Both ext4 and xfs implement seeking for the next hole or piece of data in unwritten extents by scanning the page cache, and both versions share the same bug when iterating the buffers of a page: the start offset into the page isn't taken into account, so when a page fits more than two filesystem blocks, things will go wrong. For example, on a filesystem with a block size of 1k, the following command will fail: xfs_io -f -c "falloc 0 4k" \ -c "pwrite 1k 1k" \ -c "pwrite 3k 1k" \ -c "seek -a -r 0" foo In this example, neither lseek(fd, 1024, SEEK_HOLE) nor lseek(fd, 2048, SEEK_DATA) will return the correct result. Introduce a generic vfs helper for seeking in the page cache that gets this right. The next commits will replace the filesystem specific implementations.
Unified · Split
| @@ -3492,6 +3492,130 @@ int bh_submit_read(struct buffer_head *bh) · 3491 unmodified lines | ||
| 3492 | 3492 | } |
| 3493 | 3493 | EXPORT_SYMBOL(bh_submit_read); |
| 3494 | 3494 | |
| 3495 | +/* | |
| 3496 | + * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff. | |
| 3497 | + * | |
| 3498 | + * Returns the offset within the file on success, and -ENOENT otherwise. | |
| 3499 | + */ | |
| 3500 | +static loff_t | |
| 3501 | +page_seek_hole_data(struct page *page, loff_t lastoff, int whence) | |
| 3502 | +{ | |
| 3503 | + loff_t offset = page_offset(page); | |
| 3504 | + struct buffer_head *bh, *head; | |
| 3505 | + bool seek_data = whence == SEEK_DATA; | |
| 3506 | + | |
| 3507 | + if (lastoff < offset) | |
| 3508 | + lastoff = offset; | |
| 3509 | + | |
| 3510 | + bh = head = page_buffers(page); | |
| 3511 | + do { | |
| 3512 | + offset += bh->b_size; | |
| 3513 | + if (lastoff >= offset) | |
| 3514 | + continue; | |
| 3515 | + | |
| 3516 | + /* | |
| 3517 | + * Unwritten extents that have data in the page cache covering | |
| 3518 | + * them can be identified by the BH_Unwritten state flag. | |
| 3519 | + * Pages with multiple buffers might have a mix of holes, data | |
| 3520 | + * and unwritten extents - any buffer with valid data in it | |
| 3521 | + * should have BH_Uptodate flag set on it. | |
| 3522 | + */ | |
| 3523 | + | |
| 3524 | + if ((buffer_unwritten(bh) || buffer_uptodate(bh)) == seek_data) | |
| 3525 | + return lastoff; | |
| 3526 | + | |
| 3527 | + lastoff = offset; | |
| 3528 | + } while ((bh = bh->b_this_page) != head); | |
| 3529 | + return -ENOENT; | |
| 3530 | +} | |
| 3531 | + | |
| 3532 | +/* | |
| 3533 | + * Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
| 3534 | + * | |
| 3535 | + * Within unwritten extents, the page cache determines which parts are holes | |
| 3536 | + * and which are data: unwritten and uptodate buffer heads count as data; | |
| 3537 | + * everything else counts as a hole. | |
| 3538 | + * | |
| 3539 | + * Returns the resulting offset on successs, and -ENOENT otherwise. | |
| 3540 | + */ | |
| 3541 | +loff_t | |
| 3542 | +page_cache_seek_hole_data(struct inode *inode, loff_t offset, loff_t length, | |
| 3543 | + int whence) | |
| 3544 | +{ | |
| 3545 | + pgoff_t index = offset >> PAGE_SHIFT; | |
| 3546 | + pgoff_t end = DIV_ROUND_UP(offset + length, PAGE_SIZE); | |
| 3547 | + loff_t lastoff = offset; | |
| 3548 | + struct pagevec pvec; | |
| 3549 | + | |
| 3550 | + if (length <= 0) | |
| 3551 | + return -ENOENT; | |
| 3552 | + | |
| 3553 | + pagevec_init(&pvec, 0); | |
| 3554 | + | |
| 3555 | + do { | |
| 3556 | + unsigned want, nr_pages, i; | |
| 3557 | + | |
| 3558 | + want = min_t(unsigned, end - index, PAGEVEC_SIZE); | |
| 3559 | + nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, want); | |
| 3560 | + if (nr_pages == 0) | |
| 3561 | + break; | |
| 3562 | + | |
| 3563 | + for (i = 0; i < nr_pages; i++) { | |
| 3564 | + struct page *page = pvec.pages[i]; | |
| 3565 | + | |
| 3566 | + /* | |
| 3567 | + * At this point, the page may be truncated or | |
| 3568 | + * invalidated (changing page->mapping to NULL), or | |
| 3569 | + * even swizzled back from swapper_space to tmpfs file | |
| 3570 | + * mapping. However, page->index will not change | |
| 3571 | + * because we have a reference on the page. | |
| 3572 | + * | |
| 3573 | + * If current page offset is beyond where we've ended, | |
| 3574 | + * we've found a hole. | |
| 3575 | + */ | |
| 3576 | + if (whence == SEEK_HOLE && | |
| 3577 | + lastoff < page_offset(page)) | |
| 3578 | + goto check_range; | |
| 3579 | + | |
| 3580 | + /* Searching done if the page index is out of range. */ | |
| 3581 | + if (page->index >= end) | |
| 3582 | + goto not_found; | |
| 3583 | + | |
| 3584 | + lock_page(page); | |
| 3585 | + if (likely(page->mapping == inode->i_mapping) && | |
| 3586 | + page_has_buffers(page)) { | |
| 3587 | + lastoff = page_seek_hole_data(page, lastoff, whence); | |
| 3588 | + if (lastoff >= 0) { | |
| 3589 | + unlock_page(page); | |
| 3590 | + goto check_range; | |
| 3591 | + } | |
| 3592 | + } | |
| 3593 | + unlock_page(page); | |
| 3594 | + lastoff = page_offset(page) + PAGE_SIZE; | |
| 3595 | + } | |
| 3596 | + | |
| 3597 | + /* Searching done if fewer pages returned than wanted. */ | |
| 3598 | + if (nr_pages < want) | |
| 3599 | + break; | |
| 3600 | + | |
| 3601 | + index = pvec.pages[i - 1]->index + 1; | |
| 3602 | + pagevec_release(&pvec); | |
| 3603 | + } while (index < end); | |
| 3604 | + | |
| 3605 | + /* When no page at lastoff and we are not done, we found a hole. */ | |
| 3606 | + if (whence != SEEK_HOLE) | |
| 3607 | + goto not_found; | |
| 3608 | + | |
| 3609 | +check_range: | |
| 3610 | + if (lastoff < offset + length) | |
| 3611 | + goto out; | |
| 3612 | +not_found: | |
| 3613 | + lastoff = -ENOENT; | |
| 3614 | +out: | |
| 3615 | + pagevec_release(&pvec); | |
| 3616 | + return lastoff; | |
| 3617 | +} | |
| 3618 | + | |
| 3495 | 3619 | void __init buffer_init(void) |
| 3496 | 3620 | { |
| 3497 | 3621 | unsigned long nrpages; |
| @@ -201,6 +201,8 @@ void write_boundary_block(struct block_device *bdev, · 200 unmodified lines | ||
| 201 | 201 | sector_t bblock, unsigned blocksize); |
| 202 | 202 | int bh_uptodate_or_lock(struct buffer_head *bh); |
| 203 | 203 | int bh_submit_read(struct buffer_head *bh); |
| 204 | +loff_t page_cache_seek_hole_data(struct inode *inode, loff_t offset, | |
| 205 | + loff_t length, int whence); | |
| 204 | 206 | |
| 205 | 207 | extern int buffer_heads_over_limit; |
| 206 | 208 | |