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>
24 #include <trace/events/f2fs.h>
27 * Lock ordering for the change of data block address:
30 * update block addresses in the node page
32 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
36 struct page *node_page = dn->node_page;
37 unsigned int ofs_in_node = dn->ofs_in_node;
39 wait_on_page_writeback(node_page);
41 rn = (struct f2fs_node *)page_address(node_page);
43 /* Get physical address of data block */
44 addr_array = blkaddr_in_node(rn);
45 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
46 set_page_dirty(node_page);
49 int reserve_new_block(struct dnode_of_data *dn)
51 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
53 if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
55 if (!inc_valid_block_count(sbi, dn->inode, 1))
58 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
60 __set_data_blkaddr(dn, NEW_ADDR);
61 dn->data_blkaddr = NEW_ADDR;
66 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
67 struct buffer_head *bh_result)
69 struct f2fs_inode_info *fi = F2FS_I(inode);
70 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
71 pgoff_t start_fofs, end_fofs;
72 block_t start_blkaddr;
74 read_lock(&fi->ext.ext_lock);
75 if (fi->ext.len == 0) {
76 read_unlock(&fi->ext.ext_lock);
81 start_fofs = fi->ext.fofs;
82 end_fofs = fi->ext.fofs + fi->ext.len - 1;
83 start_blkaddr = fi->ext.blk_addr;
85 if (pgofs >= start_fofs && pgofs <= end_fofs) {
86 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
89 clear_buffer_new(bh_result);
90 map_bh(bh_result, inode->i_sb,
91 start_blkaddr + pgofs - start_fofs);
92 count = end_fofs - pgofs + 1;
93 if (count < (UINT_MAX >> blkbits))
94 bh_result->b_size = (count << blkbits);
96 bh_result->b_size = UINT_MAX;
99 read_unlock(&fi->ext.ext_lock);
102 read_unlock(&fi->ext.ext_lock);
106 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
108 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
109 pgoff_t fofs, start_fofs, end_fofs;
110 block_t start_blkaddr, end_blkaddr;
112 BUG_ON(blk_addr == NEW_ADDR);
113 fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node;
115 /* Update the page address in the parent node */
116 __set_data_blkaddr(dn, blk_addr);
118 write_lock(&fi->ext.ext_lock);
120 start_fofs = fi->ext.fofs;
121 end_fofs = fi->ext.fofs + fi->ext.len - 1;
122 start_blkaddr = fi->ext.blk_addr;
123 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
125 /* Drop and initialize the matched extent */
126 if (fi->ext.len == 1 && fofs == start_fofs)
130 if (fi->ext.len == 0) {
131 if (blk_addr != NULL_ADDR) {
133 fi->ext.blk_addr = blk_addr;
140 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
148 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
153 /* Split the existing extent */
154 if (fi->ext.len > 1 &&
155 fofs >= start_fofs && fofs <= end_fofs) {
156 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
157 fi->ext.len = fofs - start_fofs;
159 fi->ext.fofs = fofs + 1;
160 fi->ext.blk_addr = start_blkaddr +
161 fofs - start_fofs + 1;
162 fi->ext.len -= fofs - start_fofs + 1;
166 write_unlock(&fi->ext.ext_lock);
170 write_unlock(&fi->ext.ext_lock);
175 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
177 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
178 struct address_space *mapping = inode->i_mapping;
179 struct dnode_of_data dn;
183 page = find_get_page(mapping, index);
184 if (page && PageUptodate(page))
186 f2fs_put_page(page, 0);
188 set_new_dnode(&dn, inode, NULL, NULL, 0);
189 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
194 if (dn.data_blkaddr == NULL_ADDR)
195 return ERR_PTR(-ENOENT);
197 /* By fallocate(), there is no cached page, but with NEW_ADDR */
198 if (dn.data_blkaddr == NEW_ADDR)
199 return ERR_PTR(-EINVAL);
201 page = grab_cache_page(mapping, index);
203 return ERR_PTR(-ENOMEM);
205 if (PageUptodate(page)) {
210 err = f2fs_readpage(sbi, page, dn.data_blkaddr,
211 sync ? READ_SYNC : READA);
213 wait_on_page_locked(page);
214 if (!PageUptodate(page)) {
215 f2fs_put_page(page, 0);
216 return ERR_PTR(-EIO);
223 * If it tries to access a hole, return an error.
224 * Because, the callers, functions in dir.c and GC, should be able to know
225 * whether this page exists or not.
227 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
229 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
230 struct address_space *mapping = inode->i_mapping;
231 struct dnode_of_data dn;
235 set_new_dnode(&dn, inode, NULL, NULL, 0);
236 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
241 if (dn.data_blkaddr == NULL_ADDR)
242 return ERR_PTR(-ENOENT);
244 page = grab_cache_page(mapping, index);
246 return ERR_PTR(-ENOMEM);
248 if (PageUptodate(page))
251 BUG_ON(dn.data_blkaddr == NEW_ADDR);
252 BUG_ON(dn.data_blkaddr == NULL_ADDR);
254 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
259 if (!PageUptodate(page)) {
260 f2fs_put_page(page, 1);
261 return ERR_PTR(-EIO);
263 if (page->mapping != mapping) {
264 f2fs_put_page(page, 1);
271 * Caller ensures that this data page is never allocated.
272 * A new zero-filled data page is allocated in the page cache.
274 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
277 struct page *get_new_data_page(struct inode *inode, pgoff_t index,
280 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
281 struct address_space *mapping = inode->i_mapping;
283 struct dnode_of_data dn;
286 set_new_dnode(&dn, inode, NULL, NULL, 0);
287 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
291 if (dn.data_blkaddr == NULL_ADDR) {
292 if (reserve_new_block(&dn)) {
294 return ERR_PTR(-ENOSPC);
299 page = grab_cache_page(mapping, index);
301 return ERR_PTR(-ENOMEM);
303 if (PageUptodate(page))
306 if (dn.data_blkaddr == NEW_ADDR) {
307 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
308 SetPageUptodate(page);
310 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
314 if (!PageUptodate(page)) {
315 f2fs_put_page(page, 1);
316 return ERR_PTR(-EIO);
318 if (page->mapping != mapping) {
319 f2fs_put_page(page, 1);
325 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
326 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
327 mark_inode_dirty_sync(inode);
332 static void read_end_io(struct bio *bio, int err)
334 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
335 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
338 struct page *page = bvec->bv_page;
340 if (--bvec >= bio->bi_io_vec)
341 prefetchw(&bvec->bv_page->flags);
344 SetPageUptodate(page);
346 ClearPageUptodate(page);
350 } while (bvec >= bio->bi_io_vec);
351 kfree(bio->bi_private);
356 * Fill the locked page with data located in the block address.
357 * Return unlocked page.
359 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
360 block_t blk_addr, int type)
362 struct block_device *bdev = sbi->sb->s_bdev;
365 trace_f2fs_readpage(page, blk_addr, type);
367 down_read(&sbi->bio_sem);
369 /* Allocate a new bio */
370 bio = f2fs_bio_alloc(bdev, 1);
372 /* Initialize the bio */
373 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
374 bio->bi_end_io = read_end_io;
376 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
377 kfree(bio->bi_private);
379 up_read(&sbi->bio_sem);
380 f2fs_put_page(page, 1);
384 submit_bio(type, bio);
385 up_read(&sbi->bio_sem);
390 * This function should be used by the data read flow only where it
391 * does not check the "create" flag that indicates block allocation.
392 * The reason for this special functionality is to exploit VFS readahead
395 static int get_data_block_ro(struct inode *inode, sector_t iblock,
396 struct buffer_head *bh_result, int create)
398 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
399 unsigned maxblocks = bh_result->b_size >> blkbits;
400 struct dnode_of_data dn;
404 /* Get the page offset from the block offset(iblock) */
405 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
407 if (check_extent_cache(inode, pgofs, bh_result)) {
408 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
412 /* When reading holes, we need its node page */
413 set_new_dnode(&dn, inode, NULL, NULL, 0);
414 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
416 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
417 return (err == -ENOENT) ? 0 : err;
420 /* It does not support data allocation */
423 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
425 unsigned int end_offset;
427 end_offset = IS_INODE(dn.node_page) ?
431 clear_buffer_new(bh_result);
433 /* Give more consecutive addresses for the read ahead */
434 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
435 if (((datablock_addr(dn.node_page,
437 != (dn.data_blkaddr + i)) || maxblocks == i)
439 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
440 bh_result->b_size = (i << blkbits);
443 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
447 static int f2fs_read_data_page(struct file *file, struct page *page)
449 return mpage_readpage(page, get_data_block_ro);
452 static int f2fs_read_data_pages(struct file *file,
453 struct address_space *mapping,
454 struct list_head *pages, unsigned nr_pages)
456 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
459 int do_write_data_page(struct page *page)
461 struct inode *inode = page->mapping->host;
462 block_t old_blk_addr, new_blk_addr;
463 struct dnode_of_data dn;
466 set_new_dnode(&dn, inode, NULL, NULL, 0);
467 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
471 old_blk_addr = dn.data_blkaddr;
473 /* This page is already truncated */
474 if (old_blk_addr == NULL_ADDR)
477 set_page_writeback(page);
480 * If current allocation needs SSR,
481 * it had better in-place writes for updated data.
483 if (old_blk_addr != NEW_ADDR && !is_cold_data(page) &&
484 need_inplace_update(inode)) {
485 rewrite_data_page(F2FS_SB(inode->i_sb), page,
488 write_data_page(inode, page, &dn,
489 old_blk_addr, &new_blk_addr);
490 update_extent_cache(new_blk_addr, &dn);
497 static int f2fs_write_data_page(struct page *page,
498 struct writeback_control *wbc)
500 struct inode *inode = page->mapping->host;
501 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
502 loff_t i_size = i_size_read(inode);
503 const pgoff_t end_index = ((unsigned long long) i_size)
506 bool need_balance_fs = false;
509 if (page->index < end_index)
513 * If the offset is out-of-range of file size,
514 * this page does not have to be written to disk.
516 offset = i_size & (PAGE_CACHE_SIZE - 1);
517 if ((page->index >= end_index + 1) || !offset) {
518 if (S_ISDIR(inode->i_mode)) {
519 dec_page_count(sbi, F2FS_DIRTY_DENTS);
520 inode_dec_dirty_dents(inode);
525 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
527 if (sbi->por_doing) {
528 err = AOP_WRITEPAGE_ACTIVATE;
532 /* Dentry blocks are controlled by checkpoint */
533 if (S_ISDIR(inode->i_mode)) {
534 dec_page_count(sbi, F2FS_DIRTY_DENTS);
535 inode_dec_dirty_dents(inode);
536 err = do_write_data_page(page);
538 int ilock = mutex_lock_op(sbi);
539 err = do_write_data_page(page);
540 mutex_unlock_op(sbi, ilock);
541 need_balance_fs = true;
548 if (wbc->for_reclaim)
549 f2fs_submit_bio(sbi, DATA, true);
551 clear_cold_data(page);
555 f2fs_balance_fs(sbi);
559 wbc->pages_skipped++;
560 set_page_dirty(page);
564 #define MAX_DESIRED_PAGES_WP 4096
566 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
569 struct address_space *mapping = data;
570 int ret = mapping->a_ops->writepage(page, wbc);
571 mapping_set_error(mapping, ret);
575 static int f2fs_write_data_pages(struct address_space *mapping,
576 struct writeback_control *wbc)
578 struct inode *inode = mapping->host;
579 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
582 long excess_nrtw = 0, desired_nrtw;
584 /* deal with chardevs and other special file */
585 if (!mapping->a_ops->writepage)
588 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
589 desired_nrtw = MAX_DESIRED_PAGES_WP;
590 excess_nrtw = desired_nrtw - wbc->nr_to_write;
591 wbc->nr_to_write = desired_nrtw;
594 if (!S_ISDIR(inode->i_mode)) {
595 mutex_lock(&sbi->writepages);
598 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
600 mutex_unlock(&sbi->writepages);
601 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
603 remove_dirty_dir_inode(inode);
605 wbc->nr_to_write -= excess_nrtw;
609 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
610 loff_t pos, unsigned len, unsigned flags,
611 struct page **pagep, void **fsdata)
613 struct inode *inode = mapping->host;
614 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
616 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
617 struct dnode_of_data dn;
621 /* for nobh_write_end */
624 f2fs_balance_fs(sbi);
626 page = grab_cache_page_write_begin(mapping, index, flags);
631 ilock = mutex_lock_op(sbi);
633 set_new_dnode(&dn, inode, NULL, NULL, 0);
634 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
638 if (dn.data_blkaddr == NULL_ADDR)
639 err = reserve_new_block(&dn);
645 mutex_unlock_op(sbi, ilock);
647 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
650 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
651 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
652 unsigned end = start + len;
654 /* Reading beyond i_size is simple: memset to zero */
655 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
659 if (dn.data_blkaddr == NEW_ADDR) {
660 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
662 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
666 if (!PageUptodate(page)) {
667 f2fs_put_page(page, 1);
670 if (page->mapping != mapping) {
671 f2fs_put_page(page, 1);
676 SetPageUptodate(page);
677 clear_cold_data(page);
681 mutex_unlock_op(sbi, ilock);
682 f2fs_put_page(page, 1);
686 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
687 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
689 struct file *file = iocb->ki_filp;
690 struct inode *inode = file->f_mapping->host;
695 /* Needs synchronization with the cleaner */
696 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
700 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
702 struct inode *inode = page->mapping->host;
703 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
704 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
705 dec_page_count(sbi, F2FS_DIRTY_DENTS);
706 inode_dec_dirty_dents(inode);
708 ClearPagePrivate(page);
711 static int f2fs_release_data_page(struct page *page, gfp_t wait)
713 ClearPagePrivate(page);
717 static int f2fs_set_data_page_dirty(struct page *page)
719 struct address_space *mapping = page->mapping;
720 struct inode *inode = mapping->host;
722 SetPageUptodate(page);
723 if (!PageDirty(page)) {
724 __set_page_dirty_nobuffers(page);
725 set_dirty_dir_page(inode, page);
731 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
733 return generic_block_bmap(mapping, block, get_data_block_ro);
736 const struct address_space_operations f2fs_dblock_aops = {
737 .readpage = f2fs_read_data_page,
738 .readpages = f2fs_read_data_pages,
739 .writepage = f2fs_write_data_page,
740 .writepages = f2fs_write_data_pages,
741 .write_begin = f2fs_write_begin,
742 .write_end = nobh_write_end,
743 .set_page_dirty = f2fs_set_data_page_dirty,
744 .invalidatepage = f2fs_invalidate_data_page,
745 .releasepage = f2fs_release_data_page,
746 .direct_IO = f2fs_direct_IO,