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 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
72 pgoff_t start_fofs, end_fofs;
73 block_t start_blkaddr;
75 read_lock(&fi->ext.ext_lock);
76 if (fi->ext.len == 0) {
77 read_unlock(&fi->ext.ext_lock);
82 start_fofs = fi->ext.fofs;
83 end_fofs = fi->ext.fofs + fi->ext.len - 1;
84 start_blkaddr = fi->ext.blk_addr;
86 if (pgofs >= start_fofs && pgofs <= end_fofs) {
87 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
90 clear_buffer_new(bh_result);
91 map_bh(bh_result, inode->i_sb,
92 start_blkaddr + pgofs - start_fofs);
93 count = end_fofs - pgofs + 1;
94 if (count < (UINT_MAX >> blkbits))
95 bh_result->b_size = (count << blkbits);
97 bh_result->b_size = UINT_MAX;
100 read_unlock(&fi->ext.ext_lock);
103 read_unlock(&fi->ext.ext_lock);
107 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
109 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
110 pgoff_t fofs, start_fofs, end_fofs;
111 block_t start_blkaddr, end_blkaddr;
113 BUG_ON(blk_addr == NEW_ADDR);
114 fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node;
116 /* Update the page address in the parent node */
117 __set_data_blkaddr(dn, blk_addr);
119 write_lock(&fi->ext.ext_lock);
121 start_fofs = fi->ext.fofs;
122 end_fofs = fi->ext.fofs + fi->ext.len - 1;
123 start_blkaddr = fi->ext.blk_addr;
124 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
126 /* Drop and initialize the matched extent */
127 if (fi->ext.len == 1 && fofs == start_fofs)
131 if (fi->ext.len == 0) {
132 if (blk_addr != NULL_ADDR) {
134 fi->ext.blk_addr = blk_addr;
141 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
149 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
154 /* Split the existing extent */
155 if (fi->ext.len > 1 &&
156 fofs >= start_fofs && fofs <= end_fofs) {
157 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
158 fi->ext.len = fofs - start_fofs;
160 fi->ext.fofs = fofs + 1;
161 fi->ext.blk_addr = start_blkaddr +
162 fofs - start_fofs + 1;
163 fi->ext.len -= fofs - start_fofs + 1;
167 write_unlock(&fi->ext.ext_lock);
171 write_unlock(&fi->ext.ext_lock);
176 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
178 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
179 struct address_space *mapping = inode->i_mapping;
180 struct dnode_of_data dn;
184 page = find_get_page(mapping, index);
185 if (page && PageUptodate(page))
187 f2fs_put_page(page, 0);
189 set_new_dnode(&dn, inode, NULL, NULL, 0);
190 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
195 if (dn.data_blkaddr == NULL_ADDR)
196 return ERR_PTR(-ENOENT);
198 /* By fallocate(), there is no cached page, but with NEW_ADDR */
199 if (dn.data_blkaddr == NEW_ADDR)
200 return ERR_PTR(-EINVAL);
202 page = grab_cache_page(mapping, index);
204 return ERR_PTR(-ENOMEM);
206 if (PageUptodate(page)) {
211 err = f2fs_readpage(sbi, page, dn.data_blkaddr,
212 sync ? READ_SYNC : READA);
214 wait_on_page_locked(page);
215 if (!PageUptodate(page)) {
216 f2fs_put_page(page, 0);
217 return ERR_PTR(-EIO);
224 * If it tries to access a hole, return an error.
225 * Because, the callers, functions in dir.c and GC, should be able to know
226 * whether this page exists or not.
228 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
230 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
231 struct address_space *mapping = inode->i_mapping;
232 struct dnode_of_data dn;
236 set_new_dnode(&dn, inode, NULL, NULL, 0);
237 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
242 if (dn.data_blkaddr == NULL_ADDR)
243 return ERR_PTR(-ENOENT);
245 page = grab_cache_page(mapping, index);
247 return ERR_PTR(-ENOMEM);
249 if (PageUptodate(page))
252 BUG_ON(dn.data_blkaddr == NEW_ADDR);
253 BUG_ON(dn.data_blkaddr == NULL_ADDR);
255 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
260 if (!PageUptodate(page)) {
261 f2fs_put_page(page, 1);
262 return ERR_PTR(-EIO);
264 if (page->mapping != mapping) {
265 f2fs_put_page(page, 1);
272 * Caller ensures that this data page is never allocated.
273 * A new zero-filled data page is allocated in the page cache.
275 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
277 * Note that, npage is set only by make_empty_dir.
279 struct page *get_new_data_page(struct inode *inode,
280 struct page *npage, pgoff_t index, bool new_i_size)
282 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
283 struct address_space *mapping = inode->i_mapping;
285 struct dnode_of_data dn;
288 set_new_dnode(&dn, inode, npage, npage, 0);
289 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
293 if (dn.data_blkaddr == NULL_ADDR) {
294 if (reserve_new_block(&dn)) {
297 return ERR_PTR(-ENOSPC);
303 page = grab_cache_page(mapping, index);
305 return ERR_PTR(-ENOMEM);
307 if (PageUptodate(page))
310 if (dn.data_blkaddr == NEW_ADDR) {
311 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
312 SetPageUptodate(page);
314 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
318 if (!PageUptodate(page)) {
319 f2fs_put_page(page, 1);
320 return ERR_PTR(-EIO);
322 if (page->mapping != mapping) {
323 f2fs_put_page(page, 1);
329 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
330 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
331 mark_inode_dirty_sync(inode);
336 static void read_end_io(struct bio *bio, int err)
338 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
339 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
342 struct page *page = bvec->bv_page;
344 if (--bvec >= bio->bi_io_vec)
345 prefetchw(&bvec->bv_page->flags);
348 SetPageUptodate(page);
350 ClearPageUptodate(page);
354 } while (bvec >= bio->bi_io_vec);
355 kfree(bio->bi_private);
360 * Fill the locked page with data located in the block address.
361 * Return unlocked page.
363 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
364 block_t blk_addr, int type)
366 struct block_device *bdev = sbi->sb->s_bdev;
369 trace_f2fs_readpage(page, blk_addr, type);
371 down_read(&sbi->bio_sem);
373 /* Allocate a new bio */
374 bio = f2fs_bio_alloc(bdev, 1);
376 /* Initialize the bio */
377 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
378 bio->bi_end_io = read_end_io;
380 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
381 kfree(bio->bi_private);
383 up_read(&sbi->bio_sem);
384 f2fs_put_page(page, 1);
388 submit_bio(type, bio);
389 up_read(&sbi->bio_sem);
394 * This function should be used by the data read flow only where it
395 * does not check the "create" flag that indicates block allocation.
396 * The reason for this special functionality is to exploit VFS readahead
399 static int get_data_block_ro(struct inode *inode, sector_t iblock,
400 struct buffer_head *bh_result, int create)
402 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
403 unsigned maxblocks = bh_result->b_size >> blkbits;
404 struct dnode_of_data dn;
408 /* Get the page offset from the block offset(iblock) */
409 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
411 if (check_extent_cache(inode, pgofs, bh_result)) {
412 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
416 /* When reading holes, we need its node page */
417 set_new_dnode(&dn, inode, NULL, NULL, 0);
418 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
420 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
421 return (err == -ENOENT) ? 0 : err;
424 /* It does not support data allocation */
427 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
429 unsigned int end_offset;
431 end_offset = IS_INODE(dn.node_page) ?
435 clear_buffer_new(bh_result);
437 /* Give more consecutive addresses for the read ahead */
438 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
439 if (((datablock_addr(dn.node_page,
441 != (dn.data_blkaddr + i)) || maxblocks == i)
443 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
444 bh_result->b_size = (i << blkbits);
447 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
451 static int f2fs_read_data_page(struct file *file, struct page *page)
453 return mpage_readpage(page, get_data_block_ro);
456 static int f2fs_read_data_pages(struct file *file,
457 struct address_space *mapping,
458 struct list_head *pages, unsigned nr_pages)
460 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
463 int do_write_data_page(struct page *page)
465 struct inode *inode = page->mapping->host;
466 block_t old_blk_addr, new_blk_addr;
467 struct dnode_of_data dn;
470 set_new_dnode(&dn, inode, NULL, NULL, 0);
471 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
475 old_blk_addr = dn.data_blkaddr;
477 /* This page is already truncated */
478 if (old_blk_addr == NULL_ADDR)
481 set_page_writeback(page);
484 * If current allocation needs SSR,
485 * it had better in-place writes for updated data.
487 if (old_blk_addr != NEW_ADDR && !is_cold_data(page) &&
488 need_inplace_update(inode)) {
489 rewrite_data_page(F2FS_SB(inode->i_sb), page,
492 write_data_page(inode, page, &dn,
493 old_blk_addr, &new_blk_addr);
494 update_extent_cache(new_blk_addr, &dn);
501 static int f2fs_write_data_page(struct page *page,
502 struct writeback_control *wbc)
504 struct inode *inode = page->mapping->host;
505 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
506 loff_t i_size = i_size_read(inode);
507 const pgoff_t end_index = ((unsigned long long) i_size)
510 bool need_balance_fs = false;
513 if (page->index < end_index)
517 * If the offset is out-of-range of file size,
518 * this page does not have to be written to disk.
520 offset = i_size & (PAGE_CACHE_SIZE - 1);
521 if ((page->index >= end_index + 1) || !offset) {
522 if (S_ISDIR(inode->i_mode)) {
523 dec_page_count(sbi, F2FS_DIRTY_DENTS);
524 inode_dec_dirty_dents(inode);
529 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
531 if (sbi->por_doing) {
532 err = AOP_WRITEPAGE_ACTIVATE;
536 /* Dentry blocks are controlled by checkpoint */
537 if (S_ISDIR(inode->i_mode)) {
538 dec_page_count(sbi, F2FS_DIRTY_DENTS);
539 inode_dec_dirty_dents(inode);
540 err = do_write_data_page(page);
542 int ilock = mutex_lock_op(sbi);
543 err = do_write_data_page(page);
544 mutex_unlock_op(sbi, ilock);
545 need_balance_fs = true;
552 if (wbc->for_reclaim)
553 f2fs_submit_bio(sbi, DATA, true);
555 clear_cold_data(page);
559 f2fs_balance_fs(sbi);
563 wbc->pages_skipped++;
564 set_page_dirty(page);
568 #define MAX_DESIRED_PAGES_WP 4096
570 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
573 struct address_space *mapping = data;
574 int ret = mapping->a_ops->writepage(page, wbc);
575 mapping_set_error(mapping, ret);
579 static int f2fs_write_data_pages(struct address_space *mapping,
580 struct writeback_control *wbc)
582 struct inode *inode = mapping->host;
583 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
586 long excess_nrtw = 0, desired_nrtw;
588 /* deal with chardevs and other special file */
589 if (!mapping->a_ops->writepage)
592 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
593 desired_nrtw = MAX_DESIRED_PAGES_WP;
594 excess_nrtw = desired_nrtw - wbc->nr_to_write;
595 wbc->nr_to_write = desired_nrtw;
598 if (!S_ISDIR(inode->i_mode)) {
599 mutex_lock(&sbi->writepages);
602 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
604 mutex_unlock(&sbi->writepages);
605 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
607 remove_dirty_dir_inode(inode);
609 wbc->nr_to_write -= excess_nrtw;
613 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
614 loff_t pos, unsigned len, unsigned flags,
615 struct page **pagep, void **fsdata)
617 struct inode *inode = mapping->host;
618 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
620 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
621 struct dnode_of_data dn;
625 /* for nobh_write_end */
628 f2fs_balance_fs(sbi);
630 page = grab_cache_page_write_begin(mapping, index, flags);
635 ilock = mutex_lock_op(sbi);
637 set_new_dnode(&dn, inode, NULL, NULL, 0);
638 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
642 if (dn.data_blkaddr == NULL_ADDR)
643 err = reserve_new_block(&dn);
649 mutex_unlock_op(sbi, ilock);
651 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
654 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
655 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
656 unsigned end = start + len;
658 /* Reading beyond i_size is simple: memset to zero */
659 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
663 if (dn.data_blkaddr == NEW_ADDR) {
664 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
666 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
670 if (!PageUptodate(page)) {
671 f2fs_put_page(page, 1);
674 if (page->mapping != mapping) {
675 f2fs_put_page(page, 1);
680 SetPageUptodate(page);
681 clear_cold_data(page);
685 mutex_unlock_op(sbi, ilock);
686 f2fs_put_page(page, 1);
690 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
691 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
693 struct file *file = iocb->ki_filp;
694 struct inode *inode = file->f_mapping->host;
699 /* Needs synchronization with the cleaner */
700 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
704 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
706 struct inode *inode = page->mapping->host;
707 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
708 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
709 dec_page_count(sbi, F2FS_DIRTY_DENTS);
710 inode_dec_dirty_dents(inode);
712 ClearPagePrivate(page);
715 static int f2fs_release_data_page(struct page *page, gfp_t wait)
717 ClearPagePrivate(page);
721 static int f2fs_set_data_page_dirty(struct page *page)
723 struct address_space *mapping = page->mapping;
724 struct inode *inode = mapping->host;
726 SetPageUptodate(page);
727 if (!PageDirty(page)) {
728 __set_page_dirty_nobuffers(page);
729 set_dirty_dir_page(inode, page);
735 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
737 return generic_block_bmap(mapping, block, get_data_block_ro);
740 const struct address_space_operations f2fs_dblock_aops = {
741 .readpage = f2fs_read_data_page,
742 .readpages = f2fs_read_data_pages,
743 .writepage = f2fs_write_data_page,
744 .writepages = f2fs_write_data_pages,
745 .write_begin = f2fs_write_begin,
746 .write_end = nobh_write_end,
747 .set_page_dirty = f2fs_set_data_page_dirty,
748 .invalidatepage = f2fs_invalidate_data_page,
749 .releasepage = f2fs_release_data_page,
750 .direct_IO = f2fs_direct_IO,