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)
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, READ_SYNC);
211 wait_on_page_locked(page);
212 if (!PageUptodate(page)) {
213 f2fs_put_page(page, 0);
214 return ERR_PTR(-EIO);
220 * If it tries to access a hole, return an error.
221 * Because, the callers, functions in dir.c and GC, should be able to know
222 * whether this page exists or not.
224 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
226 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
227 struct address_space *mapping = inode->i_mapping;
228 struct dnode_of_data dn;
232 set_new_dnode(&dn, inode, NULL, NULL, 0);
233 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
238 if (dn.data_blkaddr == NULL_ADDR)
239 return ERR_PTR(-ENOENT);
241 page = grab_cache_page(mapping, index);
243 return ERR_PTR(-ENOMEM);
245 if (PageUptodate(page))
248 BUG_ON(dn.data_blkaddr == NEW_ADDR);
249 BUG_ON(dn.data_blkaddr == NULL_ADDR);
251 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
256 if (!PageUptodate(page)) {
257 f2fs_put_page(page, 1);
258 return ERR_PTR(-EIO);
264 * Caller ensures that this data page is never allocated.
265 * A new zero-filled data page is allocated in the page cache.
267 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
270 struct page *get_new_data_page(struct inode *inode, pgoff_t index,
273 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
274 struct address_space *mapping = inode->i_mapping;
276 struct dnode_of_data dn;
279 set_new_dnode(&dn, inode, NULL, NULL, 0);
280 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
284 if (dn.data_blkaddr == NULL_ADDR) {
285 if (reserve_new_block(&dn)) {
287 return ERR_PTR(-ENOSPC);
292 page = grab_cache_page(mapping, index);
294 return ERR_PTR(-ENOMEM);
296 if (PageUptodate(page))
299 if (dn.data_blkaddr == NEW_ADDR) {
300 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
301 SetPageUptodate(page);
303 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
307 if (!PageUptodate(page)) {
308 f2fs_put_page(page, 1);
309 return ERR_PTR(-EIO);
314 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
315 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
316 mark_inode_dirty_sync(inode);
321 static void read_end_io(struct bio *bio, int err)
323 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
324 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
327 struct page *page = bvec->bv_page;
329 if (--bvec >= bio->bi_io_vec)
330 prefetchw(&bvec->bv_page->flags);
333 SetPageUptodate(page);
335 ClearPageUptodate(page);
339 } while (bvec >= bio->bi_io_vec);
340 kfree(bio->bi_private);
345 * Fill the locked page with data located in the block address.
346 * Return unlocked page.
348 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
349 block_t blk_addr, int type)
351 struct block_device *bdev = sbi->sb->s_bdev;
354 trace_f2fs_readpage(page, blk_addr, type);
356 down_read(&sbi->bio_sem);
358 /* Allocate a new bio */
359 bio = f2fs_bio_alloc(bdev, 1);
361 /* Initialize the bio */
362 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
363 bio->bi_end_io = read_end_io;
365 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
366 kfree(bio->bi_private);
368 up_read(&sbi->bio_sem);
369 f2fs_put_page(page, 1);
373 submit_bio(type, bio);
374 up_read(&sbi->bio_sem);
379 * This function should be used by the data read flow only where it
380 * does not check the "create" flag that indicates block allocation.
381 * The reason for this special functionality is to exploit VFS readahead
384 static int get_data_block_ro(struct inode *inode, sector_t iblock,
385 struct buffer_head *bh_result, int create)
387 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
388 unsigned maxblocks = bh_result->b_size >> blkbits;
389 struct dnode_of_data dn;
393 /* Get the page offset from the block offset(iblock) */
394 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
396 if (check_extent_cache(inode, pgofs, bh_result)) {
397 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
401 /* When reading holes, we need its node page */
402 set_new_dnode(&dn, inode, NULL, NULL, 0);
403 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
405 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
406 return (err == -ENOENT) ? 0 : err;
409 /* It does not support data allocation */
412 if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
414 unsigned int end_offset;
416 end_offset = IS_INODE(dn.node_page) ?
420 clear_buffer_new(bh_result);
422 /* Give more consecutive addresses for the read ahead */
423 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
424 if (((datablock_addr(dn.node_page,
426 != (dn.data_blkaddr + i)) || maxblocks == i)
428 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
429 bh_result->b_size = (i << blkbits);
432 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
436 static int f2fs_read_data_page(struct file *file, struct page *page)
438 return mpage_readpage(page, get_data_block_ro);
441 static int f2fs_read_data_pages(struct file *file,
442 struct address_space *mapping,
443 struct list_head *pages, unsigned nr_pages)
445 return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
448 int do_write_data_page(struct page *page)
450 struct inode *inode = page->mapping->host;
451 block_t old_blk_addr, new_blk_addr;
452 struct dnode_of_data dn;
455 set_new_dnode(&dn, inode, NULL, NULL, 0);
456 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
460 old_blk_addr = dn.data_blkaddr;
462 /* This page is already truncated */
463 if (old_blk_addr == NULL_ADDR)
466 set_page_writeback(page);
469 * If current allocation needs SSR,
470 * it had better in-place writes for updated data.
472 if (old_blk_addr != NEW_ADDR && !is_cold_data(page) &&
473 need_inplace_update(inode)) {
474 rewrite_data_page(F2FS_SB(inode->i_sb), page,
477 write_data_page(inode, page, &dn,
478 old_blk_addr, &new_blk_addr);
479 update_extent_cache(new_blk_addr, &dn);
486 static int f2fs_write_data_page(struct page *page,
487 struct writeback_control *wbc)
489 struct inode *inode = page->mapping->host;
490 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
491 loff_t i_size = i_size_read(inode);
492 const pgoff_t end_index = ((unsigned long long) i_size)
495 bool need_balance_fs = false;
498 if (page->index < end_index)
502 * If the offset is out-of-range of file size,
503 * this page does not have to be written to disk.
505 offset = i_size & (PAGE_CACHE_SIZE - 1);
506 if ((page->index >= end_index + 1) || !offset) {
507 if (S_ISDIR(inode->i_mode)) {
508 dec_page_count(sbi, F2FS_DIRTY_DENTS);
509 inode_dec_dirty_dents(inode);
514 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
516 if (sbi->por_doing) {
517 err = AOP_WRITEPAGE_ACTIVATE;
521 /* Dentry blocks are controlled by checkpoint */
522 if (S_ISDIR(inode->i_mode)) {
523 dec_page_count(sbi, F2FS_DIRTY_DENTS);
524 inode_dec_dirty_dents(inode);
525 err = do_write_data_page(page);
527 int ilock = mutex_lock_op(sbi);
528 err = do_write_data_page(page);
529 mutex_unlock_op(sbi, ilock);
530 need_balance_fs = true;
537 if (wbc->for_reclaim)
538 f2fs_submit_bio(sbi, DATA, true);
540 clear_cold_data(page);
544 f2fs_balance_fs(sbi);
548 wbc->pages_skipped++;
549 set_page_dirty(page);
553 #define MAX_DESIRED_PAGES_WP 4096
555 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
558 struct address_space *mapping = data;
559 int ret = mapping->a_ops->writepage(page, wbc);
560 mapping_set_error(mapping, ret);
564 static int f2fs_write_data_pages(struct address_space *mapping,
565 struct writeback_control *wbc)
567 struct inode *inode = mapping->host;
568 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
570 long excess_nrtw = 0, desired_nrtw;
572 /* deal with chardevs and other special file */
573 if (!mapping->a_ops->writepage)
576 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
577 desired_nrtw = MAX_DESIRED_PAGES_WP;
578 excess_nrtw = desired_nrtw - wbc->nr_to_write;
579 wbc->nr_to_write = desired_nrtw;
582 if (!S_ISDIR(inode->i_mode))
583 mutex_lock(&sbi->writepages);
584 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
585 if (!S_ISDIR(inode->i_mode))
586 mutex_unlock(&sbi->writepages);
587 f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
589 remove_dirty_dir_inode(inode);
591 wbc->nr_to_write -= excess_nrtw;
595 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
596 loff_t pos, unsigned len, unsigned flags,
597 struct page **pagep, void **fsdata)
599 struct inode *inode = mapping->host;
600 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
602 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
603 struct dnode_of_data dn;
607 /* for nobh_write_end */
610 f2fs_balance_fs(sbi);
612 page = grab_cache_page_write_begin(mapping, index, flags);
617 ilock = mutex_lock_op(sbi);
619 set_new_dnode(&dn, inode, NULL, NULL, 0);
620 err = get_dnode_of_data(&dn, index, ALLOC_NODE);
624 if (dn.data_blkaddr == NULL_ADDR)
625 err = reserve_new_block(&dn);
631 mutex_unlock_op(sbi, ilock);
633 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
636 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
637 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
638 unsigned end = start + len;
640 /* Reading beyond i_size is simple: memset to zero */
641 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
645 if (dn.data_blkaddr == NEW_ADDR) {
646 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
648 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
652 if (!PageUptodate(page)) {
653 f2fs_put_page(page, 1);
658 SetPageUptodate(page);
659 clear_cold_data(page);
663 mutex_unlock_op(sbi, ilock);
664 f2fs_put_page(page, 1);
668 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
669 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
671 struct file *file = iocb->ki_filp;
672 struct inode *inode = file->f_mapping->host;
677 /* Needs synchronization with the cleaner */
678 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
682 static void f2fs_invalidate_data_page(struct page *page, unsigned long offset)
684 struct inode *inode = page->mapping->host;
685 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
686 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
687 dec_page_count(sbi, F2FS_DIRTY_DENTS);
688 inode_dec_dirty_dents(inode);
690 ClearPagePrivate(page);
693 static int f2fs_release_data_page(struct page *page, gfp_t wait)
695 ClearPagePrivate(page);
699 static int f2fs_set_data_page_dirty(struct page *page)
701 struct address_space *mapping = page->mapping;
702 struct inode *inode = mapping->host;
704 SetPageUptodate(page);
705 if (!PageDirty(page)) {
706 __set_page_dirty_nobuffers(page);
707 set_dirty_dir_page(inode, page);
713 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
715 return generic_block_bmap(mapping, block, get_data_block_ro);
718 const struct address_space_operations f2fs_dblock_aops = {
719 .readpage = f2fs_read_data_page,
720 .readpages = f2fs_read_data_pages,
721 .writepage = f2fs_write_data_page,
722 .writepages = f2fs_write_data_pages,
723 .write_begin = f2fs_write_begin,
724 .write_end = nobh_write_end,
725 .set_page_dirty = f2fs_set_data_page_dirty,
726 .invalidatepage = f2fs_invalidate_data_page,
727 .releasepage = f2fs_release_data_page,
728 .direct_IO = f2fs_direct_IO,