blob: 45fc5bdb7d671a8c7f407f181a7460da014694b8 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/inode.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
21 *
Mingming Cao617ba132006-10-11 01:20:53 -070022 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
Dave Kleikampac27a0e2006-10-11 01:20:50 -070023 */
24
25#include <linux/module.h>
26#include <linux/fs.h>
27#include <linux/time.h>
Mingming Caodab291a2006-10-11 01:21:01 -070028#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070029#include <linux/highuid.h>
30#include <linux/pagemap.h>
31#include <linux/quotaops.h>
32#include <linux/string.h>
33#include <linux/buffer_head.h>
34#include <linux/writeback.h>
Alex Tomas64769242008-07-11 19:27:31 -040035#include <linux/pagevec.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070036#include <linux/mpage.h>
Duane Griffine83c1392008-12-19 20:47:15 +000037#include <linux/namei.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070038#include <linux/uio.h>
39#include <linux/bio.h>
Mingming Cao4c0425f2009-09-28 15:48:41 -040040#include <linux/workqueue.h>
Jiaying Zhang744692d2010-03-04 16:14:02 -050041#include <linux/kernel.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090042#include <linux/slab.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040043
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040044#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070045#include "xattr.h"
46#include "acl.h"
Mingming Caod2a17632008-07-14 17:52:37 -040047#include "ext4_extents.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070048
Theodore Ts'o9bffad12009-06-17 11:48:11 -040049#include <trace/events/ext4.h>
50
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040051#define MPAGE_DA_EXTENT_TAIL 0x01
52
Jan Kara678aaf42008-07-11 19:27:31 -040053static inline int ext4_begin_ordered_truncate(struct inode *inode,
54 loff_t new_size)
55{
Jan Kara7f5aa212009-02-10 11:15:34 -050056 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode->i_sb)->s_journal,
58 &EXT4_I(inode)->jinode,
59 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -040060}
61
Alex Tomas64769242008-07-11 19:27:31 -040062static void ext4_invalidatepage(struct page *page, unsigned long offset);
Theodore Ts'ocb20d512010-10-27 21:30:09 -040063static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
64 struct buffer_head *bh_result, int create);
65static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
66static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
67static int __ext4_journalled_writepage(struct page *page, unsigned int len);
68static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
Alex Tomas64769242008-07-11 19:27:31 -040069
Dave Kleikampac27a0e2006-10-11 01:20:50 -070070/*
71 * Test whether an inode is a fast symlink.
72 */
Mingming Cao617ba132006-10-11 01:20:53 -070073static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070074{
Mingming Cao617ba132006-10-11 01:20:53 -070075 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -070076 (inode->i_sb->s_blocksize >> 9) : 0;
77
78 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
79}
80
81/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -070082 * Work out how many blocks we need to proceed with the next chunk of a
83 * truncate transaction.
84 */
85static unsigned long blocks_for_truncate(struct inode *inode)
86{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -050087 ext4_lblk_t needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070088
89 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
90
91 /* Give ourselves just enough room to cope with inodes in which
92 * i_blocks is corrupt: we've seen disk corruptions in the past
93 * which resulted in random data in an inode which looked enough
Mingming Cao617ba132006-10-11 01:20:53 -070094 * like a regular file for ext4 to try to delete it. Things
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095 * will go a bit crazy if that happens, but at least we should
96 * try not to panic the whole kernel. */
97 if (needed < 2)
98 needed = 2;
99
100 /* But we need to bound the transaction so we don't overflow the
101 * journal. */
Mingming Cao617ba132006-10-11 01:20:53 -0700102 if (needed > EXT4_MAX_TRANS_DATA)
103 needed = EXT4_MAX_TRANS_DATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700104
Mingming Cao617ba132006-10-11 01:20:53 -0700105 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106}
107
108/*
109 * Truncate transactions can be complex and absolutely huge. So we need to
110 * be able to restart the transaction at a conventient checkpoint to make
111 * sure we don't overflow the journal.
112 *
113 * start_transaction gets us a new handle for a truncate transaction,
114 * and extend_transaction tries to extend the existing one a bit. If
115 * extend fails, we need to propagate the failure up and restart the
116 * transaction in the top-level truncate loop. --sct
117 */
118static handle_t *start_transaction(struct inode *inode)
119{
120 handle_t *result;
121
Mingming Cao617ba132006-10-11 01:20:53 -0700122 result = ext4_journal_start(inode, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700123 if (!IS_ERR(result))
124 return result;
125
Mingming Cao617ba132006-10-11 01:20:53 -0700126 ext4_std_error(inode->i_sb, PTR_ERR(result));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700127 return result;
128}
129
130/*
131 * Try to extend this transaction for the purposes of truncation.
132 *
133 * Returns 0 if we managed to create more room. If we can't create more
134 * room, and the transaction must be restarted we return 1.
135 */
136static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
137{
Frank Mayhar03901312009-01-07 00:06:22 -0500138 if (!ext4_handle_valid(handle))
139 return 0;
140 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700141 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700142 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700143 return 0;
144 return 1;
145}
146
147/*
148 * Restart the transaction associated with *handle. This does a commit,
149 * so before we call here everything must be consistently dirtied against
150 * this transaction.
151 */
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500152int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
Jan Kara487caee2009-08-17 22:17:20 -0400153 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700154{
Jan Kara487caee2009-08-17 22:17:20 -0400155 int ret;
156
157 /*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400158 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
Jan Kara487caee2009-08-17 22:17:20 -0400159 * moment, get_block can be called only for blocks inside i_size since
160 * page cache has been already dropped and writes are blocked by
161 * i_mutex. So we can safely drop the i_data_sem here.
162 */
Frank Mayhar03901312009-01-07 00:06:22 -0500163 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700164 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400165 up_write(&EXT4_I(inode)->i_data_sem);
166 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
167 down_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500168 ext4_discard_preallocations(inode);
Jan Kara487caee2009-08-17 22:17:20 -0400169
170 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700171}
172
173/*
174 * Called at the last iput() if i_nlink is zero.
175 */
Al Viro0930fcc2010-06-07 13:16:22 -0400176void ext4_evict_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700177{
178 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400179 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700180
Al Viro0930fcc2010-06-07 13:16:22 -0400181 if (inode->i_nlink) {
182 truncate_inode_pages(&inode->i_data, 0);
183 goto no_delete;
184 }
185
Christoph Hellwig907f4552010-03-03 09:05:06 -0500186 if (!is_bad_inode(inode))
Christoph Hellwig871a2932010-03-03 09:05:07 -0500187 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500188
Jan Kara678aaf42008-07-11 19:27:31 -0400189 if (ext4_should_order_data(inode))
190 ext4_begin_ordered_truncate(inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700191 truncate_inode_pages(&inode->i_data, 0);
192
193 if (is_bad_inode(inode))
194 goto no_delete;
195
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400196 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700197 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400198 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700199 /*
200 * If we're going to skip the normal cleanup, we still need to
201 * make sure that the in-core orphan linked list is properly
202 * cleaned up.
203 */
Mingming Cao617ba132006-10-11 01:20:53 -0700204 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700205 goto no_delete;
206 }
207
208 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500209 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400211 err = ext4_mark_inode_dirty(handle, inode);
212 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500213 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400214 "couldn't mark inode dirty (err %d)", err);
215 goto stop_handle;
216 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700217 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700218 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400219
220 /*
221 * ext4_ext_truncate() doesn't reserve any slop when it
222 * restarts journal transactions; therefore there may not be
223 * enough credits left in the handle to remove the inode from
224 * the orphan list and set the dtime field.
225 */
Frank Mayhar03901312009-01-07 00:06:22 -0500226 if (!ext4_handle_has_enough_credits(handle, 3)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400227 err = ext4_journal_extend(handle, 3);
228 if (err > 0)
229 err = ext4_journal_restart(handle, 3);
230 if (err != 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500231 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400232 "couldn't extend journal (err %d)", err);
233 stop_handle:
234 ext4_journal_stop(handle);
Theodore Ts'o45388212010-07-29 15:06:10 -0400235 ext4_orphan_del(NULL, inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400236 goto no_delete;
237 }
238 }
239
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700240 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700241 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700242 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700243 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700244 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700245 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700246 * (Well, we could do this if we need to, but heck - it works)
247 */
Mingming Cao617ba132006-10-11 01:20:53 -0700248 ext4_orphan_del(handle, inode);
249 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700250
251 /*
252 * One subtle ordering requirement: if anything has gone wrong
253 * (transaction abort, IO errors, whatever), then we can still
254 * do these next steps (the fs will already have been marked as
255 * having errors), but we can't free the inode if the mark_dirty
256 * fails.
257 */
Mingming Cao617ba132006-10-11 01:20:53 -0700258 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700259 /* If that failed, just do the required in-core inode clear. */
Al Viro0930fcc2010-06-07 13:16:22 -0400260 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700261 else
Mingming Cao617ba132006-10-11 01:20:53 -0700262 ext4_free_inode(handle, inode);
263 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700264 return;
265no_delete:
Al Viro0930fcc2010-06-07 13:16:22 -0400266 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700267}
268
269typedef struct {
270 __le32 *p;
271 __le32 key;
272 struct buffer_head *bh;
273} Indirect;
274
275static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
276{
277 p->key = *(p->p = v);
278 p->bh = bh;
279}
280
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700281/**
Mingming Cao617ba132006-10-11 01:20:53 -0700282 * ext4_block_to_path - parse the block number into array of offsets
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700283 * @inode: inode in question (we are only interested in its superblock)
284 * @i_block: block number to be parsed
285 * @offsets: array to store the offsets in
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400286 * @boundary: set this non-zero if the referred-to block is likely to be
287 * followed (on disk) by an indirect block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700288 *
Mingming Cao617ba132006-10-11 01:20:53 -0700289 * To store the locations of file's data ext4 uses a data structure common
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700290 * for UNIX filesystems - tree of pointers anchored in the inode, with
291 * data blocks at leaves and indirect blocks in intermediate nodes.
292 * This function translates the block number into path in that tree -
293 * return value is the path length and @offsets[n] is the offset of
294 * pointer to (n+1)th node in the nth one. If @block is out of range
295 * (negative or too large) warning is printed and zero returned.
296 *
297 * Note: function doesn't find node addresses, so no IO is needed. All
298 * we need to know is the capacity of indirect blocks (taken from the
299 * inode->i_sb).
300 */
301
302/*
303 * Portability note: the last comparison (check that we fit into triple
304 * indirect block) is spelled differently, because otherwise on an
305 * architecture with 32-bit longs and 8Kb pages we might get into trouble
306 * if our filesystem had 8Kb blocks. We might use long long, but that would
307 * kill us on x86. Oh, well, at least the sign propagation does not matter -
308 * i_block would have to be negative in the very beginning, so we would not
309 * get there at all.
310 */
311
Mingming Cao617ba132006-10-11 01:20:53 -0700312static int ext4_block_to_path(struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400313 ext4_lblk_t i_block,
314 ext4_lblk_t offsets[4], int *boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315{
Mingming Cao617ba132006-10-11 01:20:53 -0700316 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
317 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
318 const long direct_blocks = EXT4_NDIR_BLOCKS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700319 indirect_blocks = ptrs,
320 double_blocks = (1 << (ptrs_bits * 2));
321 int n = 0;
322 int final = 0;
323
Roel Kluinc333e072009-08-10 22:47:22 -0400324 if (i_block < direct_blocks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700325 offsets[n++] = i_block;
326 final = direct_blocks;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400327 } else if ((i_block -= direct_blocks) < indirect_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700328 offsets[n++] = EXT4_IND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700329 offsets[n++] = i_block;
330 final = ptrs;
331 } else if ((i_block -= indirect_blocks) < double_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700332 offsets[n++] = EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700333 offsets[n++] = i_block >> ptrs_bits;
334 offsets[n++] = i_block & (ptrs - 1);
335 final = ptrs;
336 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
Mingming Cao617ba132006-10-11 01:20:53 -0700337 offsets[n++] = EXT4_TIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700338 offsets[n++] = i_block >> (ptrs_bits * 2);
339 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
340 offsets[n++] = i_block & (ptrs - 1);
341 final = ptrs;
342 } else {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500343 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400344 i_block + direct_blocks +
345 indirect_blocks + double_blocks, inode->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700346 }
347 if (boundary)
348 *boundary = final - 1 - (i_block & (ptrs - 1));
349 return n;
350}
351
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400352static int __ext4_check_blockref(const char *function, unsigned int line,
353 struct inode *inode,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400354 __le32 *p, unsigned int max)
355{
Theodore Ts'o1c13d5c2010-07-27 11:56:03 -0400356 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
Thiemo Nagelf73953c2009-04-07 18:46:47 -0400357 __le32 *bref = p;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400358 unsigned int blk;
359
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400360 while (bref < p+max) {
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400361 blk = le32_to_cpu(*bref++);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400362 if (blk &&
363 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400364 blk, 1))) {
Theodore Ts'o1c13d5c2010-07-27 11:56:03 -0400365 es->s_last_error_block = cpu_to_le64(blk);
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400366 ext4_error_inode(inode, function, line, blk,
367 "invalid block");
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400368 return -EIO;
369 }
370 }
371 return 0;
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400372}
373
374
375#define ext4_check_indirect_blockref(inode, bh) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400376 __ext4_check_blockref(__func__, __LINE__, inode, \
377 (__le32 *)(bh)->b_data, \
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400378 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
379
380#define ext4_check_inode_blockref(inode) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400381 __ext4_check_blockref(__func__, __LINE__, inode, \
382 EXT4_I(inode)->i_data, \
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400383 EXT4_NDIR_BLOCKS)
384
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700385/**
Mingming Cao617ba132006-10-11 01:20:53 -0700386 * ext4_get_branch - read the chain of indirect blocks leading to data
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700387 * @inode: inode in question
388 * @depth: depth of the chain (1 - direct pointer, etc.)
389 * @offsets: offsets of pointers in inode/indirect blocks
390 * @chain: place to store the result
391 * @err: here we store the error value
392 *
393 * Function fills the array of triples <key, p, bh> and returns %NULL
394 * if everything went OK or the pointer to the last filled triple
395 * (incomplete one) otherwise. Upon the return chain[i].key contains
396 * the number of (i+1)-th block in the chain (as it is stored in memory,
397 * i.e. little-endian 32-bit), chain[i].p contains the address of that
398 * number (it points into struct inode for i==0 and into the bh->b_data
399 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
400 * block for i>0 and NULL for i==0. In other words, it holds the block
401 * numbers of the chain, addresses they were taken from (and where we can
402 * verify that chain did not change) and buffer_heads hosting these
403 * numbers.
404 *
405 * Function stops when it stumbles upon zero pointer (absent block)
406 * (pointer to last triple returned, *@err == 0)
407 * or when it gets an IO error reading an indirect block
408 * (ditto, *@err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700409 * or when it reads all @depth-1 indirect blocks successfully and finds
410 * the whole chain, all way to the data (returns %NULL, *err == 0).
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500411 *
412 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500413 * down_read(&EXT4_I(inode)->i_data_sem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700414 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500415static Indirect *ext4_get_branch(struct inode *inode, int depth,
416 ext4_lblk_t *offsets,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700417 Indirect chain[4], int *err)
418{
419 struct super_block *sb = inode->i_sb;
420 Indirect *p = chain;
421 struct buffer_head *bh;
422
423 *err = 0;
424 /* i_data is not going away, no lock needed */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400425 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700426 if (!p->key)
427 goto no_block;
428 while (--depth) {
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400429 bh = sb_getblk(sb, le32_to_cpu(p->key));
430 if (unlikely(!bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700431 goto failure;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400432
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400433 if (!bh_uptodate_or_lock(bh)) {
434 if (bh_submit_read(bh) < 0) {
435 put_bh(bh);
436 goto failure;
437 }
438 /* validate block references */
439 if (ext4_check_indirect_blockref(inode, bh)) {
440 put_bh(bh);
441 goto failure;
442 }
443 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400444
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400445 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700446 /* Reader: end */
447 if (!p->key)
448 goto no_block;
449 }
450 return NULL;
451
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700452failure:
453 *err = -EIO;
454no_block:
455 return p;
456}
457
458/**
Mingming Cao617ba132006-10-11 01:20:53 -0700459 * ext4_find_near - find a place for allocation with sufficient locality
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700460 * @inode: owner
461 * @ind: descriptor of indirect block.
462 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000463 * This function returns the preferred place for block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700464 * It is used when heuristic for sequential allocation fails.
465 * Rules are:
466 * + if there is a block to the left of our position - allocate near it.
467 * + if pointer will live in indirect block - allocate near that block.
468 * + if pointer will live in inode - allocate in the same
469 * cylinder group.
470 *
471 * In the latter case we colour the starting block by the callers PID to
472 * prevent it from clashing with concurrent allocations for a different inode
473 * in the same block group. The PID is used here so that functionally related
474 * files will be close-by on-disk.
475 *
476 * Caller must make sure that @ind is valid and will stay that way.
477 */
Mingming Cao617ba132006-10-11 01:20:53 -0700478static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700479{
Mingming Cao617ba132006-10-11 01:20:53 -0700480 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400481 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482 __le32 *p;
Mingming Cao617ba132006-10-11 01:20:53 -0700483 ext4_fsblk_t bg_start;
Valerie Clement74d34872008-02-15 13:43:07 -0500484 ext4_fsblk_t last_block;
Mingming Cao617ba132006-10-11 01:20:53 -0700485 ext4_grpblk_t colour;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400486 ext4_group_t block_group;
487 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700488
489 /* Try to find previous block */
490 for (p = ind->p - 1; p >= start; p--) {
491 if (*p)
492 return le32_to_cpu(*p);
493 }
494
495 /* No such thing, so let's try location of indirect block */
496 if (ind->bh)
497 return ind->bh->b_blocknr;
498
499 /*
500 * It is going to be referred to from the inode itself? OK, just put it
501 * into the same cylinder group then.
502 */
Theodore Ts'oa4912122009-03-12 12:18:34 -0400503 block_group = ei->i_block_group;
504 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
505 block_group &= ~(flex_size-1);
506 if (S_ISREG(inode->i_mode))
507 block_group++;
508 }
509 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
Valerie Clement74d34872008-02-15 13:43:07 -0500510 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
511
Theodore Ts'oa4912122009-03-12 12:18:34 -0400512 /*
513 * If we are doing delayed allocation, we don't need take
514 * colour into account.
515 */
516 if (test_opt(inode->i_sb, DELALLOC))
517 return bg_start;
518
Valerie Clement74d34872008-02-15 13:43:07 -0500519 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
520 colour = (current->pid % 16) *
Mingming Cao617ba132006-10-11 01:20:53 -0700521 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Valerie Clement74d34872008-02-15 13:43:07 -0500522 else
523 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700524 return bg_start + colour;
525}
526
527/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000528 * ext4_find_goal - find a preferred place for allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700529 * @inode: owner
530 * @block: block we want
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700531 * @partial: pointer to the last triple within a chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700532 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000533 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800534 * returns it.
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400535 * Because this is only used for non-extent files, we limit the block nr
536 * to 32 bits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700537 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500538static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400539 Indirect *partial)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700540{
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400541 ext4_fsblk_t goal;
542
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700543 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400544 * XXX need to get goal block from mballoc's data structures
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700545 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700546
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400547 goal = ext4_find_near(inode, partial);
548 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
549 return goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700550}
551
552/**
Mingming Cao617ba132006-10-11 01:20:53 -0700553 * ext4_blks_to_allocate: Look up the block map and count the number
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700554 * of direct blocks need to be allocated for the given branch.
555 *
556 * @branch: chain of indirect blocks
557 * @k: number of blocks need for indirect blocks
558 * @blks: number of data blocks to be mapped.
559 * @blocks_to_boundary: the offset in the indirect block
560 *
561 * return the total number of blocks to be allocate, including the
562 * direct and indirect blocks.
563 */
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500564static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400565 int blocks_to_boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700566{
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500567 unsigned int count = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700568
569 /*
570 * Simple case, [t,d]Indirect block(s) has not allocated yet
571 * then it's clear blocks on that path have not allocated
572 */
573 if (k > 0) {
574 /* right now we don't handle cross boundary allocation */
575 if (blks < blocks_to_boundary + 1)
576 count += blks;
577 else
578 count += blocks_to_boundary + 1;
579 return count;
580 }
581
582 count++;
583 while (count < blks && count <= blocks_to_boundary &&
584 le32_to_cpu(*(branch[0].p + count)) == 0) {
585 count++;
586 }
587 return count;
588}
589
590/**
Mingming Cao617ba132006-10-11 01:20:53 -0700591 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700592 * @indirect_blks: the number of blocks need to allocate for indirect
593 * blocks
594 *
595 * @new_blocks: on return it will store the new block numbers for
596 * the indirect blocks(if needed) and the first direct block,
597 * @blks: on return it will store the total number of allocated
598 * direct blocks
599 */
Mingming Cao617ba132006-10-11 01:20:53 -0700600static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400601 ext4_lblk_t iblock, ext4_fsblk_t goal,
602 int indirect_blks, int blks,
603 ext4_fsblk_t new_blocks[4], int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700604{
Theodore Ts'o815a1132009-01-01 23:59:43 -0500605 struct ext4_allocation_request ar;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700606 int target, i;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400607 unsigned long count = 0, blk_allocated = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700608 int index = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700609 ext4_fsblk_t current_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700610 int ret = 0;
611
612 /*
613 * Here we try to allocate the requested multiple blocks at once,
614 * on a best-effort basis.
615 * To build a branch, we should allocate blocks for
616 * the indirect blocks(if not allocated yet), and at least
617 * the first direct block of this branch. That's the
618 * minimum number of blocks need to allocate(required)
619 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400620 /* first we try to allocate the indirect blocks */
621 target = indirect_blks;
622 while (target > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700623 count = target;
624 /* allocating blocks for indirect blocks and direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400625 current_block = ext4_new_meta_blocks(handle, inode,
626 goal, &count, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700627 if (*err)
628 goto failed_out;
629
Frank Mayhar273df552010-03-02 11:46:09 -0500630 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
631 EXT4_ERROR_INODE(inode,
632 "current_block %llu + count %lu > %d!",
633 current_block, count,
634 EXT4_MAX_BLOCK_FILE_PHYS);
635 *err = -EIO;
636 goto failed_out;
637 }
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400638
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700639 target -= count;
640 /* allocate blocks for indirect blocks */
641 while (index < indirect_blks && count) {
642 new_blocks[index++] = current_block++;
643 count--;
644 }
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400645 if (count > 0) {
646 /*
647 * save the new block number
648 * for the first direct block
649 */
650 new_blocks[index] = current_block;
651 printk(KERN_INFO "%s returned more blocks than "
652 "requested\n", __func__);
653 WARN_ON(1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700654 break;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400655 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700656 }
657
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400658 target = blks - count ;
659 blk_allocated = count;
660 if (!target)
661 goto allocated;
662 /* Now allocate data blocks */
Theodore Ts'o815a1132009-01-01 23:59:43 -0500663 memset(&ar, 0, sizeof(ar));
664 ar.inode = inode;
665 ar.goal = goal;
666 ar.len = target;
667 ar.logical = iblock;
668 if (S_ISREG(inode->i_mode))
669 /* enable in-core preallocation only for regular files */
670 ar.flags = EXT4_MB_HINT_DATA;
671
672 current_block = ext4_mb_new_blocks(handle, &ar, err);
Frank Mayhar273df552010-03-02 11:46:09 -0500673 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
674 EXT4_ERROR_INODE(inode,
675 "current_block %llu + ar.len %d > %d!",
676 current_block, ar.len,
677 EXT4_MAX_BLOCK_FILE_PHYS);
678 *err = -EIO;
679 goto failed_out;
680 }
Theodore Ts'o815a1132009-01-01 23:59:43 -0500681
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400682 if (*err && (target == blks)) {
683 /*
684 * if the allocation failed and we didn't allocate
685 * any blocks before
686 */
687 goto failed_out;
688 }
689 if (!*err) {
690 if (target == blks) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400691 /*
692 * save the new block number
693 * for the first direct block
694 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400695 new_blocks[index] = current_block;
696 }
Theodore Ts'o815a1132009-01-01 23:59:43 -0500697 blk_allocated += ar.len;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400698 }
699allocated:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 /* total number of blocks allocated for direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400701 ret = blk_allocated;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700702 *err = 0;
703 return ret;
704failed_out:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400705 for (i = 0; i < index; i++)
Theodore Ts'oe6362602009-11-23 07:17:05 -0500706 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700707 return ret;
708}
709
710/**
Mingming Cao617ba132006-10-11 01:20:53 -0700711 * ext4_alloc_branch - allocate and set up a chain of blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700712 * @inode: owner
713 * @indirect_blks: number of allocated indirect blocks
714 * @blks: number of allocated direct blocks
715 * @offsets: offsets (in the blocks) to store the pointers to next.
716 * @branch: place to store the chain in.
717 *
718 * This function allocates blocks, zeroes out all but the last one,
719 * links them into chain and (if we are synchronous) writes them to disk.
720 * In other words, it prepares a branch that can be spliced onto the
721 * inode. It stores the information about that chain in the branch[], in
Mingming Cao617ba132006-10-11 01:20:53 -0700722 * the same format as ext4_get_branch() would do. We are calling it after
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723 * we had read the existing part of chain and partial points to the last
724 * triple of that (one with zero ->key). Upon the exit we have the same
Mingming Cao617ba132006-10-11 01:20:53 -0700725 * picture as after the successful ext4_get_block(), except that in one
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700726 * place chain is disconnected - *branch->p is still zero (we did not
727 * set the last link), but branch->key contains the number that should
728 * be placed into *branch->p to fill that gap.
729 *
730 * If allocation fails we free all blocks we've allocated (and forget
731 * their buffer_heads) and return the error value the from failed
Mingming Cao617ba132006-10-11 01:20:53 -0700732 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700733 * as described above and return 0.
734 */
Mingming Cao617ba132006-10-11 01:20:53 -0700735static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400736 ext4_lblk_t iblock, int indirect_blks,
737 int *blks, ext4_fsblk_t goal,
738 ext4_lblk_t *offsets, Indirect *branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700739{
740 int blocksize = inode->i_sb->s_blocksize;
741 int i, n = 0;
742 int err = 0;
743 struct buffer_head *bh;
744 int num;
Mingming Cao617ba132006-10-11 01:20:53 -0700745 ext4_fsblk_t new_blocks[4];
746 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700747
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400748 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700749 *blks, new_blocks, &err);
750 if (err)
751 return err;
752
753 branch[0].key = cpu_to_le32(new_blocks[0]);
754 /*
755 * metadata blocks and data blocks are allocated.
756 */
757 for (n = 1; n <= indirect_blks; n++) {
758 /*
759 * Get buffer_head for parent block, zero it out
760 * and set the pointer to new one, then send
761 * parent to disk.
762 */
763 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
Namhyung Kim87783692010-10-27 21:30:11 -0400764 if (unlikely(!bh)) {
765 err = -EIO;
766 goto failed;
767 }
768
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700769 branch[n].bh = bh;
770 lock_buffer(bh);
771 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700772 err = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700773 if (err) {
Curt Wohlgemuth6487a9d2009-07-17 10:54:08 -0400774 /* Don't brelse(bh) here; it's done in
775 * ext4_journal_forget() below */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700776 unlock_buffer(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700777 goto failed;
778 }
779
780 memset(bh->b_data, 0, blocksize);
781 branch[n].p = (__le32 *) bh->b_data + offsets[n];
782 branch[n].key = cpu_to_le32(new_blocks[n]);
783 *branch[n].p = branch[n].key;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400784 if (n == indirect_blks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700785 current_block = new_blocks[n];
786 /*
787 * End of chain, update the last new metablock of
788 * the chain to point to the new allocated
789 * data blocks numbers
790 */
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400791 for (i = 1; i < num; i++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700792 *(branch[n].p + i) = cpu_to_le32(++current_block);
793 }
794 BUFFER_TRACE(bh, "marking uptodate");
795 set_buffer_uptodate(bh);
796 unlock_buffer(bh);
797
Frank Mayhar03901312009-01-07 00:06:22 -0500798 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
799 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700800 if (err)
801 goto failed;
802 }
803 *blks = num;
804 return err;
805failed:
806 /* Allocation failed, free what we already allocated */
Theodore Ts'oe6362602009-11-23 07:17:05 -0500807 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700808 for (i = 1; i <= n ; i++) {
Theodore Ts'o60e66792010-05-17 07:00:00 -0400809 /*
Theodore Ts'oe6362602009-11-23 07:17:05 -0500810 * branch[i].bh is newly allocated, so there is no
811 * need to revoke the block, which is why we don't
812 * need to set EXT4_FREE_BLOCKS_METADATA.
Theodore Ts'ob7e57e72009-11-22 21:00:13 -0500813 */
Theodore Ts'oe6362602009-11-23 07:17:05 -0500814 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
815 EXT4_FREE_BLOCKS_FORGET);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700816 }
Theodore Ts'oe6362602009-11-23 07:17:05 -0500817 for (i = n+1; i < indirect_blks; i++)
818 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819
Theodore Ts'oe6362602009-11-23 07:17:05 -0500820 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700821
822 return err;
823}
824
825/**
Mingming Cao617ba132006-10-11 01:20:53 -0700826 * ext4_splice_branch - splice the allocated branch onto inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700827 * @inode: owner
828 * @block: (logical) number of block we are adding
829 * @chain: chain of indirect blocks (with a missing link - see
Mingming Cao617ba132006-10-11 01:20:53 -0700830 * ext4_alloc_branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831 * @where: location of missing link
832 * @num: number of indirect blocks we are adding
833 * @blks: number of direct blocks we are adding
834 *
835 * This function fills the missing link and does all housekeeping needed in
836 * inode (->i_blocks, etc.). In case of success we end up with the full
837 * chain to new block and return 0.
838 */
Mingming Cao617ba132006-10-11 01:20:53 -0700839static int ext4_splice_branch(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400840 ext4_lblk_t block, Indirect *where, int num,
841 int blks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700842{
843 int i;
844 int err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700845 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700846
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700847 /*
848 * If we're splicing into a [td]indirect block (as opposed to the
849 * inode) then we need to get write access to the [td]indirect block
850 * before the splice.
851 */
852 if (where->bh) {
853 BUFFER_TRACE(where->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700854 err = ext4_journal_get_write_access(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700855 if (err)
856 goto err_out;
857 }
858 /* That's it */
859
860 *where->p = where->key;
861
862 /*
863 * Update the host buffer_head or inode to point to more just allocated
864 * direct blocks blocks
865 */
866 if (num == 0 && blks > 1) {
867 current_block = le32_to_cpu(where->key) + 1;
868 for (i = 1; i < blks; i++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400869 *(where->p + i) = cpu_to_le32(current_block++);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870 }
871
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700872 /* We are done with atomic stuff, now do the rest of housekeeping */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700873 /* had we spliced it onto indirect block? */
874 if (where->bh) {
875 /*
876 * If we spliced it onto an indirect block, we haven't
877 * altered the inode. Note however that if it is being spliced
878 * onto an indirect block at the very end of the file (the
879 * file is growing) then we *will* alter the inode to reflect
880 * the new i_size. But that is not done here - it is done in
Mingming Cao617ba132006-10-11 01:20:53 -0700881 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700882 */
883 jbd_debug(5, "splicing indirect only\n");
Frank Mayhar03901312009-01-07 00:06:22 -0500884 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
885 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700886 if (err)
887 goto err_out;
888 } else {
889 /*
890 * OK, we spliced it into the inode itself on a direct block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700891 */
Theodore Ts'o41591752009-06-15 03:41:23 -0400892 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700893 jbd_debug(5, "splicing direct\n");
894 }
895 return err;
896
897err_out:
898 for (i = 1; i <= num; i++) {
Theodore Ts'o60e66792010-05-17 07:00:00 -0400899 /*
Theodore Ts'oe6362602009-11-23 07:17:05 -0500900 * branch[i].bh is newly allocated, so there is no
901 * need to revoke the block, which is why we don't
902 * need to set EXT4_FREE_BLOCKS_METADATA.
Theodore Ts'ob7e57e72009-11-22 21:00:13 -0500903 */
Theodore Ts'oe6362602009-11-23 07:17:05 -0500904 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
905 EXT4_FREE_BLOCKS_FORGET);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700906 }
Theodore Ts'oe6362602009-11-23 07:17:05 -0500907 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
908 blks, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909
910 return err;
911}
912
913/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400914 * The ext4_ind_map_blocks() function handles non-extents inodes
Theodore Ts'ob920c752009-05-14 00:54:29 -0400915 * (i.e., using the traditional indirect/double-indirect i_blocks
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400916 * scheme) for ext4_map_blocks().
Theodore Ts'ob920c752009-05-14 00:54:29 -0400917 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700918 * Allocation strategy is simple: if we have to allocate something, we will
919 * have to go the whole way to leaf. So let's do it before attaching anything
920 * to tree, set linkage between the newborn blocks, write them if sync is
921 * required, recheck the path, free and repeat if check fails, otherwise
922 * set the last missing link (that will protect us from any truncate-generated
923 * removals - all blocks on the path are immune now) and possibly force the
924 * write on the parent block.
925 * That has a nice additional property: no special recovery from the failed
926 * allocations is needed - we simply release blocks and do not touch anything
927 * reachable from inode.
928 *
929 * `handle' can be NULL if create == 0.
930 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 * return > 0, # of blocks mapped or allocated.
932 * return = 0, if plain lookup failed.
933 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500934 *
Theodore Ts'ob920c752009-05-14 00:54:29 -0400935 * The ext4_ind_get_blocks() function should be called with
936 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
937 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
938 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
939 * blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700940 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400941static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
942 struct ext4_map_blocks *map,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400943 int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700944{
945 int err = -EIO;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500946 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700947 Indirect chain[4];
948 Indirect *partial;
Mingming Cao617ba132006-10-11 01:20:53 -0700949 ext4_fsblk_t goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950 int indirect_blks;
951 int blocks_to_boundary = 0;
952 int depth;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953 int count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700954 ext4_fsblk_t first_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700955
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400956 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
Theodore Ts'oc2177052009-05-14 00:58:52 -0400957 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400958 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400959 &blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700960
961 if (depth == 0)
962 goto out;
963
Mingming Cao617ba132006-10-11 01:20:53 -0700964 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700965
966 /* Simplest case - block found, no allocation needed */
967 if (!partial) {
968 first_block = le32_to_cpu(chain[depth - 1].key);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 count++;
970 /*map more blocks*/
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400971 while (count < map->m_len && count <= blocks_to_boundary) {
Mingming Cao617ba132006-10-11 01:20:53 -0700972 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700973
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700974 blk = le32_to_cpu(*(chain[depth-1].p + count));
975
976 if (blk == first_block + count)
977 count++;
978 else
979 break;
980 }
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500981 goto got_it;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700982 }
983
984 /* Next simple case - plain lookup or failed read of indirect block */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400985 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700986 goto cleanup;
987
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700988 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400989 * Okay, we need to do block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700990 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400991 goal = ext4_find_goal(inode, map->m_lblk, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992
993 /* the number of blocks need to allocate for [d,t]indirect blocks */
994 indirect_blks = (chain + depth) - partial - 1;
995
996 /*
997 * Next look up the indirect map to count the totoal number of
998 * direct blocks to allocate for this branch.
999 */
Mingming Cao617ba132006-10-11 01:20:53 -07001000 count = ext4_blks_to_allocate(partial, indirect_blks,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001001 map->m_len, blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001002 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001003 * Block out ext4_truncate while we alter the tree
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001004 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001005 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001006 &count, goal,
1007 offsets + (partial - chain), partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001008
1009 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001010 * The ext4_splice_branch call will free and forget any buffers
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001011 * on the new chain if there is a failure, but that risks using
1012 * up transaction credits, especially for bitmaps where the
1013 * credits cannot be returned. Can we handle this somehow? We
1014 * may need to return -EAGAIN upwards in the worst case. --sct
1015 */
1016 if (!err)
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001017 err = ext4_splice_branch(handle, inode, map->m_lblk,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001018 partial, indirect_blks, count);
Jan Kara2bba7022009-11-23 07:24:48 -05001019 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001020 goto cleanup;
1021
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001022 map->m_flags |= EXT4_MAP_NEW;
Jan Karab436b9b2009-12-08 23:51:10 -05001023
1024 ext4_update_inode_fsync_trans(handle, inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001025got_it:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001026 map->m_flags |= EXT4_MAP_MAPPED;
1027 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1028 map->m_len = count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001029 if (count > blocks_to_boundary)
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001030 map->m_flags |= EXT4_MAP_BOUNDARY;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001031 err = count;
1032 /* Clean up and exit */
1033 partial = chain + depth - 1; /* the whole chain */
1034cleanup:
1035 while (partial > chain) {
1036 BUFFER_TRACE(partial->bh, "call brelse");
1037 brelse(partial->bh);
1038 partial--;
1039 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040out:
1041 return err;
1042}
1043
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03001044#ifdef CONFIG_QUOTA
1045qsize_t *ext4_get_reserved_space(struct inode *inode)
Mingming Cao60e58e02009-01-22 18:13:05 +01001046{
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03001047 return &EXT4_I(inode)->i_reserved_quota;
Mingming Cao60e58e02009-01-22 18:13:05 +01001048}
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03001049#endif
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001050
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001051/*
1052 * Calculate the number of metadata blocks need to reserve
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001053 * to allocate a new block at @lblocks for non extent file based file
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001054 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001055static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1056 sector_t lblock)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001057{
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001058 struct ext4_inode_info *ei = EXT4_I(inode);
Jan Karad330a5b2010-03-14 18:17:54 -04001059 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001060 int blk_bits;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001061
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001062 if (lblock < EXT4_NDIR_BLOCKS)
1063 return 0;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001064
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001065 lblock -= EXT4_NDIR_BLOCKS;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001066
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001067 if (ei->i_da_metadata_calc_len &&
1068 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1069 ei->i_da_metadata_calc_len++;
1070 return 0;
1071 }
1072 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1073 ei->i_da_metadata_calc_len = 1;
Jan Karad330a5b2010-03-14 18:17:54 -04001074 blk_bits = order_base_2(lblock);
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001075 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001076}
1077
1078/*
1079 * Calculate the number of metadata blocks need to reserve
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001080 * to allocate a block located at @lblock
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001081 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001082static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001083{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001084 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001085 return ext4_ext_calc_metadata_amount(inode, lblock);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001086
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001087 return ext4_indirect_calc_metadata_amount(inode, lblock);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001088}
1089
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001090/*
1091 * Called with i_data_sem down, which is important since we can call
1092 * ext4_discard_preallocations() from here.
1093 */
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05001094void ext4_da_update_reserve_space(struct inode *inode,
1095 int used, int quota_claim)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001096{
1097 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001098 struct ext4_inode_info *ei = EXT4_I(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001099
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001100 spin_lock(&ei->i_block_reservation_lock);
Theodore Ts'of8ec9d62010-01-01 01:00:21 -05001101 trace_ext4_da_update_reserve_space(inode, used);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001102 if (unlikely(used > ei->i_reserved_data_blocks)) {
1103 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1104 "with only %d reserved data blocks\n",
1105 __func__, inode->i_ino, used,
1106 ei->i_reserved_data_blocks);
1107 WARN_ON(1);
1108 used = ei->i_reserved_data_blocks;
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001109 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001110
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001111 /* Update per-inode reservations */
1112 ei->i_reserved_data_blocks -= used;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001113 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001114 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1115 used + ei->i_allocated_meta_blocks);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001116 ei->i_allocated_meta_blocks = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001117
1118 if (ei->i_reserved_data_blocks == 0) {
1119 /*
1120 * We can release all of the reserved metadata blocks
1121 * only when we have written all of the delayed
1122 * allocation blocks.
1123 */
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001124 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1125 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -05001126 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001127 ei->i_da_metadata_calc_len = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001128 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001129 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001130
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001131 /* Update quota subsystem for data blocks */
1132 if (quota_claim)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001133 dquot_claim_block(inode, used);
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001134 else {
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05001135 /*
1136 * We did fallocate with an offset that is already delayed
1137 * allocated. So on delayed allocated writeback we should
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001138 * not re-claim the quota for fallocated blocks.
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05001139 */
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001140 dquot_release_reservation_block(inode, used);
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05001141 }
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -04001142
1143 /*
1144 * If we have done all the pending block allocations and if
1145 * there aren't any writers on the inode, we can discard the
1146 * inode's preallocations.
1147 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001148 if ((ei->i_reserved_data_blocks == 0) &&
1149 (atomic_read(&inode->i_writecount) == 0))
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -04001150 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001151}
1152
Theodore Ts'oe29136f2010-06-29 12:54:28 -04001153static int __check_block_validity(struct inode *inode, const char *func,
Theodore Ts'oc398eda2010-07-27 11:56:40 -04001154 unsigned int line,
1155 struct ext4_map_blocks *map)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001156{
Theodore Ts'o24676da2010-05-16 21:00:00 -04001157 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1158 map->m_len)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04001159 ext4_error_inode(inode, func, line, map->m_pblk,
1160 "lblock %lu mapped to illegal pblock "
1161 "(length %d)", (unsigned long) map->m_lblk,
1162 map->m_len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001163 return -EIO;
1164 }
1165 return 0;
1166}
1167
Theodore Ts'oe29136f2010-06-29 12:54:28 -04001168#define check_block_validity(inode, map) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -04001169 __check_block_validity((inode), __func__, __LINE__, (map))
Theodore Ts'oe29136f2010-06-29 12:54:28 -04001170
Mingming Caof5ab0d12008-02-25 15:29:55 -05001171/*
Theodore Ts'o1f945332009-09-30 22:57:41 -04001172 * Return the number of contiguous dirty pages in a given inode
1173 * starting at page frame idx.
Theodore Ts'o55138e02009-09-29 13:31:31 -04001174 */
1175static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1176 unsigned int max_pages)
1177{
1178 struct address_space *mapping = inode->i_mapping;
1179 pgoff_t index;
1180 struct pagevec pvec;
1181 pgoff_t num = 0;
1182 int i, nr_pages, done = 0;
1183
1184 if (max_pages == 0)
1185 return 0;
1186 pagevec_init(&pvec, 0);
1187 while (!done) {
1188 index = idx;
1189 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1190 PAGECACHE_TAG_DIRTY,
1191 (pgoff_t)PAGEVEC_SIZE);
1192 if (nr_pages == 0)
1193 break;
1194 for (i = 0; i < nr_pages; i++) {
1195 struct page *page = pvec.pages[i];
1196 struct buffer_head *bh, *head;
1197
1198 lock_page(page);
1199 if (unlikely(page->mapping != mapping) ||
1200 !PageDirty(page) ||
1201 PageWriteback(page) ||
1202 page->index != idx) {
1203 done = 1;
1204 unlock_page(page);
1205 break;
1206 }
Theodore Ts'o1f945332009-09-30 22:57:41 -04001207 if (page_has_buffers(page)) {
1208 bh = head = page_buffers(page);
1209 do {
1210 if (!buffer_delay(bh) &&
1211 !buffer_unwritten(bh))
1212 done = 1;
1213 bh = bh->b_this_page;
1214 } while (!done && (bh != head));
1215 }
Theodore Ts'o55138e02009-09-29 13:31:31 -04001216 unlock_page(page);
1217 if (done)
1218 break;
1219 idx++;
1220 num++;
Eric Sandeen659c6002010-10-27 21:30:03 -04001221 if (num >= max_pages) {
1222 done = 1;
Theodore Ts'o55138e02009-09-29 13:31:31 -04001223 break;
Eric Sandeen659c6002010-10-27 21:30:03 -04001224 }
Theodore Ts'o55138e02009-09-29 13:31:31 -04001225 }
1226 pagevec_release(&pvec);
1227 }
1228 return num;
1229}
1230
1231/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001232 * The ext4_map_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001233 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -05001234 *
Mingming Caof5ab0d12008-02-25 15:29:55 -05001235 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1236 * and store the allocated blocks in the result buffer head and mark it
1237 * mapped.
1238 *
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001239 * If file type is extents based, it will call ext4_ext_map_blocks(),
1240 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -05001241 * based files
1242 *
1243 * On success, it returns the number of blocks being mapped or allocate.
1244 * if create==0 and the blocks are pre-allocated and uninitialized block,
1245 * the result buffer head is unmapped. If the create ==1, it will make sure
1246 * the buffer head is mapped.
1247 *
1248 * It returns 0 if plain look up failed (blocks have not been allocated), in
1249 * that casem, buffer head is unmapped
1250 *
1251 * It returns the error in case of allocation failure.
1252 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001253int ext4_map_blocks(handle_t *handle, struct inode *inode,
1254 struct ext4_map_blocks *map, int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001255{
1256 int retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -05001257
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001258 map->m_flags = 0;
1259 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1260 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1261 (unsigned long) map->m_lblk);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001262 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -04001263 * Try to see if we can get the block without requesting a new
1264 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001265 */
1266 down_read((&EXT4_I(inode)->i_data_sem));
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001267 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001268 retval = ext4_ext_map_blocks(handle, inode, map, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001269 } else {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001270 retval = ext4_ind_map_blocks(handle, inode, map, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001271 }
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001272 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -05001273
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001274 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Theodore Ts'oe29136f2010-06-29 12:54:28 -04001275 int ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001276 if (ret != 0)
1277 return ret;
1278 }
1279
Mingming Caof5ab0d12008-02-25 15:29:55 -05001280 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -04001281 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001282 return retval;
1283
1284 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001285 * Returns if the blocks have already allocated
1286 *
1287 * Note that if blocks have been preallocated
1288 * ext4_ext_get_block() returns th create = 0
1289 * with buffer head unmapped.
1290 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001291 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
Mingming Caof5ab0d12008-02-25 15:29:55 -05001292 return retval;
1293
1294 /*
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -04001295 * When we call get_blocks without the create flag, the
1296 * BH_Unwritten flag could have gotten set if the blocks
1297 * requested were part of a uninitialized extent. We need to
1298 * clear this flag now that we are committed to convert all or
1299 * part of the uninitialized extent to be an initialized
1300 * extent. This is because we need to avoid the combination
1301 * of BH_Unwritten and BH_Mapped flags being simultaneously
1302 * set on the buffer_head.
1303 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001304 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -04001305
1306 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001307 * New blocks allocate and/or writing to uninitialized extent
1308 * will possibly result in updating i_data, so we take
1309 * the write lock of i_data_sem, and call get_blocks()
1310 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001311 */
1312 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -04001313
1314 /*
1315 * if the caller is from delayed allocation writeout path
1316 * we have already reserved fs blocks for allocation
1317 * let the underlying get_block() function know to
1318 * avoid double accounting
1319 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04001320 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Mingming Caod2a17632008-07-14 17:52:37 -04001321 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001322 /*
1323 * We need to check for EXT4 here because migrate
1324 * could have changed the inode type in between
1325 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001326 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001327 retval = ext4_ext_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001328 } else {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001329 retval = ext4_ind_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001330
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001331 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001332 /*
1333 * We allocated new blocks which will result in
1334 * i_data's format changing. Force the migrate
1335 * to fail by clearing migrate flags
1336 */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001337 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001338 }
Mingming Caod2a17632008-07-14 17:52:37 -04001339
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05001340 /*
1341 * Update reserved blocks/metadata blocks after successful
1342 * block allocation which had been deferred till now. We don't
1343 * support fallocate for non extent files. So we can update
1344 * reserve space here.
1345 */
1346 if ((retval > 0) &&
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05001347 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05001348 ext4_da_update_reserve_space(inode, retval, 1);
1349 }
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001350 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Mingming Caod2a17632008-07-14 17:52:37 -04001351 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001352
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001353 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'oe35fd662010-05-16 19:00:00 -04001354 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Theodore Ts'oe29136f2010-06-29 12:54:28 -04001355 int ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001356 if (ret != 0)
1357 return ret;
1358 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001359 return retval;
1360}
1361
Mingming Caof3bd1f32008-08-19 22:16:03 -04001362/* Maximum number of blocks we map for direct IO at once. */
1363#define DIO_MAX_BLOCKS 4096
1364
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001365static int _ext4_get_block(struct inode *inode, sector_t iblock,
1366 struct buffer_head *bh, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001367{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001368 handle_t *handle = ext4_journal_current_handle();
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001369 struct ext4_map_blocks map;
Jan Kara7fb54092008-02-10 01:08:38 -05001370 int ret = 0, started = 0;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001371 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001372
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001373 map.m_lblk = iblock;
1374 map.m_len = bh->b_size >> inode->i_blkbits;
1375
1376 if (flags && !handle) {
Jan Kara7fb54092008-02-10 01:08:38 -05001377 /* Direct IO write... */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001378 if (map.m_len > DIO_MAX_BLOCKS)
1379 map.m_len = DIO_MAX_BLOCKS;
1380 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
Mingming Caof3bd1f32008-08-19 22:16:03 -04001381 handle = ext4_journal_start(inode, dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -05001382 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001383 ret = PTR_ERR(handle);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001384 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001385 }
Jan Kara7fb54092008-02-10 01:08:38 -05001386 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001387 }
1388
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001389 ret = ext4_map_blocks(handle, inode, &map, flags);
Jan Kara7fb54092008-02-10 01:08:38 -05001390 if (ret > 0) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001391 map_bh(bh, inode->i_sb, map.m_pblk);
1392 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1393 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
Jan Kara7fb54092008-02-10 01:08:38 -05001394 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001395 }
Jan Kara7fb54092008-02-10 01:08:38 -05001396 if (started)
1397 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001398 return ret;
1399}
1400
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001401int ext4_get_block(struct inode *inode, sector_t iblock,
1402 struct buffer_head *bh, int create)
1403{
1404 return _ext4_get_block(inode, iblock, bh,
1405 create ? EXT4_GET_BLOCKS_CREATE : 0);
1406}
1407
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001408/*
1409 * `handle' can be NULL if create is zero
1410 */
Mingming Cao617ba132006-10-11 01:20:53 -07001411struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001412 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001413{
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001414 struct ext4_map_blocks map;
1415 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001416 int fatal = 0, err;
1417
1418 J_ASSERT(handle != NULL || create == 0);
1419
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001420 map.m_lblk = block;
1421 map.m_len = 1;
1422 err = ext4_map_blocks(handle, inode, &map,
1423 create ? EXT4_GET_BLOCKS_CREATE : 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001424
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001425 if (err < 0)
1426 *errp = err;
1427 if (err <= 0)
1428 return NULL;
1429 *errp = 0;
1430
1431 bh = sb_getblk(inode->i_sb, map.m_pblk);
1432 if (!bh) {
1433 *errp = -EIO;
1434 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001435 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001436 if (map.m_flags & EXT4_MAP_NEW) {
1437 J_ASSERT(create != 0);
1438 J_ASSERT(handle != NULL);
1439
1440 /*
1441 * Now that we do not always journal data, we should
1442 * keep in mind whether this should always journal the
1443 * new buffer as metadata. For now, regular file
1444 * writes use ext4_get_block instead, so it's not a
1445 * problem.
1446 */
1447 lock_buffer(bh);
1448 BUFFER_TRACE(bh, "call get_create_access");
1449 fatal = ext4_journal_get_create_access(handle, bh);
1450 if (!fatal && !buffer_uptodate(bh)) {
1451 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1452 set_buffer_uptodate(bh);
1453 }
1454 unlock_buffer(bh);
1455 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1456 err = ext4_handle_dirty_metadata(handle, inode, bh);
1457 if (!fatal)
1458 fatal = err;
1459 } else {
1460 BUFFER_TRACE(bh, "not a new buffer");
1461 }
1462 if (fatal) {
1463 *errp = fatal;
1464 brelse(bh);
1465 bh = NULL;
1466 }
1467 return bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001468}
1469
Mingming Cao617ba132006-10-11 01:20:53 -07001470struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001471 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001473 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001474
Mingming Cao617ba132006-10-11 01:20:53 -07001475 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001476 if (!bh)
1477 return bh;
1478 if (buffer_uptodate(bh))
1479 return bh;
1480 ll_rw_block(READ_META, 1, &bh);
1481 wait_on_buffer(bh);
1482 if (buffer_uptodate(bh))
1483 return bh;
1484 put_bh(bh);
1485 *err = -EIO;
1486 return NULL;
1487}
1488
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001489static int walk_page_buffers(handle_t *handle,
1490 struct buffer_head *head,
1491 unsigned from,
1492 unsigned to,
1493 int *partial,
1494 int (*fn)(handle_t *handle,
1495 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001496{
1497 struct buffer_head *bh;
1498 unsigned block_start, block_end;
1499 unsigned blocksize = head->b_size;
1500 int err, ret = 0;
1501 struct buffer_head *next;
1502
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001503 for (bh = head, block_start = 0;
1504 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001505 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001506 next = bh->b_this_page;
1507 block_end = block_start + blocksize;
1508 if (block_end <= from || block_start >= to) {
1509 if (partial && !buffer_uptodate(bh))
1510 *partial = 1;
1511 continue;
1512 }
1513 err = (*fn)(handle, bh);
1514 if (!ret)
1515 ret = err;
1516 }
1517 return ret;
1518}
1519
1520/*
1521 * To preserve ordering, it is essential that the hole instantiation and
1522 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -07001523 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -07001524 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001525 * prepare_write() is the right place.
1526 *
Mingming Cao617ba132006-10-11 01:20:53 -07001527 * Also, this function can nest inside ext4_writepage() ->
1528 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001529 * has generated enough buffer credits to do the whole page. So we won't
1530 * block on the journal in that case, which is good, because the caller may
1531 * be PF_MEMALLOC.
1532 *
Mingming Cao617ba132006-10-11 01:20:53 -07001533 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001534 * quota file writes. If we were to commit the transaction while thus
1535 * reentered, there can be a deadlock - we would be holding a quota
1536 * lock, and the commit would never complete if another thread had a
1537 * transaction open and was blocking on the quota lock - a ranking
1538 * violation.
1539 *
Mingming Caodab291a2006-10-11 01:21:01 -07001540 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541 * will _not_ run commit under these circumstances because handle->h_ref
1542 * is elevated. We'll still have enough credits for the tiny quotafile
1543 * write.
1544 */
1545static int do_journal_get_write_access(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001546 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001547{
Jan Kara56d35a42010-08-05 14:41:42 -04001548 int dirty = buffer_dirty(bh);
1549 int ret;
1550
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001551 if (!buffer_mapped(bh) || buffer_freed(bh))
1552 return 0;
Jan Kara56d35a42010-08-05 14:41:42 -04001553 /*
1554 * __block_prepare_write() could have dirtied some buffers. Clean
1555 * the dirty bit as jbd2_journal_get_write_access() could complain
1556 * otherwise about fs integrity issues. Setting of the dirty bit
1557 * by __block_prepare_write() isn't a real problem here as we clear
1558 * the bit before releasing a page lock and thus writeback cannot
1559 * ever write the buffer.
1560 */
1561 if (dirty)
1562 clear_buffer_dirty(bh);
1563 ret = ext4_journal_get_write_access(handle, bh);
1564 if (!ret && dirty)
1565 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1566 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001567}
1568
Jan Karab9a42072009-12-08 21:24:33 -05001569/*
1570 * Truncate blocks that were not used by write. We have to truncate the
1571 * pagecache as well so that corresponding buffers get properly unmapped.
1572 */
1573static void ext4_truncate_failed_write(struct inode *inode)
1574{
1575 truncate_inode_pages(inode->i_mapping, inode->i_size);
1576 ext4_truncate(inode);
1577}
1578
Jiaying Zhang744692d2010-03-04 16:14:02 -05001579static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1580 struct buffer_head *bh_result, int create);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001581static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001582 loff_t pos, unsigned len, unsigned flags,
1583 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001584{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001585 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001586 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001587 handle_t *handle;
1588 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001589 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001590 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001591 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001592
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001593 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001594 /*
1595 * Reserve one block more for addition to orphan list in case
1596 * we allocate blocks but write fails for some reason
1597 */
1598 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001599 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001600 from = pos & (PAGE_CACHE_SIZE - 1);
1601 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001602
1603retry:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001604 handle = ext4_journal_start(inode, needed_blocks);
1605 if (IS_ERR(handle)) {
1606 ret = PTR_ERR(handle);
1607 goto out;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001608 }
1609
Jan Karaebd36102009-02-22 21:09:59 -05001610 /* We cannot recurse into the filesystem as the transaction is already
1611 * started */
1612 flags |= AOP_FLAG_NOFS;
1613
Nick Piggin54566b22009-01-04 12:00:53 -08001614 page = grab_cache_page_write_begin(mapping, index, flags);
Jan Karacf108bc2008-07-11 19:27:31 -04001615 if (!page) {
1616 ext4_journal_stop(handle);
1617 ret = -ENOMEM;
1618 goto out;
1619 }
1620 *pagep = page;
1621
Jiaying Zhang744692d2010-03-04 16:14:02 -05001622 if (ext4_should_dioread_nolock(inode))
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001623 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
Jiaying Zhang744692d2010-03-04 16:14:02 -05001624 else
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001625 ret = __block_write_begin(page, pos, len, ext4_get_block);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001626
1627 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628 ret = walk_page_buffers(handle, page_buffers(page),
1629 from, to, NULL, do_journal_get_write_access);
1630 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001631
1632 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001633 unlock_page(page);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001634 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001635 /*
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001636 * __block_write_begin may have instantiated a few blocks
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001637 * outside i_size. Trim these off again. Don't need
1638 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001639 *
1640 * Add inode to orphan list in case we crash before
1641 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001642 */
Jan Karaffacfa72009-07-13 16:22:22 -04001643 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001644 ext4_orphan_add(handle, inode);
1645
1646 ext4_journal_stop(handle);
1647 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001648 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001649 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001650 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001651 * still be on the orphan list; we need to
1652 * make sure the inode is removed from the
1653 * orphan list in that case.
1654 */
1655 if (inode->i_nlink)
1656 ext4_orphan_del(NULL, inode);
1657 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001658 }
1659
Mingming Cao617ba132006-10-11 01:20:53 -07001660 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001661 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001662out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001663 return ret;
1664}
1665
Nick Pigginbfc1af62007-10-16 01:25:05 -07001666/* For write_end() in data=journal mode */
1667static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001668{
1669 if (!buffer_mapped(bh) || buffer_freed(bh))
1670 return 0;
1671 set_buffer_uptodate(bh);
Frank Mayhar03901312009-01-07 00:06:22 -05001672 return ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001673}
1674
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001675static int ext4_generic_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001676 struct address_space *mapping,
1677 loff_t pos, unsigned len, unsigned copied,
1678 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001679{
1680 int i_size_changed = 0;
1681 struct inode *inode = mapping->host;
1682 handle_t *handle = ext4_journal_current_handle();
1683
1684 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1685
1686 /*
1687 * No need to use i_size_read() here, the i_size
1688 * cannot change under us because we hold i_mutex.
1689 *
1690 * But it's important to update i_size while still holding page lock:
1691 * page writeout could otherwise come in and zero beyond i_size.
1692 */
1693 if (pos + copied > inode->i_size) {
1694 i_size_write(inode, pos + copied);
1695 i_size_changed = 1;
1696 }
1697
1698 if (pos + copied > EXT4_I(inode)->i_disksize) {
1699 /* We need to mark inode dirty even if
1700 * new_i_size is less that inode->i_size
1701 * bu greater than i_disksize.(hint delalloc)
1702 */
1703 ext4_update_i_disksize(inode, (pos + copied));
1704 i_size_changed = 1;
1705 }
1706 unlock_page(page);
1707 page_cache_release(page);
1708
1709 /*
1710 * Don't mark the inode dirty under page lock. First, it unnecessarily
1711 * makes the holding time of page lock longer. Second, it forces lock
1712 * ordering of page lock and transaction start for journaling
1713 * filesystems.
1714 */
1715 if (i_size_changed)
1716 ext4_mark_inode_dirty(handle, inode);
1717
1718 return copied;
1719}
1720
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001721/*
1722 * We need to pick up the new inode size which generic_commit_write gave us
1723 * `file' can be NULL - eg, when called from page_symlink().
1724 *
Mingming Cao617ba132006-10-11 01:20:53 -07001725 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001726 * buffers are managed internally.
1727 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001728static int ext4_ordered_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001729 struct address_space *mapping,
1730 loff_t pos, unsigned len, unsigned copied,
1731 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001732{
Mingming Cao617ba132006-10-11 01:20:53 -07001733 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001734 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001735 int ret = 0, ret2;
1736
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001737 trace_ext4_ordered_write_end(inode, pos, len, copied);
Jan Kara678aaf42008-07-11 19:27:31 -04001738 ret = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001739
1740 if (ret == 0) {
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001741 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001742 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001743 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001744 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001745 /* if we have allocated more blocks and copied
1746 * less. We will have blocks allocated outside
1747 * inode->i_size. So truncate them
1748 */
1749 ext4_orphan_add(handle, inode);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001750 if (ret2 < 0)
1751 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752 }
Mingming Cao617ba132006-10-11 01:20:53 -07001753 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001754 if (!ret)
1755 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001756
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001757 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001758 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001759 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001760 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001761 * on the orphan list; we need to make sure the inode
1762 * is removed from the orphan list in that case.
1763 */
1764 if (inode->i_nlink)
1765 ext4_orphan_del(NULL, inode);
1766 }
1767
1768
Nick Pigginbfc1af62007-10-16 01:25:05 -07001769 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001770}
1771
Nick Pigginbfc1af62007-10-16 01:25:05 -07001772static int ext4_writeback_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001773 struct address_space *mapping,
1774 loff_t pos, unsigned len, unsigned copied,
1775 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001776{
Mingming Cao617ba132006-10-11 01:20:53 -07001777 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001778 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001779 int ret = 0, ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001780
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001781 trace_ext4_writeback_write_end(inode, pos, len, copied);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001782 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001783 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001784 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001785 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001786 /* if we have allocated more blocks and copied
1787 * less. We will have blocks allocated outside
1788 * inode->i_size. So truncate them
1789 */
1790 ext4_orphan_add(handle, inode);
1791
Roel Kluinf8a87d82008-04-29 22:01:18 -04001792 if (ret2 < 0)
1793 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001794
Mingming Cao617ba132006-10-11 01:20:53 -07001795 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001796 if (!ret)
1797 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001798
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001799 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001800 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001801 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001802 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001803 * on the orphan list; we need to make sure the inode
1804 * is removed from the orphan list in that case.
1805 */
1806 if (inode->i_nlink)
1807 ext4_orphan_del(NULL, inode);
1808 }
1809
Nick Pigginbfc1af62007-10-16 01:25:05 -07001810 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001811}
1812
Nick Pigginbfc1af62007-10-16 01:25:05 -07001813static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001814 struct address_space *mapping,
1815 loff_t pos, unsigned len, unsigned copied,
1816 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001817{
Mingming Cao617ba132006-10-11 01:20:53 -07001818 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001819 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001820 int ret = 0, ret2;
1821 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001822 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001823 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001824
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001825 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001826 from = pos & (PAGE_CACHE_SIZE - 1);
1827 to = from + len;
1828
1829 if (copied < len) {
1830 if (!PageUptodate(page))
1831 copied = 0;
1832 page_zero_new_buffers(page, from+copied, to);
1833 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001834
1835 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001836 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001837 if (!partial)
1838 SetPageUptodate(page);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001839 new_i_size = pos + copied;
1840 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001841 i_size_write(inode, pos+copied);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001842 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001843 if (new_i_size > EXT4_I(inode)->i_disksize) {
1844 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001845 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001846 if (!ret)
1847 ret = ret2;
1848 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001849
Jan Karacf108bc2008-07-11 19:27:31 -04001850 unlock_page(page);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001851 page_cache_release(page);
Jan Karaffacfa72009-07-13 16:22:22 -04001852 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001853 /* if we have allocated more blocks and copied
1854 * less. We will have blocks allocated outside
1855 * inode->i_size. So truncate them
1856 */
1857 ext4_orphan_add(handle, inode);
1858
Mingming Cao617ba132006-10-11 01:20:53 -07001859 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001860 if (!ret)
1861 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001862 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001863 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001864 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001865 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001866 * on the orphan list; we need to make sure the inode
1867 * is removed from the orphan list in that case.
1868 */
1869 if (inode->i_nlink)
1870 ext4_orphan_del(NULL, inode);
1871 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001872
1873 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001874}
Mingming Caod2a17632008-07-14 17:52:37 -04001875
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001876/*
1877 * Reserve a single block located at lblock
1878 */
1879static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -04001880{
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001881 int retries = 0;
Mingming Cao60e58e02009-01-22 18:13:05 +01001882 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001883 struct ext4_inode_info *ei = EXT4_I(inode);
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001884 unsigned long md_needed;
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001885 int ret;
Mingming Caod2a17632008-07-14 17:52:37 -04001886
1887 /*
1888 * recalculate the amount of metadata blocks to reserve
1889 * in order to allocate nrblocks
1890 * worse case is one extent per block
1891 */
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001892repeat:
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001893 spin_lock(&ei->i_block_reservation_lock);
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001894 md_needed = ext4_calc_metadata_amount(inode, lblock);
Theodore Ts'of8ec9d62010-01-01 01:00:21 -05001895 trace_ext4_da_reserve_space(inode, md_needed);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001896 spin_unlock(&ei->i_block_reservation_lock);
Mingming Caod2a17632008-07-14 17:52:37 -04001897
Mingming Cao60e58e02009-01-22 18:13:05 +01001898 /*
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001899 * We will charge metadata quota at writeout time; this saves
1900 * us from metadata over-estimation, though we may go over by
1901 * a small amount in the end. Here we just reserve for data.
Mingming Cao60e58e02009-01-22 18:13:05 +01001902 */
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001903 ret = dquot_reserve_block(inode, 1);
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001904 if (ret)
1905 return ret;
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001906 /*
1907 * We do still charge estimated metadata to the sb though;
1908 * we cannot afford to run out of free blocks.
1909 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001910 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001911 dquot_release_reservation_block(inode, 1);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001912 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1913 yield();
1914 goto repeat;
1915 }
Mingming Caod2a17632008-07-14 17:52:37 -04001916 return -ENOSPC;
1917 }
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001918 spin_lock(&ei->i_block_reservation_lock);
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001919 ei->i_reserved_data_blocks++;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001920 ei->i_reserved_meta_blocks += md_needed;
1921 spin_unlock(&ei->i_block_reservation_lock);
Dmitry Monakhov39bc6802009-12-10 16:36:27 +00001922
Mingming Caod2a17632008-07-14 17:52:37 -04001923 return 0; /* success */
1924}
1925
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001926static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001927{
1928 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001929 struct ext4_inode_info *ei = EXT4_I(inode);
Mingming Caod2a17632008-07-14 17:52:37 -04001930
Mingming Caocd213222008-08-19 22:16:59 -04001931 if (!to_free)
1932 return; /* Nothing to release, exit */
1933
Mingming Caod2a17632008-07-14 17:52:37 -04001934 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001935
Li Zefan5a58ec82010-05-17 02:00:00 -04001936 trace_ext4_da_release_space(inode, to_free);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001937 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
Mingming Caocd213222008-08-19 22:16:59 -04001938 /*
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001939 * if there aren't enough reserved blocks, then the
1940 * counter is messed up somewhere. Since this
1941 * function is called from invalidate page, it's
1942 * harmless to return without any action.
Mingming Caocd213222008-08-19 22:16:59 -04001943 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001944 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1945 "ino %lu, to_free %d with only %d reserved "
1946 "data blocks\n", inode->i_ino, to_free,
1947 ei->i_reserved_data_blocks);
1948 WARN_ON(1);
1949 to_free = ei->i_reserved_data_blocks;
1950 }
1951 ei->i_reserved_data_blocks -= to_free;
1952
1953 if (ei->i_reserved_data_blocks == 0) {
1954 /*
1955 * We can release all of the reserved metadata blocks
1956 * only when we have written all of the delayed
1957 * allocation blocks.
1958 */
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001959 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1960 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -05001961 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001962 ei->i_da_metadata_calc_len = 0;
Mingming Caocd213222008-08-19 22:16:59 -04001963 }
1964
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001965 /* update fs dirty data blocks counter */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001966 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
Mingming Caod2a17632008-07-14 17:52:37 -04001967
Mingming Caod2a17632008-07-14 17:52:37 -04001968 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001969
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001970 dquot_release_reservation_block(inode, to_free);
Mingming Caod2a17632008-07-14 17:52:37 -04001971}
1972
1973static void ext4_da_page_release_reservation(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001974 unsigned long offset)
Mingming Caod2a17632008-07-14 17:52:37 -04001975{
1976 int to_release = 0;
1977 struct buffer_head *head, *bh;
1978 unsigned int curr_off = 0;
1979
1980 head = page_buffers(page);
1981 bh = head;
1982 do {
1983 unsigned int next_off = curr_off + bh->b_size;
1984
1985 if ((offset <= curr_off) && (buffer_delay(bh))) {
1986 to_release++;
1987 clear_buffer_delay(bh);
1988 }
1989 curr_off = next_off;
1990 } while ((bh = bh->b_this_page) != head);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001991 ext4_da_release_space(page->mapping->host, to_release);
Mingming Caod2a17632008-07-14 17:52:37 -04001992}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001993
1994/*
Alex Tomas64769242008-07-11 19:27:31 -04001995 * Delayed allocation stuff
1996 */
1997
Alex Tomas64769242008-07-11 19:27:31 -04001998/*
1999 * mpage_da_submit_io - walks through extent of pages and try to write
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002000 * them with writepage() call back
Alex Tomas64769242008-07-11 19:27:31 -04002001 *
2002 * @mpd->inode: inode
2003 * @mpd->first_page: first page of the extent
2004 * @mpd->next_page: page after the last page of the extent
Alex Tomas64769242008-07-11 19:27:31 -04002005 *
2006 * By the time mpage_da_submit_io() is called we expect all blocks
2007 * to be allocated. this may be wrong if allocation failed.
2008 *
2009 * As pages are already locked by write_cache_pages(), we can't use it
2010 */
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002011static int mpage_da_submit_io(struct mpage_da_data *mpd,
2012 struct ext4_map_blocks *map)
Alex Tomas64769242008-07-11 19:27:31 -04002013{
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002014 struct pagevec pvec;
2015 unsigned long index, end;
2016 int ret = 0, err, nr_pages, i;
2017 struct inode *inode = mpd->inode;
2018 struct address_space *mapping = inode->i_mapping;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002019 loff_t size = i_size_read(inode);
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04002020 unsigned int len, block_start;
2021 struct buffer_head *bh, *page_bufs = NULL;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002022 int journal_data = ext4_should_journal_data(inode);
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002023 sector_t pblock = 0, cur_logical = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002024 struct ext4_io_submit io_submit;
Alex Tomas64769242008-07-11 19:27:31 -04002025
2026 BUG_ON(mpd->next_page <= mpd->first_page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002027 memset(&io_submit, 0, sizeof(io_submit));
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002028 /*
2029 * We need to start from the first_page to the next_page - 1
2030 * to make sure we also write the mapped dirty buffer_heads.
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002031 * If we look at mpd->b_blocknr we would only be looking
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002032 * at the currently mapped buffer_heads.
2033 */
Alex Tomas64769242008-07-11 19:27:31 -04002034 index = mpd->first_page;
2035 end = mpd->next_page - 1;
2036
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002037 pagevec_init(&pvec, 0);
Alex Tomas64769242008-07-11 19:27:31 -04002038 while (index <= end) {
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002039 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
Alex Tomas64769242008-07-11 19:27:31 -04002040 if (nr_pages == 0)
2041 break;
2042 for (i = 0; i < nr_pages; i++) {
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002043 int commit_write = 0, redirty_page = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002044 struct page *page = pvec.pages[i];
2045
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002046 index = page->index;
2047 if (index > end)
2048 break;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002049
2050 if (index == size >> PAGE_CACHE_SHIFT)
2051 len = size & ~PAGE_CACHE_MASK;
2052 else
2053 len = PAGE_CACHE_SIZE;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002054 if (map) {
2055 cur_logical = index << (PAGE_CACHE_SHIFT -
2056 inode->i_blkbits);
2057 pblock = map->m_pblk + (cur_logical -
2058 map->m_lblk);
2059 }
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002060 index++;
2061
2062 BUG_ON(!PageLocked(page));
2063 BUG_ON(PageWriteback(page));
2064
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002065 /*
2066 * If the page does not have buffers (for
2067 * whatever reason), try to create them using
2068 * block_prepare_write. If this fails,
2069 * redirty the page and move on.
2070 */
2071 if (!page_has_buffers(page)) {
2072 if (block_prepare_write(page, 0, len,
2073 noalloc_get_block_write)) {
2074 redirty_page:
2075 redirty_page_for_writepage(mpd->wbc,
2076 page);
2077 unlock_page(page);
2078 continue;
2079 }
2080 commit_write = 1;
2081 }
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04002082
2083 bh = page_bufs = page_buffers(page);
2084 block_start = 0;
2085 do {
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002086 if (!bh)
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04002087 goto redirty_page;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002088 if (map && (cur_logical >= map->m_lblk) &&
2089 (cur_logical <= (map->m_lblk +
2090 (map->m_len - 1)))) {
2091 if (buffer_delay(bh)) {
2092 clear_buffer_delay(bh);
2093 bh->b_blocknr = pblock;
2094 }
2095 if (buffer_unwritten(bh) ||
2096 buffer_mapped(bh))
2097 BUG_ON(bh->b_blocknr != pblock);
2098 if (map->m_flags & EXT4_MAP_UNINIT)
2099 set_buffer_uninit(bh);
2100 clear_buffer_unwritten(bh);
2101 }
2102
2103 /* redirty page if block allocation undone */
2104 if (buffer_delay(bh) || buffer_unwritten(bh))
2105 redirty_page = 1;
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04002106 bh = bh->b_this_page;
2107 block_start += bh->b_size;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002108 cur_logical++;
2109 pblock++;
2110 } while (bh != page_bufs);
2111
2112 if (redirty_page)
2113 goto redirty_page;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002114
2115 if (commit_write)
2116 /* mark the buffer_heads as dirty & uptodate */
2117 block_commit_write(page, 0, len);
2118
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002119 /*
2120 * Delalloc doesn't support data journalling,
2121 * but eventually maybe we'll lift this
2122 * restriction.
2123 */
2124 if (unlikely(journal_data && PageChecked(page)))
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002125 err = __ext4_journalled_writepage(page, len);
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002126 else
2127 err = ext4_bio_write_page(&io_submit, page,
2128 len, mpd->wbc);
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002129
2130 if (!err)
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002131 mpd->pages_written++;
Alex Tomas64769242008-07-11 19:27:31 -04002132 /*
2133 * In error case, we have to continue because
2134 * remaining pages are still locked
Alex Tomas64769242008-07-11 19:27:31 -04002135 */
2136 if (ret == 0)
2137 ret = err;
2138 }
2139 pagevec_release(&pvec);
2140 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002141 ext4_io_submit(&io_submit);
Alex Tomas64769242008-07-11 19:27:31 -04002142 return ret;
2143}
2144
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002145static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2146 sector_t logical, long blk_cnt)
2147{
2148 int nr_pages, i;
2149 pgoff_t index, end;
2150 struct pagevec pvec;
2151 struct inode *inode = mpd->inode;
2152 struct address_space *mapping = inode->i_mapping;
2153
2154 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2155 end = (logical + blk_cnt - 1) >>
2156 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2157 while (index <= end) {
2158 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2159 if (nr_pages == 0)
2160 break;
2161 for (i = 0; i < nr_pages; i++) {
2162 struct page *page = pvec.pages[i];
Jan Kara9b1d09982010-03-03 16:19:32 -05002163 if (page->index > end)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002164 break;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002165 BUG_ON(!PageLocked(page));
2166 BUG_ON(PageWriteback(page));
2167 block_invalidatepage(page, 0);
2168 ClearPageUptodate(page);
2169 unlock_page(page);
2170 }
Jan Kara9b1d09982010-03-03 16:19:32 -05002171 index = pvec.pages[nr_pages - 1]->index + 1;
2172 pagevec_release(&pvec);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002173 }
2174 return;
2175}
2176
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002177static void ext4_print_free_blocks(struct inode *inode)
2178{
2179 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o16939182009-09-26 17:43:59 -04002180 printk(KERN_CRIT "Total free blocks count %lld\n",
2181 ext4_count_free_blocks(inode->i_sb));
2182 printk(KERN_CRIT "Free/Dirty block details\n");
2183 printk(KERN_CRIT "free_blocks=%lld\n",
2184 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2185 printk(KERN_CRIT "dirty_blocks=%lld\n",
2186 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2187 printk(KERN_CRIT "Block reservation details\n");
2188 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2189 EXT4_I(inode)->i_reserved_data_blocks);
2190 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2191 EXT4_I(inode)->i_reserved_meta_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002192 return;
2193}
2194
Theodore Ts'ob920c752009-05-14 00:54:29 -04002195/*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002196 * mpage_da_map_and_submit - go through given space, map them
2197 * if necessary, and then submit them for I/O
Alex Tomas64769242008-07-11 19:27:31 -04002198 *
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002199 * @mpd - bh describing space
Alex Tomas64769242008-07-11 19:27:31 -04002200 *
2201 * The function skips space we know is already mapped to disk blocks.
2202 *
Alex Tomas64769242008-07-11 19:27:31 -04002203 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002204static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04002205{
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002206 int err, blks, get_blocks_flags;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002207 struct ext4_map_blocks map, *mapp = NULL;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002208 sector_t next = mpd->b_blocknr;
2209 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2210 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2211 handle_t *handle = NULL;
Alex Tomas64769242008-07-11 19:27:31 -04002212
2213 /*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002214 * If the blocks are mapped already, or we couldn't accumulate
2215 * any blocks, then proceed immediately to the submission stage.
Alex Tomas64769242008-07-11 19:27:31 -04002216 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002217 if ((mpd->b_size == 0) ||
2218 ((mpd->b_state & (1 << BH_Mapped)) &&
2219 !(mpd->b_state & (1 << BH_Delay)) &&
2220 !(mpd->b_state & (1 << BH_Unwritten))))
2221 goto submit_io;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002222
2223 handle = ext4_journal_current_handle();
2224 BUG_ON(!handle);
2225
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04002226 /*
Eric Sandeen79e83032010-07-27 11:56:07 -04002227 * Call ext4_map_blocks() to allocate any delayed allocation
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002228 * blocks, or to convert an uninitialized extent to be
2229 * initialized (in the case where we have written into
2230 * one or more preallocated blocks).
2231 *
2232 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2233 * indicate that we are on the delayed allocation path. This
2234 * affects functions in many different parts of the allocation
2235 * call path. This flag exists primarily because we don't
Eric Sandeen79e83032010-07-27 11:56:07 -04002236 * want to change *many* call functions, so ext4_map_blocks()
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002237 * will set the magic i_delalloc_reserved_flag once the
2238 * inode's allocation semaphore is taken.
2239 *
2240 * If the blocks in questions were delalloc blocks, set
2241 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2242 * variables are updated after the blocks have been allocated.
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04002243 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002244 map.m_lblk = next;
2245 map.m_len = max_blocks;
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05002246 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
Jiaying Zhang744692d2010-03-04 16:14:02 -05002247 if (ext4_should_dioread_nolock(mpd->inode))
2248 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002249 if (mpd->b_state & (1 << BH_Delay))
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05002250 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2251
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002252 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002253 if (blks < 0) {
Eric Sandeene3570632010-07-27 11:56:08 -04002254 struct super_block *sb = mpd->inode->i_sb;
2255
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002256 err = blks;
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002257 /*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002258 * If get block returns EAGAIN or ENOSPC and there
2259 * appears to be free blocks we will call
2260 * ext4_writepage() for all of the pages which will
2261 * just redirty the pages.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002262 */
2263 if (err == -EAGAIN)
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002264 goto submit_io;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002265
2266 if (err == -ENOSPC &&
Eric Sandeene3570632010-07-27 11:56:08 -04002267 ext4_count_free_blocks(sb)) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002268 mpd->retval = err;
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002269 goto submit_io;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002270 }
2271
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002272 /*
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002273 * get block failure will cause us to loop in
2274 * writepages, because a_ops->writepage won't be able
2275 * to make progress. The page will be redirtied by
2276 * writepage and writepages will again try to write
2277 * the same.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002278 */
Eric Sandeene3570632010-07-27 11:56:08 -04002279 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2280 ext4_msg(sb, KERN_CRIT,
2281 "delayed block allocation failed for inode %lu "
2282 "at logical offset %llu with max blocks %zd "
2283 "with error %d", mpd->inode->i_ino,
2284 (unsigned long long) next,
2285 mpd->b_size >> mpd->inode->i_blkbits, err);
2286 ext4_msg(sb, KERN_CRIT,
2287 "This should not happen!! Data will be lost\n");
2288 if (err == -ENOSPC)
2289 ext4_print_free_blocks(mpd->inode);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002290 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002291 /* invalidate all the pages */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002292 ext4_da_block_invalidatepages(mpd, next,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002293 mpd->b_size >> mpd->inode->i_blkbits);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002294 return;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002295 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002296 BUG_ON(blks == 0);
2297
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002298 mapp = &map;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002299 if (map.m_flags & EXT4_MAP_NEW) {
2300 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2301 int i;
Alex Tomas64769242008-07-11 19:27:31 -04002302
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002303 for (i = 0; i < map.m_len; i++)
2304 unmap_underlying_metadata(bdev, map.m_pblk + i);
2305 }
Alex Tomas64769242008-07-11 19:27:31 -04002306
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002307 if (ext4_should_order_data(mpd->inode)) {
2308 err = ext4_jbd2_file_inode(handle, mpd->inode);
2309 if (err)
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002310 /* This only happens if the journal is aborted */
2311 return;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002312 }
2313
2314 /*
Jan Kara03f5d8b2009-06-09 00:17:05 -04002315 * Update on-disk size along with block allocation.
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002316 */
2317 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2318 if (disksize > i_size_read(mpd->inode))
2319 disksize = i_size_read(mpd->inode);
2320 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2321 ext4_update_i_disksize(mpd->inode, disksize);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002322 err = ext4_mark_inode_dirty(handle, mpd->inode);
2323 if (err)
2324 ext4_error(mpd->inode->i_sb,
2325 "Failed to mark inode %lu dirty",
2326 mpd->inode->i_ino);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002327 }
2328
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002329submit_io:
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04002330 mpage_da_submit_io(mpd, mapp);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002331 mpd->io_done = 1;
Alex Tomas64769242008-07-11 19:27:31 -04002332}
2333
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04002334#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2335 (1 << BH_Delay) | (1 << BH_Unwritten))
Alex Tomas64769242008-07-11 19:27:31 -04002336
2337/*
2338 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2339 *
2340 * @mpd->lbh - extent of blocks
2341 * @logical - logical number of the block in the file
2342 * @bh - bh of the block (used to access block's state)
2343 *
2344 * the function is used to collect contig. blocks in same state
2345 */
2346static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002347 sector_t logical, size_t b_size,
2348 unsigned long b_state)
Alex Tomas64769242008-07-11 19:27:31 -04002349{
Alex Tomas64769242008-07-11 19:27:31 -04002350 sector_t next;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002351 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
Alex Tomas64769242008-07-11 19:27:31 -04002352
Eric Sandeenc445e3e2010-05-16 04:00:00 -04002353 /*
2354 * XXX Don't go larger than mballoc is willing to allocate
2355 * This is a stopgap solution. We eventually need to fold
2356 * mpage_da_submit_io() into this function and then call
Eric Sandeen79e83032010-07-27 11:56:07 -04002357 * ext4_map_blocks() multiple times in a loop
Eric Sandeenc445e3e2010-05-16 04:00:00 -04002358 */
2359 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2360 goto flush_it;
2361
Mingming Cao525f4ed2008-08-19 22:15:58 -04002362 /* check if thereserved journal credits might overflow */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002363 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
Mingming Cao525f4ed2008-08-19 22:15:58 -04002364 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2365 /*
2366 * With non-extent format we are limited by the journal
2367 * credit available. Total credit needed to insert
2368 * nrblocks contiguous blocks is dependent on the
2369 * nrblocks. So limit nrblocks.
2370 */
2371 goto flush_it;
2372 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2373 EXT4_MAX_TRANS_DATA) {
2374 /*
2375 * Adding the new buffer_head would make it cross the
2376 * allowed limit for which we have journal credit
2377 * reserved. So limit the new bh->b_size
2378 */
2379 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2380 mpd->inode->i_blkbits;
2381 /* we will do mpage_da_submit_io in the next loop */
2382 }
2383 }
Alex Tomas64769242008-07-11 19:27:31 -04002384 /*
2385 * First block in the extent
2386 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002387 if (mpd->b_size == 0) {
2388 mpd->b_blocknr = logical;
2389 mpd->b_size = b_size;
2390 mpd->b_state = b_state & BH_FLAGS;
Alex Tomas64769242008-07-11 19:27:31 -04002391 return;
2392 }
2393
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002394 next = mpd->b_blocknr + nrblocks;
Alex Tomas64769242008-07-11 19:27:31 -04002395 /*
2396 * Can we merge the block to our big extent?
2397 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002398 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2399 mpd->b_size += b_size;
Alex Tomas64769242008-07-11 19:27:31 -04002400 return;
2401 }
2402
Mingming Cao525f4ed2008-08-19 22:15:58 -04002403flush_it:
Alex Tomas64769242008-07-11 19:27:31 -04002404 /*
2405 * We couldn't merge the block to our extent, so we
2406 * need to flush current extent and start new one
2407 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002408 mpage_da_map_and_submit(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002409 return;
Alex Tomas64769242008-07-11 19:27:31 -04002410}
2411
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002412static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002413{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002414 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002415}
2416
Alex Tomas64769242008-07-11 19:27:31 -04002417/*
2418 * __mpage_da_writepage - finds extent of pages and blocks
2419 *
2420 * @page: page to consider
2421 * @wbc: not used, we just follow rules
2422 * @data: context
2423 *
2424 * The function finds extents of pages and scan them for all blocks.
2425 */
2426static int __mpage_da_writepage(struct page *page,
Eric Sandeenbbd08342010-10-27 21:30:12 -04002427 struct writeback_control *wbc,
2428 struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04002429{
Alex Tomas64769242008-07-11 19:27:31 -04002430 struct inode *inode = mpd->inode;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002431 struct buffer_head *bh, *head;
Alex Tomas64769242008-07-11 19:27:31 -04002432 sector_t logical;
2433
2434 /*
2435 * Can we merge this page to current extent?
2436 */
2437 if (mpd->next_page != page->index) {
2438 /*
2439 * Nope, we can't. So, we map non-allocated blocks
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002440 * and start IO on them
Alex Tomas64769242008-07-11 19:27:31 -04002441 */
2442 if (mpd->next_page != mpd->first_page) {
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002443 mpage_da_map_and_submit(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002444 /*
2445 * skip rest of the page in the page_vec
2446 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002447 redirty_page_for_writepage(wbc, page);
2448 unlock_page(page);
2449 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002450 }
2451
2452 /*
2453 * Start next extent of pages ...
2454 */
2455 mpd->first_page = page->index;
2456
2457 /*
2458 * ... and blocks
2459 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002460 mpd->b_size = 0;
2461 mpd->b_state = 0;
2462 mpd->b_blocknr = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002463 }
2464
2465 mpd->next_page = page->index + 1;
2466 logical = (sector_t) page->index <<
2467 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2468
2469 if (!page_has_buffers(page)) {
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002470 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2471 (1 << BH_Dirty) | (1 << BH_Uptodate));
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002472 if (mpd->io_done)
2473 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002474 } else {
2475 /*
2476 * Page with regular buffer heads, just add all dirty ones
2477 */
2478 head = page_buffers(page);
2479 bh = head;
2480 do {
2481 BUG_ON(buffer_locked(bh));
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002482 /*
2483 * We need to try to allocate
2484 * unmapped blocks in the same page.
2485 * Otherwise we won't make progress
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002486 * with the page in ext4_writepage
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002487 */
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002488 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002489 mpage_add_bh_to_extent(mpd, logical,
2490 bh->b_size,
2491 bh->b_state);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002492 if (mpd->io_done)
2493 return MPAGE_DA_EXTENT_TAIL;
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002494 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2495 /*
2496 * mapped dirty buffer. We need to update
2497 * the b_state because we look at
2498 * b_state in mpage_da_map_blocks. We don't
2499 * update b_size because if we find an
2500 * unmapped buffer_head later we need to
2501 * use the b_state flag of that buffer_head.
2502 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002503 if (mpd->b_size == 0)
2504 mpd->b_state = bh->b_state & BH_FLAGS;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002505 }
Alex Tomas64769242008-07-11 19:27:31 -04002506 logical++;
2507 } while ((bh = bh->b_this_page) != head);
2508 }
2509
2510 return 0;
2511}
2512
2513/*
Theodore Ts'ob920c752009-05-14 00:54:29 -04002514 * This is a special get_blocks_t callback which is used by
2515 * ext4_da_write_begin(). It will either return mapped block or
2516 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002517 *
2518 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2519 * We also have b_blocknr = -1 and b_bdev initialized properly
2520 *
2521 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2522 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2523 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04002524 */
2525static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002526 struct buffer_head *bh, int create)
Alex Tomas64769242008-07-11 19:27:31 -04002527{
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002528 struct ext4_map_blocks map;
Alex Tomas64769242008-07-11 19:27:31 -04002529 int ret = 0;
Aneesh Kumar K.V33b98172009-05-12 14:40:37 -04002530 sector_t invalid_block = ~((sector_t) 0xffff);
2531
2532 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2533 invalid_block = ~0;
Alex Tomas64769242008-07-11 19:27:31 -04002534
2535 BUG_ON(create == 0);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002536 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2537
2538 map.m_lblk = iblock;
2539 map.m_len = 1;
Alex Tomas64769242008-07-11 19:27:31 -04002540
2541 /*
2542 * first, we need to know whether the block is allocated already
2543 * preallocated blocks are unmapped but should treated
2544 * the same as allocated blocks.
2545 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002546 ret = ext4_map_blocks(NULL, inode, &map, 0);
2547 if (ret < 0)
2548 return ret;
2549 if (ret == 0) {
2550 if (buffer_delay(bh))
2551 return 0; /* Not sure this could or should happen */
Alex Tomas64769242008-07-11 19:27:31 -04002552 /*
2553 * XXX: __block_prepare_write() unmaps passed block,
2554 * is it OK?
2555 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -05002556 ret = ext4_da_reserve_space(inode, iblock);
Mingming Caod2a17632008-07-14 17:52:37 -04002557 if (ret)
2558 /* not enough space to reserve */
2559 return ret;
2560
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002561 map_bh(bh, inode->i_sb, invalid_block);
2562 set_buffer_new(bh);
2563 set_buffer_delay(bh);
2564 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04002565 }
2566
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002567 map_bh(bh, inode->i_sb, map.m_pblk);
2568 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2569
2570 if (buffer_unwritten(bh)) {
2571 /* A delayed write to unwritten bh should be marked
2572 * new and mapped. Mapped ensures that we don't do
2573 * get_block multiple times when we write to the same
2574 * offset and new ensures that we do proper zero out
2575 * for partial write.
2576 */
2577 set_buffer_new(bh);
2578 set_buffer_mapped(bh);
2579 }
2580 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04002581}
Mingming Cao61628a32008-07-11 19:27:31 -04002582
Theodore Ts'ob920c752009-05-14 00:54:29 -04002583/*
2584 * This function is used as a standard get_block_t calback function
2585 * when there is no desire to allocate any blocks. It is used as a
Christoph Hellwig206f7ab2010-06-14 14:42:49 -04002586 * callback function for block_prepare_write() and block_write_full_page().
2587 * These functions should only try to map a single block at a time.
Theodore Ts'ob920c752009-05-14 00:54:29 -04002588 *
2589 * Since this function doesn't do block allocations even if the caller
2590 * requests it by passing in create=1, it is critically important that
2591 * any caller checks to make sure that any buffer heads are returned
2592 * by this function are either all already mapped or marked for
Christoph Hellwig206f7ab2010-06-14 14:42:49 -04002593 * delayed allocation before calling block_write_full_page(). Otherwise,
2594 * b_blocknr could be left unitialized, and the page write functions will
2595 * be taken by surprise.
Theodore Ts'ob920c752009-05-14 00:54:29 -04002596 */
2597static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002598 struct buffer_head *bh_result, int create)
2599{
Theodore Ts'oa2dc52b2009-05-12 13:51:29 -04002600 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002601 return _ext4_get_block(inode, iblock, bh_result, 0);
Mingming Cao61628a32008-07-11 19:27:31 -04002602}
2603
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002604static int bget_one(handle_t *handle, struct buffer_head *bh)
2605{
2606 get_bh(bh);
2607 return 0;
2608}
2609
2610static int bput_one(handle_t *handle, struct buffer_head *bh)
2611{
2612 put_bh(bh);
2613 return 0;
2614}
2615
2616static int __ext4_journalled_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002617 unsigned int len)
2618{
2619 struct address_space *mapping = page->mapping;
2620 struct inode *inode = mapping->host;
2621 struct buffer_head *page_bufs;
2622 handle_t *handle = NULL;
2623 int ret = 0;
2624 int err;
2625
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002626 ClearPageChecked(page);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002627 page_bufs = page_buffers(page);
2628 BUG_ON(!page_bufs);
2629 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2630 /* As soon as we unlock the page, it can go away, but we have
2631 * references to buffers so we are safe */
2632 unlock_page(page);
2633
2634 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2635 if (IS_ERR(handle)) {
2636 ret = PTR_ERR(handle);
2637 goto out;
2638 }
2639
2640 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2641 do_journal_get_write_access);
2642
2643 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2644 write_end_fn);
2645 if (ret == 0)
2646 ret = err;
2647 err = ext4_journal_stop(handle);
2648 if (!ret)
2649 ret = err;
2650
2651 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002652 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002653out:
2654 return ret;
2655}
2656
Jiaying Zhang744692d2010-03-04 16:14:02 -05002657static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2658static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2659
Mingming Cao61628a32008-07-11 19:27:31 -04002660/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002661 * Note that we don't need to start a transaction unless we're journaling data
2662 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2663 * need to file the inode to the transaction's list in ordered mode because if
2664 * we are writing back data added by write(), the inode is already there and if
2665 * we are writing back data modified via mmap(), noone guarantees in which
2666 * transaction the data will hit the disk. In case we are journaling data, we
2667 * cannot start transaction directly because transaction start ranks above page
2668 * lock so we have to do some magic.
2669 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04002670 * This function can get called via...
2671 * - ext4_da_writepages after taking page lock (have journal handle)
2672 * - journal_submit_inode_data_buffers (no journal handle)
2673 * - shrink_page_list via pdflush (no journal handle)
2674 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002675 *
2676 * We don't do any block allocation in this function. If we have page with
2677 * multiple blocks we need to write those buffer_heads that are mapped. This
2678 * is important for mmaped based write. So if we do with blocksize 1K
2679 * truncate(f, 1024);
2680 * a = mmap(f, 0, 4096);
2681 * a[0] = 'a';
2682 * truncate(f, 4096);
2683 * we have in the page first buffer_head mapped via page_mkwrite call back
2684 * but other bufer_heads would be unmapped but dirty(dirty done via the
2685 * do_wp_page). So writepage should write the first block. If we modify
2686 * the mmap area beyond 1024 we will again get a page_fault and the
2687 * page_mkwrite callback will do the block allocation and mark the
2688 * buffer_heads mapped.
2689 *
2690 * We redirty the page if we have any buffer_heads that is either delay or
2691 * unwritten in the page.
2692 *
2693 * We can get recursively called as show below.
2694 *
2695 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2696 * ext4_writepage()
2697 *
2698 * But since we don't do any block allocation we should not deadlock.
2699 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04002700 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002701static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002702 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002703{
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002704 int ret = 0, commit_write = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002705 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002706 unsigned int len;
Jiaying Zhang744692d2010-03-04 16:14:02 -05002707 struct buffer_head *page_bufs = NULL;
Mingming Cao61628a32008-07-11 19:27:31 -04002708 struct inode *inode = page->mapping->host;
Alex Tomas64769242008-07-11 19:27:31 -04002709
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002710 trace_ext4_writepage(inode, page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002711 size = i_size_read(inode);
2712 if (page->index == size >> PAGE_CACHE_SHIFT)
2713 len = size & ~PAGE_CACHE_MASK;
2714 else
2715 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04002716
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002717 /*
2718 * If the page does not have buffers (for whatever reason),
2719 * try to create them using block_prepare_write. If this
2720 * fails, redirty the page and move on.
2721 */
2722 if (!page_buffers(page)) {
2723 if (block_prepare_write(page, 0, len,
2724 noalloc_get_block_write)) {
2725 redirty_page:
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002726 redirty_page_for_writepage(wbc, page);
2727 unlock_page(page);
2728 return 0;
2729 }
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002730 commit_write = 1;
2731 }
2732 page_bufs = page_buffers(page);
2733 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2734 ext4_bh_delay_or_unwritten)) {
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002735 /*
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002736 * We don't want to do block allocation So redirty the
2737 * page and return We may reach here when we do a
2738 * journal commit via
2739 * journal_submit_inode_data_buffers. If we don't
2740 * have mapping block we just ignore them. We can also
2741 * reach here via shrink_page_list
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002742 */
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002743 goto redirty_page;
2744 }
2745 if (commit_write)
Aneesh Kumar K.Ved9b3e32008-11-07 09:06:45 -05002746 /* now mark the buffer_heads as dirty and uptodate */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04002747 block_commit_write(page, 0, len);
Alex Tomas64769242008-07-11 19:27:31 -04002748
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002749 if (PageChecked(page) && ext4_should_journal_data(inode))
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002750 /*
2751 * It's mmapped pagecache. Add buffers and journal it. There
2752 * doesn't seem much point in redirtying the page here.
2753 */
Wu Fengguang3f0ca302009-11-24 11:15:44 -05002754 return __ext4_journalled_writepage(page, len);
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002755
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002756 if (buffer_uninit(page_bufs)) {
Jiaying Zhang744692d2010-03-04 16:14:02 -05002757 ext4_set_bh_endio(page_bufs, inode);
2758 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2759 wbc, ext4_end_io_buffer_write);
2760 } else
Theodore Ts'ob920c752009-05-14 00:54:29 -04002761 ret = block_write_full_page(page, noalloc_get_block_write,
2762 wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002763
Alex Tomas64769242008-07-11 19:27:31 -04002764 return ret;
2765}
2766
Mingming Cao61628a32008-07-11 19:27:31 -04002767/*
Mingming Cao525f4ed2008-08-19 22:15:58 -04002768 * This is called via ext4_da_writepages() to
2769 * calulate the total number of credits to reserve to fit
2770 * a single extent allocation into a single transaction,
2771 * ext4_da_writpeages() will loop calling this before
2772 * the block allocation.
Mingming Cao61628a32008-07-11 19:27:31 -04002773 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002774
2775static int ext4_da_writepages_trans_blocks(struct inode *inode)
2776{
2777 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2778
2779 /*
2780 * With non-extent format the journal credit needed to
2781 * insert nrblocks contiguous block is dependent on
2782 * number of contiguous block. So we will limit
2783 * number of contiguous block to a sane value
2784 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002785 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
Mingming Cao525f4ed2008-08-19 22:15:58 -04002786 (max_blocks > EXT4_MAX_TRANS_DATA))
2787 max_blocks = EXT4_MAX_TRANS_DATA;
2788
2789 return ext4_chunk_trans_blocks(inode, max_blocks);
2790}
Mingming Cao61628a32008-07-11 19:27:31 -04002791
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002792/*
2793 * write_cache_pages_da - walk the list of dirty pages of the given
2794 * address space and call the callback function (which usually writes
2795 * the pages).
2796 *
2797 * This is a forked version of write_cache_pages(). Differences:
2798 * Range cyclic is ignored.
2799 * no_nrwrite_index_update is always presumed true
2800 */
2801static int write_cache_pages_da(struct address_space *mapping,
2802 struct writeback_control *wbc,
Eric Sandeen72f84e62010-10-27 21:30:13 -04002803 struct mpage_da_data *mpd,
2804 pgoff_t *done_index)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002805{
2806 int ret = 0;
2807 int done = 0;
2808 struct pagevec pvec;
Eric Sandeen72f84e62010-10-27 21:30:13 -04002809 unsigned nr_pages;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002810 pgoff_t index;
2811 pgoff_t end; /* Inclusive */
2812 long nr_to_write = wbc->nr_to_write;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002813 int tag;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002814
2815 pagevec_init(&pvec, 0);
2816 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2817 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2818
Eric Sandeen5b41d922010-10-27 21:30:13 -04002819 if (wbc->sync_mode == WB_SYNC_ALL)
2820 tag = PAGECACHE_TAG_TOWRITE;
2821 else
2822 tag = PAGECACHE_TAG_DIRTY;
2823
Eric Sandeen72f84e62010-10-27 21:30:13 -04002824 *done_index = index;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002825 while (!done && (index <= end)) {
2826 int i;
2827
Eric Sandeen5b41d922010-10-27 21:30:13 -04002828 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002829 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2830 if (nr_pages == 0)
2831 break;
2832
2833 for (i = 0; i < nr_pages; i++) {
2834 struct page *page = pvec.pages[i];
2835
2836 /*
2837 * At this point, the page may be truncated or
2838 * invalidated (changing page->mapping to NULL), or
2839 * even swizzled back from swapper_space to tmpfs file
2840 * mapping. However, page->index will not change
2841 * because we have a reference on the page.
2842 */
2843 if (page->index > end) {
2844 done = 1;
2845 break;
2846 }
2847
Eric Sandeen72f84e62010-10-27 21:30:13 -04002848 *done_index = page->index + 1;
2849
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002850 lock_page(page);
2851
2852 /*
2853 * Page truncated or invalidated. We can freely skip it
2854 * then, even for data integrity operations: the page
2855 * has disappeared concurrently, so there could be no
2856 * real expectation of this data interity operation
2857 * even if there is now a new, dirty page at the same
2858 * pagecache address.
2859 */
2860 if (unlikely(page->mapping != mapping)) {
2861continue_unlock:
2862 unlock_page(page);
2863 continue;
2864 }
2865
2866 if (!PageDirty(page)) {
2867 /* someone wrote it for us */
2868 goto continue_unlock;
2869 }
2870
2871 if (PageWriteback(page)) {
2872 if (wbc->sync_mode != WB_SYNC_NONE)
2873 wait_on_page_writeback(page);
2874 else
2875 goto continue_unlock;
2876 }
2877
2878 BUG_ON(PageWriteback(page));
2879 if (!clear_page_dirty_for_io(page))
2880 goto continue_unlock;
2881
2882 ret = __mpage_da_writepage(page, wbc, mpd);
2883 if (unlikely(ret)) {
2884 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2885 unlock_page(page);
2886 ret = 0;
2887 } else {
2888 done = 1;
2889 break;
2890 }
2891 }
2892
2893 if (nr_to_write > 0) {
2894 nr_to_write--;
2895 if (nr_to_write == 0 &&
2896 wbc->sync_mode == WB_SYNC_NONE) {
2897 /*
2898 * We stop writing back only if we are
2899 * not doing integrity sync. In case of
2900 * integrity sync we have to keep going
2901 * because someone may be concurrently
2902 * dirtying pages, and we might have
2903 * synced a lot of newly appeared dirty
2904 * pages, but have not synced all of the
2905 * old dirty pages.
2906 */
2907 done = 1;
2908 break;
2909 }
2910 }
2911 }
2912 pagevec_release(&pvec);
2913 cond_resched();
2914 }
2915 return ret;
2916}
2917
2918
Alex Tomas64769242008-07-11 19:27:31 -04002919static int ext4_da_writepages(struct address_space *mapping,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002920 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002921{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002922 pgoff_t index;
2923 int range_whole = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002924 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002925 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002926 struct inode *inode = mapping->host;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002927 int pages_written = 0;
2928 long pages_skipped;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002929 unsigned int max_pages;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002930 int range_cyclic, cycled = 1, io_done = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002931 int needed_blocks, ret = 0;
2932 long desired_nr_to_write, nr_to_writebump = 0;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002933 loff_t range_start = wbc->range_start;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002934 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Eric Sandeen72f84e62010-10-27 21:30:13 -04002935 pgoff_t done_index = 0;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002936 pgoff_t end;
Mingming Cao61628a32008-07-11 19:27:31 -04002937
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002938 trace_ext4_da_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002939
Mingming Cao61628a32008-07-11 19:27:31 -04002940 /*
2941 * No pages to write? This is mainly a kludge to avoid starting
2942 * a transaction for special inodes like journal inode on last iput()
2943 * because that could violate lock ordering on umount
2944 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002945 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Mingming Cao61628a32008-07-11 19:27:31 -04002946 return 0;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002947
2948 /*
2949 * If the filesystem has aborted, it is read-only, so return
2950 * right away instead of dumping stack traces later on that
2951 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002952 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002953 * the latter could be true if the filesystem is mounted
2954 * read-only, and in that case, ext4_da_writepages should
2955 * *never* be called, so if that ever happens, we would want
2956 * the stack trace.
2957 */
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002958 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002959 return -EROFS;
2960
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002961 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2962 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002963
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002964 range_cyclic = wbc->range_cyclic;
2965 if (wbc->range_cyclic) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002966 index = mapping->writeback_index;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002967 if (index)
2968 cycled = 0;
2969 wbc->range_start = index << PAGE_CACHE_SHIFT;
2970 wbc->range_end = LLONG_MAX;
2971 wbc->range_cyclic = 0;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002972 end = -1;
2973 } else {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002974 index = wbc->range_start >> PAGE_CACHE_SHIFT;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002975 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2976 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002977
Theodore Ts'o55138e02009-09-29 13:31:31 -04002978 /*
2979 * This works around two forms of stupidity. The first is in
2980 * the writeback code, which caps the maximum number of pages
2981 * written to be 1024 pages. This is wrong on multiple
2982 * levels; different architectues have a different page size,
2983 * which changes the maximum amount of data which gets
2984 * written. Secondly, 4 megabytes is way too small. XFS
2985 * forces this value to be 16 megabytes by multiplying
2986 * nr_to_write parameter by four, and then relies on its
2987 * allocator to allocate larger extents to make them
2988 * contiguous. Unfortunately this brings us to the second
2989 * stupidity, which is that ext4's mballoc code only allocates
2990 * at most 2048 blocks. So we force contiguous writes up to
2991 * the number of dirty blocks in the inode, or
2992 * sbi->max_writeback_mb_bump whichever is smaller.
2993 */
2994 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
Eric Sandeenb443e732010-10-27 21:30:03 -04002995 if (!range_cyclic && range_whole) {
2996 if (wbc->nr_to_write == LONG_MAX)
2997 desired_nr_to_write = wbc->nr_to_write;
2998 else
2999 desired_nr_to_write = wbc->nr_to_write * 8;
3000 } else
Theodore Ts'o55138e02009-09-29 13:31:31 -04003001 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
3002 max_pages);
3003 if (desired_nr_to_write > max_pages)
3004 desired_nr_to_write = max_pages;
3005
3006 if (wbc->nr_to_write < desired_nr_to_write) {
3007 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
3008 wbc->nr_to_write = desired_nr_to_write;
3009 }
3010
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04003011 mpd.wbc = wbc;
3012 mpd.inode = mapping->host;
3013
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003014 pages_skipped = wbc->pages_skipped;
3015
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05003016retry:
Eric Sandeen5b41d922010-10-27 21:30:13 -04003017 if (wbc->sync_mode == WB_SYNC_ALL)
3018 tag_pages_for_writeback(mapping, index, end);
3019
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003020 while (!ret && wbc->nr_to_write > 0) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04003021
3022 /*
3023 * we insert one extent at a time. So we need
3024 * credit needed for single extent allocation.
3025 * journalled mode is currently not supported
3026 * by delalloc
3027 */
3028 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04003029 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04003030
Mingming Cao61628a32008-07-11 19:27:31 -04003031 /* start a new transaction*/
3032 handle = ext4_journal_start(inode, needed_blocks);
3033 if (IS_ERR(handle)) {
3034 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04003035 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04003036 "%ld pages, ino %lu; err %d", __func__,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04003037 wbc->nr_to_write, inode->i_ino, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04003038 goto out_writepages;
3039 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05003040
3041 /*
3042 * Now call __mpage_da_writepage to find the next
3043 * contiguous region of logical blocks that need
3044 * blocks to be allocated by ext4. We don't actually
3045 * submit the blocks for I/O here, even though
3046 * write_cache_pages thinks it will, and will set the
3047 * pages as clean for write before calling
3048 * __mpage_da_writepage().
3049 */
3050 mpd.b_size = 0;
3051 mpd.b_state = 0;
3052 mpd.b_blocknr = 0;
3053 mpd.first_page = 0;
3054 mpd.next_page = 0;
3055 mpd.io_done = 0;
3056 mpd.pages_written = 0;
3057 mpd.retval = 0;
Eric Sandeen72f84e62010-10-27 21:30:13 -04003058 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
Theodore Ts'of63e6002009-02-23 16:42:39 -05003059 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003060 * If we have a contiguous extent of pages and we
Theodore Ts'of63e6002009-02-23 16:42:39 -05003061 * haven't done the I/O yet, map the blocks and submit
3062 * them for I/O.
3063 */
3064 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04003065 mpage_da_map_and_submit(&mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05003066 ret = MPAGE_DA_EXTENT_TAIL;
3067 }
Theodore Ts'ob3a3ca82009-08-31 23:13:11 -04003068 trace_ext4_da_write_pages(inode, &mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05003069 wbc->nr_to_write -= mpd.pages_written;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04003070
Mingming Cao61628a32008-07-11 19:27:31 -04003071 ext4_journal_stop(handle);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04003072
Eric Sandeen8f64b322009-02-26 00:57:35 -05003073 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003074 /* commit the transaction which would
3075 * free blocks released in the transaction
3076 * and try again
3077 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04003078 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003079 wbc->pages_skipped = pages_skipped;
3080 ret = 0;
3081 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04003082 /*
3083 * got one extent now try with
3084 * rest of the pages
3085 */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003086 pages_written += mpd.pages_written;
3087 wbc->pages_skipped = pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04003088 ret = 0;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05003089 io_done = 1;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003090 } else if (wbc->nr_to_write)
Mingming Cao61628a32008-07-11 19:27:31 -04003091 /*
3092 * There is no more writeout needed
3093 * or we requested for a noblocking writeout
3094 * and we found the device congested
3095 */
Mingming Cao61628a32008-07-11 19:27:31 -04003096 break;
Mingming Cao61628a32008-07-11 19:27:31 -04003097 }
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05003098 if (!io_done && !cycled) {
3099 cycled = 1;
3100 index = 0;
3101 wbc->range_start = index << PAGE_CACHE_SHIFT;
3102 wbc->range_end = mapping->writeback_index - 1;
3103 goto retry;
3104 }
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003105 if (pages_skipped != wbc->pages_skipped)
Theodore Ts'o16939182009-09-26 17:43:59 -04003106 ext4_msg(inode->i_sb, KERN_CRIT,
3107 "This should not happen leaving %s "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04003108 "with nr_to_write = %ld ret = %d",
Theodore Ts'o16939182009-09-26 17:43:59 -04003109 __func__, wbc->nr_to_write, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04003110
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003111 /* Update index */
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05003112 wbc->range_cyclic = range_cyclic;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003113 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3114 /*
3115 * set the writeback_index so that range_cyclic
3116 * mode will write it back later
3117 */
Eric Sandeen72f84e62010-10-27 21:30:13 -04003118 mapping->writeback_index = done_index;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04003119
Mingming Cao61628a32008-07-11 19:27:31 -04003120out_writepages:
Richard Kennedy2faf2e12009-12-25 15:46:07 -05003121 wbc->nr_to_write -= nr_to_writebump;
Theodore Ts'ode89de62009-08-31 17:00:59 -04003122 wbc->range_start = range_start;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003123 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
Mingming Cao61628a32008-07-11 19:27:31 -04003124 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04003125}
3126
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003127#define FALL_BACK_TO_NONDELALLOC 1
3128static int ext4_nonda_switch(struct super_block *sb)
3129{
3130 s64 free_blocks, dirty_blocks;
3131 struct ext4_sb_info *sbi = EXT4_SB(sb);
3132
3133 /*
3134 * switch to non delalloc mode if we are running low
3135 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08003136 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003137 * accumulated on each CPU without updating global counters
3138 * Delalloc need an accurate free block accounting. So switch
3139 * to non delalloc when we are near to error range.
3140 */
3141 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3142 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3143 if (2 * free_blocks < 3 * dirty_blocks ||
3144 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3145 /*
Eric Sandeenc8afb442009-12-23 07:58:12 -05003146 * free block count is less than 150% of dirty blocks
3147 * or free blocks is less than watermark
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003148 */
3149 return 1;
3150 }
Eric Sandeenc8afb442009-12-23 07:58:12 -05003151 /*
3152 * Even if we don't switch but are nearing capacity,
3153 * start pushing delalloc when 1/2 of free blocks are dirty.
3154 */
3155 if (free_blocks < 2 * dirty_blocks)
3156 writeback_inodes_sb_if_idle(sb);
3157
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003158 return 0;
3159}
3160
Alex Tomas64769242008-07-11 19:27:31 -04003161static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003162 loff_t pos, unsigned len, unsigned flags,
3163 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04003164{
Eric Sandeen72b8ab92010-05-16 11:00:00 -04003165 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04003166 struct page *page;
3167 pgoff_t index;
Alex Tomas64769242008-07-11 19:27:31 -04003168 struct inode *inode = mapping->host;
3169 handle_t *handle;
3170
3171 index = pos >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003172
3173 if (ext4_nonda_switch(inode->i_sb)) {
3174 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3175 return ext4_write_begin(file, mapping, pos,
3176 len, flags, pagep, fsdata);
3177 }
3178 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003179 trace_ext4_da_write_begin(inode, pos, len, flags);
Mingming Caod2a17632008-07-14 17:52:37 -04003180retry:
Alex Tomas64769242008-07-11 19:27:31 -04003181 /*
3182 * With delayed allocation, we don't log the i_disksize update
3183 * if there is delayed block allocation. But we still need
3184 * to journalling the i_disksize update if writes to the end
3185 * of file which has an already mapped buffer.
3186 */
3187 handle = ext4_journal_start(inode, 1);
3188 if (IS_ERR(handle)) {
3189 ret = PTR_ERR(handle);
3190 goto out;
3191 }
Jan Karaebd36102009-02-22 21:09:59 -05003192 /* We cannot recurse into the filesystem as the transaction is already
3193 * started */
3194 flags |= AOP_FLAG_NOFS;
Alex Tomas64769242008-07-11 19:27:31 -04003195
Nick Piggin54566b22009-01-04 12:00:53 -08003196 page = grab_cache_page_write_begin(mapping, index, flags);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04003197 if (!page) {
3198 ext4_journal_stop(handle);
3199 ret = -ENOMEM;
3200 goto out;
3201 }
Alex Tomas64769242008-07-11 19:27:31 -04003202 *pagep = page;
3203
Christoph Hellwig6e1db882010-06-04 11:29:57 +02003204 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04003205 if (ret < 0) {
3206 unlock_page(page);
3207 ext4_journal_stop(handle);
3208 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04003209 /*
3210 * block_write_begin may have instantiated a few blocks
3211 * outside i_size. Trim these off again. Don't need
3212 * i_size_read because we hold i_mutex.
3213 */
3214 if (pos + len > inode->i_size)
Jan Karab9a42072009-12-08 21:24:33 -05003215 ext4_truncate_failed_write(inode);
Alex Tomas64769242008-07-11 19:27:31 -04003216 }
3217
Mingming Caod2a17632008-07-14 17:52:37 -04003218 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3219 goto retry;
Alex Tomas64769242008-07-11 19:27:31 -04003220out:
3221 return ret;
3222}
3223
Mingming Cao632eaea2008-07-11 19:27:31 -04003224/*
3225 * Check if we should update i_disksize
3226 * when write to the end of file but not require block allocation
3227 */
3228static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003229 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04003230{
3231 struct buffer_head *bh;
3232 struct inode *inode = page->mapping->host;
3233 unsigned int idx;
3234 int i;
3235
3236 bh = page_buffers(page);
3237 idx = offset >> inode->i_blkbits;
3238
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003239 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04003240 bh = bh->b_this_page;
3241
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04003242 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04003243 return 0;
3244 return 1;
3245}
3246
Alex Tomas64769242008-07-11 19:27:31 -04003247static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003248 struct address_space *mapping,
3249 loff_t pos, unsigned len, unsigned copied,
3250 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04003251{
3252 struct inode *inode = mapping->host;
3253 int ret = 0, ret2;
3254 handle_t *handle = ext4_journal_current_handle();
3255 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04003256 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003257 int write_mode = (int)(unsigned long)fsdata;
3258
3259 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3260 if (ext4_should_order_data(inode)) {
3261 return ext4_ordered_write_end(file, mapping, pos,
3262 len, copied, page, fsdata);
3263 } else if (ext4_should_writeback_data(inode)) {
3264 return ext4_writeback_write_end(file, mapping, pos,
3265 len, copied, page, fsdata);
3266 } else {
3267 BUG();
3268 }
3269 }
Mingming Cao632eaea2008-07-11 19:27:31 -04003270
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003271 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04003272 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003273 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04003274
3275 /*
3276 * generic_write_end() will run mark_inode_dirty() if i_size
3277 * changes. So let's piggyback the i_disksize mark_inode_dirty
3278 * into that.
3279 */
3280
3281 new_i_size = pos + copied;
Mingming Cao632eaea2008-07-11 19:27:31 -04003282 if (new_i_size > EXT4_I(inode)->i_disksize) {
3283 if (ext4_da_should_update_i_disksize(page, end)) {
3284 down_write(&EXT4_I(inode)->i_data_sem);
3285 if (new_i_size > EXT4_I(inode)->i_disksize) {
3286 /*
3287 * Updating i_disksize when extending file
3288 * without needing block allocation
3289 */
3290 if (ext4_should_order_data(inode))
3291 ret = ext4_jbd2_file_inode(handle,
3292 inode);
Alex Tomas64769242008-07-11 19:27:31 -04003293
Mingming Cao632eaea2008-07-11 19:27:31 -04003294 EXT4_I(inode)->i_disksize = new_i_size;
3295 }
3296 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04003297 /* We need to mark inode dirty even if
3298 * new_i_size is less that inode->i_size
3299 * bu greater than i_disksize.(hint delalloc)
3300 */
3301 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04003302 }
Mingming Cao632eaea2008-07-11 19:27:31 -04003303 }
Alex Tomas64769242008-07-11 19:27:31 -04003304 ret2 = generic_write_end(file, mapping, pos, len, copied,
3305 page, fsdata);
3306 copied = ret2;
3307 if (ret2 < 0)
3308 ret = ret2;
3309 ret2 = ext4_journal_stop(handle);
3310 if (!ret)
3311 ret = ret2;
3312
3313 return ret ? ret : copied;
3314}
3315
3316static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3317{
Alex Tomas64769242008-07-11 19:27:31 -04003318 /*
3319 * Drop reserved blocks
3320 */
3321 BUG_ON(!PageLocked(page));
3322 if (!page_has_buffers(page))
3323 goto out;
3324
Mingming Caod2a17632008-07-14 17:52:37 -04003325 ext4_da_page_release_reservation(page, offset);
Alex Tomas64769242008-07-11 19:27:31 -04003326
3327out:
3328 ext4_invalidatepage(page, offset);
3329
3330 return;
3331}
3332
Theodore Ts'occd25062009-02-26 01:04:07 -05003333/*
3334 * Force all delayed allocation blocks to be allocated for a given inode.
3335 */
3336int ext4_alloc_da_blocks(struct inode *inode)
3337{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04003338 trace_ext4_alloc_da_blocks(inode);
3339
Theodore Ts'occd25062009-02-26 01:04:07 -05003340 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3341 !EXT4_I(inode)->i_reserved_meta_blocks)
3342 return 0;
3343
3344 /*
3345 * We do something simple for now. The filemap_flush() will
3346 * also start triggering a write of the data blocks, which is
3347 * not strictly speaking necessary (and for users of
3348 * laptop_mode, not even desirable). However, to do otherwise
3349 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003350 *
Theodore Ts'occd25062009-02-26 01:04:07 -05003351 * ext4_da_writepages() ->
3352 * write_cache_pages() ---> (via passed in callback function)
3353 * __mpage_da_writepage() -->
3354 * mpage_add_bh_to_extent()
3355 * mpage_da_map_blocks()
3356 *
3357 * The problem is that write_cache_pages(), located in
3358 * mm/page-writeback.c, marks pages clean in preparation for
3359 * doing I/O, which is not desirable if we're not planning on
3360 * doing I/O at all.
3361 *
3362 * We could call write_cache_pages(), and then redirty all of
3363 * the pages by calling redirty_page_for_writeback() but that
3364 * would be ugly in the extreme. So instead we would need to
3365 * replicate parts of the code in the above functions,
3366 * simplifying them becuase we wouldn't actually intend to
3367 * write out the pages, but rather only collect contiguous
3368 * logical block extents, call the multi-block allocator, and
3369 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003370 *
Theodore Ts'occd25062009-02-26 01:04:07 -05003371 * For now, though, we'll cheat by calling filemap_flush(),
3372 * which will map the blocks, and start the I/O, but not
3373 * actually wait for the I/O to complete.
3374 */
3375 return filemap_flush(inode->i_mapping);
3376}
Alex Tomas64769242008-07-11 19:27:31 -04003377
3378/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003379 * bmap() is special. It gets used by applications such as lilo and by
3380 * the swapper to find the on-disk block of a specific piece of data.
3381 *
3382 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07003383 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003384 * filesystem and enables swap, then they may get a nasty shock when the
3385 * data getting swapped to that swapfile suddenly gets overwritten by
3386 * the original zero's written out previously to the journal and
3387 * awaiting writeback in the kernel's buffer cache.
3388 *
3389 * So, if we see any bmap calls here on a modified, data-journaled file,
3390 * take extra steps to flush any blocks which might be in the cache.
3391 */
Mingming Cao617ba132006-10-11 01:20:53 -07003392static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003393{
3394 struct inode *inode = mapping->host;
3395 journal_t *journal;
3396 int err;
3397
Alex Tomas64769242008-07-11 19:27:31 -04003398 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3399 test_opt(inode->i_sb, DELALLOC)) {
3400 /*
3401 * With delalloc we want to sync the file
3402 * so that we can make sure we allocate
3403 * blocks for file
3404 */
3405 filemap_write_and_wait(mapping);
3406 }
3407
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003408 if (EXT4_JOURNAL(inode) &&
3409 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003410 /*
3411 * This is a REALLY heavyweight approach, but the use of
3412 * bmap on dirty files is expected to be extremely rare:
3413 * only if we run lilo or swapon on a freshly made file
3414 * do we expect this to happen.
3415 *
3416 * (bmap requires CAP_SYS_RAWIO so this does not
3417 * represent an unprivileged user DOS attack --- we'd be
3418 * in trouble if mortal users could trigger this path at
3419 * will.)
3420 *
Mingming Cao617ba132006-10-11 01:20:53 -07003421 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003422 * regular files. If somebody wants to bmap a directory
3423 * or symlink and gets confused because the buffer
3424 * hasn't yet been flushed to disk, they deserve
3425 * everything they get.
3426 */
3427
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003428 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
Mingming Cao617ba132006-10-11 01:20:53 -07003429 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07003430 jbd2_journal_lock_updates(journal);
3431 err = jbd2_journal_flush(journal);
3432 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003433
3434 if (err)
3435 return 0;
3436 }
3437
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003438 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003439}
3440
Mingming Cao617ba132006-10-11 01:20:53 -07003441static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003442{
Mingming Cao617ba132006-10-11 01:20:53 -07003443 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003444}
3445
3446static int
Mingming Cao617ba132006-10-11 01:20:53 -07003447ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003448 struct list_head *pages, unsigned nr_pages)
3449{
Mingming Cao617ba132006-10-11 01:20:53 -07003450 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003451}
3452
Jiaying Zhang744692d2010-03-04 16:14:02 -05003453static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3454{
3455 struct buffer_head *head, *bh;
3456 unsigned int curr_off = 0;
3457
3458 if (!page_has_buffers(page))
3459 return;
3460 head = bh = page_buffers(page);
3461 do {
3462 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3463 && bh->b_private) {
3464 ext4_free_io_end(bh->b_private);
3465 bh->b_private = NULL;
3466 bh->b_end_io = NULL;
3467 }
3468 curr_off = curr_off + bh->b_size;
3469 bh = bh->b_this_page;
3470 } while (bh != head);
3471}
3472
Mingming Cao617ba132006-10-11 01:20:53 -07003473static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003474{
Mingming Cao617ba132006-10-11 01:20:53 -07003475 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003476
3477 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05003478 * free any io_end structure allocated for buffers to be discarded
3479 */
3480 if (ext4_should_dioread_nolock(page->mapping->host))
3481 ext4_invalidatepage_free_endio(page, offset);
3482 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003483 * If it's a full truncate we just forget about the pending dirtying
3484 */
3485 if (offset == 0)
3486 ClearPageChecked(page);
3487
Frank Mayhar03901312009-01-07 00:06:22 -05003488 if (journal)
3489 jbd2_journal_invalidatepage(journal, page, offset);
3490 else
3491 block_invalidatepage(page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003492}
3493
Mingming Cao617ba132006-10-11 01:20:53 -07003494static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003495{
Mingming Cao617ba132006-10-11 01:20:53 -07003496 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003497
3498 WARN_ON(PageChecked(page));
3499 if (!page_has_buffers(page))
3500 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05003501 if (journal)
3502 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3503 else
3504 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003505}
3506
3507/*
Mingming Cao4c0425f2009-09-28 15:48:41 -04003508 * O_DIRECT for ext3 (or indirect map) based files
3509 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003510 * If the O_DIRECT write will extend the file then add this inode to the
3511 * orphan list. So recovery will truncate it back to the original size
3512 * if the machine crashes during the write.
3513 *
3514 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Kara7fb54092008-02-10 01:08:38 -05003515 * crashes then stale disk data _may_ be exposed inside the file. But current
3516 * VFS code falls back into buffered path in that case so we are safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003517 */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003518static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003519 const struct iovec *iov, loff_t offset,
3520 unsigned long nr_segs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003521{
3522 struct file *file = iocb->ki_filp;
3523 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07003524 struct ext4_inode_info *ei = EXT4_I(inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003525 handle_t *handle;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003526 ssize_t ret;
3527 int orphan = 0;
3528 size_t count = iov_length(iov, nr_segs);
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003529 int retries = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003530
3531 if (rw == WRITE) {
3532 loff_t final_size = offset + count;
3533
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003534 if (final_size > inode->i_size) {
Jan Kara7fb54092008-02-10 01:08:38 -05003535 /* Credits for sb + inode write */
3536 handle = ext4_journal_start(inode, 2);
3537 if (IS_ERR(handle)) {
3538 ret = PTR_ERR(handle);
3539 goto out;
3540 }
Mingming Cao617ba132006-10-11 01:20:53 -07003541 ret = ext4_orphan_add(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003542 if (ret) {
3543 ext4_journal_stop(handle);
3544 goto out;
3545 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003546 orphan = 1;
3547 ei->i_disksize = inode->i_size;
Jan Kara7fb54092008-02-10 01:08:38 -05003548 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003549 }
3550 }
3551
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003552retry:
Jiaying Zhangb7adc1f2010-03-02 13:26:36 -05003553 if (rw == READ && ext4_should_dioread_nolock(inode))
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02003554 ret = __blockdev_direct_IO(rw, iocb, inode,
Jiaying Zhangb7adc1f2010-03-02 13:26:36 -05003555 inode->i_sb->s_bdev, iov,
3556 offset, nr_segs,
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02003557 ext4_get_block, NULL, NULL, 0);
3558 else {
Jiaying Zhangb7adc1f2010-03-02 13:26:36 -05003559 ret = blockdev_direct_IO(rw, iocb, inode,
3560 inode->i_sb->s_bdev, iov,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003561 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07003562 ext4_get_block, NULL);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02003563
3564 if (unlikely((rw & WRITE) && ret < 0)) {
3565 loff_t isize = i_size_read(inode);
3566 loff_t end = offset + iov_length(iov, nr_segs);
3567
3568 if (end > isize)
3569 vmtruncate(inode, isize);
3570 }
3571 }
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003572 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3573 goto retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003574
Jan Kara7fb54092008-02-10 01:08:38 -05003575 if (orphan) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003576 int err;
3577
Jan Kara7fb54092008-02-10 01:08:38 -05003578 /* Credits for sb + inode write */
3579 handle = ext4_journal_start(inode, 2);
3580 if (IS_ERR(handle)) {
3581 /* This is really bad luck. We've written the data
3582 * but cannot extend i_size. Bail out and pretend
3583 * the write failed... */
3584 ret = PTR_ERR(handle);
Dmitry Monakhovda1dafc2010-03-01 23:15:02 -05003585 if (inode->i_nlink)
3586 ext4_orphan_del(NULL, inode);
3587
Jan Kara7fb54092008-02-10 01:08:38 -05003588 goto out;
3589 }
3590 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003591 ext4_orphan_del(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003592 if (ret > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003593 loff_t end = offset + ret;
3594 if (end > inode->i_size) {
3595 ei->i_disksize = end;
3596 i_size_write(inode, end);
3597 /*
3598 * We're going to return a positive `ret'
3599 * here due to non-zero-length I/O, so there's
3600 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07003601 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003602 * ignore it.
3603 */
Mingming Cao617ba132006-10-11 01:20:53 -07003604 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003605 }
3606 }
Mingming Cao617ba132006-10-11 01:20:53 -07003607 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003608 if (ret == 0)
3609 ret = err;
3610 }
3611out:
3612 return ret;
3613}
3614
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003615/*
3616 * ext4_get_block used when preparing for a DIO write or buffer write.
3617 * We allocate an uinitialized extent if blocks haven't been allocated.
3618 * The extent will be converted to initialized after the IO is complete.
3619 */
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003620static int ext4_get_block_write(struct inode *inode, sector_t iblock,
Mingming Cao4c0425f2009-09-28 15:48:41 -04003621 struct buffer_head *bh_result, int create)
3622{
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003623 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003624 inode->i_ino, create);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003625 return _ext4_get_block(inode, iblock, bh_result,
3626 EXT4_GET_BLOCKS_IO_CREATE_EXT);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003627}
3628
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003629static void dump_completed_IO(struct inode * inode)
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003630{
3631#ifdef EXT4_DEBUG
3632 struct list_head *cur, *before, *after;
3633 ext4_io_end_t *io, *io0, *io1;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003634 unsigned long flags;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003635
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003636 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3637 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003638 return;
3639 }
3640
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003641 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003642 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003643 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003644 cur = &io->list;
3645 before = cur->prev;
3646 io0 = container_of(before, ext4_io_end_t, list);
3647 after = cur->next;
3648 io1 = container_of(after, ext4_io_end_t, list);
3649
3650 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3651 io, inode->i_ino, io0, io1);
3652 }
Jiaying Zhang744692d2010-03-04 16:14:02 -05003653 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003654#endif
3655}
Mingming Cao4c0425f2009-09-28 15:48:41 -04003656
3657/*
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003658 * This function is called from ext4_sync_file().
3659 *
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003660 * When IO is completed, the work to convert unwritten extents to
3661 * written is queued on workqueue but may not get immediately
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003662 * scheduled. When fsync is called, we need to ensure the
3663 * conversion is complete before fsync returns.
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003664 * The inode keeps track of a list of pending/completed IO that
3665 * might needs to do the conversion. This function walks through
3666 * the list and convert the related unwritten extents for completed IO
3667 * to written.
3668 * The function return the number of pending IOs on success.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003669 */
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003670int flush_completed_IO(struct inode *inode)
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003671{
3672 ext4_io_end_t *io;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003673 struct ext4_inode_info *ei = EXT4_I(inode);
3674 unsigned long flags;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003675 int ret = 0;
3676 int ret2 = 0;
3677
Jiaying Zhang744692d2010-03-04 16:14:02 -05003678 if (list_empty(&ei->i_completed_io_list))
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003679 return ret;
3680
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003681 dump_completed_IO(inode);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003682 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3683 while (!list_empty(&ei->i_completed_io_list)){
3684 io = list_entry(ei->i_completed_io_list.next,
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003685 ext4_io_end_t, list);
3686 /*
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003687 * Calling ext4_end_io_nolock() to convert completed
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003688 * IO to written.
3689 *
3690 * When ext4_sync_file() is called, run_queue() may already
3691 * about to flush the work corresponding to this io structure.
3692 * It will be upset if it founds the io structure related
3693 * to the work-to-be schedule is freed.
3694 *
3695 * Thus we need to keep the io structure still valid here after
3696 * convertion finished. The io structure has a flag to
3697 * avoid double converting from both fsync and background work
3698 * queue work.
3699 */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003700 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003701 ret = ext4_end_io_nolock(io);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003702 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003703 if (ret < 0)
3704 ret2 = ret;
3705 else
3706 list_del_init(&io->list);
3707 }
Jiaying Zhang744692d2010-03-04 16:14:02 -05003708 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003709 return (ret2 < 0) ? ret2 : 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003710}
3711
Mingming Cao4c0425f2009-09-28 15:48:41 -04003712static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
Christoph Hellwig552ef8022010-07-27 11:56:06 -04003713 ssize_t size, void *private, int ret,
3714 bool is_async)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003715{
3716 ext4_io_end_t *io_end = iocb->private;
3717 struct workqueue_struct *wq;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003718 unsigned long flags;
3719 struct ext4_inode_info *ei;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003720
Mingming4b70df12009-11-03 14:44:54 -05003721 /* if not async direct IO or dio with 0 bytes write, just return */
3722 if (!io_end || !size)
Christoph Hellwig552ef8022010-07-27 11:56:06 -04003723 goto out;
Mingming4b70df12009-11-03 14:44:54 -05003724
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003725 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3726 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3727 iocb->private, io_end->inode->i_ino, iocb, offset,
3728 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003729
3730 /* if not aio dio with unwritten extents, just free io and return */
Theodore Ts'obd2d0212010-10-27 21:30:10 -04003731 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003732 ext4_free_io_end(io_end);
3733 iocb->private = NULL;
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04003734out:
3735 if (is_async)
3736 aio_complete(iocb, ret, 0);
3737 return;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003738 }
3739
Mingming Cao4c0425f2009-09-28 15:48:41 -04003740 io_end->offset = offset;
3741 io_end->size = size;
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04003742 if (is_async) {
3743 io_end->iocb = iocb;
3744 io_end->result = ret;
3745 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04003746 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3747
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003748 /* Add the io_end to per-inode completed aio dio list*/
Jiaying Zhang744692d2010-03-04 16:14:02 -05003749 ei = EXT4_I(io_end->inode);
3750 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3751 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3752 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
Eric Sandeenc999af22010-10-27 21:30:07 -04003753
3754 /* queue the work to convert unwritten extents to written */
3755 queue_work(wq, &io_end->work);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003756 iocb->private = NULL;
3757}
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003758
Jiaying Zhang744692d2010-03-04 16:14:02 -05003759static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3760{
3761 ext4_io_end_t *io_end = bh->b_private;
3762 struct workqueue_struct *wq;
3763 struct inode *inode;
3764 unsigned long flags;
3765
3766 if (!test_clear_buffer_uninit(bh) || !io_end)
3767 goto out;
3768
3769 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3770 printk("sb umounted, discard end_io request for inode %lu\n",
3771 io_end->inode->i_ino);
3772 ext4_free_io_end(io_end);
3773 goto out;
3774 }
3775
Theodore Ts'obd2d0212010-10-27 21:30:10 -04003776 io_end->flag = EXT4_IO_END_UNWRITTEN;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003777 inode = io_end->inode;
3778
3779 /* Add the io_end to per-inode completed io list*/
3780 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3781 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3782 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3783
3784 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3785 /* queue the work to convert unwritten extents to written */
3786 queue_work(wq, &io_end->work);
3787out:
3788 bh->b_private = NULL;
3789 bh->b_end_io = NULL;
3790 clear_buffer_uninit(bh);
3791 end_buffer_async_write(bh, uptodate);
3792}
3793
3794static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3795{
3796 ext4_io_end_t *io_end;
3797 struct page *page = bh->b_page;
3798 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3799 size_t size = bh->b_size;
3800
3801retry:
3802 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3803 if (!io_end) {
3804 if (printk_ratelimit())
3805 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3806 schedule();
3807 goto retry;
3808 }
3809 io_end->offset = offset;
3810 io_end->size = size;
3811 /*
3812 * We need to hold a reference to the page to make sure it
3813 * doesn't get evicted before ext4_end_io_work() has a chance
3814 * to convert the extent from written to unwritten.
3815 */
3816 io_end->page = page;
3817 get_page(io_end->page);
3818
3819 bh->b_private = io_end;
3820 bh->b_end_io = ext4_end_io_buffer_write;
3821 return 0;
3822}
3823
Mingming Cao4c0425f2009-09-28 15:48:41 -04003824/*
3825 * For ext4 extent files, ext4 will do direct-io write to holes,
3826 * preallocated extents, and those write extend the file, no need to
3827 * fall back to buffered IO.
3828 *
3829 * For holes, we fallocate those blocks, mark them as unintialized
3830 * If those blocks were preallocated, we mark sure they are splited, but
3831 * still keep the range to write as unintialized.
3832 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003833 * The unwrritten extents will be converted to written when DIO is completed.
3834 * For async direct IO, since the IO may still pending when return, we
3835 * set up an end_io call back function, which will do the convertion
3836 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003837 *
3838 * If the O_DIRECT write will extend the file then add this inode to the
3839 * orphan list. So recovery will truncate it back to the original size
3840 * if the machine crashes during the write.
3841 *
3842 */
3843static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3844 const struct iovec *iov, loff_t offset,
3845 unsigned long nr_segs)
3846{
3847 struct file *file = iocb->ki_filp;
3848 struct inode *inode = file->f_mapping->host;
3849 ssize_t ret;
3850 size_t count = iov_length(iov, nr_segs);
3851
3852 loff_t final_size = offset + count;
3853 if (rw == WRITE && final_size <= inode->i_size) {
3854 /*
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003855 * We could direct write to holes and fallocate.
3856 *
3857 * Allocated blocks to fill the hole are marked as uninitialized
Mingming Cao4c0425f2009-09-28 15:48:41 -04003858 * to prevent paralel buffered read to expose the stale data
3859 * before DIO complete the data IO.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003860 *
3861 * As to previously fallocated extents, ext4 get_block
Mingming Cao4c0425f2009-09-28 15:48:41 -04003862 * will just simply mark the buffer mapped but still
3863 * keep the extents uninitialized.
3864 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003865 * for non AIO case, we will convert those unwritten extents
3866 * to written after return back from blockdev_direct_IO.
3867 *
3868 * for async DIO, the conversion needs to be defered when
3869 * the IO is completed. The ext4 end_io callback function
3870 * will be called to take care of the conversion work.
3871 * Here for async case, we allocate an io_end structure to
3872 * hook to the iocb.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003873 */
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003874 iocb->private = NULL;
3875 EXT4_I(inode)->cur_aio_dio = NULL;
3876 if (!is_sync_kiocb(iocb)) {
Jiaying Zhang744692d2010-03-04 16:14:02 -05003877 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003878 if (!iocb->private)
3879 return -ENOMEM;
3880 /*
3881 * we save the io structure for current async
Eric Sandeen79e83032010-07-27 11:56:07 -04003882 * direct IO, so that later ext4_map_blocks()
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003883 * could flag the io structure whether there
3884 * is a unwritten extents needs to be converted
3885 * when IO is completed.
3886 */
3887 EXT4_I(inode)->cur_aio_dio = iocb->private;
3888 }
3889
Mingming Cao4c0425f2009-09-28 15:48:41 -04003890 ret = blockdev_direct_IO(rw, iocb, inode,
3891 inode->i_sb->s_bdev, iov,
3892 offset, nr_segs,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003893 ext4_get_block_write,
Mingming Cao4c0425f2009-09-28 15:48:41 -04003894 ext4_end_io_dio);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003895 if (iocb->private)
3896 EXT4_I(inode)->cur_aio_dio = NULL;
3897 /*
3898 * The io_end structure takes a reference to the inode,
3899 * that structure needs to be destroyed and the
3900 * reference to the inode need to be dropped, when IO is
3901 * complete, even with 0 byte write, or failed.
3902 *
3903 * In the successful AIO DIO case, the io_end structure will be
3904 * desctroyed and the reference to the inode will be dropped
3905 * after the end_io call back function is called.
3906 *
3907 * In the case there is 0 byte write, or error case, since
3908 * VFS direct IO won't invoke the end_io call back function,
3909 * we need to free the end_io structure here.
3910 */
3911 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3912 ext4_free_io_end(iocb->private);
3913 iocb->private = NULL;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003914 } else if (ret > 0 && ext4_test_inode_state(inode,
3915 EXT4_STATE_DIO_UNWRITTEN)) {
Mingming109f5562009-11-10 10:48:08 -05003916 int err;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003917 /*
3918 * for non AIO case, since the IO is already
3919 * completed, we could do the convertion right here
3920 */
Mingming109f5562009-11-10 10:48:08 -05003921 err = ext4_convert_unwritten_extents(inode,
3922 offset, ret);
3923 if (err < 0)
3924 ret = err;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003925 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Mingming109f5562009-11-10 10:48:08 -05003926 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04003927 return ret;
3928 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003929
3930 /* for write the the end of file case, we fall back to old way */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003931 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3932}
3933
3934static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3935 const struct iovec *iov, loff_t offset,
3936 unsigned long nr_segs)
3937{
3938 struct file *file = iocb->ki_filp;
3939 struct inode *inode = file->f_mapping->host;
3940
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003941 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Mingming Cao4c0425f2009-09-28 15:48:41 -04003942 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3943
3944 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3945}
3946
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003947/*
Mingming Cao617ba132006-10-11 01:20:53 -07003948 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003949 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3950 * much here because ->set_page_dirty is called under VFS locks. The page is
3951 * not necessarily locked.
3952 *
3953 * We cannot just dirty the page and leave attached buffers clean, because the
3954 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3955 * or jbddirty because all the journalling code will explode.
3956 *
3957 * So what we do is to mark the page "pending dirty" and next time writepage
3958 * is called, propagate that into the buffers appropriately.
3959 */
Mingming Cao617ba132006-10-11 01:20:53 -07003960static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003961{
3962 SetPageChecked(page);
3963 return __set_page_dirty_nobuffers(page);
3964}
3965
Mingming Cao617ba132006-10-11 01:20:53 -07003966static const struct address_space_operations ext4_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003967 .readpage = ext4_readpage,
3968 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003969 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003970 .sync_page = block_sync_page,
3971 .write_begin = ext4_write_begin,
3972 .write_end = ext4_ordered_write_end,
3973 .bmap = ext4_bmap,
3974 .invalidatepage = ext4_invalidatepage,
3975 .releasepage = ext4_releasepage,
3976 .direct_IO = ext4_direct_IO,
3977 .migratepage = buffer_migrate_page,
3978 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003979 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003980};
3981
Mingming Cao617ba132006-10-11 01:20:53 -07003982static const struct address_space_operations ext4_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003983 .readpage = ext4_readpage,
3984 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003985 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003986 .sync_page = block_sync_page,
3987 .write_begin = ext4_write_begin,
3988 .write_end = ext4_writeback_write_end,
3989 .bmap = ext4_bmap,
3990 .invalidatepage = ext4_invalidatepage,
3991 .releasepage = ext4_releasepage,
3992 .direct_IO = ext4_direct_IO,
3993 .migratepage = buffer_migrate_page,
3994 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003995 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003996};
3997
Mingming Cao617ba132006-10-11 01:20:53 -07003998static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003999 .readpage = ext4_readpage,
4000 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04004001 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07004002 .sync_page = block_sync_page,
4003 .write_begin = ext4_write_begin,
4004 .write_end = ext4_journalled_write_end,
4005 .set_page_dirty = ext4_journalled_set_page_dirty,
4006 .bmap = ext4_bmap,
4007 .invalidatepage = ext4_invalidatepage,
4008 .releasepage = ext4_releasepage,
4009 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02004010 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004011};
4012
Alex Tomas64769242008-07-11 19:27:31 -04004013static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07004014 .readpage = ext4_readpage,
4015 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04004016 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07004017 .writepages = ext4_da_writepages,
4018 .sync_page = block_sync_page,
4019 .write_begin = ext4_da_write_begin,
4020 .write_end = ext4_da_write_end,
4021 .bmap = ext4_bmap,
4022 .invalidatepage = ext4_da_invalidatepage,
4023 .releasepage = ext4_releasepage,
4024 .direct_IO = ext4_direct_IO,
4025 .migratepage = buffer_migrate_page,
4026 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02004027 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04004028};
4029
Mingming Cao617ba132006-10-11 01:20:53 -07004030void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004031{
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04004032 if (ext4_should_order_data(inode) &&
4033 test_opt(inode->i_sb, DELALLOC))
4034 inode->i_mapping->a_ops = &ext4_da_aops;
4035 else if (ext4_should_order_data(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07004036 inode->i_mapping->a_ops = &ext4_ordered_aops;
Alex Tomas64769242008-07-11 19:27:31 -04004037 else if (ext4_should_writeback_data(inode) &&
4038 test_opt(inode->i_sb, DELALLOC))
4039 inode->i_mapping->a_ops = &ext4_da_aops;
Mingming Cao617ba132006-10-11 01:20:53 -07004040 else if (ext4_should_writeback_data(inode))
4041 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004042 else
Mingming Cao617ba132006-10-11 01:20:53 -07004043 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004044}
4045
4046/*
Mingming Cao617ba132006-10-11 01:20:53 -07004047 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004048 * up to the end of the block which corresponds to `from'.
4049 * This required during truncate. We need to physically zero the tail end
4050 * of that block so it doesn't yield old data if the file is later grown.
4051 */
Jan Karacf108bc2008-07-11 19:27:31 -04004052int ext4_block_truncate_page(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004053 struct address_space *mapping, loff_t from)
4054{
Mingming Cao617ba132006-10-11 01:20:53 -07004055 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004056 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004057 unsigned blocksize, length, pos;
4058 ext4_lblk_t iblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004059 struct inode *inode = mapping->host;
4060 struct buffer_head *bh;
Jan Karacf108bc2008-07-11 19:27:31 -04004061 struct page *page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004062 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004063
Theodore Ts'of4a01012009-07-05 22:08:16 -04004064 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4065 mapping_gfp_mask(mapping) & ~__GFP_FS);
Jan Karacf108bc2008-07-11 19:27:31 -04004066 if (!page)
4067 return -EINVAL;
4068
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004069 blocksize = inode->i_sb->s_blocksize;
4070 length = blocksize - (offset & (blocksize - 1));
4071 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4072
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004073 if (!page_has_buffers(page))
4074 create_empty_buffers(page, blocksize, 0);
4075
4076 /* Find the buffer that contains "offset" */
4077 bh = page_buffers(page);
4078 pos = blocksize;
4079 while (offset >= pos) {
4080 bh = bh->b_this_page;
4081 iblock++;
4082 pos += blocksize;
4083 }
4084
4085 err = 0;
4086 if (buffer_freed(bh)) {
4087 BUFFER_TRACE(bh, "freed: skip");
4088 goto unlock;
4089 }
4090
4091 if (!buffer_mapped(bh)) {
4092 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07004093 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004094 /* unmapped? It's a hole - nothing to do */
4095 if (!buffer_mapped(bh)) {
4096 BUFFER_TRACE(bh, "still unmapped");
4097 goto unlock;
4098 }
4099 }
4100
4101 /* Ok, it's mapped. Make sure it's up-to-date */
4102 if (PageUptodate(page))
4103 set_buffer_uptodate(bh);
4104
4105 if (!buffer_uptodate(bh)) {
4106 err = -EIO;
4107 ll_rw_block(READ, 1, &bh);
4108 wait_on_buffer(bh);
4109 /* Uhhuh. Read error. Complain and punt. */
4110 if (!buffer_uptodate(bh))
4111 goto unlock;
4112 }
4113
Mingming Cao617ba132006-10-11 01:20:53 -07004114 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004115 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07004116 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004117 if (err)
4118 goto unlock;
4119 }
4120
Christoph Lametereebd2aa2008-02-04 22:28:29 -08004121 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004122
4123 BUFFER_TRACE(bh, "zeroed end of block");
4124
4125 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07004126 if (ext4_should_journal_data(inode)) {
Frank Mayhar03901312009-01-07 00:06:22 -05004127 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004128 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004129 if (ext4_should_order_data(inode))
Jan Kara678aaf42008-07-11 19:27:31 -04004130 err = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004131 mark_buffer_dirty(bh);
4132 }
4133
4134unlock:
4135 unlock_page(page);
4136 page_cache_release(page);
4137 return err;
4138}
4139
4140/*
4141 * Probably it should be a library function... search for first non-zero word
4142 * or memcmp with zero_page, whatever is better for particular architecture.
4143 * Linus?
4144 */
4145static inline int all_zeroes(__le32 *p, __le32 *q)
4146{
4147 while (p < q)
4148 if (*p++)
4149 return 0;
4150 return 1;
4151}
4152
4153/**
Mingming Cao617ba132006-10-11 01:20:53 -07004154 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004155 * @inode: inode in question
4156 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07004157 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004158 * @chain: place to store the pointers to partial indirect blocks
4159 * @top: place to the (detached) top of branch
4160 *
Mingming Cao617ba132006-10-11 01:20:53 -07004161 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004162 *
4163 * When we do truncate() we may have to clean the ends of several
4164 * indirect blocks but leave the blocks themselves alive. Block is
4165 * partially truncated if some data below the new i_size is refered
4166 * from it (and it is on the path to the first completely truncated
4167 * data block, indeed). We have to free the top of that path along
4168 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07004169 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004170 * finishes, we may safely do the latter, but top of branch may
4171 * require special attention - pageout below the truncation point
4172 * might try to populate it.
4173 *
4174 * We atomically detach the top of branch from the tree, store the
4175 * block number of its root in *@top, pointers to buffer_heads of
4176 * partially truncated blocks - in @chain[].bh and pointers to
4177 * their last elements that should not be removed - in
4178 * @chain[].p. Return value is the pointer to last filled element
4179 * of @chain.
4180 *
4181 * The work left to caller to do the actual freeing of subtrees:
4182 * a) free the subtree starting from *@top
4183 * b) free the subtrees whose roots are stored in
4184 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4185 * c) free the subtrees growing from the inode past the @chain[0].
4186 * (no partially truncated stuff there). */
4187
Mingming Cao617ba132006-10-11 01:20:53 -07004188static Indirect *ext4_find_shared(struct inode *inode, int depth,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004189 ext4_lblk_t offsets[4], Indirect chain[4],
4190 __le32 *top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004191{
4192 Indirect *partial, *p;
4193 int k, err;
4194
4195 *top = 0;
Uwe Kleine-Königbf48aab2009-10-28 20:11:03 +01004196 /* Make k index the deepest non-null offset + 1 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004197 for (k = depth; k > 1 && !offsets[k-1]; k--)
4198 ;
Mingming Cao617ba132006-10-11 01:20:53 -07004199 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004200 /* Writer: pointers */
4201 if (!partial)
4202 partial = chain + k-1;
4203 /*
4204 * If the branch acquired continuation since we've looked at it -
4205 * fine, it should all survive and (new) top doesn't belong to us.
4206 */
4207 if (!partial->key && *partial->p)
4208 /* Writer: end */
4209 goto no_top;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004210 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004211 ;
4212 /*
4213 * OK, we've found the last block that must survive. The rest of our
4214 * branch should be detached before unlocking. However, if that rest
4215 * of branch is all ours and does not grow immediately from the inode
4216 * it's easier to cheat and just decrement partial->p.
4217 */
4218 if (p == chain + k - 1 && p > chain) {
4219 p->p--;
4220 } else {
4221 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07004222 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004223#if 0
4224 *p->p = 0;
4225#endif
4226 }
4227 /* Writer: end */
4228
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004229 while (partial > p) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004230 brelse(partial->bh);
4231 partial--;
4232 }
4233no_top:
4234 return partial;
4235}
4236
4237/*
4238 * Zero a number of block pointers in either an inode or an indirect block.
4239 * If we restart the transaction we must again get write access to the
4240 * indirect block for further modification.
4241 *
4242 * We release `count' blocks on disk, but (last - first) may be greater
4243 * than `count' because there can be holes in there.
4244 */
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004245static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4246 struct buffer_head *bh,
4247 ext4_fsblk_t block_to_free,
4248 unsigned long count, __le32 *first,
4249 __le32 *last)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004250{
4251 __le32 *p;
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004252 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
Theodore Ts'oe6362602009-11-23 07:17:05 -05004253
4254 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4255 flags |= EXT4_FREE_BLOCKS_METADATA;
Theodore Ts'o50689692009-11-23 07:17:34 -05004256
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004257 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4258 count)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04004259 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4260 "blocks %llu len %lu",
4261 (unsigned long long) block_to_free, count);
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004262 return 1;
4263 }
4264
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004265 if (try_to_extend_transaction(handle, inode)) {
4266 if (bh) {
Frank Mayhar03901312009-01-07 00:06:22 -05004267 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4268 ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004269 }
Mingming Cao617ba132006-10-11 01:20:53 -07004270 ext4_mark_inode_dirty(handle, inode);
Jan Kara487caee2009-08-17 22:17:20 -04004271 ext4_truncate_restart_trans(handle, inode,
4272 blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004273 if (bh) {
4274 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07004275 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004276 }
4277 }
4278
Theodore Ts'oe6362602009-11-23 07:17:05 -05004279 for (p = first; p < last; p++)
4280 *p = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004281
Theodore Ts'oe6362602009-11-23 07:17:05 -05004282 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004283 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004284}
4285
4286/**
Mingming Cao617ba132006-10-11 01:20:53 -07004287 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004288 * @handle: handle for this transaction
4289 * @inode: inode we are dealing with
4290 * @this_bh: indirect buffer_head which contains *@first and *@last
4291 * @first: array of block numbers
4292 * @last: points immediately past the end of array
4293 *
4294 * We are freeing all blocks refered from that array (numbers are stored as
4295 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4296 *
4297 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4298 * blocks are contiguous then releasing them at one time will only affect one
4299 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4300 * actually use a lot of journal space.
4301 *
4302 * @this_bh will be %NULL if @first and @last point into the inode's direct
4303 * block pointers.
4304 */
Mingming Cao617ba132006-10-11 01:20:53 -07004305static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004306 struct buffer_head *this_bh,
4307 __le32 *first, __le32 *last)
4308{
Mingming Cao617ba132006-10-11 01:20:53 -07004309 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004310 unsigned long count = 0; /* Number of blocks in the run */
4311 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4312 corresponding to
4313 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07004314 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004315 __le32 *p; /* Pointer into inode/ind
4316 for current block */
4317 int err;
4318
4319 if (this_bh) { /* For indirect block */
4320 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004321 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004322 /* Important: if we can't update the indirect pointers
4323 * to the blocks, we can't free them. */
4324 if (err)
4325 return;
4326 }
4327
4328 for (p = first; p < last; p++) {
4329 nr = le32_to_cpu(*p);
4330 if (nr) {
4331 /* accumulate blocks to free if they're contiguous */
4332 if (count == 0) {
4333 block_to_free = nr;
4334 block_to_free_p = p;
4335 count = 1;
4336 } else if (nr == block_to_free + count) {
4337 count++;
4338 } else {
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004339 if (ext4_clear_blocks(handle, inode, this_bh,
4340 block_to_free, count,
4341 block_to_free_p, p))
4342 break;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004343 block_to_free = nr;
4344 block_to_free_p = p;
4345 count = 1;
4346 }
4347 }
4348 }
4349
4350 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07004351 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004352 count, block_to_free_p, p);
4353
4354 if (this_bh) {
Frank Mayhar03901312009-01-07 00:06:22 -05004355 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
Duane Griffin71dc8fb2008-07-11 19:27:31 -04004356
4357 /*
4358 * The buffer head should have an attached journal head at this
4359 * point. However, if the data is corrupted and an indirect
4360 * block pointed to itself, it would have been detached when
4361 * the block was cleared. Check for this instead of OOPSing.
4362 */
Theodore Ts'oe7f07962009-01-20 09:50:19 -05004363 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
Frank Mayhar03901312009-01-07 00:06:22 -05004364 ext4_handle_dirty_metadata(handle, inode, this_bh);
Duane Griffin71dc8fb2008-07-11 19:27:31 -04004365 else
Theodore Ts'o24676da2010-05-16 21:00:00 -04004366 EXT4_ERROR_INODE(inode,
4367 "circular indirect block detected at "
4368 "block %llu",
4369 (unsigned long long) this_bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004370 }
4371}
4372
4373/**
Mingming Cao617ba132006-10-11 01:20:53 -07004374 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004375 * @handle: JBD handle for this transaction
4376 * @inode: inode we are dealing with
4377 * @parent_bh: the buffer_head which contains *@first and *@last
4378 * @first: array of block numbers
4379 * @last: pointer immediately past the end of array
4380 * @depth: depth of the branches to free
4381 *
4382 * We are freeing all blocks refered from these branches (numbers are
4383 * stored as little-endian 32-bit) and updating @inode->i_blocks
4384 * appropriately.
4385 */
Mingming Cao617ba132006-10-11 01:20:53 -07004386static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004387 struct buffer_head *parent_bh,
4388 __le32 *first, __le32 *last, int depth)
4389{
Mingming Cao617ba132006-10-11 01:20:53 -07004390 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004391 __le32 *p;
4392
Frank Mayhar03901312009-01-07 00:06:22 -05004393 if (ext4_handle_is_aborted(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004394 return;
4395
4396 if (depth--) {
4397 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07004398 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004399 p = last;
4400 while (--p >= first) {
4401 nr = le32_to_cpu(*p);
4402 if (!nr)
4403 continue; /* A hole */
4404
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004405 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4406 nr, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04004407 EXT4_ERROR_INODE(inode,
4408 "invalid indirect mapped "
4409 "block %lu (level %d)",
4410 (unsigned long) nr, depth);
Theodore Ts'o1f2acb62010-01-22 17:40:42 -05004411 break;
4412 }
4413
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004414 /* Go read the buffer for the next level down */
4415 bh = sb_bread(inode->i_sb, nr);
4416
4417 /*
4418 * A read failure? Report error and clear slot
4419 * (should be rare).
4420 */
4421 if (!bh) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004422 EXT4_ERROR_INODE_BLOCK(inode, nr,
4423 "Read failure");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004424 continue;
4425 }
4426
4427 /* This zaps the entire block. Bottom up. */
4428 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07004429 ext4_free_branches(handle, inode, bh,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004430 (__le32 *) bh->b_data,
4431 (__le32 *) bh->b_data + addr_per_block,
4432 depth);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004433
4434 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004435 * Everything below this this pointer has been
4436 * released. Now let this top-of-subtree go.
4437 *
4438 * We want the freeing of this indirect block to be
4439 * atomic in the journal with the updating of the
4440 * bitmap block which owns it. So make some room in
4441 * the journal.
4442 *
4443 * We zero the parent pointer *after* freeing its
4444 * pointee in the bitmaps, so if extend_transaction()
4445 * for some reason fails to put the bitmap changes and
4446 * the release into the same transaction, recovery
4447 * will merely complain about releasing a free block,
4448 * rather than leaking blocks.
4449 */
Frank Mayhar03901312009-01-07 00:06:22 -05004450 if (ext4_handle_is_aborted(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004451 return;
4452 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004453 ext4_mark_inode_dirty(handle, inode);
Jan Kara487caee2009-08-17 22:17:20 -04004454 ext4_truncate_restart_trans(handle, inode,
4455 blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004456 }
4457
Amir G40389682010-07-27 11:56:05 -04004458 /*
4459 * The forget flag here is critical because if
4460 * we are journaling (and not doing data
4461 * journaling), we have to make sure a revoke
4462 * record is written to prevent the journal
4463 * replay from overwriting the (former)
4464 * indirect block if it gets reallocated as a
4465 * data block. This must happen in the same
4466 * transaction where the data blocks are
4467 * actually freed.
4468 */
Theodore Ts'oe6362602009-11-23 07:17:05 -05004469 ext4_free_blocks(handle, inode, 0, nr, 1,
Amir G40389682010-07-27 11:56:05 -04004470 EXT4_FREE_BLOCKS_METADATA|
4471 EXT4_FREE_BLOCKS_FORGET);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004472
4473 if (parent_bh) {
4474 /*
4475 * The block which we have just freed is
4476 * pointed to by an indirect block: journal it
4477 */
4478 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004479 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004480 parent_bh)){
4481 *p = 0;
4482 BUFFER_TRACE(parent_bh,
Frank Mayhar03901312009-01-07 00:06:22 -05004483 "call ext4_handle_dirty_metadata");
4484 ext4_handle_dirty_metadata(handle,
4485 inode,
4486 parent_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004487 }
4488 }
4489 }
4490 } else {
4491 /* We have reached the bottom of the tree. */
4492 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07004493 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004494 }
4495}
4496
Duane Griffin91ef4ca2008-07-11 19:27:31 -04004497int ext4_can_truncate(struct inode *inode)
4498{
4499 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4500 return 0;
4501 if (S_ISREG(inode->i_mode))
4502 return 1;
4503 if (S_ISDIR(inode->i_mode))
4504 return 1;
4505 if (S_ISLNK(inode->i_mode))
4506 return !ext4_inode_is_fast_symlink(inode);
4507 return 0;
4508}
4509
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004510/*
Mingming Cao617ba132006-10-11 01:20:53 -07004511 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004512 *
Mingming Cao617ba132006-10-11 01:20:53 -07004513 * We block out ext4_get_block() block instantiations across the entire
4514 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004515 * simultaneously on behalf of the same inode.
4516 *
4517 * As we work through the truncate and commmit bits of it to the journal there
4518 * is one core, guiding principle: the file's tree must always be consistent on
4519 * disk. We must be able to restart the truncate after a crash.
4520 *
4521 * The file's tree may be transiently inconsistent in memory (although it
4522 * probably isn't), but whenever we close off and commit a journal transaction,
4523 * the contents of (the filesystem + the journal) must be consistent and
4524 * restartable. It's pretty simple, really: bottom up, right to left (although
4525 * left-to-right works OK too).
4526 *
4527 * Note that at recovery time, journal replay occurs *before* the restart of
4528 * truncate against the orphan inode list.
4529 *
4530 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07004531 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004532 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07004533 * ext4_truncate() to have another go. So there will be instantiated blocks
4534 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004535 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07004536 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004537 */
Mingming Cao617ba132006-10-11 01:20:53 -07004538void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004539{
4540 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07004541 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004542 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07004543 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004544 struct address_space *mapping = inode->i_mapping;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004545 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004546 Indirect chain[4];
4547 Indirect *partial;
4548 __le32 nr = 0;
4549 int n;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004550 ext4_lblk_t last_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004551 unsigned blocksize = inode->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004552
Duane Griffin91ef4ca2008-07-11 19:27:31 -04004553 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004554 return;
4555
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004556 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004557
Theodore Ts'o5534fb52009-09-17 09:34:16 -04004558 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004559 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05004560
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004561 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Jan Karacf108bc2008-07-11 19:27:31 -04004562 ext4_ext_truncate(inode);
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004563 return;
4564 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004565
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004566 handle = start_transaction(inode);
Jan Karacf108bc2008-07-11 19:27:31 -04004567 if (IS_ERR(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004568 return; /* AKPM: return what? */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004569
4570 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07004571 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004572
Jan Karacf108bc2008-07-11 19:27:31 -04004573 if (inode->i_size & (blocksize - 1))
4574 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4575 goto out_stop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004576
Mingming Cao617ba132006-10-11 01:20:53 -07004577 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004578 if (n == 0)
4579 goto out_stop; /* error */
4580
4581 /*
4582 * OK. This truncate is going to happen. We add the inode to the
4583 * orphan list, so that if this truncate spans multiple transactions,
4584 * and we crash, we will resume the truncate when the filesystem
4585 * recovers. It also marks the inode dirty, to catch the new size.
4586 *
4587 * Implication: the file must always be in a sane, consistent
4588 * truncatable state while each transaction commits.
4589 */
Mingming Cao617ba132006-10-11 01:20:53 -07004590 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004591 goto out_stop;
4592
4593 /*
Mingming Cao632eaea2008-07-11 19:27:31 -04004594 * From here we block out all ext4_get_block() callers who want to
4595 * modify the block allocation tree.
4596 */
4597 down_write(&ei->i_data_sem);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04004598
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004599 ext4_discard_preallocations(inode);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04004600
Mingming Cao632eaea2008-07-11 19:27:31 -04004601 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004602 * The orphan list entry will now protect us from any crash which
4603 * occurs before the truncate completes, so it is now safe to propagate
4604 * the new, shorter inode size (held for now in i_size) into the
4605 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07004606 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004607 */
4608 ei->i_disksize = inode->i_size;
4609
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004610 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07004611 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4612 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004613 goto do_indirects;
4614 }
4615
Mingming Cao617ba132006-10-11 01:20:53 -07004616 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004617 /* Kill the top of shared branch (not detached) */
4618 if (nr) {
4619 if (partial == chain) {
4620 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07004621 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004622 &nr, &nr+1, (chain+n-1) - partial);
4623 *partial->p = 0;
4624 /*
4625 * We mark the inode dirty prior to restart,
4626 * and prior to stop. No need for it here.
4627 */
4628 } else {
4629 /* Shared branch grows from an indirect block */
4630 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004631 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004632 partial->p,
4633 partial->p+1, (chain+n-1) - partial);
4634 }
4635 }
4636 /* Clear the ends of indirect blocks on the shared branch */
4637 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07004638 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004639 (__le32*)partial->bh->b_data+addr_per_block,
4640 (chain+n-1) - partial);
4641 BUFFER_TRACE(partial->bh, "call brelse");
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004642 brelse(partial->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004643 partial--;
4644 }
4645do_indirects:
4646 /* Kill the remaining (whole) subtrees */
4647 switch (offsets[0]) {
4648 default:
Mingming Cao617ba132006-10-11 01:20:53 -07004649 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004650 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004651 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4652 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004653 }
Mingming Cao617ba132006-10-11 01:20:53 -07004654 case EXT4_IND_BLOCK:
4655 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004656 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004657 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4658 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004659 }
Mingming Cao617ba132006-10-11 01:20:53 -07004660 case EXT4_DIND_BLOCK:
4661 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004662 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004663 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4664 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004665 }
Mingming Cao617ba132006-10-11 01:20:53 -07004666 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004667 ;
4668 }
4669
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004670 up_write(&ei->i_data_sem);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004671 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07004672 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004673
4674 /*
4675 * In a multi-transaction truncate, we only make the final transaction
4676 * synchronous
4677 */
4678 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05004679 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004680out_stop:
4681 /*
4682 * If this was a simple ftruncate(), and the file will remain alive
4683 * then we need to clear up the orphan record which we created above.
4684 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07004685 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004686 * orphan info for us.
4687 */
4688 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004689 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004690
Mingming Cao617ba132006-10-11 01:20:53 -07004691 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004692}
4693
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004694/*
Mingming Cao617ba132006-10-11 01:20:53 -07004695 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004696 * underlying buffer_head on success. If 'in_mem' is true, we have all
4697 * data in memory that is needed to recreate the on-disk version of this
4698 * inode.
4699 */
Mingming Cao617ba132006-10-11 01:20:53 -07004700static int __ext4_get_inode_loc(struct inode *inode,
4701 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004702{
Theodore Ts'o240799c2008-10-09 23:53:47 -04004703 struct ext4_group_desc *gdp;
4704 struct buffer_head *bh;
4705 struct super_block *sb = inode->i_sb;
4706 ext4_fsblk_t block;
4707 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004708
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004709 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04004710 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004711 return -EIO;
4712
Theodore Ts'o240799c2008-10-09 23:53:47 -04004713 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4714 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4715 if (!gdp)
4716 return -EIO;
4717
4718 /*
4719 * Figure out the offset within the block group inode table
4720 */
4721 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4722 inode_offset = ((inode->i_ino - 1) %
4723 EXT4_INODES_PER_GROUP(sb));
4724 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4725 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4726
4727 bh = sb_getblk(sb, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004728 if (!bh) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004729 EXT4_ERROR_INODE_BLOCK(inode, block,
4730 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004731 return -EIO;
4732 }
4733 if (!buffer_uptodate(bh)) {
4734 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04004735
4736 /*
4737 * If the buffer has the write error flag, we have failed
4738 * to write out another inode in the same block. In this
4739 * case, we don't have to read the block because we may
4740 * read the old inode data successfully.
4741 */
4742 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4743 set_buffer_uptodate(bh);
4744
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004745 if (buffer_uptodate(bh)) {
4746 /* someone brought it uptodate while we waited */
4747 unlock_buffer(bh);
4748 goto has_buffer;
4749 }
4750
4751 /*
4752 * If we have all information of the inode in memory and this
4753 * is the only valid inode in the block, we need not read the
4754 * block.
4755 */
4756 if (in_mem) {
4757 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04004758 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004759
Theodore Ts'o240799c2008-10-09 23:53:47 -04004760 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004761
4762 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04004763 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004764 if (!bitmap_bh)
4765 goto make_io;
4766
4767 /*
4768 * If the inode bitmap isn't in cache then the
4769 * optimisation may end up performing two reads instead
4770 * of one, so skip it.
4771 */
4772 if (!buffer_uptodate(bitmap_bh)) {
4773 brelse(bitmap_bh);
4774 goto make_io;
4775 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04004776 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004777 if (i == inode_offset)
4778 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07004779 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004780 break;
4781 }
4782 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04004783 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004784 /* all other inodes are free, so skip I/O */
4785 memset(bh->b_data, 0, bh->b_size);
4786 set_buffer_uptodate(bh);
4787 unlock_buffer(bh);
4788 goto has_buffer;
4789 }
4790 }
4791
4792make_io:
4793 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04004794 * If we need to do any I/O, try to pre-readahead extra
4795 * blocks from the inode table.
4796 */
4797 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4798 ext4_fsblk_t b, end, table;
4799 unsigned num;
4800
4801 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04004802 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o240799c2008-10-09 23:53:47 -04004803 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4804 if (table > b)
4805 b = table;
4806 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4807 num = EXT4_INODES_PER_GROUP(sb);
4808 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4809 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05004810 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04004811 table += num / inodes_per_block;
4812 if (end > table)
4813 end = table;
4814 while (b <= end)
4815 sb_breadahead(sb, b++);
4816 }
4817
4818 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004819 * There are other valid inodes in the buffer, this inode
4820 * has in-inode xattrs, or we don't have this inode in memory.
4821 * Read the block from disk.
4822 */
4823 get_bh(bh);
4824 bh->b_end_io = end_buffer_read_sync;
4825 submit_bh(READ_META, bh);
4826 wait_on_buffer(bh);
4827 if (!buffer_uptodate(bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004828 EXT4_ERROR_INODE_BLOCK(inode, block,
4829 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004830 brelse(bh);
4831 return -EIO;
4832 }
4833 }
4834has_buffer:
4835 iloc->bh = bh;
4836 return 0;
4837}
4838
Mingming Cao617ba132006-10-11 01:20:53 -07004839int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004840{
4841 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07004842 return __ext4_get_inode_loc(inode, iloc,
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004843 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004844}
4845
Mingming Cao617ba132006-10-11 01:20:53 -07004846void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004847{
Mingming Cao617ba132006-10-11 01:20:53 -07004848 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004849
4850 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07004851 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004852 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07004853 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004854 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07004855 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004856 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07004857 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004858 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07004859 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004860 inode->i_flags |= S_DIRSYNC;
4861}
4862
Jan Karaff9ddf72007-07-18 09:24:20 -04004863/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4864void ext4_get_inode_flags(struct ext4_inode_info *ei)
4865{
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004866 unsigned int vfs_fl;
4867 unsigned long old_fl, new_fl;
Jan Karaff9ddf72007-07-18 09:24:20 -04004868
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004869 do {
4870 vfs_fl = ei->vfs_inode.i_flags;
4871 old_fl = ei->i_flags;
4872 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4873 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4874 EXT4_DIRSYNC_FL);
4875 if (vfs_fl & S_SYNC)
4876 new_fl |= EXT4_SYNC_FL;
4877 if (vfs_fl & S_APPEND)
4878 new_fl |= EXT4_APPEND_FL;
4879 if (vfs_fl & S_IMMUTABLE)
4880 new_fl |= EXT4_IMMUTABLE_FL;
4881 if (vfs_fl & S_NOATIME)
4882 new_fl |= EXT4_NOATIME_FL;
4883 if (vfs_fl & S_DIRSYNC)
4884 new_fl |= EXT4_DIRSYNC_FL;
4885 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
Jan Karaff9ddf72007-07-18 09:24:20 -04004886}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004887
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004888static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004889 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004890{
4891 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004892 struct inode *inode = &(ei->vfs_inode);
4893 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004894
4895 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4896 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4897 /* we are using combined 48 bit field */
4898 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4899 le32_to_cpu(raw_inode->i_blocks_lo);
Theodore Ts'o07a03822010-06-14 09:54:48 -04004900 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004901 /* i_blocks represent file system block size */
4902 return i_blocks << (inode->i_blkbits - 9);
4903 } else {
4904 return i_blocks;
4905 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004906 } else {
4907 return le32_to_cpu(raw_inode->i_blocks_lo);
4908 }
4909}
Jan Karaff9ddf72007-07-18 09:24:20 -04004910
David Howells1d1fe1e2008-02-07 00:15:37 -08004911struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004912{
Mingming Cao617ba132006-10-11 01:20:53 -07004913 struct ext4_iloc iloc;
4914 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004915 struct ext4_inode_info *ei;
David Howells1d1fe1e2008-02-07 00:15:37 -08004916 struct inode *inode;
Jan Karab436b9b2009-12-08 23:51:10 -05004917 journal_t *journal = EXT4_SB(sb)->s_journal;
David Howells1d1fe1e2008-02-07 00:15:37 -08004918 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004919 int block;
4920
David Howells1d1fe1e2008-02-07 00:15:37 -08004921 inode = iget_locked(sb, ino);
4922 if (!inode)
4923 return ERR_PTR(-ENOMEM);
4924 if (!(inode->i_state & I_NEW))
4925 return inode;
4926
4927 ei = EXT4_I(inode);
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004928 iloc.bh = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004929
David Howells1d1fe1e2008-02-07 00:15:37 -08004930 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4931 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004932 goto bad_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07004933 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004934 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4935 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4936 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004937 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004938 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4939 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4940 }
4941 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004942
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004943 ei->i_state_flags = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004944 ei->i_dir_start_lookup = 0;
4945 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4946 /* We now have enough fields to check if the inode was active or not.
4947 * This is needed because nfsd might try to access dead inodes
4948 * the test is that same one that e2fsck uses
4949 * NeilBrown 1999oct15
4950 */
4951 if (inode->i_nlink == 0) {
4952 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07004953 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004954 /* this inode is deleted */
David Howells1d1fe1e2008-02-07 00:15:37 -08004955 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004956 goto bad_inode;
4957 }
4958 /* The only unlinked inodes we let through here have
4959 * valid i_mode and are being read by the orphan
4960 * recovery code: that's fine, we're about to complete
4961 * the process of deleting those. */
4962 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004963 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004964 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004965 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04004966 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004967 ei->i_file_acl |=
4968 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004969 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004970 ei->i_disksize = inode->i_size;
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03004971#ifdef CONFIG_QUOTA
4972 ei->i_reserved_quota = 0;
4973#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004974 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4975 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04004976 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004977 /*
4978 * NOTE! The in-memory inode i_data array is in little-endian order
4979 * even on big-endian machines: we do NOT byteswap the block numbers!
4980 */
Mingming Cao617ba132006-10-11 01:20:53 -07004981 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004982 ei->i_data[block] = raw_inode->i_block[block];
4983 INIT_LIST_HEAD(&ei->i_orphan);
4984
Jan Karab436b9b2009-12-08 23:51:10 -05004985 /*
4986 * Set transaction id's of transactions that have to be committed
4987 * to finish f[data]sync. We set them to currently running transaction
4988 * as we cannot be sure that the inode or some of its metadata isn't
4989 * part of the transaction - the inode could have been reclaimed and
4990 * now it is reread from disk.
4991 */
4992 if (journal) {
4993 transaction_t *transaction;
4994 tid_t tid;
4995
Theodore Ts'oa931da62010-08-03 21:35:12 -04004996 read_lock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004997 if (journal->j_running_transaction)
4998 transaction = journal->j_running_transaction;
4999 else
5000 transaction = journal->j_committing_transaction;
5001 if (transaction)
5002 tid = transaction->t_tid;
5003 else
5004 tid = journal->j_commit_sequence;
Theodore Ts'oa931da62010-08-03 21:35:12 -04005005 read_unlock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05005006 ei->i_sync_tid = tid;
5007 ei->i_datasync_tid = tid;
5008 }
5009
Eric Sandeen0040d982008-02-05 22:36:43 -05005010 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005011 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07005012 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07005013 EXT4_INODE_SIZE(inode->i_sb)) {
David Howells1d1fe1e2008-02-07 00:15:37 -08005014 ret = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005015 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07005016 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005017 if (ei->i_extra_isize == 0) {
5018 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07005019 ei->i_extra_isize = sizeof(struct ext4_inode) -
5020 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005021 } else {
5022 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07005023 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005024 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07005025 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005026 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005027 }
5028 } else
5029 ei->i_extra_isize = 0;
5030
Kalpak Shahef7f3832007-07-18 09:15:20 -04005031 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5032 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5033 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5034 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5035
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005036 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5037 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5038 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5039 inode->i_version |=
5040 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5041 }
5042
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04005043 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04005044 if (ei->i_file_acl &&
Theodore Ts'o10329882009-11-15 15:29:56 -05005045 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04005046 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5047 ei->i_file_acl);
Theodore Ts'o485c26e2009-04-24 13:43:20 -04005048 ret = -EIO;
5049 goto bad_inode;
Theodore Ts'o07a03822010-06-14 09:54:48 -04005050 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04005051 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5052 (S_ISLNK(inode->i_mode) &&
5053 !ext4_inode_is_fast_symlink(inode)))
5054 /* Validate extent which is part of inode */
5055 ret = ext4_ext_check_inode(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005056 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04005057 (S_ISLNK(inode->i_mode) &&
5058 !ext4_inode_is_fast_symlink(inode))) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005059 /* Validate block references which are part of inode */
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04005060 ret = ext4_check_inode_blockref(inode);
5061 }
Theodore Ts'o567f3e92009-11-14 08:19:05 -05005062 if (ret)
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005063 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04005064
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005065 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07005066 inode->i_op = &ext4_file_inode_operations;
5067 inode->i_fop = &ext4_file_operations;
5068 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005069 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07005070 inode->i_op = &ext4_dir_inode_operations;
5071 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005072 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00005073 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07005074 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00005075 nd_terminate_link(ei->i_data, inode->i_size,
5076 sizeof(ei->i_data) - 1);
5077 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07005078 inode->i_op = &ext4_symlink_inode_operations;
5079 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005080 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04005081 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5082 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07005083 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005084 if (raw_inode->i_block[0])
5085 init_special_inode(inode, inode->i_mode,
5086 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5087 else
5088 init_special_inode(inode, inode->i_mode,
5089 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Theodore Ts'o563bdd62009-03-26 00:06:19 -04005090 } else {
Theodore Ts'o563bdd62009-03-26 00:06:19 -04005091 ret = -EIO;
Theodore Ts'o24676da2010-05-16 21:00:00 -04005092 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04005093 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005094 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005095 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07005096 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08005097 unlock_new_inode(inode);
5098 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005099
5100bad_inode:
Theodore Ts'o567f3e92009-11-14 08:19:05 -05005101 brelse(iloc.bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08005102 iget_failed(inode);
5103 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005104}
5105
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005106static int ext4_inode_blocks_set(handle_t *handle,
5107 struct ext4_inode *raw_inode,
5108 struct ext4_inode_info *ei)
5109{
5110 struct inode *inode = &(ei->vfs_inode);
5111 u64 i_blocks = inode->i_blocks;
5112 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005113
5114 if (i_blocks <= ~0U) {
5115 /*
5116 * i_blocks can be represnted in a 32 bit variable
5117 * as multiple of 512 bytes
5118 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05005119 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005120 raw_inode->i_blocks_high = 0;
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04005121 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Theodore Ts'of287a1a2008-10-16 22:50:48 -04005122 return 0;
5123 }
5124 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5125 return -EFBIG;
5126
5127 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005128 /*
5129 * i_blocks can be represented in a 48 bit variable
5130 * as multiple of 512 bytes
5131 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05005132 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005133 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04005134 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005135 } else {
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04005136 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05005137 /* i_block is stored in file system block size */
5138 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5139 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5140 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005141 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04005142 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005143}
5144
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005145/*
5146 * Post the struct inode info into an on-disk inode location in the
5147 * buffer-cache. This gobbles the caller's reference to the
5148 * buffer_head in the inode location struct.
5149 *
5150 * The caller must have write access to iloc->bh.
5151 */
Mingming Cao617ba132006-10-11 01:20:53 -07005152static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005153 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04005154 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005155{
Mingming Cao617ba132006-10-11 01:20:53 -07005156 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5157 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005158 struct buffer_head *bh = iloc->bh;
5159 int err = 0, rc, block;
5160
5161 /* For fields not not tracking in the in-memory inode,
5162 * initialise them to zero for new inodes. */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005163 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
Mingming Cao617ba132006-10-11 01:20:53 -07005164 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005165
Jan Karaff9ddf72007-07-18 09:24:20 -04005166 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005167 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005168 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005169 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5170 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5171/*
5172 * Fix up interoperability with old kernels. Otherwise, old inodes get
5173 * re-used with the upper 16 bits of the uid/gid intact
5174 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005175 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005176 raw_inode->i_uid_high =
5177 cpu_to_le16(high_16_bits(inode->i_uid));
5178 raw_inode->i_gid_high =
5179 cpu_to_le16(high_16_bits(inode->i_gid));
5180 } else {
5181 raw_inode->i_uid_high = 0;
5182 raw_inode->i_gid_high = 0;
5183 }
5184 } else {
5185 raw_inode->i_uid_low =
5186 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5187 raw_inode->i_gid_low =
5188 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5189 raw_inode->i_uid_high = 0;
5190 raw_inode->i_gid_high = 0;
5191 }
5192 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04005193
5194 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5195 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5196 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5197 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5198
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005199 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5200 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005201 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o1b9c12f2009-09-17 08:32:22 -04005202 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07005203 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5204 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07005205 raw_inode->i_file_acl_high =
5206 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05005207 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005208 ext4_isize_set(raw_inode, ei->i_disksize);
5209 if (ei->i_disksize > 0x7fffffffULL) {
5210 struct super_block *sb = inode->i_sb;
5211 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5212 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5213 EXT4_SB(sb)->s_es->s_rev_level ==
5214 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5215 /* If this is the first large file
5216 * created, add a flag to the superblock.
5217 */
5218 err = ext4_journal_get_write_access(handle,
5219 EXT4_SB(sb)->s_sbh);
5220 if (err)
5221 goto out_brelse;
5222 ext4_update_dynamic_rev(sb);
5223 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07005224 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005225 sb->s_dirt = 1;
Frank Mayhar03901312009-01-07 00:06:22 -05005226 ext4_handle_sync(handle);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05005227 err = ext4_handle_dirty_metadata(handle, NULL,
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005228 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005229 }
5230 }
5231 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5232 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5233 if (old_valid_dev(inode->i_rdev)) {
5234 raw_inode->i_block[0] =
5235 cpu_to_le32(old_encode_dev(inode->i_rdev));
5236 raw_inode->i_block[1] = 0;
5237 } else {
5238 raw_inode->i_block[0] = 0;
5239 raw_inode->i_block[1] =
5240 cpu_to_le32(new_encode_dev(inode->i_rdev));
5241 raw_inode->i_block[2] = 0;
5242 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005243 } else
5244 for (block = 0; block < EXT4_N_BLOCKS; block++)
5245 raw_inode->i_block[block] = ei->i_data[block];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005246
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005247 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5248 if (ei->i_extra_isize) {
5249 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5250 raw_inode->i_version_hi =
5251 cpu_to_le32(inode->i_version >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005252 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005253 }
5254
Frank Mayhar830156c2009-09-29 10:07:47 -04005255 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05005256 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
Frank Mayhar830156c2009-09-29 10:07:47 -04005257 if (!err)
5258 err = rc;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005259 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005260
Jan Karab436b9b2009-12-08 23:51:10 -05005261 ext4_update_inode_fsync_trans(handle, inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005262out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005263 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07005264 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005265 return err;
5266}
5267
5268/*
Mingming Cao617ba132006-10-11 01:20:53 -07005269 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005270 *
5271 * We are called from a few places:
5272 *
5273 * - Within generic_file_write() for O_SYNC files.
5274 * Here, there will be no transaction running. We wait for any running
5275 * trasnaction to commit.
5276 *
5277 * - Within sys_sync(), kupdate and such.
5278 * We wait on commit, if tol to.
5279 *
5280 * - Within prune_icache() (PF_MEMALLOC == true)
5281 * Here we simply return. We can't afford to block kswapd on the
5282 * journal commit.
5283 *
5284 * In all cases it is actually safe for us to return without doing anything,
5285 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07005286 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005287 * knfsd.
5288 *
5289 * Note that we are absolutely dependent upon all inode dirtiers doing the
5290 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5291 * which we are interested.
5292 *
5293 * It would be a bug for them to not do this. The code:
5294 *
5295 * mark_inode_dirty(inode)
5296 * stuff();
5297 * inode->i_size = expr;
5298 *
5299 * is in error because a kswapd-driven write_inode() could occur while
5300 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5301 * will no longer be on the superblock's dirty inode list.
5302 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005303int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005304{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005305 int err;
5306
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005307 if (current->flags & PF_MEMALLOC)
5308 return 0;
5309
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005310 if (EXT4_SB(inode->i_sb)->s_journal) {
5311 if (ext4_journal_current_handle()) {
5312 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5313 dump_stack();
5314 return -EIO;
5315 }
5316
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005317 if (wbc->sync_mode != WB_SYNC_ALL)
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005318 return 0;
5319
5320 err = ext4_force_commit(inode->i_sb);
5321 } else {
5322 struct ext4_iloc iloc;
5323
Curt Wohlgemuth8b472d72010-04-03 16:45:06 -04005324 err = __ext4_get_inode_loc(inode, &iloc, 0);
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005325 if (err)
5326 return err;
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005327 if (wbc->sync_mode == WB_SYNC_ALL)
Frank Mayhar830156c2009-09-29 10:07:47 -04005328 sync_dirty_buffer(iloc.bh);
5329 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04005330 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5331 "IO error syncing inode");
Frank Mayhar830156c2009-09-29 10:07:47 -04005332 err = -EIO;
5333 }
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04005334 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005335 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005336 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005337}
5338
5339/*
Mingming Cao617ba132006-10-11 01:20:53 -07005340 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005341 *
5342 * Called from notify_change.
5343 *
5344 * We want to trap VFS attempts to truncate the file as soon as
5345 * possible. In particular, we want to make sure that when the VFS
5346 * shrinks i_size, we put the inode on the orphan list and modify
5347 * i_disksize immediately, so that during the subsequent flushing of
5348 * dirty pages and freeing of disk blocks, we can guarantee that any
5349 * commit will leave the blocks being flushed in an unused state on
5350 * disk. (On recovery, the inode will get truncated and the blocks will
5351 * be freed, so we have a strong guarantee that no future commit will
5352 * leave these blocks visible to the user.)
5353 *
Jan Kara678aaf42008-07-11 19:27:31 -04005354 * Another thing we have to assure is that if we are in ordered mode
5355 * and inode is still attached to the committing transaction, we must
5356 * we start writeout of all the dirty pages which are being truncated.
5357 * This way we are sure that all the data written in the previous
5358 * transaction are already on disk (truncate waits for pages under
5359 * writeback).
5360 *
5361 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005362 */
Mingming Cao617ba132006-10-11 01:20:53 -07005363int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005364{
5365 struct inode *inode = dentry->d_inode;
5366 int error, rc = 0;
5367 const unsigned int ia_valid = attr->ia_valid;
5368
5369 error = inode_change_ok(inode, attr);
5370 if (error)
5371 return error;
5372
Dmitry Monakhov12755622010-04-08 22:04:20 +04005373 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05005374 dquot_initialize(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005375 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5376 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5377 handle_t *handle;
5378
5379 /* (user+group)*(old+new) structure, inode write (sb,
5380 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05005381 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
Dmitry Monakhov194074a2009-12-08 22:42:28 -05005382 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005383 if (IS_ERR(handle)) {
5384 error = PTR_ERR(handle);
5385 goto err_out;
5386 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05005387 error = dquot_transfer(inode, attr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005388 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07005389 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005390 return error;
5391 }
5392 /* Update corresponding info in inode so that everything is in
5393 * one transaction */
5394 if (attr->ia_valid & ATTR_UID)
5395 inode->i_uid = attr->ia_uid;
5396 if (attr->ia_valid & ATTR_GID)
5397 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07005398 error = ext4_mark_inode_dirty(handle, inode);
5399 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005400 }
5401
Eric Sandeene2b46572008-01-28 23:58:27 -05005402 if (attr->ia_valid & ATTR_SIZE) {
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005403 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -05005404 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5405
Theodore Ts'o0c095c72010-07-27 11:56:06 -04005406 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5407 return -EFBIG;
Eric Sandeene2b46572008-01-28 23:58:27 -05005408 }
5409 }
5410
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005411 if (S_ISREG(inode->i_mode) &&
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05005412 attr->ia_valid & ATTR_SIZE &&
5413 (attr->ia_size < inode->i_size ||
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005414 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005415 handle_t *handle;
5416
Mingming Cao617ba132006-10-11 01:20:53 -07005417 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005418 if (IS_ERR(handle)) {
5419 error = PTR_ERR(handle);
5420 goto err_out;
5421 }
5422
Mingming Cao617ba132006-10-11 01:20:53 -07005423 error = ext4_orphan_add(handle, inode);
5424 EXT4_I(inode)->i_disksize = attr->ia_size;
5425 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005426 if (!error)
5427 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07005428 ext4_journal_stop(handle);
Jan Kara678aaf42008-07-11 19:27:31 -04005429
5430 if (ext4_should_order_data(inode)) {
5431 error = ext4_begin_ordered_truncate(inode,
5432 attr->ia_size);
5433 if (error) {
5434 /* Do as much error cleanup as possible */
5435 handle = ext4_journal_start(inode, 3);
5436 if (IS_ERR(handle)) {
5437 ext4_orphan_del(NULL, inode);
5438 goto err_out;
5439 }
5440 ext4_orphan_del(handle, inode);
5441 ext4_journal_stop(handle);
5442 goto err_out;
5443 }
5444 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05005445 /* ext4_truncate will clear the flag */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005446 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05005447 ext4_truncate(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005448 }
5449
Christoph Hellwig10257742010-06-04 11:30:02 +02005450 if ((attr->ia_valid & ATTR_SIZE) &&
5451 attr->ia_size != i_size_read(inode))
5452 rc = vmtruncate(inode, attr->ia_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005453
Christoph Hellwig10257742010-06-04 11:30:02 +02005454 if (!rc) {
5455 setattr_copy(inode, attr);
5456 mark_inode_dirty(inode);
5457 }
5458
5459 /*
5460 * If the call to ext4_truncate failed to get a transaction handle at
5461 * all, we need to clean up the in-core orphan list manually.
5462 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005463 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07005464 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005465
5466 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07005467 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005468
5469err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07005470 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005471 if (!error)
5472 error = rc;
5473 return error;
5474}
5475
Mingming Cao3e3398a2008-07-11 19:27:31 -04005476int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5477 struct kstat *stat)
5478{
5479 struct inode *inode;
5480 unsigned long delalloc_blocks;
5481
5482 inode = dentry->d_inode;
5483 generic_fillattr(inode, stat);
5484
5485 /*
5486 * We can't update i_blocks if the block allocation is delayed
5487 * otherwise in the case of system crash before the real block
5488 * allocation is done, we will have i_blocks inconsistent with
5489 * on-disk file blocks.
5490 * We always keep i_blocks updated together with real
5491 * allocation. But to not confuse with user, stat
5492 * will return the blocks that include the delayed allocation
5493 * blocks for this file.
5494 */
5495 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5496 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5497 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5498
5499 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5500 return 0;
5501}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005502
Mingming Caoa02908f2008-08-19 22:16:07 -04005503static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5504 int chunk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005505{
Mingming Caoa02908f2008-08-19 22:16:07 -04005506 int indirects;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005507
Mingming Caoa02908f2008-08-19 22:16:07 -04005508 /* if nrblocks are contiguous */
5509 if (chunk) {
5510 /*
5511 * With N contiguous data blocks, it need at most
5512 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5513 * 2 dindirect blocks
5514 * 1 tindirect block
5515 */
5516 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5517 return indirects + 3;
5518 }
5519 /*
5520 * if nrblocks are not contiguous, worse case, each block touch
5521 * a indirect block, and each indirect block touch a double indirect
5522 * block, plus a triple indirect block
5523 */
5524 indirects = nrblocks * 2 + 1;
5525 return indirects;
5526}
Alex Tomasa86c6182006-10-11 01:21:03 -07005527
Mingming Caoa02908f2008-08-19 22:16:07 -04005528static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5529{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005530 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Theodore Ts'oac51d832008-11-06 16:49:36 -05005531 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5532 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
Mingming Caoa02908f2008-08-19 22:16:07 -04005533}
Theodore Ts'oac51d832008-11-06 16:49:36 -05005534
Mingming Caoa02908f2008-08-19 22:16:07 -04005535/*
5536 * Account for index blocks, block groups bitmaps and block group
5537 * descriptor blocks if modify datablocks and index blocks
5538 * worse case, the indexs blocks spread over different block groups
5539 *
5540 * If datablocks are discontiguous, they are possible to spread over
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02005541 * different block groups too. If they are contiuguous, with flexbg,
Mingming Caoa02908f2008-08-19 22:16:07 -04005542 * they could still across block group boundary.
5543 *
5544 * Also account for superblock, inode, quota and xattr blocks
5545 */
5546int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5547{
Theodore Ts'o8df96752009-05-01 08:50:38 -04005548 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5549 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04005550 int idxblocks;
5551 int ret = 0;
5552
5553 /*
5554 * How many index blocks need to touch to modify nrblocks?
5555 * The "Chunk" flag indicating whether the nrblocks is
5556 * physically contiguous on disk
5557 *
5558 * For Direct IO and fallocate, they calls get_block to allocate
5559 * one single extent at a time, so they could set the "Chunk" flag
5560 */
5561 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5562
5563 ret = idxblocks;
5564
5565 /*
5566 * Now let's see how many group bitmaps and group descriptors need
5567 * to account
5568 */
5569 groups = idxblocks;
5570 if (chunk)
5571 groups += 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005572 else
Mingming Caoa02908f2008-08-19 22:16:07 -04005573 groups += nrblocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005574
Mingming Caoa02908f2008-08-19 22:16:07 -04005575 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04005576 if (groups > ngroups)
5577 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04005578 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5579 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5580
5581 /* bitmaps and block group descriptor blocks */
5582 ret += groups + gdpblocks;
5583
5584 /* Blocks for super block, inode, quota and xattr blocks */
5585 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005586
5587 return ret;
5588}
5589
5590/*
Mingming Caoa02908f2008-08-19 22:16:07 -04005591 * Calulate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04005592 * the modification of a single pages into a single transaction,
5593 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04005594 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04005595 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04005596 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04005597 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04005598 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04005599 */
5600int ext4_writepage_trans_blocks(struct inode *inode)
5601{
5602 int bpp = ext4_journal_blocks_per_page(inode);
5603 int ret;
5604
5605 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5606
5607 /* Account for data blocks for journalled mode */
5608 if (ext4_should_journal_data(inode))
5609 ret += bpp;
5610 return ret;
5611}
Mingming Caof3bd1f32008-08-19 22:16:03 -04005612
5613/*
5614 * Calculate the journal credits for a chunk of data modification.
5615 *
5616 * This is called from DIO, fallocate or whoever calling
Eric Sandeen79e83032010-07-27 11:56:07 -04005617 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04005618 *
5619 * journal buffers for data blocks are not included here, as DIO
5620 * and fallocate do no need to journal data buffers.
5621 */
5622int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5623{
5624 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5625}
5626
Mingming Caoa02908f2008-08-19 22:16:07 -04005627/*
Mingming Cao617ba132006-10-11 01:20:53 -07005628 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005629 * Give this, we know that the caller already has write access to iloc->bh.
5630 */
Mingming Cao617ba132006-10-11 01:20:53 -07005631int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005632 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005633{
5634 int err = 0;
5635
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005636 if (test_opt(inode->i_sb, I_VERSION))
5637 inode_inc_iversion(inode);
5638
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005639 /* the do_update_inode consumes one bh->b_count */
5640 get_bh(iloc->bh);
5641
Mingming Caodab291a2006-10-11 01:21:01 -07005642 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04005643 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005644 put_bh(iloc->bh);
5645 return err;
5646}
5647
5648/*
5649 * On success, We end up with an outstanding reference count against
5650 * iloc->bh. This _must_ be cleaned up later.
5651 */
5652
5653int
Mingming Cao617ba132006-10-11 01:20:53 -07005654ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5655 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005656{
Frank Mayhar03901312009-01-07 00:06:22 -05005657 int err;
5658
5659 err = ext4_get_inode_loc(inode, iloc);
5660 if (!err) {
5661 BUFFER_TRACE(iloc->bh, "get_write_access");
5662 err = ext4_journal_get_write_access(handle, iloc->bh);
5663 if (err) {
5664 brelse(iloc->bh);
5665 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005666 }
5667 }
Mingming Cao617ba132006-10-11 01:20:53 -07005668 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005669 return err;
5670}
5671
5672/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005673 * Expand an inode by new_extra_isize bytes.
5674 * Returns 0 on success or negative error number on failure.
5675 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05005676static int ext4_expand_extra_isize(struct inode *inode,
5677 unsigned int new_extra_isize,
5678 struct ext4_iloc iloc,
5679 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005680{
5681 struct ext4_inode *raw_inode;
5682 struct ext4_xattr_ibody_header *header;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005683
5684 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5685 return 0;
5686
5687 raw_inode = ext4_raw_inode(&iloc);
5688
5689 header = IHDR(inode, raw_inode);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005690
5691 /* No extended attributes present */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005692 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5693 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005694 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5695 new_extra_isize);
5696 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5697 return 0;
5698 }
5699
5700 /* try to expand with EAs present */
5701 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5702 raw_inode, handle);
5703}
5704
5705/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005706 * What we do here is to mark the in-core inode as clean with respect to inode
5707 * dirtiness (it may still be data-dirty).
5708 * This means that the in-core inode may be reaped by prune_icache
5709 * without having to perform any I/O. This is a very good thing,
5710 * because *any* task may call prune_icache - even ones which
5711 * have a transaction open against a different journal.
5712 *
5713 * Is this cheating? Not really. Sure, we haven't written the
5714 * inode out, but prune_icache isn't a user-visible syncing function.
5715 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5716 * we start and wait on commits.
5717 *
5718 * Is this efficient/effective? Well, we're being nice to the system
5719 * by cleaning up our inodes proactively so they can be reaped
5720 * without I/O. But we are potentially leaving up to five seconds'
5721 * worth of inodes floating about which prune_icache wants us to
5722 * write out. One way to fix that would be to get prune_icache()
5723 * to do a write_super() to free up some memory. It has the desired
5724 * effect.
5725 */
Mingming Cao617ba132006-10-11 01:20:53 -07005726int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005727{
Mingming Cao617ba132006-10-11 01:20:53 -07005728 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005729 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5730 static unsigned int mnt_count;
5731 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005732
5733 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07005734 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05005735 if (ext4_handle_valid(handle) &&
5736 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005737 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005738 /*
5739 * We need extra buffer credits since we may write into EA block
5740 * with this same handle. If journal_extend fails, then it will
5741 * only result in a minor loss of functionality for that inode.
5742 * If this is felt to be critical, then e2fsck should be run to
5743 * force a large enough s_min_extra_isize.
5744 */
5745 if ((jbd2_journal_extend(handle,
5746 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5747 ret = ext4_expand_extra_isize(inode,
5748 sbi->s_want_extra_isize,
5749 iloc, handle);
5750 if (ret) {
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005751 ext4_set_inode_state(inode,
5752 EXT4_STATE_NO_EXPAND);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005753 if (mnt_count !=
5754 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05005755 ext4_warning(inode->i_sb,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005756 "Unable to expand inode %lu. Delete"
5757 " some EAs or run e2fsck.",
5758 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005759 mnt_count =
5760 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005761 }
5762 }
5763 }
5764 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005765 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07005766 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005767 return err;
5768}
5769
5770/*
Mingming Cao617ba132006-10-11 01:20:53 -07005771 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005772 *
5773 * We're really interested in the case where a file is being extended.
5774 * i_size has been changed by generic_commit_write() and we thus need
5775 * to include the updated inode in the current transaction.
5776 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005777 * Also, dquot_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005778 * are allocated to the file.
5779 *
5780 * If the inode is marked synchronous, we don't honour that here - doing
5781 * so would cause a commit on atime updates, which we don't bother doing.
5782 * We handle synchronous inodes at the highest possible level.
5783 */
Mingming Cao617ba132006-10-11 01:20:53 -07005784void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005785{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005786 handle_t *handle;
5787
Mingming Cao617ba132006-10-11 01:20:53 -07005788 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005789 if (IS_ERR(handle))
5790 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005791
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005792 ext4_mark_inode_dirty(handle, inode);
5793
Mingming Cao617ba132006-10-11 01:20:53 -07005794 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005795out:
5796 return;
5797}
5798
5799#if 0
5800/*
5801 * Bind an inode's backing buffer_head into this transaction, to prevent
5802 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07005803 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005804 * returns no iloc structure, so the caller needs to repeat the iloc
5805 * lookup to mark the inode dirty later.
5806 */
Mingming Cao617ba132006-10-11 01:20:53 -07005807static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005808{
Mingming Cao617ba132006-10-11 01:20:53 -07005809 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005810
5811 int err = 0;
5812 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07005813 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005814 if (!err) {
5815 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07005816 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005817 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05005818 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05005819 NULL,
Frank Mayhar03901312009-01-07 00:06:22 -05005820 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005821 brelse(iloc.bh);
5822 }
5823 }
Mingming Cao617ba132006-10-11 01:20:53 -07005824 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005825 return err;
5826}
5827#endif
5828
Mingming Cao617ba132006-10-11 01:20:53 -07005829int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005830{
5831 journal_t *journal;
5832 handle_t *handle;
5833 int err;
5834
5835 /*
5836 * We have to be very careful here: changing a data block's
5837 * journaling status dynamically is dangerous. If we write a
5838 * data block to the journal, change the status and then delete
5839 * that block, we risk forgetting to revoke the old log record
5840 * from the journal and so a subsequent replay can corrupt data.
5841 * So, first we make sure that the journal is empty and that
5842 * nobody is changing anything.
5843 */
5844
Mingming Cao617ba132006-10-11 01:20:53 -07005845 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005846 if (!journal)
5847 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04005848 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005849 return -EROFS;
5850
Mingming Caodab291a2006-10-11 01:21:01 -07005851 jbd2_journal_lock_updates(journal);
5852 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005853
5854 /*
5855 * OK, there are no updates running now, and all cached data is
5856 * synced to disk. We are now in a completely consistent state
5857 * which doesn't have anything in the journal, and we know that
5858 * no filesystem updates are running, so it is safe to modify
5859 * the inode's in-core data-journaling state flag now.
5860 */
5861
5862 if (val)
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005863 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005864 else
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005865 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Mingming Cao617ba132006-10-11 01:20:53 -07005866 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005867
Mingming Caodab291a2006-10-11 01:21:01 -07005868 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005869
5870 /* Finally we can mark the inode as dirty. */
5871
Mingming Cao617ba132006-10-11 01:20:53 -07005872 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005873 if (IS_ERR(handle))
5874 return PTR_ERR(handle);
5875
Mingming Cao617ba132006-10-11 01:20:53 -07005876 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005877 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07005878 ext4_journal_stop(handle);
5879 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005880
5881 return err;
5882}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005883
5884static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5885{
5886 return !buffer_mapped(bh);
5887}
5888
Nick Pigginc2ec1752009-03-31 15:23:21 -07005889int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005890{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005891 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005892 loff_t size;
5893 unsigned long len;
5894 int ret = -EINVAL;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005895 void *fsdata;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005896 struct file *file = vma->vm_file;
5897 struct inode *inode = file->f_path.dentry->d_inode;
5898 struct address_space *mapping = inode->i_mapping;
5899
5900 /*
5901 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5902 * get i_mutex because we are already holding mmap_sem.
5903 */
5904 down_read(&inode->i_alloc_sem);
5905 size = i_size_read(inode);
5906 if (page->mapping != mapping || size <= page_offset(page)
5907 || !PageUptodate(page)) {
5908 /* page got truncated from under us? */
5909 goto out_unlock;
5910 }
5911 ret = 0;
5912 if (PageMappedToDisk(page))
5913 goto out_unlock;
5914
5915 if (page->index == size >> PAGE_CACHE_SHIFT)
5916 len = size & ~PAGE_CACHE_MASK;
5917 else
5918 len = PAGE_CACHE_SIZE;
5919
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005920 lock_page(page);
5921 /*
5922 * return if we have all the buffers mapped. This avoid
5923 * the need to call write_begin/write_end which does a
5924 * journal_start/journal_stop which can block and take
5925 * long time
5926 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005927 if (page_has_buffers(page)) {
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005928 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005929 ext4_bh_unmapped)) {
5930 unlock_page(page);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005931 goto out_unlock;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005932 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005933 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005934 unlock_page(page);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005935 /*
5936 * OK, we need to fill the hole... Do write_begin write_end
5937 * to do block allocation/reservation.We are not holding
5938 * inode.i__mutex here. That allow * parallel write_begin,
5939 * write_end call. lock_page prevent this from happening
5940 * on the same page though
5941 */
5942 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005943 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005944 if (ret < 0)
5945 goto out_unlock;
5946 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005947 len, len, page, fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005948 if (ret < 0)
5949 goto out_unlock;
5950 ret = 0;
5951out_unlock:
Nick Pigginc2ec1752009-03-31 15:23:21 -07005952 if (ret)
5953 ret = VM_FAULT_SIGBUS;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005954 up_read(&inode->i_alloc_sem);
5955 return ret;
5956}