4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/blkdev.h>
18 #include <linux/bio.h>
19 #include <linux/prefetch.h>
26 * Lock ordering for the change of data block address:
29 * update block addresses in the node page
31 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
35 struct page *node_page = dn->node_page;
36 unsigned int ofs_in_node = dn->ofs_in_node;
38 wait_on_page_writeback(node_page);
40 rn = (struct f2fs_node *)page_address(node_page);
42 /* Get physical address of data block */
43 addr_array = blkaddr_in_node(rn);
44 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
45 set_page_dirty(node_page);
48 int reserve_new_block(struct dnode_of_data *dn)
50 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
52 if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
54 if (!inc_valid_block_count(sbi, dn->inode, 1))
57 __set_data_blkaddr(dn, NEW_ADDR);
58 dn->data_blkaddr = NEW_ADDR;
63 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
64 struct buffer_head *bh_result)
66 struct f2fs_inode_info *fi = F2FS_I(inode);
67 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
68 pgoff_t start_fofs, end_fofs;
69 block_t start_blkaddr;
71 read_lock(&fi->ext.ext_lock);
72 if (fi->ext.len == 0) {
73 read_unlock(&fi->ext.ext_lock);
78 start_fofs = fi->ext.fofs;
79 end_fofs = fi->ext.fofs + fi->ext.len - 1;
80 start_blkaddr = fi->ext.blk_addr;
82 if (pgofs >= start_fofs && pgofs <= end_fofs) {
83 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
86 clear_buffer_new(bh_result);
87 map_bh(bh_result, inode->i_sb,
88 start_blkaddr + pgofs - start_fofs);
89 count = end_fofs - pgofs + 1;
90 if (count < (UINT_MAX >> blkbits))
91 bh_result->b_size = (count << blkbits);
93 bh_result->b_size = UINT_MAX;
96 read_unlock(&fi->ext.ext_lock);
99 read_unlock(&fi->ext.ext_lock);
103 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
105 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
106 pgoff_t fofs, start_fofs, end_fofs;
107 block_t start_blkaddr, end_blkaddr;
109 BUG_ON(blk_addr == NEW_ADDR);
110 fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node;
112 /* Update the page address in the parent node */
113 __set_data_blkaddr(dn, blk_addr);
115 write_lock(&fi->ext.ext_lock);
117 start_fofs = fi->ext.fofs;
118 end_fofs = fi->ext.fofs + fi->ext.len - 1;
119 start_blkaddr = fi->ext.blk_addr;
120 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
122 /* Drop and initialize the matched extent */
123 if (fi->ext.len == 1 && fofs == start_fofs)
127 if (fi->ext.len == 0) {
128 if (blk_addr != NULL_ADDR) {
130 fi->ext.blk_addr = blk_addr;
137 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
145 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
150 /* Split the existing extent */
151 if (fi->ext.len > 1 &&
152 fofs >= start_fofs && fofs <= end_fofs) {
153 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
154 fi->ext.len = fofs - start_fofs;
156 fi->ext.fofs = fofs + 1;
157 fi->ext.blk_addr = start_blkaddr +
158 fofs - start_fofs + 1;
159 fi->ext.len -= fofs - start_fofs + 1;
163 write_unlock(&fi->ext.ext_lock);
167 write_unlock(&fi->ext.ext_lock);
172 struct page *find_data_page(struct inode *inode, pgoff_t index)
174 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
175 struct address_space *mapping = inode->i_mapping;
176 struct dnode_of_data dn;
180 page = find_get_page(mapping, index);
181 if (page && PageUptodate(page))
183 f2fs_put_page(page, 0);
185 set_new_dnode(&dn, inode, NULL, NULL, 0);
186 err = get_dnode_of_data(&dn, index, RDONLY_NODE);
191 if (dn.data_blkaddr == NULL_ADDR)
192 return ERR_PTR(-ENOENT);
194 /* By fallocate(), there is no cached page, but with NEW_ADDR */
195 if (dn.data_blkaddr == NEW_ADDR)
196 return ERR_PTR(-EINVAL);
198 page = grab_cache_page(mapping, index);
200 return ERR_PTR(-ENOMEM);
202 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
204 f2fs_put_page(page, 1);
212 * If it tries to access a hole, return an error.
213 * Because, the callers, functions in dir.c and GC, should be able to know
214 * whether this page exists or not.
216 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
218 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
219 struct address_space *mapping = inode->i_mapping;
220 struct dnode_of_data dn;
224 set_new_dnode(&dn, inode, NULL, NULL, 0);
225 err = get_dnode_of_data(&dn, index, RDONLY_NODE);
230 if (dn.data_blkaddr == NULL_ADDR)
231 return ERR_PTR(-ENOENT);
233 page = grab_cache_page(mapping, index);
235 return ERR_PTR(-ENOMEM);
237 if (PageUptodate(page))
240 BUG_ON(dn.data_blkaddr == NEW_ADDR);
241 BUG_ON(dn.data_blkaddr == NULL_ADDR);
243 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
245 f2fs_put_page(page, 1);
252 * Caller ensures that this data page is never allocated.
253 * A new zero-filled data page is allocated in the page cache.
255 struct page *get_new_data_page(struct inode *inode, pgoff_t index,
258 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
259 struct address_space *mapping = inode->i_mapping;
261 struct dnode_of_data dn;
264 set_new_dnode(&dn, inode, NULL, NULL, 0);
265 err = get_dnode_of_data(&dn, index, 0);
269 if (dn.data_blkaddr == NULL_ADDR) {
270 if (reserve_new_block(&dn)) {
272 return ERR_PTR(-ENOSPC);
277 page = grab_cache_page(mapping, index);
279 return ERR_PTR(-ENOMEM);
281 if (PageUptodate(page))
284 if (dn.data_blkaddr == NEW_ADDR) {
285 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
287 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
289 f2fs_put_page(page, 1);
293 SetPageUptodate(page);
296 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
297 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
298 mark_inode_dirty_sync(inode);
303 static void read_end_io(struct bio *bio, int err)
305 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
306 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
309 struct page *page = bvec->bv_page;
311 if (--bvec >= bio->bi_io_vec)
312 prefetchw(&bvec->bv_page->flags);
315 SetPageUptodate(page);
317 ClearPageUptodate(page);
321 } while (bvec >= bio->bi_io_vec);
322 kfree(bio->bi_private);
327 * Fill the locked page with data located in the block address.
328 * Read operation is synchronous, and caller must unlock the page.
330 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
331 block_t blk_addr, int type)
333 struct block_device *bdev = sbi->sb->s_bdev;
334 bool sync = (type == READ_SYNC);
337 /* This page can be already read by other threads */
338 if (PageUptodate(page)) {
344 down_read(&sbi->bio_sem);
346 /* Allocate a new bio */
347 bio = f2fs_bio_alloc(bdev, 1);
349 /* Initialize the bio */
350 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
351 bio->bi_end_io = read_end_io;
353 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
354 kfree(bio->bi_private);
356 up_read(&sbi->bio_sem);
360 submit_bio(type, bio);
361 up_read(&sbi->bio_sem);
363 /* wait for read completion if sync */
373 * This function should be used by the data read flow only where it
374 * does not check the "create" flag that indicates block allocation.
375 * The reason for this special functionality is to exploit VFS readahead
378 static int get_data_block_ro(struct inode *inode, sector_t iblock,
379 struct buffer_head *bh_result, int create)
381 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
382 unsigned maxblocks = bh_result->b_size >> blkbits;
383 struct dnode_of_data dn;
387 /* Get the page offset from the block offset(iblock) */
388 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
390 if (check_extent_cache(inode, pgofs, bh_result))
393 /* When reading holes, we need its node page */
394 set_new_dnode(&dn, inode, NULL, NULL, 0);
395 err = get_dnode_of_data(&dn, pgofs, RDONLY_NODE);
397 return (err == -ENOENT) ? 0 : err;
399 /* It does not support data allocation */
402 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
404 unsigned int end_offset;
406 end_offset = IS_INODE(dn.node_page) ?
410 clear_buffer_new(bh_result);
412 /* Give more consecutive addresses for the read ahead */
413 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
414 if (((datablock_addr(dn.node_page,
416 != (dn.data_blkaddr + i)) || maxblocks == i)
418 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
419 bh_result->b_size = (i << blkbits);
425 static int f2fs_read_data_page(struct file *file, struct page *page)
427 return mpage_readpage(page, get_data_block_ro);
430 static int f2fs_read_data_pages(struct file *file,
431 struct address_space *mapping,
432 struct list_head *pages, unsigned nr_pages)
434 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
437 int do_write_data_page(struct page *page)
439 struct inode *inode = page->mapping->host;
440 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
441 block_t old_blk_addr, new_blk_addr;
442 struct dnode_of_data dn;
445 set_new_dnode(&dn, inode, NULL, NULL, 0);
446 err = get_dnode_of_data(&dn, page->index, RDONLY_NODE);
450 old_blk_addr = dn.data_blkaddr;
452 /* This page is already truncated */
453 if (old_blk_addr == NULL_ADDR)
456 set_page_writeback(page);
459 * If current allocation needs SSR,
460 * it had better in-place writes for updated data.
462 if (old_blk_addr != NEW_ADDR && !is_cold_data(page) &&
463 need_inplace_update(inode)) {
464 rewrite_data_page(F2FS_SB(inode->i_sb), page,
467 write_data_page(inode, page, &dn,
468 old_blk_addr, &new_blk_addr);
469 update_extent_cache(new_blk_addr, &dn);
470 F2FS_I(inode)->data_version =
471 le64_to_cpu(F2FS_CKPT(sbi)->checkpoint_ver);
478 static int f2fs_write_data_page(struct page *page,
479 struct writeback_control *wbc)
481 struct inode *inode = page->mapping->host;
482 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
483 loff_t i_size = i_size_read(inode);
484 const pgoff_t end_index = ((unsigned long long) i_size)
489 if (page->index < end_index)
493 * If the offset is out-of-range of file size,
494 * this page does not have to be written to disk.
496 offset = i_size & (PAGE_CACHE_SIZE - 1);
497 if ((page->index >= end_index + 1) || !offset) {
498 if (S_ISDIR(inode->i_mode)) {
499 dec_page_count(sbi, F2FS_DIRTY_DENTS);
500 inode_dec_dirty_dents(inode);
505 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
510 if (wbc->for_reclaim && !S_ISDIR(inode->i_mode) && !is_cold_data(page))
513 mutex_lock_op(sbi, DATA_WRITE);
514 if (S_ISDIR(inode->i_mode)) {
515 dec_page_count(sbi, F2FS_DIRTY_DENTS);
516 inode_dec_dirty_dents(inode);
518 err = do_write_data_page(page);
519 if (err && err != -ENOENT) {
520 wbc->pages_skipped++;
521 set_page_dirty(page);
523 mutex_unlock_op(sbi, DATA_WRITE);
525 if (wbc->for_reclaim)
526 f2fs_submit_bio(sbi, DATA, true);
531 clear_cold_data(page);
534 if (!wbc->for_reclaim && !S_ISDIR(inode->i_mode))
535 f2fs_balance_fs(sbi);
540 return (err == -ENOENT) ? 0 : err;
543 wbc->pages_skipped++;
544 set_page_dirty(page);
545 return AOP_WRITEPAGE_ACTIVATE;
548 #define MAX_DESIRED_PAGES_WP 4096
550 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
553 struct address_space *mapping = data;
554 int ret = mapping->a_ops->writepage(page, wbc);
555 mapping_set_error(mapping, ret);
559 static int f2fs_write_data_pages(struct address_space *mapping,
560 struct writeback_control *wbc)
562 struct inode *inode = mapping->host;
563 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
565 long excess_nrtw = 0, desired_nrtw;
567 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
568 desired_nrtw = MAX_DESIRED_PAGES_WP;
569 excess_nrtw = desired_nrtw - wbc->nr_to_write;
570 wbc->nr_to_write = desired_nrtw;
573 if (!S_ISDIR(inode->i_mode))
574 mutex_lock(&sbi->writepages);
575 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
576 if (!S_ISDIR(inode->i_mode))
577 mutex_unlock(&sbi->writepages);
578 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
580 remove_dirty_dir_inode(inode);
582 wbc->nr_to_write -= excess_nrtw;
586 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
587 loff_t pos, unsigned len, unsigned flags,
588 struct page **pagep, void **fsdata)
590 struct inode *inode = mapping->host;
591 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
593 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
594 struct dnode_of_data dn;
597 /* for nobh_write_end */
600 f2fs_balance_fs(sbi);
602 page = grab_cache_page_write_begin(mapping, index, flags);
607 mutex_lock_op(sbi, DATA_NEW);
609 set_new_dnode(&dn, inode, NULL, NULL, 0);
610 err = get_dnode_of_data(&dn, index, 0);
612 mutex_unlock_op(sbi, DATA_NEW);
613 f2fs_put_page(page, 1);
617 if (dn.data_blkaddr == NULL_ADDR) {
618 err = reserve_new_block(&dn);
621 mutex_unlock_op(sbi, DATA_NEW);
622 f2fs_put_page(page, 1);
628 mutex_unlock_op(sbi, DATA_NEW);
630 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
633 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
634 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
635 unsigned end = start + len;
637 /* Reading beyond i_size is simple: memset to zero */
638 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
642 if (dn.data_blkaddr == NEW_ADDR) {
643 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
645 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
647 f2fs_put_page(page, 1);
651 SetPageUptodate(page);
652 clear_cold_data(page);
656 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
657 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
659 struct file *file = iocb->ki_filp;
660 struct inode *inode = file->f_mapping->host;
665 /* Needs synchronization with the cleaner */
666 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
670 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
672 struct inode *inode = page->mapping->host;
673 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
674 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
675 dec_page_count(sbi, F2FS_DIRTY_DENTS);
676 inode_dec_dirty_dents(inode);
678 ClearPagePrivate(page);
681 static int f2fs_release_data_page(struct page *page, gfp_t wait)
683 ClearPagePrivate(page);
687 static int f2fs_set_data_page_dirty(struct page *page)
689 struct address_space *mapping = page->mapping;
690 struct inode *inode = mapping->host;
692 SetPageUptodate(page);
693 if (!PageDirty(page)) {
694 __set_page_dirty_nobuffers(page);
695 set_dirty_dir_page(inode, page);
701 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
703 return generic_block_bmap(mapping, block, get_data_block_ro);
706 const struct address_space_operations f2fs_dblock_aops = {
707 .readpage = f2fs_read_data_page,
708 .readpages = f2fs_read_data_pages,
709 .writepage = f2fs_write_data_page,
710 .writepages = f2fs_write_data_pages,
711 .write_begin = f2fs_write_begin,
712 .write_end = nobh_write_end,
713 .set_page_dirty = f2fs_set_data_page_dirty,
714 .invalidatepage = f2fs_invalidate_data_page,
715 .releasepage = f2fs_release_data_page,
716 .direct_IO = f2fs_direct_IO,