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/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
25 #include <trace/events/f2fs.h>
28 * Lock ordering for the change of data block address:
31 * update block addresses in the node page
33 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
37 struct page *node_page = dn->node_page;
38 unsigned int ofs_in_node = dn->ofs_in_node;
40 wait_on_page_writeback(node_page);
42 rn = (struct f2fs_node *)page_address(node_page);
44 /* Get physical address of data block */
45 addr_array = blkaddr_in_node(rn);
46 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
47 set_page_dirty(node_page);
50 int reserve_new_block(struct dnode_of_data *dn)
52 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
54 if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
56 if (!inc_valid_block_count(sbi, dn->inode, 1))
59 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
61 __set_data_blkaddr(dn, NEW_ADDR);
62 dn->data_blkaddr = NEW_ADDR;
67 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
68 struct buffer_head *bh_result)
70 struct f2fs_inode_info *fi = F2FS_I(inode);
71 #ifdef CONFIG_F2FS_STAT_FS
72 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
74 pgoff_t start_fofs, end_fofs;
75 block_t start_blkaddr;
77 read_lock(&fi->ext.ext_lock);
78 if (fi->ext.len == 0) {
79 read_unlock(&fi->ext.ext_lock);
83 #ifdef CONFIG_F2FS_STAT_FS
86 start_fofs = fi->ext.fofs;
87 end_fofs = fi->ext.fofs + fi->ext.len - 1;
88 start_blkaddr = fi->ext.blk_addr;
90 if (pgofs >= start_fofs && pgofs <= end_fofs) {
91 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
94 clear_buffer_new(bh_result);
95 map_bh(bh_result, inode->i_sb,
96 start_blkaddr + pgofs - start_fofs);
97 count = end_fofs - pgofs + 1;
98 if (count < (UINT_MAX >> blkbits))
99 bh_result->b_size = (count << blkbits);
101 bh_result->b_size = UINT_MAX;
103 #ifdef CONFIG_F2FS_STAT_FS
106 read_unlock(&fi->ext.ext_lock);
109 read_unlock(&fi->ext.ext_lock);
113 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
115 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
116 pgoff_t fofs, start_fofs, end_fofs;
117 block_t start_blkaddr, end_blkaddr;
119 BUG_ON(blk_addr == NEW_ADDR);
120 fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node;
122 /* Update the page address in the parent node */
123 __set_data_blkaddr(dn, blk_addr);
125 write_lock(&fi->ext.ext_lock);
127 start_fofs = fi->ext.fofs;
128 end_fofs = fi->ext.fofs + fi->ext.len - 1;
129 start_blkaddr = fi->ext.blk_addr;
130 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
132 /* Drop and initialize the matched extent */
133 if (fi->ext.len == 1 && fofs == start_fofs)
137 if (fi->ext.len == 0) {
138 if (blk_addr != NULL_ADDR) {
140 fi->ext.blk_addr = blk_addr;
147 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
155 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
160 /* Split the existing extent */
161 if (fi->ext.len > 1 &&
162 fofs >= start_fofs && fofs <= end_fofs) {
163 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
164 fi->ext.len = fofs - start_fofs;
166 fi->ext.fofs = fofs + 1;
167 fi->ext.blk_addr = start_blkaddr +
168 fofs - start_fofs + 1;
169 fi->ext.len -= fofs - start_fofs + 1;
173 write_unlock(&fi->ext.ext_lock);
177 write_unlock(&fi->ext.ext_lock);
182 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
184 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
185 struct address_space *mapping = inode->i_mapping;
186 struct dnode_of_data dn;
190 page = find_get_page(mapping, index);
191 if (page && PageUptodate(page))
193 f2fs_put_page(page, 0);
195 set_new_dnode(&dn, inode, NULL, NULL, 0);
196 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
201 if (dn.data_blkaddr == NULL_ADDR)
202 return ERR_PTR(-ENOENT);
204 /* By fallocate(), there is no cached page, but with NEW_ADDR */
205 if (dn.data_blkaddr == NEW_ADDR)
206 return ERR_PTR(-EINVAL);
208 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
210 return ERR_PTR(-ENOMEM);
212 if (PageUptodate(page)) {
217 err = f2fs_readpage(sbi, page, dn.data_blkaddr,
218 sync ? READ_SYNC : READA);
220 wait_on_page_locked(page);
221 if (!PageUptodate(page)) {
222 f2fs_put_page(page, 0);
223 return ERR_PTR(-EIO);
230 * If it tries to access a hole, return an error.
231 * Because, the callers, functions in dir.c and GC, should be able to know
232 * whether this page exists or not.
234 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
236 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
237 struct address_space *mapping = inode->i_mapping;
238 struct dnode_of_data dn;
243 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
245 return ERR_PTR(-ENOMEM);
247 set_new_dnode(&dn, inode, NULL, NULL, 0);
248 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
250 f2fs_put_page(page, 1);
255 if (dn.data_blkaddr == NULL_ADDR) {
256 f2fs_put_page(page, 1);
257 return ERR_PTR(-ENOENT);
260 if (PageUptodate(page))
263 BUG_ON(dn.data_blkaddr == NEW_ADDR);
264 BUG_ON(dn.data_blkaddr == NULL_ADDR);
266 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
271 if (!PageUptodate(page)) {
272 f2fs_put_page(page, 1);
273 return ERR_PTR(-EIO);
275 if (page->mapping != mapping) {
276 f2fs_put_page(page, 1);
283 * Caller ensures that this data page is never allocated.
284 * A new zero-filled data page is allocated in the page cache.
286 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
288 * Note that, npage is set only by make_empty_dir.
290 struct page *get_new_data_page(struct inode *inode,
291 struct page *npage, pgoff_t index, bool new_i_size)
293 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
294 struct address_space *mapping = inode->i_mapping;
296 struct dnode_of_data dn;
299 set_new_dnode(&dn, inode, npage, npage, 0);
300 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
304 if (dn.data_blkaddr == NULL_ADDR) {
305 if (reserve_new_block(&dn)) {
308 return ERR_PTR(-ENOSPC);
314 page = grab_cache_page(mapping, index);
316 return ERR_PTR(-ENOMEM);
318 if (PageUptodate(page))
321 if (dn.data_blkaddr == NEW_ADDR) {
322 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
323 SetPageUptodate(page);
325 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
329 if (!PageUptodate(page)) {
330 f2fs_put_page(page, 1);
331 return ERR_PTR(-EIO);
333 if (page->mapping != mapping) {
334 f2fs_put_page(page, 1);
340 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
341 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
342 mark_inode_dirty_sync(inode);
347 static void read_end_io(struct bio *bio, int err)
349 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
350 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
353 struct page *page = bvec->bv_page;
355 if (--bvec >= bio->bi_io_vec)
356 prefetchw(&bvec->bv_page->flags);
359 SetPageUptodate(page);
361 ClearPageUptodate(page);
365 } while (bvec >= bio->bi_io_vec);
366 kfree(bio->bi_private);
371 * Fill the locked page with data located in the block address.
372 * Return unlocked page.
374 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
375 block_t blk_addr, int type)
377 struct block_device *bdev = sbi->sb->s_bdev;
380 trace_f2fs_readpage(page, blk_addr, type);
382 down_read(&sbi->bio_sem);
384 /* Allocate a new bio */
385 bio = f2fs_bio_alloc(bdev, 1);
387 /* Initialize the bio */
388 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
389 bio->bi_end_io = read_end_io;
391 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
392 kfree(bio->bi_private);
394 up_read(&sbi->bio_sem);
395 f2fs_put_page(page, 1);
399 submit_bio(type, bio);
400 up_read(&sbi->bio_sem);
405 * This function should be used by the data read flow only where it
406 * does not check the "create" flag that indicates block allocation.
407 * The reason for this special functionality is to exploit VFS readahead
410 static int get_data_block_ro(struct inode *inode, sector_t iblock,
411 struct buffer_head *bh_result, int create)
413 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
414 unsigned maxblocks = bh_result->b_size >> blkbits;
415 struct dnode_of_data dn;
419 /* Get the page offset from the block offset(iblock) */
420 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
422 if (check_extent_cache(inode, pgofs, bh_result)) {
423 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
427 /* When reading holes, we need its node page */
428 set_new_dnode(&dn, inode, NULL, NULL, 0);
429 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
431 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
432 return (err == -ENOENT) ? 0 : err;
435 /* It does not support data allocation */
438 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
440 unsigned int end_offset;
442 end_offset = IS_INODE(dn.node_page) ?
446 clear_buffer_new(bh_result);
448 /* Give more consecutive addresses for the read ahead */
449 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
450 if (((datablock_addr(dn.node_page,
452 != (dn.data_blkaddr + i)) || maxblocks == i)
454 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
455 bh_result->b_size = (i << blkbits);
458 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
462 static int f2fs_read_data_page(struct file *file, struct page *page)
464 return mpage_readpage(page, get_data_block_ro);
467 static int f2fs_read_data_pages(struct file *file,
468 struct address_space *mapping,
469 struct list_head *pages, unsigned nr_pages)
471 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
474 int do_write_data_page(struct page *page)
476 struct inode *inode = page->mapping->host;
477 block_t old_blk_addr, new_blk_addr;
478 struct dnode_of_data dn;
481 set_new_dnode(&dn, inode, NULL, NULL, 0);
482 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
486 old_blk_addr = dn.data_blkaddr;
488 /* This page is already truncated */
489 if (old_blk_addr == NULL_ADDR)
492 set_page_writeback(page);
495 * If current allocation needs SSR,
496 * it had better in-place writes for updated data.
498 if (old_blk_addr != NEW_ADDR && !is_cold_data(page) &&
499 need_inplace_update(inode)) {
500 rewrite_data_page(F2FS_SB(inode->i_sb), page,
503 write_data_page(inode, page, &dn,
504 old_blk_addr, &new_blk_addr);
505 update_extent_cache(new_blk_addr, &dn);
512 static int f2fs_write_data_page(struct page *page,
513 struct writeback_control *wbc)
515 struct inode *inode = page->mapping->host;
516 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
517 loff_t i_size = i_size_read(inode);
518 const pgoff_t end_index = ((unsigned long long) i_size)
521 bool need_balance_fs = false;
524 if (page->index < end_index)
528 * If the offset is out-of-range of file size,
529 * this page does not have to be written to disk.
531 offset = i_size & (PAGE_CACHE_SIZE - 1);
532 if ((page->index >= end_index + 1) || !offset) {
533 if (S_ISDIR(inode->i_mode)) {
534 dec_page_count(sbi, F2FS_DIRTY_DENTS);
535 inode_dec_dirty_dents(inode);
540 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
542 if (sbi->por_doing) {
543 err = AOP_WRITEPAGE_ACTIVATE;
547 /* Dentry blocks are controlled by checkpoint */
548 if (S_ISDIR(inode->i_mode)) {
549 dec_page_count(sbi, F2FS_DIRTY_DENTS);
550 inode_dec_dirty_dents(inode);
551 err = do_write_data_page(page);
553 int ilock = mutex_lock_op(sbi);
554 err = do_write_data_page(page);
555 mutex_unlock_op(sbi, ilock);
556 need_balance_fs = true;
563 if (wbc->for_reclaim)
564 f2fs_submit_bio(sbi, DATA, true);
566 clear_cold_data(page);
570 f2fs_balance_fs(sbi);
574 wbc->pages_skipped++;
575 set_page_dirty(page);
579 #define MAX_DESIRED_PAGES_WP 4096
581 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
584 struct address_space *mapping = data;
585 int ret = mapping->a_ops->writepage(page, wbc);
586 mapping_set_error(mapping, ret);
590 static int f2fs_write_data_pages(struct address_space *mapping,
591 struct writeback_control *wbc)
593 struct inode *inode = mapping->host;
594 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
597 long excess_nrtw = 0, desired_nrtw;
599 /* deal with chardevs and other special file */
600 if (!mapping->a_ops->writepage)
603 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
604 desired_nrtw = MAX_DESIRED_PAGES_WP;
605 excess_nrtw = desired_nrtw - wbc->nr_to_write;
606 wbc->nr_to_write = desired_nrtw;
609 if (!S_ISDIR(inode->i_mode)) {
610 mutex_lock(&sbi->writepages);
613 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
615 mutex_unlock(&sbi->writepages);
616 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
618 remove_dirty_dir_inode(inode);
620 wbc->nr_to_write -= excess_nrtw;
624 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
625 loff_t pos, unsigned len, unsigned flags,
626 struct page **pagep, void **fsdata)
628 struct inode *inode = mapping->host;
629 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
631 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
632 struct dnode_of_data dn;
636 /* for nobh_write_end */
639 f2fs_balance_fs(sbi);
641 page = grab_cache_page_write_begin(mapping, index, flags);
646 ilock = mutex_lock_op(sbi);
648 set_new_dnode(&dn, inode, NULL, NULL, 0);
649 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
653 if (dn.data_blkaddr == NULL_ADDR)
654 err = reserve_new_block(&dn);
660 mutex_unlock_op(sbi, ilock);
662 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
665 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
666 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
667 unsigned end = start + len;
669 /* Reading beyond i_size is simple: memset to zero */
670 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
674 if (dn.data_blkaddr == NEW_ADDR) {
675 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
677 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
681 if (!PageUptodate(page)) {
682 f2fs_put_page(page, 1);
685 if (page->mapping != mapping) {
686 f2fs_put_page(page, 1);
691 SetPageUptodate(page);
692 clear_cold_data(page);
696 mutex_unlock_op(sbi, ilock);
697 f2fs_put_page(page, 1);
701 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
702 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
704 struct file *file = iocb->ki_filp;
705 struct inode *inode = file->f_mapping->host;
710 /* Needs synchronization with the cleaner */
711 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
715 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
717 struct inode *inode = page->mapping->host;
718 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
719 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
720 dec_page_count(sbi, F2FS_DIRTY_DENTS);
721 inode_dec_dirty_dents(inode);
723 ClearPagePrivate(page);
726 static int f2fs_release_data_page(struct page *page, gfp_t wait)
728 ClearPagePrivate(page);
732 static int f2fs_set_data_page_dirty(struct page *page)
734 struct address_space *mapping = page->mapping;
735 struct inode *inode = mapping->host;
737 SetPageUptodate(page);
738 if (!PageDirty(page)) {
739 __set_page_dirty_nobuffers(page);
740 set_dirty_dir_page(inode, page);
746 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
748 return generic_block_bmap(mapping, block, get_data_block_ro);
751 const struct address_space_operations f2fs_dblock_aops = {
752 .readpage = f2fs_read_data_page,
753 .readpages = f2fs_read_data_pages,
754 .writepage = f2fs_write_data_page,
755 .writepages = f2fs_write_data_pages,
756 .write_begin = f2fs_write_begin,
757 .write_end = nobh_write_end,
758 .set_page_dirty = f2fs_set_data_page_dirty,
759 .invalidatepage = f2fs_invalidate_data_page,
760 .releasepage = f2fs_release_data_page,
761 .direct_IO = f2fs_direct_IO,