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 f2fs_wait_on_page_writeback(node_page, NODE, false);
42 rn = F2FS_NODE(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), fi) +
123 /* Update the page address in the parent node */
124 __set_data_blkaddr(dn, blk_addr);
126 write_lock(&fi->ext.ext_lock);
128 start_fofs = fi->ext.fofs;
129 end_fofs = fi->ext.fofs + fi->ext.len - 1;
130 start_blkaddr = fi->ext.blk_addr;
131 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
133 /* Drop and initialize the matched extent */
134 if (fi->ext.len == 1 && fofs == start_fofs)
138 if (fi->ext.len == 0) {
139 if (blk_addr != NULL_ADDR) {
141 fi->ext.blk_addr = blk_addr;
148 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
156 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
161 /* Split the existing extent */
162 if (fi->ext.len > 1 &&
163 fofs >= start_fofs && fofs <= end_fofs) {
164 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
165 fi->ext.len = fofs - start_fofs;
167 fi->ext.fofs = fofs + 1;
168 fi->ext.blk_addr = start_blkaddr +
169 fofs - start_fofs + 1;
170 fi->ext.len -= fofs - start_fofs + 1;
174 write_unlock(&fi->ext.ext_lock);
178 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))
264 * A new dentry page is allocated but not able to be written, since its
265 * new inode page couldn't be allocated due to -ENOSPC.
266 * In such the case, its blkaddr can be remained as NEW_ADDR.
267 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
269 if (dn.data_blkaddr == NEW_ADDR) {
270 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
271 SetPageUptodate(page);
275 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
280 if (!PageUptodate(page)) {
281 f2fs_put_page(page, 1);
282 return ERR_PTR(-EIO);
284 if (page->mapping != mapping) {
285 f2fs_put_page(page, 1);
292 * Caller ensures that this data page is never allocated.
293 * A new zero-filled data page is allocated in the page cache.
295 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
297 * Note that, npage is set only by make_empty_dir.
299 struct page *get_new_data_page(struct inode *inode,
300 struct page *npage, pgoff_t index, bool new_i_size)
302 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
303 struct address_space *mapping = inode->i_mapping;
305 struct dnode_of_data dn;
308 set_new_dnode(&dn, inode, npage, npage, 0);
309 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
313 if (dn.data_blkaddr == NULL_ADDR) {
314 if (reserve_new_block(&dn)) {
317 return ERR_PTR(-ENOSPC);
323 page = grab_cache_page(mapping, index);
325 return ERR_PTR(-ENOMEM);
327 if (PageUptodate(page))
330 if (dn.data_blkaddr == NEW_ADDR) {
331 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
332 SetPageUptodate(page);
334 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
338 if (!PageUptodate(page)) {
339 f2fs_put_page(page, 1);
340 return ERR_PTR(-EIO);
342 if (page->mapping != mapping) {
343 f2fs_put_page(page, 1);
349 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
350 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
351 /* Only the directory inode sets new_i_size */
352 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
353 mark_inode_dirty_sync(inode);
358 static void read_end_io(struct bio *bio, int err)
360 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
361 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
364 struct page *page = bvec->bv_page;
366 if (--bvec >= bio->bi_io_vec)
367 prefetchw(&bvec->bv_page->flags);
370 SetPageUptodate(page);
372 ClearPageUptodate(page);
376 } while (bvec >= bio->bi_io_vec);
381 * Fill the locked page with data located in the block address.
382 * Return unlocked page.
384 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
385 block_t blk_addr, int type)
387 struct block_device *bdev = sbi->sb->s_bdev;
390 trace_f2fs_readpage(page, blk_addr, type);
392 down_read(&sbi->bio_sem);
394 /* Allocate a new bio */
395 bio = f2fs_bio_alloc(bdev, 1);
397 /* Initialize the bio */
398 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
399 bio->bi_end_io = read_end_io;
401 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
403 up_read(&sbi->bio_sem);
404 f2fs_put_page(page, 1);
408 submit_bio(type, bio);
409 up_read(&sbi->bio_sem);
414 * This function should be used by the data read flow only where it
415 * does not check the "create" flag that indicates block allocation.
416 * The reason for this special functionality is to exploit VFS readahead
419 static int get_data_block_ro(struct inode *inode, sector_t iblock,
420 struct buffer_head *bh_result, int create)
422 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
423 unsigned maxblocks = bh_result->b_size >> blkbits;
424 struct dnode_of_data dn;
428 /* Get the page offset from the block offset(iblock) */
429 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
431 if (check_extent_cache(inode, pgofs, bh_result)) {
432 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
436 /* When reading holes, we need its node page */
437 set_new_dnode(&dn, inode, NULL, NULL, 0);
438 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
440 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
441 return (err == -ENOENT) ? 0 : err;
444 /* It does not support data allocation */
447 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
449 unsigned int end_offset;
451 end_offset = IS_INODE(dn.node_page) ?
452 ADDRS_PER_INODE(F2FS_I(inode)) :
455 clear_buffer_new(bh_result);
457 /* Give more consecutive addresses for the read ahead */
458 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
459 if (((datablock_addr(dn.node_page,
461 != (dn.data_blkaddr + i)) || maxblocks == i)
463 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
464 bh_result->b_size = (i << blkbits);
467 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
471 static int f2fs_read_data_page(struct file *file, struct page *page)
473 return mpage_readpage(page, get_data_block_ro);
476 static int f2fs_read_data_pages(struct file *file,
477 struct address_space *mapping,
478 struct list_head *pages, unsigned nr_pages)
480 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
483 int do_write_data_page(struct page *page)
485 struct inode *inode = page->mapping->host;
486 block_t old_blk_addr, new_blk_addr;
487 struct dnode_of_data dn;
490 set_new_dnode(&dn, inode, NULL, NULL, 0);
491 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
495 old_blk_addr = dn.data_blkaddr;
497 /* This page is already truncated */
498 if (old_blk_addr == NULL_ADDR)
501 set_page_writeback(page);
504 * If current allocation needs SSR,
505 * it had better in-place writes for updated data.
507 if (unlikely(old_blk_addr != NEW_ADDR &&
508 !is_cold_data(page) &&
509 need_inplace_update(inode))) {
510 rewrite_data_page(F2FS_SB(inode->i_sb), page,
513 write_data_page(inode, page, &dn,
514 old_blk_addr, &new_blk_addr);
515 update_extent_cache(new_blk_addr, &dn);
522 static int f2fs_write_data_page(struct page *page,
523 struct writeback_control *wbc)
525 struct inode *inode = page->mapping->host;
526 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
527 loff_t i_size = i_size_read(inode);
528 const pgoff_t end_index = ((unsigned long long) i_size)
531 bool need_balance_fs = false;
534 if (page->index < end_index)
538 * If the offset is out-of-range of file size,
539 * this page does not have to be written to disk.
541 offset = i_size & (PAGE_CACHE_SIZE - 1);
542 if ((page->index >= end_index + 1) || !offset) {
543 if (S_ISDIR(inode->i_mode)) {
544 dec_page_count(sbi, F2FS_DIRTY_DENTS);
545 inode_dec_dirty_dents(inode);
550 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
552 if (sbi->por_doing) {
553 err = AOP_WRITEPAGE_ACTIVATE;
557 /* Dentry blocks are controlled by checkpoint */
558 if (S_ISDIR(inode->i_mode)) {
559 dec_page_count(sbi, F2FS_DIRTY_DENTS);
560 inode_dec_dirty_dents(inode);
561 err = do_write_data_page(page);
564 err = do_write_data_page(page);
566 need_balance_fs = true;
573 if (wbc->for_reclaim)
574 f2fs_submit_bio(sbi, DATA, true);
576 clear_cold_data(page);
580 f2fs_balance_fs(sbi);
584 wbc->pages_skipped++;
585 set_page_dirty(page);
589 #define MAX_DESIRED_PAGES_WP 4096
591 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
594 struct address_space *mapping = data;
595 int ret = mapping->a_ops->writepage(page, wbc);
596 mapping_set_error(mapping, ret);
600 static int f2fs_write_data_pages(struct address_space *mapping,
601 struct writeback_control *wbc)
603 struct inode *inode = mapping->host;
604 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
607 long excess_nrtw = 0, desired_nrtw;
609 /* deal with chardevs and other special file */
610 if (!mapping->a_ops->writepage)
613 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
614 desired_nrtw = MAX_DESIRED_PAGES_WP;
615 excess_nrtw = desired_nrtw - wbc->nr_to_write;
616 wbc->nr_to_write = desired_nrtw;
619 if (!S_ISDIR(inode->i_mode)) {
620 mutex_lock(&sbi->writepages);
623 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
625 mutex_unlock(&sbi->writepages);
626 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
628 remove_dirty_dir_inode(inode);
630 wbc->nr_to_write -= excess_nrtw;
634 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
635 loff_t pos, unsigned len, unsigned flags,
636 struct page **pagep, void **fsdata)
638 struct inode *inode = mapping->host;
639 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
641 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
642 struct dnode_of_data dn;
645 f2fs_balance_fs(sbi);
647 page = grab_cache_page_write_begin(mapping, index, flags);
654 set_new_dnode(&dn, inode, NULL, NULL, 0);
655 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
659 if (dn.data_blkaddr == NULL_ADDR)
660 err = reserve_new_block(&dn);
668 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
671 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
672 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
673 unsigned end = start + len;
675 /* Reading beyond i_size is simple: memset to zero */
676 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
680 if (dn.data_blkaddr == NEW_ADDR) {
681 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
683 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
687 if (!PageUptodate(page)) {
688 f2fs_put_page(page, 1);
691 if (page->mapping != mapping) {
692 f2fs_put_page(page, 1);
697 SetPageUptodate(page);
698 clear_cold_data(page);
703 f2fs_put_page(page, 1);
707 static int f2fs_write_end(struct file *file,
708 struct address_space *mapping,
709 loff_t pos, unsigned len, unsigned copied,
710 struct page *page, void *fsdata)
712 struct inode *inode = page->mapping->host;
714 SetPageUptodate(page);
715 set_page_dirty(page);
717 if (pos + copied > i_size_read(inode)) {
718 i_size_write(inode, pos + copied);
719 mark_inode_dirty(inode);
720 update_inode_page(inode);
724 page_cache_release(page);
728 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
729 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
731 struct file *file = iocb->ki_filp;
732 struct inode *inode = file->f_mapping->host;
737 /* Needs synchronization with the cleaner */
738 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
742 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
745 struct inode *inode = page->mapping->host;
746 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
747 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
748 dec_page_count(sbi, F2FS_DIRTY_DENTS);
749 inode_dec_dirty_dents(inode);
751 ClearPagePrivate(page);
754 static int f2fs_release_data_page(struct page *page, gfp_t wait)
756 ClearPagePrivate(page);
760 static int f2fs_set_data_page_dirty(struct page *page)
762 struct address_space *mapping = page->mapping;
763 struct inode *inode = mapping->host;
765 SetPageUptodate(page);
766 if (!PageDirty(page)) {
767 __set_page_dirty_nobuffers(page);
768 set_dirty_dir_page(inode, page);
774 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
776 return generic_block_bmap(mapping, block, get_data_block_ro);
779 const struct address_space_operations f2fs_dblock_aops = {
780 .readpage = f2fs_read_data_page,
781 .readpages = f2fs_read_data_pages,
782 .writepage = f2fs_write_data_page,
783 .writepages = f2fs_write_data_pages,
784 .write_begin = f2fs_write_begin,
785 .write_end = f2fs_write_end,
786 .set_page_dirty = f2fs_set_data_page_dirty,
787 .invalidatepage = f2fs_invalidate_data_page,
788 .releasepage = f2fs_release_data_page,
789 .direct_IO = f2fs_direct_IO,