]>
Commit | Line | Data |
---|---|---|
d7b4be12 GKH |
1 | From tytso@mit.edu Mon Apr 19 10:19:40 2010 |
2 | From: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> | |
3 | Date: Mon, 15 Mar 2010 20:25:55 -0400 | |
4 | Subject: ext4: invalidate pages if delalloc block allocation fails. | |
5 | To: stable@kernel.org | |
6 | Cc: Ext4 Developers List <linux-ext4@vger.kernel.org>, "Theodore Ts'o" <tytso@mit.edu>, "Jayson R. King" <dev@jaysonking.com>, "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com> | |
7 | Message-ID: <1268699165-17461-2-git-send-email-tytso@mit.edu> | |
8 | ||
9 | ||
10 | From: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> | |
11 | ||
12 | commit c4a0c46ec92c194c873232b88debce4e1a448483 upstream. | |
13 | ||
14 | We are a bit agressive in invalidating all the pages. But | |
15 | it is ok because we really don't know why the block allocation | |
16 | failed and it is better to come of the writeback path | |
17 | so that user can look for more info. | |
18 | ||
19 | Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> | |
20 | Signed-off-by: Jayson R. King <dev@jaysonking.com> | |
21 | Signed-off-by: Theodore Ts'o <tytso@mit.edu> | |
22 | Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> | |
23 | ||
24 | --- | |
25 | fs/ext4/inode.c | 85 ++++++++++++++++++++++++++++++++++++++++++++++++-------- | |
26 | 1 file changed, 73 insertions(+), 12 deletions(-) | |
27 | ||
28 | --- a/fs/ext4/inode.c | |
29 | +++ b/fs/ext4/inode.c | |
30 | @@ -1821,6 +1821,39 @@ static inline void __unmap_underlying_bl | |
31 | unmap_underlying_metadata(bdev, bh->b_blocknr + i); | |
32 | } | |
33 | ||
34 | +static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, | |
35 | + sector_t logical, long blk_cnt) | |
36 | +{ | |
37 | + int nr_pages, i; | |
38 | + pgoff_t index, end; | |
39 | + struct pagevec pvec; | |
40 | + struct inode *inode = mpd->inode; | |
41 | + struct address_space *mapping = inode->i_mapping; | |
42 | + | |
43 | + index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
44 | + end = (logical + blk_cnt - 1) >> | |
45 | + (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
46 | + while (index <= end) { | |
47 | + nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
48 | + if (nr_pages == 0) | |
49 | + break; | |
50 | + for (i = 0; i < nr_pages; i++) { | |
51 | + struct page *page = pvec.pages[i]; | |
52 | + index = page->index; | |
53 | + if (index > end) | |
54 | + break; | |
55 | + index++; | |
56 | + | |
57 | + BUG_ON(!PageLocked(page)); | |
58 | + BUG_ON(PageWriteback(page)); | |
59 | + block_invalidatepage(page, 0); | |
60 | + ClearPageUptodate(page); | |
61 | + unlock_page(page); | |
62 | + } | |
63 | + } | |
64 | + return; | |
65 | +} | |
66 | + | |
67 | /* | |
68 | * mpage_da_map_blocks - go through given space | |
69 | * | |
70 | @@ -1830,7 +1863,7 @@ static inline void __unmap_underlying_bl | |
71 | * The function skips space we know is already mapped to disk blocks. | |
72 | * | |
73 | */ | |
74 | -static void mpage_da_map_blocks(struct mpage_da_data *mpd) | |
75 | +static int mpage_da_map_blocks(struct mpage_da_data *mpd) | |
76 | { | |
77 | int err = 0; | |
78 | struct buffer_head *lbh = &mpd->lbh; | |
79 | @@ -1841,7 +1874,7 @@ static void mpage_da_map_blocks(struct m | |
80 | * We consider only non-mapped and non-allocated blocks | |
81 | */ | |
82 | if (buffer_mapped(lbh) && !buffer_delay(lbh)) | |
83 | - return; | |
84 | + return 0; | |
85 | ||
86 | new.b_state = lbh->b_state; | |
87 | new.b_blocknr = 0; | |
88 | @@ -1852,10 +1885,38 @@ static void mpage_da_map_blocks(struct m | |
89 | * to write simply return | |
90 | */ | |
91 | if (!new.b_size) | |
92 | - return; | |
93 | + return 0; | |
94 | err = mpd->get_block(mpd->inode, next, &new, 1); | |
95 | - if (err) | |
96 | - return; | |
97 | + if (err) { | |
98 | + | |
99 | + /* If get block returns with error | |
100 | + * we simply return. Later writepage | |
101 | + * will redirty the page and writepages | |
102 | + * will find the dirty page again | |
103 | + */ | |
104 | + if (err == -EAGAIN) | |
105 | + return 0; | |
106 | + /* | |
107 | + * get block failure will cause us | |
108 | + * to loop in writepages. Because | |
109 | + * a_ops->writepage won't be able to | |
110 | + * make progress. The page will be redirtied | |
111 | + * by writepage and writepages will again | |
112 | + * try to write the same. | |
113 | + */ | |
114 | + printk(KERN_EMERG "%s block allocation failed for inode %lu " | |
115 | + "at logical offset %llu with max blocks " | |
116 | + "%zd with error %d\n", | |
117 | + __func__, mpd->inode->i_ino, | |
118 | + (unsigned long long)next, | |
119 | + lbh->b_size >> mpd->inode->i_blkbits, err); | |
120 | + printk(KERN_EMERG "This should not happen.!! " | |
121 | + "Data will be lost\n"); | |
122 | + /* invlaidate all the pages */ | |
123 | + ext4_da_block_invalidatepages(mpd, next, | |
124 | + lbh->b_size >> mpd->inode->i_blkbits); | |
125 | + return err; | |
126 | + } | |
127 | BUG_ON(new.b_size == 0); | |
128 | ||
129 | if (buffer_new(&new)) | |
130 | @@ -1868,7 +1929,7 @@ static void mpage_da_map_blocks(struct m | |
131 | if (buffer_delay(lbh) || buffer_unwritten(lbh)) | |
132 | mpage_put_bnr_to_bhs(mpd, next, &new); | |
133 | ||
134 | - return; | |
135 | + return 0; | |
136 | } | |
137 | ||
138 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ | |
139 | @@ -1937,8 +1998,8 @@ flush_it: | |
140 | * We couldn't merge the block to our extent, so we | |
141 | * need to flush current extent and start new one | |
142 | */ | |
143 | - mpage_da_map_blocks(mpd); | |
144 | - mpage_da_submit_io(mpd); | |
145 | + if (mpage_da_map_blocks(mpd) == 0) | |
146 | + mpage_da_submit_io(mpd); | |
147 | mpd->io_done = 1; | |
148 | return; | |
149 | } | |
150 | @@ -1980,8 +2041,8 @@ static int __mpage_da_writepage(struct p | |
151 | * and start IO on them using writepage() | |
152 | */ | |
153 | if (mpd->next_page != mpd->first_page) { | |
154 | - mpage_da_map_blocks(mpd); | |
155 | - mpage_da_submit_io(mpd); | |
156 | + if (mpage_da_map_blocks(mpd) == 0) | |
157 | + mpage_da_submit_io(mpd); | |
158 | /* | |
159 | * skip rest of the page in the page_vec | |
160 | */ | |
161 | @@ -2102,8 +2163,8 @@ static int mpage_da_writepages(struct ad | |
162 | * Handle last extent of pages | |
163 | */ | |
164 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
165 | - mpage_da_map_blocks(&mpd); | |
166 | - mpage_da_submit_io(&mpd); | |
167 | + if (mpage_da_map_blocks(&mpd) == 0) | |
168 | + mpage_da_submit_io(&mpd); | |
169 | } | |
170 | ||
171 | wbc->nr_to_write = to_write - mpd.pages_written; |