_end_defer(io_end); |
| 333 | ext4_finish_bio(bio); |
| 334 | bio_put(bio); |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | void ext4_io_submit(struct ext4_io_submit *io) |
| 339 | { |
| 340 | struct bio *bio = io->io_bio; |
| 341 | |
| 342 | if (bio) { |
| 343 | bio_get(io->io_bio); |
| 344 | submit_bio(io->io_op, io->io_bio); |
| 345 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); |
| 346 | bio_put(io->io_bio); |
| 347 | } |
| 348 | io->io_bio = NULL; |
| 349 | } |
| 350 | |
| 351 | void ext4_io_submit_init(struct ext4_io_submit *io, |
| 352 | struct writeback_control *wbc) |
| 353 | { |
| 354 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); |
| 355 | io->io_bio = NULL; |
| 356 | io->io_end = NULL; |
| 357 | } |
| 358 | |
| 359 | static int io_submit_init_bio(struct ext4_io_submit *io, |
| 360 | struct buffer_head *bh) |
| 361 | { |
| 362 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
| 363 | struct bio *bio; |
| 364 | |
| 365 | bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES)); |
| 366 | if (!bio) |
| 367 | return -ENOMEM; |
| 368 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
| 369 | bio->bi_bdev = bh->b_bdev; |
| 370 | bio->bi_end_io = ext4_end_bio; |
| 371 | bio->bi_private = ext4_get_io_end(io->io_end); |
| 372 | io->io_bio = bio; |
| 373 | io->io_next_block = bh->b_blocknr; |
| 374 | return 0; |
| 375 | } |
| 376 | |
| 377 | static int io_submit_add_bh(struct ext4_io_submit *io, |
| 378 | struct inode *inode, |
| 379 | struct buffer_head *bh) |
| 380 | { |
| 381 | int ret; |
| 382 | |
| 383 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { |
| 384 | submit_and_retry: |
| 385 | ext4_io_submit(io); |
| 386 | } |
| 387 | if (io->io_bio == NULL) { |
| 388 | ret = io_submit_init_bio(io, bh); |
| 389 | if (ret) |
| 390 | return ret; |
| 391 | } |
| 392 | ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); |
| 393 | if (ret != bh->b_size) |
| 394 | goto submit_and_retry; |
| 395 | io->io_next_block++; |
| 396 | return 0; |
| 397 | } |
| 398 | |
| 399 | int ext4_bio_write_page(struct ext4_io_submit *io, |
| 400 | struct page *page, |
| 401 | int len, |
| 402 | struct writeback_control *wbc) |
| 403 | { |
| 404 | struct inode *inode = page->mapping->host; |
| 405 | unsigned block_start, blocksize; |
| 406 | struct buffer_head *bh, *head; |
| 407 | int ret = 0; |
| 408 | int nr_submitted = 0; |
| 409 | |
| 410 | blocksize = 1 << inode->i_blkbits; |
| 411 | |
| 412 | BUG_ON(!PageLocked(page)); |
| 413 | BUG_ON(PageWriteback(page)); |
| 414 | |
| 415 | set_page_writeback(page); |
| 416 | ClearPageError(page); |
| 417 | |
| 418 | /* |
| 419 | * In the first loop we prepare and mark buffers to submit. We have to |
| 420 | * mark all buffers in the page before submitting so that |
| 421 | * end_page_writeback() cannot be called from ext4_bio_end_io() when IO |
| 422 | * on the first buffer finishes and we are still working on submitting |
| 423 | * the second buffer. |
| 424 | */ |
| 425 | bh = head = page_buffers(page); |
| 426 | do { |
| 427 | block_start = bh_offset(bh); |
| 428 | if (block_start >= len) { |
| 429 | /* |
| 430 | * Comments copied from block_write_full_page_endio: |
| 431 | * |
| 432 | * The page straddles i_size. It must be zeroed out on |
| 433 | * each and every writepage invocation because it may |
| 434 | * be mmapped. "A file is mapped in multiples of the |
| 435 | * page size. For a file that is not a multiple of |
| 436 | * the page size, the remaining memory is zeroed when |
| 437 | * mapped, and writes to that region are not written |
| 438 | * out to the file." |
| 439 | */ |
| 440 | zero_user_segment(page, block_start, |
| 441 | block_start + blocksize); |
| 442 | clear_buffer_dirty(bh); |
| 443 | set_buffer_uptodate(bh); |
| 444 | continue; |
| 445 | } |
| 446 | if (!buffer_dirty(bh) || buffer_delay(bh) || |
| 447 | !buffer_mapped(bh) || buffer_unwritten(bh)) { |
| 448 | /* A hole? We can safely clear the dirty bit */ |
| 449 | if (!buffer_mapped(bh)) |
| 450 | clear_buffer_dirty(bh); |
| 451 | if (io->io_bio) |
| 452 | ext4_io_submit(io); |
| 453 | continue; |
| 454 | } |
| 455 | if (buffer_new(bh)) { |
| 456 | clear_buffer_new(bh); |
| 457 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); |
| 458 | } |
| 459 | set_buffer_async_write(bh); |
| 460 | } while ((bh = bh->b_this_page) != head); |
| 461 | |
| 462 | /* Now submit buffers to write */ |
| 463 | bh = head = page_buffers(page); |
| 464 | do { |
| 465 | if (!buffer_async_write(bh)) |
| 466 | continue; |
| 467 | ret = io_submit_add_bh(io, inode, bh); |
| 468 | if (ret) { |
| 469 | /* |
| 470 | * We only get here on ENOMEM. Not much else |
| 471 | * we can do but mark the page as dirty, and |
| 472 | * better luck next time. |
| 473 | */ |
| 474 | redirty_page_for_writepage(wbc, page); |
| 475 | break; |
| 476 | } |
| 477 | nr_submitted++; |
| 478 | clear_buffer_dirty(bh); |
| 479 | } while ((bh = bh->b_this_page) != head); |
| 480 | |
| 481 | /* Error stopped previous loop? Clean up buffers... */ |
| 482 | if (ret) { |
| 483 | do { |
| 484 | clear_buffer_async_write(bh); |
| 485 | bh = bh->b_this_page; |
| 486 | } while (bh != head); |
| 487 | } |
| 488 | unlock_page(page); |
| 489 | /* Nothing submitted - we have to end page writeback */ |
| 490 | if (!nr_submitted) |
| 491 | end_page_writeback(page); |
| 492 | return ret; |
| 493 | } |