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>
25 * Lock ordering for the change of data block address:
28 * update block addresses in the node page
30 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
34 struct page *node_page = dn->node_page;
35 unsigned int ofs_in_node = dn->ofs_in_node;
37 wait_on_page_writeback(node_page);
39 rn = (struct f2fs_node *)page_address(node_page);
41 /* Get physical address of data block */
42 addr_array = blkaddr_in_node(rn);
43 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
44 set_page_dirty(node_page);
47 int reserve_new_block(struct dnode_of_data *dn)
49 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
51 if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
53 if (!inc_valid_block_count(sbi, dn->inode, 1))
56 __set_data_blkaddr(dn, NEW_ADDR);
57 dn->data_blkaddr = NEW_ADDR;
62 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
63 struct buffer_head *bh_result)
65 struct f2fs_inode_info *fi = F2FS_I(inode);
66 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
67 pgoff_t start_fofs, end_fofs;
68 block_t start_blkaddr;
70 read_lock(&fi->ext.ext_lock);
71 if (fi->ext.len == 0) {
72 read_unlock(&fi->ext.ext_lock);
77 start_fofs = fi->ext.fofs;
78 end_fofs = fi->ext.fofs + fi->ext.len - 1;
79 start_blkaddr = fi->ext.blk_addr;
81 if (pgofs >= start_fofs && pgofs <= end_fofs) {
82 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
85 clear_buffer_new(bh_result);
86 map_bh(bh_result, inode->i_sb,
87 start_blkaddr + pgofs - start_fofs);
88 count = end_fofs - pgofs + 1;
89 if (count < (UINT_MAX >> blkbits))
90 bh_result->b_size = (count << blkbits);
92 bh_result->b_size = UINT_MAX;
95 read_unlock(&fi->ext.ext_lock);
98 read_unlock(&fi->ext.ext_lock);
102 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
104 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
105 pgoff_t fofs, start_fofs, end_fofs;
106 block_t start_blkaddr, end_blkaddr;
108 BUG_ON(blk_addr == NEW_ADDR);
109 fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node;
111 /* Update the page address in the parent node */
112 __set_data_blkaddr(dn, blk_addr);
114 write_lock(&fi->ext.ext_lock);
116 start_fofs = fi->ext.fofs;
117 end_fofs = fi->ext.fofs + fi->ext.len - 1;
118 start_blkaddr = fi->ext.blk_addr;
119 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
121 /* Drop and initialize the matched extent */
122 if (fi->ext.len == 1 && fofs == start_fofs)
126 if (fi->ext.len == 0) {
127 if (blk_addr != NULL_ADDR) {
129 fi->ext.blk_addr = blk_addr;
136 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
144 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
149 /* Split the existing extent */
150 if (fi->ext.len > 1 &&
151 fofs >= start_fofs && fofs <= end_fofs) {
152 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
153 fi->ext.len = fofs - start_fofs;
155 fi->ext.fofs = fofs + 1;
156 fi->ext.blk_addr = start_blkaddr +
157 fofs - start_fofs + 1;
158 fi->ext.len -= fofs - start_fofs + 1;
162 write_unlock(&fi->ext.ext_lock);
166 write_unlock(&fi->ext.ext_lock);
171 struct page *find_data_page(struct inode *inode, pgoff_t index)
173 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
174 struct address_space *mapping = inode->i_mapping;
175 struct dnode_of_data dn;
179 page = find_get_page(mapping, index);
180 if (page && PageUptodate(page))
182 f2fs_put_page(page, 0);
184 set_new_dnode(&dn, inode, NULL, NULL, 0);
185 err = get_dnode_of_data(&dn, index, RDONLY_NODE);
190 if (dn.data_blkaddr == NULL_ADDR)
191 return ERR_PTR(-ENOENT);
193 /* By fallocate(), there is no cached page, but with NEW_ADDR */
194 if (dn.data_blkaddr == NEW_ADDR)
195 return ERR_PTR(-EINVAL);
197 page = grab_cache_page(mapping, index);
199 return ERR_PTR(-ENOMEM);
201 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
203 f2fs_put_page(page, 1);
211 * If it tries to access a hole, return an error.
212 * Because, the callers, functions in dir.c and GC, should be able to know
213 * whether this page exists or not.
215 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
217 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
218 struct address_space *mapping = inode->i_mapping;
219 struct dnode_of_data dn;
223 set_new_dnode(&dn, inode, NULL, NULL, 0);
224 err = get_dnode_of_data(&dn, index, RDONLY_NODE);
229 if (dn.data_blkaddr == NULL_ADDR)
230 return ERR_PTR(-ENOENT);
232 page = grab_cache_page(mapping, index);
234 return ERR_PTR(-ENOMEM);
236 if (PageUptodate(page))
239 BUG_ON(dn.data_blkaddr == NEW_ADDR);
240 BUG_ON(dn.data_blkaddr == NULL_ADDR);
242 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
244 f2fs_put_page(page, 1);
251 * Caller ensures that this data page is never allocated.
252 * A new zero-filled data page is allocated in the page cache.
254 struct page *get_new_data_page(struct inode *inode, pgoff_t index,
257 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
258 struct address_space *mapping = inode->i_mapping;
260 struct dnode_of_data dn;
263 set_new_dnode(&dn, inode, NULL, NULL, 0);
264 err = get_dnode_of_data(&dn, index, 0);
268 if (dn.data_blkaddr == NULL_ADDR) {
269 if (reserve_new_block(&dn)) {
271 return ERR_PTR(-ENOSPC);
276 page = grab_cache_page(mapping, index);
278 return ERR_PTR(-ENOMEM);
280 if (PageUptodate(page))
283 if (dn.data_blkaddr == NEW_ADDR) {
284 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
286 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
288 f2fs_put_page(page, 1);
292 SetPageUptodate(page);
295 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
296 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
297 mark_inode_dirty_sync(inode);
302 static void read_end_io(struct bio *bio, int err)
304 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
305 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
308 struct page *page = bvec->bv_page;
310 if (--bvec >= bio->bi_io_vec)
311 prefetchw(&bvec->bv_page->flags);
314 SetPageUptodate(page);
316 ClearPageUptodate(page);
320 } while (bvec >= bio->bi_io_vec);
321 kfree(bio->bi_private);
326 * Fill the locked page with data located in the block address.
327 * Read operation is synchronous, and caller must unlock the page.
329 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
330 block_t blk_addr, int type)
332 struct block_device *bdev = sbi->sb->s_bdev;
333 bool sync = (type == READ_SYNC);
336 /* This page can be already read by other threads */
337 if (PageUptodate(page)) {
343 down_read(&sbi->bio_sem);
345 /* Allocate a new bio */
346 bio = f2fs_bio_alloc(bdev, blk_addr << (sbi->log_blocksize - 9),
347 1, GFP_NOFS | __GFP_HIGH);
349 /* Initialize the bio */
350 bio->bi_end_io = read_end_io;
351 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
352 kfree(bio->bi_private);
354 up_read(&sbi->bio_sem);
358 submit_bio(type, bio);
359 up_read(&sbi->bio_sem);
361 /* wait for read completion if sync */
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, RDONLY_NODE);
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, RDONLY_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_write_data_pages(struct address_space *mapping,
549 struct writeback_control *wbc)
551 struct inode *inode = mapping->host;
552 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
554 long excess_nrtw = 0, desired_nrtw;
556 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
557 desired_nrtw = MAX_DESIRED_PAGES_WP;
558 excess_nrtw = desired_nrtw - wbc->nr_to_write;
559 wbc->nr_to_write = desired_nrtw;
562 if (!S_ISDIR(inode->i_mode))
563 mutex_lock(&sbi->writepages);
564 ret = generic_writepages(mapping, wbc);
565 if (!S_ISDIR(inode->i_mode))
566 mutex_unlock(&sbi->writepages);
567 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
569 remove_dirty_dir_inode(inode);
571 wbc->nr_to_write -= excess_nrtw;
575 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
576 loff_t pos, unsigned len, unsigned flags,
577 struct page **pagep, void **fsdata)
579 struct inode *inode = mapping->host;
580 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
582 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
583 struct dnode_of_data dn;
586 /* for nobh_write_end */
589 f2fs_balance_fs(sbi);
591 page = grab_cache_page_write_begin(mapping, index, flags);
596 mutex_lock_op(sbi, DATA_NEW);
598 set_new_dnode(&dn, inode, NULL, NULL, 0);
599 err = get_dnode_of_data(&dn, index, 0);
601 mutex_unlock_op(sbi, DATA_NEW);
602 f2fs_put_page(page, 1);
606 if (dn.data_blkaddr == NULL_ADDR) {
607 err = reserve_new_block(&dn);
610 mutex_unlock_op(sbi, DATA_NEW);
611 f2fs_put_page(page, 1);
617 mutex_unlock_op(sbi, DATA_NEW);
619 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
622 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
623 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
624 unsigned end = start + len;
626 /* Reading beyond i_size is simple: memset to zero */
627 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
631 if (dn.data_blkaddr == NEW_ADDR) {
632 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
634 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
636 f2fs_put_page(page, 1);
640 SetPageUptodate(page);
641 clear_cold_data(page);
645 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
646 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
648 struct file *file = iocb->ki_filp;
649 struct inode *inode = file->f_mapping->host;
654 /* Needs synchronization with the cleaner */
655 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
659 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
661 struct inode *inode = page->mapping->host;
662 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
663 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
664 dec_page_count(sbi, F2FS_DIRTY_DENTS);
665 inode_dec_dirty_dents(inode);
667 ClearPagePrivate(page);
670 static int f2fs_release_data_page(struct page *page, gfp_t wait)
672 ClearPagePrivate(page);
676 static int f2fs_set_data_page_dirty(struct page *page)
678 struct address_space *mapping = page->mapping;
679 struct inode *inode = mapping->host;
681 SetPageUptodate(page);
682 if (!PageDirty(page)) {
683 __set_page_dirty_nobuffers(page);
684 set_dirty_dir_page(inode, page);
690 const struct address_space_operations f2fs_dblock_aops = {
691 .readpage = f2fs_read_data_page,
692 .readpages = f2fs_read_data_pages,
693 .writepage = f2fs_write_data_page,
694 .writepages = f2fs_write_data_pages,
695 .write_begin = f2fs_write_begin,
696 .write_end = nobh_write_end,
697 .set_page_dirty = f2fs_set_data_page_dirty,
698 .invalidatepage = f2fs_invalidate_data_page,
699 .releasepage = f2fs_release_data_page,
700 .direct_IO = f2fs_direct_IO,