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, LOOKUP_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 if (PageUptodate(page)) {
207 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
208 wait_on_page_locked(page);
209 if (!PageUptodate(page)) {
210 f2fs_put_page(page, 0);
211 return ERR_PTR(-EIO);
217 * If it tries to access a hole, return an error.
218 * Because, the callers, functions in dir.c and GC, should be able to know
219 * whether this page exists or not.
221 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
223 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
224 struct address_space *mapping = inode->i_mapping;
225 struct dnode_of_data dn;
229 set_new_dnode(&dn, inode, NULL, NULL, 0);
230 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
235 if (dn.data_blkaddr == NULL_ADDR)
236 return ERR_PTR(-ENOENT);
238 page = grab_cache_page(mapping, index);
240 return ERR_PTR(-ENOMEM);
242 if (PageUptodate(page))
245 BUG_ON(dn.data_blkaddr == NEW_ADDR);
246 BUG_ON(dn.data_blkaddr == NULL_ADDR);
248 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
253 if (!PageUptodate(page)) {
254 f2fs_put_page(page, 1);
255 return ERR_PTR(-EIO);
261 * Caller ensures that this data page is never allocated.
262 * A new zero-filled data page is allocated in the page cache.
264 struct page *get_new_data_page(struct inode *inode, pgoff_t index,
267 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
268 struct address_space *mapping = inode->i_mapping;
270 struct dnode_of_data dn;
273 set_new_dnode(&dn, inode, NULL, NULL, 0);
274 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
278 if (dn.data_blkaddr == NULL_ADDR) {
279 if (reserve_new_block(&dn)) {
281 return ERR_PTR(-ENOSPC);
286 page = grab_cache_page(mapping, index);
288 return ERR_PTR(-ENOMEM);
290 if (PageUptodate(page))
293 if (dn.data_blkaddr == NEW_ADDR) {
294 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
295 SetPageUptodate(page);
297 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
301 if (!PageUptodate(page)) {
302 f2fs_put_page(page, 1);
303 return ERR_PTR(-EIO);
308 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
309 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
310 mark_inode_dirty_sync(inode);
315 static void read_end_io(struct bio *bio, int err)
317 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
318 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
321 struct page *page = bvec->bv_page;
323 if (--bvec >= bio->bi_io_vec)
324 prefetchw(&bvec->bv_page->flags);
327 SetPageUptodate(page);
329 ClearPageUptodate(page);
333 } while (bvec >= bio->bi_io_vec);
334 kfree(bio->bi_private);
339 * Fill the locked page with data located in the block address.
340 * Return unlocked page.
342 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
343 block_t blk_addr, int type)
345 struct block_device *bdev = sbi->sb->s_bdev;
348 down_read(&sbi->bio_sem);
350 /* Allocate a new bio */
351 bio = f2fs_bio_alloc(bdev, 1);
353 /* Initialize the bio */
354 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
355 bio->bi_end_io = read_end_io;
357 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
358 kfree(bio->bi_private);
360 up_read(&sbi->bio_sem);
361 f2fs_put_page(page, 1);
365 submit_bio(type, bio);
366 up_read(&sbi->bio_sem);
371 * This function should be used by the data read flow only where it
372 * does not check the "create" flag that indicates block allocation.
373 * The reason for this special functionality is to exploit VFS readahead
376 static int get_data_block_ro(struct inode *inode, sector_t iblock,
377 struct buffer_head *bh_result, int create)
379 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
380 unsigned maxblocks = bh_result->b_size >> blkbits;
381 struct dnode_of_data dn;
385 /* Get the page offset from the block offset(iblock) */
386 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
388 if (check_extent_cache(inode, pgofs, bh_result))
391 /* When reading holes, we need its node page */
392 set_new_dnode(&dn, inode, NULL, NULL, 0);
393 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
395 return (err == -ENOENT) ? 0 : err;
397 /* It does not support data allocation */
400 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
402 unsigned int end_offset;
404 end_offset = IS_INODE(dn.node_page) ?
408 clear_buffer_new(bh_result);
410 /* Give more consecutive addresses for the read ahead */
411 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
412 if (((datablock_addr(dn.node_page,
414 != (dn.data_blkaddr + i)) || maxblocks == i)
416 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
417 bh_result->b_size = (i << blkbits);
423 static int f2fs_read_data_page(struct file *file, struct page *page)
425 return mpage_readpage(page, get_data_block_ro);
428 static int f2fs_read_data_pages(struct file *file,
429 struct address_space *mapping,
430 struct list_head *pages, unsigned nr_pages)
432 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
435 int do_write_data_page(struct page *page)
437 struct inode *inode = page->mapping->host;
438 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
439 block_t old_blk_addr, new_blk_addr;
440 struct dnode_of_data dn;
443 set_new_dnode(&dn, inode, NULL, NULL, 0);
444 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
448 old_blk_addr = dn.data_blkaddr;
450 /* This page is already truncated */
451 if (old_blk_addr == NULL_ADDR)
454 set_page_writeback(page);
457 * If current allocation needs SSR,
458 * it had better in-place writes for updated data.
460 if (old_blk_addr != NEW_ADDR && !is_cold_data(page) &&
461 need_inplace_update(inode)) {
462 rewrite_data_page(F2FS_SB(inode->i_sb), page,
465 write_data_page(inode, page, &dn,
466 old_blk_addr, &new_blk_addr);
467 update_extent_cache(new_blk_addr, &dn);
468 F2FS_I(inode)->data_version =
469 le64_to_cpu(F2FS_CKPT(sbi)->checkpoint_ver);
476 static int f2fs_write_data_page(struct page *page,
477 struct writeback_control *wbc)
479 struct inode *inode = page->mapping->host;
480 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
481 loff_t i_size = i_size_read(inode);
482 const pgoff_t end_index = ((unsigned long long) i_size)
487 if (page->index < end_index)
491 * If the offset is out-of-range of file size,
492 * this page does not have to be written to disk.
494 offset = i_size & (PAGE_CACHE_SIZE - 1);
495 if ((page->index >= end_index + 1) || !offset) {
496 if (S_ISDIR(inode->i_mode)) {
497 dec_page_count(sbi, F2FS_DIRTY_DENTS);
498 inode_dec_dirty_dents(inode);
503 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
508 if (wbc->for_reclaim && !S_ISDIR(inode->i_mode) && !is_cold_data(page))
511 mutex_lock_op(sbi, DATA_WRITE);
512 if (S_ISDIR(inode->i_mode)) {
513 dec_page_count(sbi, F2FS_DIRTY_DENTS);
514 inode_dec_dirty_dents(inode);
516 err = do_write_data_page(page);
517 if (err && err != -ENOENT) {
518 wbc->pages_skipped++;
519 set_page_dirty(page);
521 mutex_unlock_op(sbi, DATA_WRITE);
523 if (wbc->for_reclaim)
524 f2fs_submit_bio(sbi, DATA, true);
529 clear_cold_data(page);
532 if (!wbc->for_reclaim && !S_ISDIR(inode->i_mode))
533 f2fs_balance_fs(sbi);
538 return (err == -ENOENT) ? 0 : err;
541 wbc->pages_skipped++;
542 set_page_dirty(page);
543 return AOP_WRITEPAGE_ACTIVATE;
546 #define MAX_DESIRED_PAGES_WP 4096
548 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
551 struct address_space *mapping = data;
552 int ret = mapping->a_ops->writepage(page, wbc);
553 mapping_set_error(mapping, ret);
557 static int f2fs_write_data_pages(struct address_space *mapping,
558 struct writeback_control *wbc)
560 struct inode *inode = mapping->host;
561 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
563 long excess_nrtw = 0, desired_nrtw;
565 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
566 desired_nrtw = MAX_DESIRED_PAGES_WP;
567 excess_nrtw = desired_nrtw - wbc->nr_to_write;
568 wbc->nr_to_write = desired_nrtw;
571 if (!S_ISDIR(inode->i_mode))
572 mutex_lock(&sbi->writepages);
573 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
574 if (!S_ISDIR(inode->i_mode))
575 mutex_unlock(&sbi->writepages);
576 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
578 remove_dirty_dir_inode(inode);
580 wbc->nr_to_write -= excess_nrtw;
584 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
585 loff_t pos, unsigned len, unsigned flags,
586 struct page **pagep, void **fsdata)
588 struct inode *inode = mapping->host;
589 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
591 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
592 struct dnode_of_data dn;
595 /* for nobh_write_end */
598 f2fs_balance_fs(sbi);
600 page = grab_cache_page_write_begin(mapping, index, flags);
605 mutex_lock_op(sbi, DATA_NEW);
607 set_new_dnode(&dn, inode, NULL, NULL, 0);
608 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
610 mutex_unlock_op(sbi, DATA_NEW);
611 f2fs_put_page(page, 1);
615 if (dn.data_blkaddr == NULL_ADDR) {
616 err = reserve_new_block(&dn);
619 mutex_unlock_op(sbi, DATA_NEW);
620 f2fs_put_page(page, 1);
626 mutex_unlock_op(sbi, DATA_NEW);
628 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
631 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
632 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
633 unsigned end = start + len;
635 /* Reading beyond i_size is simple: memset to zero */
636 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
640 if (dn.data_blkaddr == NEW_ADDR) {
641 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
643 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
647 if (!PageUptodate(page)) {
648 f2fs_put_page(page, 1);
653 SetPageUptodate(page);
654 clear_cold_data(page);
658 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
659 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
661 struct file *file = iocb->ki_filp;
662 struct inode *inode = file->f_mapping->host;
667 /* Needs synchronization with the cleaner */
668 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
672 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
674 struct inode *inode = page->mapping->host;
675 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
676 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
677 dec_page_count(sbi, F2FS_DIRTY_DENTS);
678 inode_dec_dirty_dents(inode);
680 ClearPagePrivate(page);
683 static int f2fs_release_data_page(struct page *page, gfp_t wait)
685 ClearPagePrivate(page);
689 static int f2fs_set_data_page_dirty(struct page *page)
691 struct address_space *mapping = page->mapping;
692 struct inode *inode = mapping->host;
694 SetPageUptodate(page);
695 if (!PageDirty(page)) {
696 __set_page_dirty_nobuffers(page);
697 set_dirty_dir_page(inode, page);
703 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
705 return generic_block_bmap(mapping, block, get_data_block_ro);
708 const struct address_space_operations f2fs_dblock_aops = {
709 .readpage = f2fs_read_data_page,
710 .readpages = f2fs_read_data_pages,
711 .writepage = f2fs_write_data_page,
712 .writepages = f2fs_write_data_pages,
713 .write_begin = f2fs_write_begin,
714 .write_end = nobh_write_end,
715 .set_page_dirty = f2fs_set_data_page_dirty,
716 .invalidatepage = f2fs_invalidate_data_page,
717 .releasepage = f2fs_release_data_page,
718 .direct_IO = f2fs_direct_IO,