blob: 5c5bc5dafff818991b26f5fd7ebdf6ebf4e83c67 [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>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040041
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040042#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070043#include "xattr.h"
44#include "acl.h"
Mingming Caod2a17632008-07-14 17:52:37 -040045#include "ext4_extents.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070046
Theodore Ts'o9bffad12009-06-17 11:48:11 -040047#include <trace/events/ext4.h>
48
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040049#define MPAGE_DA_EXTENT_TAIL 0x01
50
Jan Kara678aaf42008-07-11 19:27:31 -040051static inline int ext4_begin_ordered_truncate(struct inode *inode,
52 loff_t new_size)
53{
Jan Kara7f5aa212009-02-10 11:15:34 -050054 return jbd2_journal_begin_ordered_truncate(
55 EXT4_SB(inode->i_sb)->s_journal,
56 &EXT4_I(inode)->jinode,
57 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -040058}
59
Alex Tomas64769242008-07-11 19:27:31 -040060static void ext4_invalidatepage(struct page *page, unsigned long offset);
61
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062/*
63 * Test whether an inode is a fast symlink.
64 */
Mingming Cao617ba132006-10-11 01:20:53 -070065static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066{
Mingming Cao617ba132006-10-11 01:20:53 -070067 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -070068 (inode->i_sb->s_blocksize >> 9) : 0;
69
70 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
71}
72
73/*
Mingming Cao617ba132006-10-11 01:20:53 -070074 * The ext4 forget function must perform a revoke if we are freeing data
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075 * which has been journaled. Metadata (eg. indirect blocks) must be
76 * revoked in all cases.
77 *
78 * "bh" may be NULL: a metadata block may have been freed from memory
79 * but there may still be a record of it in the journal, and that record
80 * still needs to be revoked.
Frank Mayhar03901312009-01-07 00:06:22 -050081 *
Curt Wohlgemuthe6b5d302009-07-13 09:07:20 -040082 * If the handle isn't valid we're not journaling, but we still need to
83 * call into ext4_journal_revoke() to put the buffer head.
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 */
Mingming Cao617ba132006-10-11 01:20:53 -070085int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -040086 struct buffer_head *bh, ext4_fsblk_t blocknr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070087{
88 int err;
89
90 might_sleep();
91
92 BUFFER_TRACE(bh, "enter");
93
94 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
Theodore Ts'o7f4520c2009-06-13 10:09:41 -040095 "data mode %x\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 bh, is_metadata, inode->i_mode,
97 test_opt(inode->i_sb, DATA_FLAGS));
98
99 /* Never use the revoke function if we are doing full data
100 * journaling: there is no need to, and a V1 superblock won't
101 * support it. Otherwise, only skip the revoke on un-journaled
102 * data blocks. */
103
Mingming Cao617ba132006-10-11 01:20:53 -0700104 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
105 (!is_metadata && !ext4_should_journal_data(inode))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 if (bh) {
Mingming Caodab291a2006-10-11 01:21:01 -0700107 BUFFER_TRACE(bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700108 return ext4_journal_forget(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 }
110 return 0;
111 }
112
113 /*
114 * data!=journal && (is_metadata || should_journal_data(inode))
115 */
Mingming Cao617ba132006-10-11 01:20:53 -0700116 BUFFER_TRACE(bh, "call ext4_journal_revoke");
117 err = ext4_journal_revoke(handle, blocknr, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700118 if (err)
Harvey Harrison46e665e2008-04-17 10:38:59 -0400119 ext4_abort(inode->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120 "error %d when attempting revoke", err);
121 BUFFER_TRACE(bh, "exit");
122 return err;
123}
124
125/*
126 * Work out how many blocks we need to proceed with the next chunk of a
127 * truncate transaction.
128 */
129static unsigned long blocks_for_truncate(struct inode *inode)
130{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500131 ext4_lblk_t needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700132
133 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
134
135 /* Give ourselves just enough room to cope with inodes in which
136 * i_blocks is corrupt: we've seen disk corruptions in the past
137 * which resulted in random data in an inode which looked enough
Mingming Cao617ba132006-10-11 01:20:53 -0700138 * like a regular file for ext4 to try to delete it. Things
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700139 * will go a bit crazy if that happens, but at least we should
140 * try not to panic the whole kernel. */
141 if (needed < 2)
142 needed = 2;
143
144 /* But we need to bound the transaction so we don't overflow the
145 * journal. */
Mingming Cao617ba132006-10-11 01:20:53 -0700146 if (needed > EXT4_MAX_TRANS_DATA)
147 needed = EXT4_MAX_TRANS_DATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700148
Mingming Cao617ba132006-10-11 01:20:53 -0700149 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700150}
151
152/*
153 * Truncate transactions can be complex and absolutely huge. So we need to
154 * be able to restart the transaction at a conventient checkpoint to make
155 * sure we don't overflow the journal.
156 *
157 * start_transaction gets us a new handle for a truncate transaction,
158 * and extend_transaction tries to extend the existing one a bit. If
159 * extend fails, we need to propagate the failure up and restart the
160 * transaction in the top-level truncate loop. --sct
161 */
162static handle_t *start_transaction(struct inode *inode)
163{
164 handle_t *result;
165
Mingming Cao617ba132006-10-11 01:20:53 -0700166 result = ext4_journal_start(inode, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700167 if (!IS_ERR(result))
168 return result;
169
Mingming Cao617ba132006-10-11 01:20:53 -0700170 ext4_std_error(inode->i_sb, PTR_ERR(result));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700171 return result;
172}
173
174/*
175 * Try to extend this transaction for the purposes of truncation.
176 *
177 * Returns 0 if we managed to create more room. If we can't create more
178 * room, and the transaction must be restarted we return 1.
179 */
180static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
181{
Frank Mayhar03901312009-01-07 00:06:22 -0500182 if (!ext4_handle_valid(handle))
183 return 0;
184 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700185 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700186 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700187 return 0;
188 return 1;
189}
190
191/*
192 * Restart the transaction associated with *handle. This does a commit,
193 * so before we call here everything must be consistently dirtied against
194 * this transaction.
195 */
Jan Kara487caee2009-08-17 22:17:20 -0400196 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
197 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700198{
Jan Kara487caee2009-08-17 22:17:20 -0400199 int ret;
200
201 /*
202 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
203 * moment, get_block can be called only for blocks inside i_size since
204 * page cache has been already dropped and writes are blocked by
205 * i_mutex. So we can safely drop the i_data_sem here.
206 */
Frank Mayhar03901312009-01-07 00:06:22 -0500207 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700208 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400209 up_write(&EXT4_I(inode)->i_data_sem);
210 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
211 down_write(&EXT4_I(inode)->i_data_sem);
212
213 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214}
215
216/*
217 * Called at the last iput() if i_nlink is zero.
218 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400219void ext4_delete_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700220{
221 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400222 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700223
Jan Kara678aaf42008-07-11 19:27:31 -0400224 if (ext4_should_order_data(inode))
225 ext4_begin_ordered_truncate(inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226 truncate_inode_pages(&inode->i_data, 0);
227
228 if (is_bad_inode(inode))
229 goto no_delete;
230
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400231 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400233 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 /*
235 * If we're going to skip the normal cleanup, we still need to
236 * make sure that the in-core orphan linked list is properly
237 * cleaned up.
238 */
Mingming Cao617ba132006-10-11 01:20:53 -0700239 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700240 goto no_delete;
241 }
242
243 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500244 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700245 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400246 err = ext4_mark_inode_dirty(handle, inode);
247 if (err) {
248 ext4_warning(inode->i_sb, __func__,
249 "couldn't mark inode dirty (err %d)", err);
250 goto stop_handle;
251 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700252 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700253 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400254
255 /*
256 * ext4_ext_truncate() doesn't reserve any slop when it
257 * restarts journal transactions; therefore there may not be
258 * enough credits left in the handle to remove the inode from
259 * the orphan list and set the dtime field.
260 */
Frank Mayhar03901312009-01-07 00:06:22 -0500261 if (!ext4_handle_has_enough_credits(handle, 3)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400262 err = ext4_journal_extend(handle, 3);
263 if (err > 0)
264 err = ext4_journal_restart(handle, 3);
265 if (err != 0) {
266 ext4_warning(inode->i_sb, __func__,
267 "couldn't extend journal (err %d)", err);
268 stop_handle:
269 ext4_journal_stop(handle);
270 goto no_delete;
271 }
272 }
273
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700274 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700275 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700276 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700277 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700278 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700279 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700280 * (Well, we could do this if we need to, but heck - it works)
281 */
Mingming Cao617ba132006-10-11 01:20:53 -0700282 ext4_orphan_del(handle, inode);
283 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700284
285 /*
286 * One subtle ordering requirement: if anything has gone wrong
287 * (transaction abort, IO errors, whatever), then we can still
288 * do these next steps (the fs will already have been marked as
289 * having errors), but we can't free the inode if the mark_dirty
290 * fails.
291 */
Mingming Cao617ba132006-10-11 01:20:53 -0700292 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700293 /* If that failed, just do the required in-core inode clear. */
294 clear_inode(inode);
295 else
Mingming Cao617ba132006-10-11 01:20:53 -0700296 ext4_free_inode(handle, inode);
297 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700298 return;
299no_delete:
300 clear_inode(inode); /* We must guarantee clearing of inode... */
301}
302
303typedef struct {
304 __le32 *p;
305 __le32 key;
306 struct buffer_head *bh;
307} Indirect;
308
309static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
310{
311 p->key = *(p->p = v);
312 p->bh = bh;
313}
314
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315/**
Mingming Cao617ba132006-10-11 01:20:53 -0700316 * ext4_block_to_path - parse the block number into array of offsets
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700317 * @inode: inode in question (we are only interested in its superblock)
318 * @i_block: block number to be parsed
319 * @offsets: array to store the offsets in
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400320 * @boundary: set this non-zero if the referred-to block is likely to be
321 * followed (on disk) by an indirect block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700322 *
Mingming Cao617ba132006-10-11 01:20:53 -0700323 * To store the locations of file's data ext4 uses a data structure common
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700324 * for UNIX filesystems - tree of pointers anchored in the inode, with
325 * data blocks at leaves and indirect blocks in intermediate nodes.
326 * This function translates the block number into path in that tree -
327 * return value is the path length and @offsets[n] is the offset of
328 * pointer to (n+1)th node in the nth one. If @block is out of range
329 * (negative or too large) warning is printed and zero returned.
330 *
331 * Note: function doesn't find node addresses, so no IO is needed. All
332 * we need to know is the capacity of indirect blocks (taken from the
333 * inode->i_sb).
334 */
335
336/*
337 * Portability note: the last comparison (check that we fit into triple
338 * indirect block) is spelled differently, because otherwise on an
339 * architecture with 32-bit longs and 8Kb pages we might get into trouble
340 * if our filesystem had 8Kb blocks. We might use long long, but that would
341 * kill us on x86. Oh, well, at least the sign propagation does not matter -
342 * i_block would have to be negative in the very beginning, so we would not
343 * get there at all.
344 */
345
Mingming Cao617ba132006-10-11 01:20:53 -0700346static int ext4_block_to_path(struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400347 ext4_lblk_t i_block,
348 ext4_lblk_t offsets[4], int *boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700349{
Mingming Cao617ba132006-10-11 01:20:53 -0700350 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
351 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
352 const long direct_blocks = EXT4_NDIR_BLOCKS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700353 indirect_blocks = ptrs,
354 double_blocks = (1 << (ptrs_bits * 2));
355 int n = 0;
356 int final = 0;
357
Roel Kluinc333e072009-08-10 22:47:22 -0400358 if (i_block < direct_blocks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700359 offsets[n++] = i_block;
360 final = direct_blocks;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400361 } else if ((i_block -= direct_blocks) < indirect_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700362 offsets[n++] = EXT4_IND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700363 offsets[n++] = i_block;
364 final = ptrs;
365 } else if ((i_block -= indirect_blocks) < double_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700366 offsets[n++] = EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700367 offsets[n++] = i_block >> ptrs_bits;
368 offsets[n++] = i_block & (ptrs - 1);
369 final = ptrs;
370 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
Mingming Cao617ba132006-10-11 01:20:53 -0700371 offsets[n++] = EXT4_TIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700372 offsets[n++] = i_block >> (ptrs_bits * 2);
373 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
374 offsets[n++] = i_block & (ptrs - 1);
375 final = ptrs;
376 } else {
Eric Sandeene2b46572008-01-28 23:58:27 -0500377 ext4_warning(inode->i_sb, "ext4_block_to_path",
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400378 "block %lu > max in inode %lu",
379 i_block + direct_blocks +
380 indirect_blocks + double_blocks, inode->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700381 }
382 if (boundary)
383 *boundary = final - 1 - (i_block & (ptrs - 1));
384 return n;
385}
386
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400387static int __ext4_check_blockref(const char *function, struct inode *inode,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400388 __le32 *p, unsigned int max)
389{
Thiemo Nagelf73953c2009-04-07 18:46:47 -0400390 __le32 *bref = p;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400391 unsigned int blk;
392
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400393 while (bref < p+max) {
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400394 blk = le32_to_cpu(*bref++);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400395 if (blk &&
396 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400397 blk, 1))) {
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400398 ext4_error(inode->i_sb, function,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400399 "invalid block reference %u "
400 "in inode #%lu", blk, inode->i_ino);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400401 return -EIO;
402 }
403 }
404 return 0;
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400405}
406
407
408#define ext4_check_indirect_blockref(inode, bh) \
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400409 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400410 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
411
412#define ext4_check_inode_blockref(inode) \
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400413 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400414 EXT4_NDIR_BLOCKS)
415
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700416/**
Mingming Cao617ba132006-10-11 01:20:53 -0700417 * ext4_get_branch - read the chain of indirect blocks leading to data
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700418 * @inode: inode in question
419 * @depth: depth of the chain (1 - direct pointer, etc.)
420 * @offsets: offsets of pointers in inode/indirect blocks
421 * @chain: place to store the result
422 * @err: here we store the error value
423 *
424 * Function fills the array of triples <key, p, bh> and returns %NULL
425 * if everything went OK or the pointer to the last filled triple
426 * (incomplete one) otherwise. Upon the return chain[i].key contains
427 * the number of (i+1)-th block in the chain (as it is stored in memory,
428 * i.e. little-endian 32-bit), chain[i].p contains the address of that
429 * number (it points into struct inode for i==0 and into the bh->b_data
430 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
431 * block for i>0 and NULL for i==0. In other words, it holds the block
432 * numbers of the chain, addresses they were taken from (and where we can
433 * verify that chain did not change) and buffer_heads hosting these
434 * numbers.
435 *
436 * Function stops when it stumbles upon zero pointer (absent block)
437 * (pointer to last triple returned, *@err == 0)
438 * or when it gets an IO error reading an indirect block
439 * (ditto, *@err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700440 * or when it reads all @depth-1 indirect blocks successfully and finds
441 * the whole chain, all way to the data (returns %NULL, *err == 0).
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500442 *
443 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500444 * down_read(&EXT4_I(inode)->i_data_sem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700445 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500446static Indirect *ext4_get_branch(struct inode *inode, int depth,
447 ext4_lblk_t *offsets,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700448 Indirect chain[4], int *err)
449{
450 struct super_block *sb = inode->i_sb;
451 Indirect *p = chain;
452 struct buffer_head *bh;
453
454 *err = 0;
455 /* i_data is not going away, no lock needed */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400456 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700457 if (!p->key)
458 goto no_block;
459 while (--depth) {
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400460 bh = sb_getblk(sb, le32_to_cpu(p->key));
461 if (unlikely(!bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700462 goto failure;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400463
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400464 if (!bh_uptodate_or_lock(bh)) {
465 if (bh_submit_read(bh) < 0) {
466 put_bh(bh);
467 goto failure;
468 }
469 /* validate block references */
470 if (ext4_check_indirect_blockref(inode, bh)) {
471 put_bh(bh);
472 goto failure;
473 }
474 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400475
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400476 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700477 /* Reader: end */
478 if (!p->key)
479 goto no_block;
480 }
481 return NULL;
482
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700483failure:
484 *err = -EIO;
485no_block:
486 return p;
487}
488
489/**
Mingming Cao617ba132006-10-11 01:20:53 -0700490 * ext4_find_near - find a place for allocation with sufficient locality
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700491 * @inode: owner
492 * @ind: descriptor of indirect block.
493 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000494 * This function returns the preferred place for block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495 * It is used when heuristic for sequential allocation fails.
496 * Rules are:
497 * + if there is a block to the left of our position - allocate near it.
498 * + if pointer will live in indirect block - allocate near that block.
499 * + if pointer will live in inode - allocate in the same
500 * cylinder group.
501 *
502 * In the latter case we colour the starting block by the callers PID to
503 * prevent it from clashing with concurrent allocations for a different inode
504 * in the same block group. The PID is used here so that functionally related
505 * files will be close-by on-disk.
506 *
507 * Caller must make sure that @ind is valid and will stay that way.
508 */
Mingming Cao617ba132006-10-11 01:20:53 -0700509static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700510{
Mingming Cao617ba132006-10-11 01:20:53 -0700511 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400512 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700513 __le32 *p;
Mingming Cao617ba132006-10-11 01:20:53 -0700514 ext4_fsblk_t bg_start;
Valerie Clement74d34872008-02-15 13:43:07 -0500515 ext4_fsblk_t last_block;
Mingming Cao617ba132006-10-11 01:20:53 -0700516 ext4_grpblk_t colour;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400517 ext4_group_t block_group;
518 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700519
520 /* Try to find previous block */
521 for (p = ind->p - 1; p >= start; p--) {
522 if (*p)
523 return le32_to_cpu(*p);
524 }
525
526 /* No such thing, so let's try location of indirect block */
527 if (ind->bh)
528 return ind->bh->b_blocknr;
529
530 /*
531 * It is going to be referred to from the inode itself? OK, just put it
532 * into the same cylinder group then.
533 */
Theodore Ts'oa4912122009-03-12 12:18:34 -0400534 block_group = ei->i_block_group;
535 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
536 block_group &= ~(flex_size-1);
537 if (S_ISREG(inode->i_mode))
538 block_group++;
539 }
540 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
Valerie Clement74d34872008-02-15 13:43:07 -0500541 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
542
Theodore Ts'oa4912122009-03-12 12:18:34 -0400543 /*
544 * If we are doing delayed allocation, we don't need take
545 * colour into account.
546 */
547 if (test_opt(inode->i_sb, DELALLOC))
548 return bg_start;
549
Valerie Clement74d34872008-02-15 13:43:07 -0500550 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
551 colour = (current->pid % 16) *
Mingming Cao617ba132006-10-11 01:20:53 -0700552 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Valerie Clement74d34872008-02-15 13:43:07 -0500553 else
554 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 return bg_start + colour;
556}
557
558/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000559 * ext4_find_goal - find a preferred place for allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700560 * @inode: owner
561 * @block: block we want
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700562 * @partial: pointer to the last triple within a chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700563 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000564 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800565 * returns it.
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400566 * Because this is only used for non-extent files, we limit the block nr
567 * to 32 bits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700568 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500569static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400570 Indirect *partial)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700571{
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400572 ext4_fsblk_t goal;
573
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700574 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400575 * XXX need to get goal block from mballoc's data structures
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700576 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400578 goal = ext4_find_near(inode, partial);
579 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
580 return goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581}
582
583/**
Mingming Cao617ba132006-10-11 01:20:53 -0700584 * ext4_blks_to_allocate: Look up the block map and count the number
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700585 * of direct blocks need to be allocated for the given branch.
586 *
587 * @branch: chain of indirect blocks
588 * @k: number of blocks need for indirect blocks
589 * @blks: number of data blocks to be mapped.
590 * @blocks_to_boundary: the offset in the indirect block
591 *
592 * return the total number of blocks to be allocate, including the
593 * direct and indirect blocks.
594 */
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500595static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400596 int blocks_to_boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700597{
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500598 unsigned int count = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700599
600 /*
601 * Simple case, [t,d]Indirect block(s) has not allocated yet
602 * then it's clear blocks on that path have not allocated
603 */
604 if (k > 0) {
605 /* right now we don't handle cross boundary allocation */
606 if (blks < blocks_to_boundary + 1)
607 count += blks;
608 else
609 count += blocks_to_boundary + 1;
610 return count;
611 }
612
613 count++;
614 while (count < blks && count <= blocks_to_boundary &&
615 le32_to_cpu(*(branch[0].p + count)) == 0) {
616 count++;
617 }
618 return count;
619}
620
621/**
Mingming Cao617ba132006-10-11 01:20:53 -0700622 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700623 * @indirect_blks: the number of blocks need to allocate for indirect
624 * blocks
625 *
626 * @new_blocks: on return it will store the new block numbers for
627 * the indirect blocks(if needed) and the first direct block,
628 * @blks: on return it will store the total number of allocated
629 * direct blocks
630 */
Mingming Cao617ba132006-10-11 01:20:53 -0700631static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400632 ext4_lblk_t iblock, ext4_fsblk_t goal,
633 int indirect_blks, int blks,
634 ext4_fsblk_t new_blocks[4], int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700635{
Theodore Ts'o815a1132009-01-01 23:59:43 -0500636 struct ext4_allocation_request ar;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700637 int target, i;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400638 unsigned long count = 0, blk_allocated = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700639 int index = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700640 ext4_fsblk_t current_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700641 int ret = 0;
642
643 /*
644 * Here we try to allocate the requested multiple blocks at once,
645 * on a best-effort basis.
646 * To build a branch, we should allocate blocks for
647 * the indirect blocks(if not allocated yet), and at least
648 * the first direct block of this branch. That's the
649 * minimum number of blocks need to allocate(required)
650 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400651 /* first we try to allocate the indirect blocks */
652 target = indirect_blks;
653 while (target > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700654 count = target;
655 /* allocating blocks for indirect blocks and direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400656 current_block = ext4_new_meta_blocks(handle, inode,
657 goal, &count, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700658 if (*err)
659 goto failed_out;
660
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400661 BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
662
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700663 target -= count;
664 /* allocate blocks for indirect blocks */
665 while (index < indirect_blks && count) {
666 new_blocks[index++] = current_block++;
667 count--;
668 }
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400669 if (count > 0) {
670 /*
671 * save the new block number
672 * for the first direct block
673 */
674 new_blocks[index] = current_block;
675 printk(KERN_INFO "%s returned more blocks than "
676 "requested\n", __func__);
677 WARN_ON(1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700678 break;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400679 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700680 }
681
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400682 target = blks - count ;
683 blk_allocated = count;
684 if (!target)
685 goto allocated;
686 /* Now allocate data blocks */
Theodore Ts'o815a1132009-01-01 23:59:43 -0500687 memset(&ar, 0, sizeof(ar));
688 ar.inode = inode;
689 ar.goal = goal;
690 ar.len = target;
691 ar.logical = iblock;
692 if (S_ISREG(inode->i_mode))
693 /* enable in-core preallocation only for regular files */
694 ar.flags = EXT4_MB_HINT_DATA;
695
696 current_block = ext4_mb_new_blocks(handle, &ar, err);
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400697 BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
Theodore Ts'o815a1132009-01-01 23:59:43 -0500698
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400699 if (*err && (target == blks)) {
700 /*
701 * if the allocation failed and we didn't allocate
702 * any blocks before
703 */
704 goto failed_out;
705 }
706 if (!*err) {
707 if (target == blks) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400708 /*
709 * save the new block number
710 * for the first direct block
711 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400712 new_blocks[index] = current_block;
713 }
Theodore Ts'o815a1132009-01-01 23:59:43 -0500714 blk_allocated += ar.len;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400715 }
716allocated:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700717 /* total number of blocks allocated for direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400718 ret = blk_allocated;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 *err = 0;
720 return ret;
721failed_out:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400722 for (i = 0; i < index; i++)
Alex Tomasc9de5602008-01-29 00:19:52 -0500723 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700724 return ret;
725}
726
727/**
Mingming Cao617ba132006-10-11 01:20:53 -0700728 * ext4_alloc_branch - allocate and set up a chain of blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700729 * @inode: owner
730 * @indirect_blks: number of allocated indirect blocks
731 * @blks: number of allocated direct blocks
732 * @offsets: offsets (in the blocks) to store the pointers to next.
733 * @branch: place to store the chain in.
734 *
735 * This function allocates blocks, zeroes out all but the last one,
736 * links them into chain and (if we are synchronous) writes them to disk.
737 * In other words, it prepares a branch that can be spliced onto the
738 * inode. It stores the information about that chain in the branch[], in
Mingming Cao617ba132006-10-11 01:20:53 -0700739 * the same format as ext4_get_branch() would do. We are calling it after
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700740 * we had read the existing part of chain and partial points to the last
741 * triple of that (one with zero ->key). Upon the exit we have the same
Mingming Cao617ba132006-10-11 01:20:53 -0700742 * picture as after the successful ext4_get_block(), except that in one
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700743 * place chain is disconnected - *branch->p is still zero (we did not
744 * set the last link), but branch->key contains the number that should
745 * be placed into *branch->p to fill that gap.
746 *
747 * If allocation fails we free all blocks we've allocated (and forget
748 * their buffer_heads) and return the error value the from failed
Mingming Cao617ba132006-10-11 01:20:53 -0700749 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700750 * as described above and return 0.
751 */
Mingming Cao617ba132006-10-11 01:20:53 -0700752static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400753 ext4_lblk_t iblock, int indirect_blks,
754 int *blks, ext4_fsblk_t goal,
755 ext4_lblk_t *offsets, Indirect *branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700756{
757 int blocksize = inode->i_sb->s_blocksize;
758 int i, n = 0;
759 int err = 0;
760 struct buffer_head *bh;
761 int num;
Mingming Cao617ba132006-10-11 01:20:53 -0700762 ext4_fsblk_t new_blocks[4];
763 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700764
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400765 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700766 *blks, new_blocks, &err);
767 if (err)
768 return err;
769
770 branch[0].key = cpu_to_le32(new_blocks[0]);
771 /*
772 * metadata blocks and data blocks are allocated.
773 */
774 for (n = 1; n <= indirect_blks; n++) {
775 /*
776 * Get buffer_head for parent block, zero it out
777 * and set the pointer to new one, then send
778 * parent to disk.
779 */
780 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
781 branch[n].bh = bh;
782 lock_buffer(bh);
783 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700784 err = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700785 if (err) {
Curt Wohlgemuth6487a9d2009-07-17 10:54:08 -0400786 /* Don't brelse(bh) here; it's done in
787 * ext4_journal_forget() below */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700788 unlock_buffer(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700789 goto failed;
790 }
791
792 memset(bh->b_data, 0, blocksize);
793 branch[n].p = (__le32 *) bh->b_data + offsets[n];
794 branch[n].key = cpu_to_le32(new_blocks[n]);
795 *branch[n].p = branch[n].key;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400796 if (n == indirect_blks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700797 current_block = new_blocks[n];
798 /*
799 * End of chain, update the last new metablock of
800 * the chain to point to the new allocated
801 * data blocks numbers
802 */
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400803 for (i = 1; i < num; i++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804 *(branch[n].p + i) = cpu_to_le32(++current_block);
805 }
806 BUFFER_TRACE(bh, "marking uptodate");
807 set_buffer_uptodate(bh);
808 unlock_buffer(bh);
809
Frank Mayhar03901312009-01-07 00:06:22 -0500810 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
811 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700812 if (err)
813 goto failed;
814 }
815 *blks = num;
816 return err;
817failed:
818 /* Allocation failed, free what we already allocated */
819 for (i = 1; i <= n ; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700820 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700821 ext4_journal_forget(handle, branch[i].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700822 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400823 for (i = 0; i < indirect_blks; i++)
Alex Tomasc9de5602008-01-29 00:19:52 -0500824 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700825
Alex Tomasc9de5602008-01-29 00:19:52 -0500826 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700827
828 return err;
829}
830
831/**
Mingming Cao617ba132006-10-11 01:20:53 -0700832 * ext4_splice_branch - splice the allocated branch onto inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700833 * @inode: owner
834 * @block: (logical) number of block we are adding
835 * @chain: chain of indirect blocks (with a missing link - see
Mingming Cao617ba132006-10-11 01:20:53 -0700836 * ext4_alloc_branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837 * @where: location of missing link
838 * @num: number of indirect blocks we are adding
839 * @blks: number of direct blocks we are adding
840 *
841 * This function fills the missing link and does all housekeeping needed in
842 * inode (->i_blocks, etc.). In case of success we end up with the full
843 * chain to new block and return 0.
844 */
Mingming Cao617ba132006-10-11 01:20:53 -0700845static int ext4_splice_branch(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400846 ext4_lblk_t block, Indirect *where, int num,
847 int blks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700848{
849 int i;
850 int err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700851 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700852
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700853 /*
854 * If we're splicing into a [td]indirect block (as opposed to the
855 * inode) then we need to get write access to the [td]indirect block
856 * before the splice.
857 */
858 if (where->bh) {
859 BUFFER_TRACE(where->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700860 err = ext4_journal_get_write_access(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861 if (err)
862 goto err_out;
863 }
864 /* That's it */
865
866 *where->p = where->key;
867
868 /*
869 * Update the host buffer_head or inode to point to more just allocated
870 * direct blocks blocks
871 */
872 if (num == 0 && blks > 1) {
873 current_block = le32_to_cpu(where->key) + 1;
874 for (i = 1; i < blks; i++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400875 *(where->p + i) = cpu_to_le32(current_block++);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700876 }
877
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700878 /* We are done with atomic stuff, now do the rest of housekeeping */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879 /* had we spliced it onto indirect block? */
880 if (where->bh) {
881 /*
882 * If we spliced it onto an indirect block, we haven't
883 * altered the inode. Note however that if it is being spliced
884 * onto an indirect block at the very end of the file (the
885 * file is growing) then we *will* alter the inode to reflect
886 * the new i_size. But that is not done here - it is done in
Mingming Cao617ba132006-10-11 01:20:53 -0700887 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888 */
889 jbd_debug(5, "splicing indirect only\n");
Frank Mayhar03901312009-01-07 00:06:22 -0500890 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
891 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700892 if (err)
893 goto err_out;
894 } else {
895 /*
896 * OK, we spliced it into the inode itself on a direct block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700897 */
Theodore Ts'o41591752009-06-15 03:41:23 -0400898 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700899 jbd_debug(5, "splicing direct\n");
900 }
901 return err;
902
903err_out:
904 for (i = 1; i <= num; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700905 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700906 ext4_journal_forget(handle, where[i].bh);
Alex Tomasc9de5602008-01-29 00:19:52 -0500907 ext4_free_blocks(handle, inode,
908 le32_to_cpu(where[i-1].key), 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909 }
Alex Tomasc9de5602008-01-29 00:19:52 -0500910 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911
912 return err;
913}
914
915/*
Theodore Ts'ob920c752009-05-14 00:54:29 -0400916 * The ext4_ind_get_blocks() function handles non-extents inodes
917 * (i.e., using the traditional indirect/double-indirect i_blocks
918 * scheme) for ext4_get_blocks().
919 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920 * Allocation strategy is simple: if we have to allocate something, we will
921 * have to go the whole way to leaf. So let's do it before attaching anything
922 * to tree, set linkage between the newborn blocks, write them if sync is
923 * required, recheck the path, free and repeat if check fails, otherwise
924 * set the last missing link (that will protect us from any truncate-generated
925 * removals - all blocks on the path are immune now) and possibly force the
926 * write on the parent block.
927 * That has a nice additional property: no special recovery from the failed
928 * allocations is needed - we simply release blocks and do not touch anything
929 * reachable from inode.
930 *
931 * `handle' can be NULL if create == 0.
932 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700933 * return > 0, # of blocks mapped or allocated.
934 * return = 0, if plain lookup failed.
935 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500936 *
Theodore Ts'ob920c752009-05-14 00:54:29 -0400937 * The ext4_ind_get_blocks() function should be called with
938 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
939 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
940 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
941 * blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700942 */
Theodore Ts'oe4d996c2009-05-12 00:25:28 -0400943static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400944 ext4_lblk_t iblock, unsigned int maxblocks,
945 struct buffer_head *bh_result,
946 int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700947{
948 int err = -EIO;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500949 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950 Indirect chain[4];
951 Indirect *partial;
Mingming Cao617ba132006-10-11 01:20:53 -0700952 ext4_fsblk_t goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953 int indirect_blks;
954 int blocks_to_boundary = 0;
955 int depth;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700956 int count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700957 ext4_fsblk_t first_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700958
Alex Tomasa86c6182006-10-11 01:21:03 -0700959 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
Theodore Ts'oc2177052009-05-14 00:58:52 -0400960 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500961 depth = ext4_block_to_path(inode, iblock, offsets,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400962 &blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700963
964 if (depth == 0)
965 goto out;
966
Mingming Cao617ba132006-10-11 01:20:53 -0700967 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700968
969 /* Simplest case - block found, no allocation needed */
970 if (!partial) {
971 first_block = le32_to_cpu(chain[depth - 1].key);
972 clear_buffer_new(bh_result);
973 count++;
974 /*map more blocks*/
975 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao617ba132006-10-11 01:20:53 -0700976 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700977
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 blk = le32_to_cpu(*(chain[depth-1].p + count));
979
980 if (blk == first_block + count)
981 count++;
982 else
983 break;
984 }
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500985 goto got_it;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700986 }
987
988 /* Next simple case - plain lookup or failed read of indirect block */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400989 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700990 goto cleanup;
991
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400993 * Okay, we need to do block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994 */
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800995 goal = ext4_find_goal(inode, iblock, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700996
997 /* the number of blocks need to allocate for [d,t]indirect blocks */
998 indirect_blks = (chain + depth) - partial - 1;
999
1000 /*
1001 * Next look up the indirect map to count the totoal number of
1002 * direct blocks to allocate for this branch.
1003 */
Mingming Cao617ba132006-10-11 01:20:53 -07001004 count = ext4_blks_to_allocate(partial, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001005 maxblocks, blocks_to_boundary);
1006 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001007 * Block out ext4_truncate while we alter the tree
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001008 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -04001009 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001010 &count, goal,
1011 offsets + (partial - chain), partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001012
1013 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001014 * The ext4_splice_branch call will free and forget any buffers
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015 * on the new chain if there is a failure, but that risks using
1016 * up transaction credits, especially for bitmaps where the
1017 * credits cannot be returned. Can we handle this somehow? We
1018 * may need to return -EAGAIN upwards in the worst case. --sct
1019 */
1020 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07001021 err = ext4_splice_branch(handle, inode, iblock,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001022 partial, indirect_blks, count);
1023 else
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001024 goto cleanup;
1025
1026 set_buffer_new(bh_result);
1027got_it:
1028 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1029 if (count > blocks_to_boundary)
1030 set_buffer_boundary(bh_result);
1031 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 }
1040 BUFFER_TRACE(bh_result, "returned");
1041out:
1042 return err;
1043}
1044
Mingming Cao60e58e02009-01-22 18:13:05 +01001045qsize_t ext4_get_reserved_space(struct inode *inode)
1046{
1047 unsigned long long total;
1048
1049 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1050 total = EXT4_I(inode)->i_reserved_data_blocks +
1051 EXT4_I(inode)->i_reserved_meta_blocks;
1052 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1053
1054 return total;
1055}
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001056/*
1057 * Calculate the number of metadata blocks need to reserve
1058 * to allocate @blocks for non extent file based file
1059 */
1060static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1061{
1062 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1063 int ind_blks, dind_blks, tind_blks;
1064
1065 /* number of new indirect blocks needed */
1066 ind_blks = (blocks + icap - 1) / icap;
1067
1068 dind_blks = (ind_blks + icap - 1) / icap;
1069
1070 tind_blks = 1;
1071
1072 return ind_blks + dind_blks + tind_blks;
1073}
1074
1075/*
1076 * Calculate the number of metadata blocks need to reserve
1077 * to allocate given number of blocks
1078 */
1079static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1080{
Mingming Caocd213222008-08-19 22:16:59 -04001081 if (!blocks)
1082 return 0;
1083
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001084 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1085 return ext4_ext_calc_metadata_amount(inode, blocks);
1086
1087 return ext4_indirect_calc_metadata_amount(inode, blocks);
1088}
1089
1090static void ext4_da_update_reserve_space(struct inode *inode, int used)
1091{
1092 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1093 int total, mdb, mdb_free;
1094
1095 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1096 /* recalculate the number of metablocks still need to be reserved */
1097 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1098 mdb = ext4_calc_metadata_amount(inode, total);
1099
1100 /* figure out how many metablocks to release */
1101 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1102 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1103
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001104 if (mdb_free) {
1105 /* Account for allocated meta_blocks */
1106 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001107
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001108 /* update fs dirty blocks counter */
1109 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1110 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1111 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1112 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001113
1114 /* update per-inode reservations */
1115 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1116 EXT4_I(inode)->i_reserved_data_blocks -= used;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001117 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001118
1119 /*
1120 * free those over-booking quota for metadata blocks
1121 */
Mingming Cao60e58e02009-01-22 18:13:05 +01001122 if (mdb_free)
1123 vfs_dq_release_reservation_block(inode, mdb_free);
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -04001124
1125 /*
1126 * If we have done all the pending block allocations and if
1127 * there aren't any writers on the inode, we can discard the
1128 * inode's preallocations.
1129 */
1130 if (!total && (atomic_read(&inode->i_writecount) == 0))
1131 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001132}
1133
Theodore Ts'o80e42462009-09-08 08:21:26 -04001134static int check_block_validity(struct inode *inode, const char *msg,
1135 sector_t logical, sector_t phys, int len)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001136{
1137 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
Theodore Ts'o80e42462009-09-08 08:21:26 -04001138 ext4_error(inode->i_sb, msg,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001139 "inode #%lu logical block %llu mapped to %llu "
1140 "(size %d)", inode->i_ino,
1141 (unsigned long long) logical,
1142 (unsigned long long) phys, len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001143 return -EIO;
1144 }
1145 return 0;
1146}
1147
Mingming Caof5ab0d12008-02-25 15:29:55 -05001148/*
Theodore Ts'o1f945332009-09-30 22:57:41 -04001149 * Return the number of contiguous dirty pages in a given inode
1150 * starting at page frame idx.
Theodore Ts'o55138e02009-09-29 13:31:31 -04001151 */
1152static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1153 unsigned int max_pages)
1154{
1155 struct address_space *mapping = inode->i_mapping;
1156 pgoff_t index;
1157 struct pagevec pvec;
1158 pgoff_t num = 0;
1159 int i, nr_pages, done = 0;
1160
1161 if (max_pages == 0)
1162 return 0;
1163 pagevec_init(&pvec, 0);
1164 while (!done) {
1165 index = idx;
1166 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1167 PAGECACHE_TAG_DIRTY,
1168 (pgoff_t)PAGEVEC_SIZE);
1169 if (nr_pages == 0)
1170 break;
1171 for (i = 0; i < nr_pages; i++) {
1172 struct page *page = pvec.pages[i];
1173 struct buffer_head *bh, *head;
1174
1175 lock_page(page);
1176 if (unlikely(page->mapping != mapping) ||
1177 !PageDirty(page) ||
1178 PageWriteback(page) ||
1179 page->index != idx) {
1180 done = 1;
1181 unlock_page(page);
1182 break;
1183 }
Theodore Ts'o1f945332009-09-30 22:57:41 -04001184 if (page_has_buffers(page)) {
1185 bh = head = page_buffers(page);
1186 do {
1187 if (!buffer_delay(bh) &&
1188 !buffer_unwritten(bh))
1189 done = 1;
1190 bh = bh->b_this_page;
1191 } while (!done && (bh != head));
1192 }
Theodore Ts'o55138e02009-09-29 13:31:31 -04001193 unlock_page(page);
1194 if (done)
1195 break;
1196 idx++;
1197 num++;
1198 if (num >= max_pages)
1199 break;
1200 }
1201 pagevec_release(&pvec);
1202 }
1203 return num;
1204}
1205
1206/*
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04001207 * The ext4_get_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001208 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -05001209 *
Mingming Caof5ab0d12008-02-25 15:29:55 -05001210 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1211 * and store the allocated blocks in the result buffer head and mark it
1212 * mapped.
1213 *
1214 * If file type is extents based, it will call ext4_ext_get_blocks(),
Theodore Ts'oe4d996c2009-05-12 00:25:28 -04001215 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -05001216 * based files
1217 *
1218 * On success, it returns the number of blocks being mapped or allocate.
1219 * if create==0 and the blocks are pre-allocated and uninitialized block,
1220 * the result buffer head is unmapped. If the create ==1, it will make sure
1221 * the buffer head is mapped.
1222 *
1223 * It returns 0 if plain look up failed (blocks have not been allocated), in
1224 * that casem, buffer head is unmapped
1225 *
1226 * It returns the error in case of allocation failure.
1227 */
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04001228int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1229 unsigned int max_blocks, struct buffer_head *bh,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001230 int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001231{
1232 int retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -05001233
1234 clear_buffer_mapped(bh);
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -04001235 clear_buffer_unwritten(bh);
Mingming Caof5ab0d12008-02-25 15:29:55 -05001236
Mingming Cao00314622009-09-28 15:49:08 -04001237 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1238 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1239 (unsigned long)block);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001240 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -04001241 * Try to see if we can get the block without requesting a new
1242 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001243 */
1244 down_read((&EXT4_I(inode)->i_data_sem));
1245 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1246 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001247 bh, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001248 } else {
Theodore Ts'oe4d996c2009-05-12 00:25:28 -04001249 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001250 bh, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001251 }
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001252 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -05001253
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001254 if (retval > 0 && buffer_mapped(bh)) {
Theodore Ts'o80e42462009-09-08 08:21:26 -04001255 int ret = check_block_validity(inode, "file system corruption",
1256 block, bh->b_blocknr, retval);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001257 if (ret != 0)
1258 return ret;
1259 }
1260
Mingming Caof5ab0d12008-02-25 15:29:55 -05001261 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -04001262 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001263 return retval;
1264
1265 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001266 * Returns if the blocks have already allocated
1267 *
1268 * Note that if blocks have been preallocated
1269 * ext4_ext_get_block() returns th create = 0
1270 * with buffer head unmapped.
1271 */
1272 if (retval > 0 && buffer_mapped(bh))
1273 return retval;
1274
1275 /*
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -04001276 * When we call get_blocks without the create flag, the
1277 * BH_Unwritten flag could have gotten set if the blocks
1278 * requested were part of a uninitialized extent. We need to
1279 * clear this flag now that we are committed to convert all or
1280 * part of the uninitialized extent to be an initialized
1281 * extent. This is because we need to avoid the combination
1282 * of BH_Unwritten and BH_Mapped flags being simultaneously
1283 * set on the buffer_head.
1284 */
1285 clear_buffer_unwritten(bh);
1286
1287 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001288 * New blocks allocate and/or writing to uninitialized extent
1289 * will possibly result in updating i_data, so we take
1290 * the write lock of i_data_sem, and call get_blocks()
1291 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001292 */
1293 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -04001294
1295 /*
1296 * if the caller is from delayed allocation writeout path
1297 * we have already reserved fs blocks for allocation
1298 * let the underlying get_block() function know to
1299 * avoid double accounting
1300 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04001301 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Mingming Caod2a17632008-07-14 17:52:37 -04001302 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001303 /*
1304 * We need to check for EXT4 here because migrate
1305 * could have changed the inode type in between
1306 */
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001307 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1308 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001309 bh, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001310 } else {
Theodore Ts'oe4d996c2009-05-12 00:25:28 -04001311 retval = ext4_ind_get_blocks(handle, inode, block,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001312 max_blocks, bh, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001313
1314 if (retval > 0 && buffer_new(bh)) {
1315 /*
1316 * We allocated new blocks which will result in
1317 * i_data's format changing. Force the migrate
1318 * to fail by clearing migrate flags
1319 */
Theodore Ts'o1b9c12f2009-09-17 08:32:22 -04001320 EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001321 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001322 }
Mingming Caod2a17632008-07-14 17:52:37 -04001323
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001324 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Mingming Caod2a17632008-07-14 17:52:37 -04001325 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001326
1327 /*
1328 * Update reserved blocks/metadata blocks after successful
1329 * block allocation which had been deferred till now.
1330 */
1331 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1332 ext4_da_update_reserve_space(inode, retval);
Mingming Caod2a17632008-07-14 17:52:37 -04001333
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001334 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001335 if (retval > 0 && buffer_mapped(bh)) {
Theodore Ts'o80e42462009-09-08 08:21:26 -04001336 int ret = check_block_validity(inode, "file system "
1337 "corruption after allocation",
1338 block, bh->b_blocknr, retval);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001339 if (ret != 0)
1340 return ret;
1341 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001342 return retval;
1343}
1344
Mingming Caof3bd1f32008-08-19 22:16:03 -04001345/* Maximum number of blocks we map for direct IO at once. */
1346#define DIO_MAX_BLOCKS 4096
1347
Eric Sandeen6873fa02008-10-07 00:46:36 -04001348int ext4_get_block(struct inode *inode, sector_t iblock,
1349 struct buffer_head *bh_result, int create)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001350{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001351 handle_t *handle = ext4_journal_current_handle();
Jan Kara7fb54092008-02-10 01:08:38 -05001352 int ret = 0, started = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001353 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001354 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001355
Jan Kara7fb54092008-02-10 01:08:38 -05001356 if (create && !handle) {
1357 /* Direct IO write... */
1358 if (max_blocks > DIO_MAX_BLOCKS)
1359 max_blocks = DIO_MAX_BLOCKS;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001360 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1361 handle = ext4_journal_start(inode, dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -05001362 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001363 ret = PTR_ERR(handle);
Jan Kara7fb54092008-02-10 01:08:38 -05001364 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001365 }
Jan Kara7fb54092008-02-10 01:08:38 -05001366 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001367 }
1368
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04001369 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001370 create ? EXT4_GET_BLOCKS_CREATE : 0);
Jan Kara7fb54092008-02-10 01:08:38 -05001371 if (ret > 0) {
1372 bh_result->b_size = (ret << inode->i_blkbits);
1373 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001374 }
Jan Kara7fb54092008-02-10 01:08:38 -05001375 if (started)
1376 ext4_journal_stop(handle);
1377out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001378 return ret;
1379}
1380
1381/*
1382 * `handle' can be NULL if create is zero
1383 */
Mingming Cao617ba132006-10-11 01:20:53 -07001384struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001385 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001386{
1387 struct buffer_head dummy;
1388 int fatal = 0, err;
Jan Kara03f5d8b2009-06-09 00:17:05 -04001389 int flags = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001390
1391 J_ASSERT(handle != NULL || create == 0);
1392
1393 dummy.b_state = 0;
1394 dummy.b_blocknr = -1000;
1395 buffer_trace_init(&dummy.b_history);
Theodore Ts'oc2177052009-05-14 00:58:52 -04001396 if (create)
1397 flags |= EXT4_GET_BLOCKS_CREATE;
1398 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001399 /*
Theodore Ts'oc2177052009-05-14 00:58:52 -04001400 * ext4_get_blocks() returns number of blocks mapped. 0 in
1401 * case of a HOLE.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001402 */
1403 if (err > 0) {
1404 if (err > 1)
1405 WARN_ON(1);
1406 err = 0;
1407 }
1408 *errp = err;
1409 if (!err && buffer_mapped(&dummy)) {
1410 struct buffer_head *bh;
1411 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1412 if (!bh) {
1413 *errp = -EIO;
1414 goto err;
1415 }
1416 if (buffer_new(&dummy)) {
1417 J_ASSERT(create != 0);
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -04001418 J_ASSERT(handle != NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001419
1420 /*
1421 * Now that we do not always journal data, we should
1422 * keep in mind whether this should always journal the
1423 * new buffer as metadata. For now, regular file
Mingming Cao617ba132006-10-11 01:20:53 -07001424 * writes use ext4_get_block instead, so it's not a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001425 * problem.
1426 */
1427 lock_buffer(bh);
1428 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001429 fatal = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001430 if (!fatal && !buffer_uptodate(bh)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001431 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001432 set_buffer_uptodate(bh);
1433 }
1434 unlock_buffer(bh);
Frank Mayhar03901312009-01-07 00:06:22 -05001435 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1436 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001437 if (!fatal)
1438 fatal = err;
1439 } else {
1440 BUFFER_TRACE(bh, "not a new buffer");
1441 }
1442 if (fatal) {
1443 *errp = fatal;
1444 brelse(bh);
1445 bh = NULL;
1446 }
1447 return bh;
1448 }
1449err:
1450 return NULL;
1451}
1452
Mingming Cao617ba132006-10-11 01:20:53 -07001453struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001454 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001456 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001457
Mingming Cao617ba132006-10-11 01:20:53 -07001458 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001459 if (!bh)
1460 return bh;
1461 if (buffer_uptodate(bh))
1462 return bh;
1463 ll_rw_block(READ_META, 1, &bh);
1464 wait_on_buffer(bh);
1465 if (buffer_uptodate(bh))
1466 return bh;
1467 put_bh(bh);
1468 *err = -EIO;
1469 return NULL;
1470}
1471
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001472static int walk_page_buffers(handle_t *handle,
1473 struct buffer_head *head,
1474 unsigned from,
1475 unsigned to,
1476 int *partial,
1477 int (*fn)(handle_t *handle,
1478 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001479{
1480 struct buffer_head *bh;
1481 unsigned block_start, block_end;
1482 unsigned blocksize = head->b_size;
1483 int err, ret = 0;
1484 struct buffer_head *next;
1485
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001486 for (bh = head, block_start = 0;
1487 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001488 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001489 next = bh->b_this_page;
1490 block_end = block_start + blocksize;
1491 if (block_end <= from || block_start >= to) {
1492 if (partial && !buffer_uptodate(bh))
1493 *partial = 1;
1494 continue;
1495 }
1496 err = (*fn)(handle, bh);
1497 if (!ret)
1498 ret = err;
1499 }
1500 return ret;
1501}
1502
1503/*
1504 * To preserve ordering, it is essential that the hole instantiation and
1505 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -07001506 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -07001507 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001508 * prepare_write() is the right place.
1509 *
Mingming Cao617ba132006-10-11 01:20:53 -07001510 * Also, this function can nest inside ext4_writepage() ->
1511 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001512 * has generated enough buffer credits to do the whole page. So we won't
1513 * block on the journal in that case, which is good, because the caller may
1514 * be PF_MEMALLOC.
1515 *
Mingming Cao617ba132006-10-11 01:20:53 -07001516 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001517 * quota file writes. If we were to commit the transaction while thus
1518 * reentered, there can be a deadlock - we would be holding a quota
1519 * lock, and the commit would never complete if another thread had a
1520 * transaction open and was blocking on the quota lock - a ranking
1521 * violation.
1522 *
Mingming Caodab291a2006-10-11 01:21:01 -07001523 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001524 * will _not_ run commit under these circumstances because handle->h_ref
1525 * is elevated. We'll still have enough credits for the tiny quotafile
1526 * write.
1527 */
1528static int do_journal_get_write_access(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001529 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001530{
1531 if (!buffer_mapped(bh) || buffer_freed(bh))
1532 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001533 return ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001534}
1535
Nick Pigginbfc1af62007-10-16 01:25:05 -07001536static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001537 loff_t pos, unsigned len, unsigned flags,
1538 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001539{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001540 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001541 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001542 handle_t *handle;
1543 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001544 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001545 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001546 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001547
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001548 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001549 /*
1550 * Reserve one block more for addition to orphan list in case
1551 * we allocate blocks but write fails for some reason
1552 */
1553 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001554 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001555 from = pos & (PAGE_CACHE_SIZE - 1);
1556 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001557
1558retry:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001559 handle = ext4_journal_start(inode, needed_blocks);
1560 if (IS_ERR(handle)) {
1561 ret = PTR_ERR(handle);
1562 goto out;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001563 }
1564
Jan Karaebd36102009-02-22 21:09:59 -05001565 /* We cannot recurse into the filesystem as the transaction is already
1566 * started */
1567 flags |= AOP_FLAG_NOFS;
1568
Nick Piggin54566b22009-01-04 12:00:53 -08001569 page = grab_cache_page_write_begin(mapping, index, flags);
Jan Karacf108bc2008-07-11 19:27:31 -04001570 if (!page) {
1571 ext4_journal_stop(handle);
1572 ret = -ENOMEM;
1573 goto out;
1574 }
1575 *pagep = page;
1576
Nick Pigginbfc1af62007-10-16 01:25:05 -07001577 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
Jan Karaebd36102009-02-22 21:09:59 -05001578 ext4_get_block);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001579
1580 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001581 ret = walk_page_buffers(handle, page_buffers(page),
1582 from, to, NULL, do_journal_get_write_access);
1583 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001584
1585 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001586 unlock_page(page);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001587 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001588 /*
1589 * block_write_begin may have instantiated a few blocks
1590 * outside i_size. Trim these off again. Don't need
1591 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001592 *
1593 * Add inode to orphan list in case we crash before
1594 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001595 */
Jan Karaffacfa72009-07-13 16:22:22 -04001596 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001597 ext4_orphan_add(handle, inode);
1598
1599 ext4_journal_stop(handle);
1600 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001601 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001602 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001603 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001604 * still be on the orphan list; we need to
1605 * make sure the inode is removed from the
1606 * orphan list in that case.
1607 */
1608 if (inode->i_nlink)
1609 ext4_orphan_del(NULL, inode);
1610 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001611 }
1612
Mingming Cao617ba132006-10-11 01:20:53 -07001613 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001614 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001615out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001616 return ret;
1617}
1618
Nick Pigginbfc1af62007-10-16 01:25:05 -07001619/* For write_end() in data=journal mode */
1620static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001621{
1622 if (!buffer_mapped(bh) || buffer_freed(bh))
1623 return 0;
1624 set_buffer_uptodate(bh);
Frank Mayhar03901312009-01-07 00:06:22 -05001625 return ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001626}
1627
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001628static int ext4_generic_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001629 struct address_space *mapping,
1630 loff_t pos, unsigned len, unsigned copied,
1631 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001632{
1633 int i_size_changed = 0;
1634 struct inode *inode = mapping->host;
1635 handle_t *handle = ext4_journal_current_handle();
1636
1637 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1638
1639 /*
1640 * No need to use i_size_read() here, the i_size
1641 * cannot change under us because we hold i_mutex.
1642 *
1643 * But it's important to update i_size while still holding page lock:
1644 * page writeout could otherwise come in and zero beyond i_size.
1645 */
1646 if (pos + copied > inode->i_size) {
1647 i_size_write(inode, pos + copied);
1648 i_size_changed = 1;
1649 }
1650
1651 if (pos + copied > EXT4_I(inode)->i_disksize) {
1652 /* We need to mark inode dirty even if
1653 * new_i_size is less that inode->i_size
1654 * bu greater than i_disksize.(hint delalloc)
1655 */
1656 ext4_update_i_disksize(inode, (pos + copied));
1657 i_size_changed = 1;
1658 }
1659 unlock_page(page);
1660 page_cache_release(page);
1661
1662 /*
1663 * Don't mark the inode dirty under page lock. First, it unnecessarily
1664 * makes the holding time of page lock longer. Second, it forces lock
1665 * ordering of page lock and transaction start for journaling
1666 * filesystems.
1667 */
1668 if (i_size_changed)
1669 ext4_mark_inode_dirty(handle, inode);
1670
1671 return copied;
1672}
1673
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001674/*
1675 * We need to pick up the new inode size which generic_commit_write gave us
1676 * `file' can be NULL - eg, when called from page_symlink().
1677 *
Mingming Cao617ba132006-10-11 01:20:53 -07001678 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001679 * buffers are managed internally.
1680 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001681static int ext4_ordered_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001682 struct address_space *mapping,
1683 loff_t pos, unsigned len, unsigned copied,
1684 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001685{
Mingming Cao617ba132006-10-11 01:20:53 -07001686 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001687 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001688 int ret = 0, ret2;
1689
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001690 trace_ext4_ordered_write_end(inode, pos, len, copied);
Jan Kara678aaf42008-07-11 19:27:31 -04001691 ret = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001692
1693 if (ret == 0) {
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001694 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001695 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001696 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001697 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001698 /* if we have allocated more blocks and copied
1699 * less. We will have blocks allocated outside
1700 * inode->i_size. So truncate them
1701 */
1702 ext4_orphan_add(handle, inode);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001703 if (ret2 < 0)
1704 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001705 }
Mingming Cao617ba132006-10-11 01:20:53 -07001706 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001707 if (!ret)
1708 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001709
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001710 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001711 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001712 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001713 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001714 * on the orphan list; we need to make sure the inode
1715 * is removed from the orphan list in that case.
1716 */
1717 if (inode->i_nlink)
1718 ext4_orphan_del(NULL, inode);
1719 }
1720
1721
Nick Pigginbfc1af62007-10-16 01:25:05 -07001722 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001723}
1724
Nick Pigginbfc1af62007-10-16 01:25:05 -07001725static int ext4_writeback_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001726 struct address_space *mapping,
1727 loff_t pos, unsigned len, unsigned copied,
1728 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001729{
Mingming Cao617ba132006-10-11 01:20:53 -07001730 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001731 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001732 int ret = 0, ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001733
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001734 trace_ext4_writeback_write_end(inode, pos, len, copied);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001735 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001736 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001737 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001738 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001739 /* if we have allocated more blocks and copied
1740 * less. We will have blocks allocated outside
1741 * inode->i_size. So truncate them
1742 */
1743 ext4_orphan_add(handle, inode);
1744
Roel Kluinf8a87d82008-04-29 22:01:18 -04001745 if (ret2 < 0)
1746 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001747
Mingming Cao617ba132006-10-11 01:20:53 -07001748 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001749 if (!ret)
1750 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001751
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001752 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001753 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001754 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001755 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001756 * on the orphan list; we need to make sure the inode
1757 * is removed from the orphan list in that case.
1758 */
1759 if (inode->i_nlink)
1760 ext4_orphan_del(NULL, inode);
1761 }
1762
Nick Pigginbfc1af62007-10-16 01:25:05 -07001763 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764}
1765
Nick Pigginbfc1af62007-10-16 01:25:05 -07001766static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001767 struct address_space *mapping,
1768 loff_t pos, unsigned len, unsigned copied,
1769 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001770{
Mingming Cao617ba132006-10-11 01:20:53 -07001771 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001772 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001773 int ret = 0, ret2;
1774 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001775 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001776 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001777
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001778 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001779 from = pos & (PAGE_CACHE_SIZE - 1);
1780 to = from + len;
1781
1782 if (copied < len) {
1783 if (!PageUptodate(page))
1784 copied = 0;
1785 page_zero_new_buffers(page, from+copied, to);
1786 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001787
1788 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001789 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001790 if (!partial)
1791 SetPageUptodate(page);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001792 new_i_size = pos + copied;
1793 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001794 i_size_write(inode, pos+copied);
Mingming Cao617ba132006-10-11 01:20:53 -07001795 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001796 if (new_i_size > EXT4_I(inode)->i_disksize) {
1797 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001798 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001799 if (!ret)
1800 ret = ret2;
1801 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001802
Jan Karacf108bc2008-07-11 19:27:31 -04001803 unlock_page(page);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001804 page_cache_release(page);
Jan Karaffacfa72009-07-13 16:22:22 -04001805 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001806 /* if we have allocated more blocks and copied
1807 * less. We will have blocks allocated outside
1808 * inode->i_size. So truncate them
1809 */
1810 ext4_orphan_add(handle, inode);
1811
Mingming Cao617ba132006-10-11 01:20:53 -07001812 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001813 if (!ret)
1814 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001815 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001816 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001817 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001818 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001819 * on the orphan list; we need to make sure the inode
1820 * is removed from the orphan list in that case.
1821 */
1822 if (inode->i_nlink)
1823 ext4_orphan_del(NULL, inode);
1824 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001825
1826 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827}
Mingming Caod2a17632008-07-14 17:52:37 -04001828
1829static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1830{
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001831 int retries = 0;
Mingming Cao60e58e02009-01-22 18:13:05 +01001832 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1833 unsigned long md_needed, mdblocks, total = 0;
Mingming Caod2a17632008-07-14 17:52:37 -04001834
1835 /*
1836 * recalculate the amount of metadata blocks to reserve
1837 * in order to allocate nrblocks
1838 * worse case is one extent per block
1839 */
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001840repeat:
Mingming Caod2a17632008-07-14 17:52:37 -04001841 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1842 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1843 mdblocks = ext4_calc_metadata_amount(inode, total);
1844 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1845
1846 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1847 total = md_needed + nrblocks;
1848
Mingming Cao60e58e02009-01-22 18:13:05 +01001849 /*
1850 * Make quota reservation here to prevent quota overflow
1851 * later. Real quota accounting is done at pages writeout
1852 * time.
1853 */
1854 if (vfs_dq_reserve_block(inode, total)) {
1855 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1856 return -EDQUOT;
1857 }
1858
Aneesh Kumar K.Va30d542a2008-10-09 10:56:23 -04001859 if (ext4_claim_free_blocks(sbi, total)) {
Mingming Caod2a17632008-07-14 17:52:37 -04001860 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao9f0ccfd2009-09-28 15:49:52 -04001861 vfs_dq_release_reservation_block(inode, total);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001862 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1863 yield();
1864 goto repeat;
1865 }
Mingming Caod2a17632008-07-14 17:52:37 -04001866 return -ENOSPC;
1867 }
Mingming Caod2a17632008-07-14 17:52:37 -04001868 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1869 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1870
1871 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1872 return 0; /* success */
1873}
1874
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001875static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001876{
1877 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1878 int total, mdb, mdb_free, release;
1879
Mingming Caocd213222008-08-19 22:16:59 -04001880 if (!to_free)
1881 return; /* Nothing to release, exit */
1882
Mingming Caod2a17632008-07-14 17:52:37 -04001883 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001884
1885 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1886 /*
1887 * if there is no reserved blocks, but we try to free some
1888 * then the counter is messed up somewhere.
1889 * but since this function is called from invalidate
1890 * page, it's harmless to return without any action
1891 */
1892 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1893 "blocks for inode %lu, but there is no reserved "
1894 "data blocks\n", to_free, inode->i_ino);
1895 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1896 return;
1897 }
1898
Mingming Caod2a17632008-07-14 17:52:37 -04001899 /* recalculate the number of metablocks still need to be reserved */
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001900 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
Mingming Caod2a17632008-07-14 17:52:37 -04001901 mdb = ext4_calc_metadata_amount(inode, total);
1902
1903 /* figure out how many metablocks to release */
1904 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1905 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1906
Mingming Caod2a17632008-07-14 17:52:37 -04001907 release = to_free + mdb_free;
1908
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001909 /* update fs dirty blocks counter for truncate case */
1910 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
Mingming Caod2a17632008-07-14 17:52:37 -04001911
1912 /* update per-inode reservations */
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001913 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1914 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
Mingming Caod2a17632008-07-14 17:52:37 -04001915
1916 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1917 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
Mingming Caod2a17632008-07-14 17:52:37 -04001918 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001919
1920 vfs_dq_release_reservation_block(inode, release);
Mingming Caod2a17632008-07-14 17:52:37 -04001921}
1922
1923static void ext4_da_page_release_reservation(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001924 unsigned long offset)
Mingming Caod2a17632008-07-14 17:52:37 -04001925{
1926 int to_release = 0;
1927 struct buffer_head *head, *bh;
1928 unsigned int curr_off = 0;
1929
1930 head = page_buffers(page);
1931 bh = head;
1932 do {
1933 unsigned int next_off = curr_off + bh->b_size;
1934
1935 if ((offset <= curr_off) && (buffer_delay(bh))) {
1936 to_release++;
1937 clear_buffer_delay(bh);
1938 }
1939 curr_off = next_off;
1940 } while ((bh = bh->b_this_page) != head);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001941 ext4_da_release_space(page->mapping->host, to_release);
Mingming Caod2a17632008-07-14 17:52:37 -04001942}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001943
1944/*
Alex Tomas64769242008-07-11 19:27:31 -04001945 * Delayed allocation stuff
1946 */
1947
Alex Tomas64769242008-07-11 19:27:31 -04001948/*
1949 * mpage_da_submit_io - walks through extent of pages and try to write
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001950 * them with writepage() call back
Alex Tomas64769242008-07-11 19:27:31 -04001951 *
1952 * @mpd->inode: inode
1953 * @mpd->first_page: first page of the extent
1954 * @mpd->next_page: page after the last page of the extent
Alex Tomas64769242008-07-11 19:27:31 -04001955 *
1956 * By the time mpage_da_submit_io() is called we expect all blocks
1957 * to be allocated. this may be wrong if allocation failed.
1958 *
1959 * As pages are already locked by write_cache_pages(), we can't use it
1960 */
1961static int mpage_da_submit_io(struct mpage_da_data *mpd)
1962{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001963 long pages_skipped;
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001964 struct pagevec pvec;
1965 unsigned long index, end;
1966 int ret = 0, err, nr_pages, i;
1967 struct inode *inode = mpd->inode;
1968 struct address_space *mapping = inode->i_mapping;
Alex Tomas64769242008-07-11 19:27:31 -04001969
1970 BUG_ON(mpd->next_page <= mpd->first_page);
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001971 /*
1972 * We need to start from the first_page to the next_page - 1
1973 * to make sure we also write the mapped dirty buffer_heads.
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001974 * If we look at mpd->b_blocknr we would only be looking
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001975 * at the currently mapped buffer_heads.
1976 */
Alex Tomas64769242008-07-11 19:27:31 -04001977 index = mpd->first_page;
1978 end = mpd->next_page - 1;
1979
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001980 pagevec_init(&pvec, 0);
Alex Tomas64769242008-07-11 19:27:31 -04001981 while (index <= end) {
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001982 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
Alex Tomas64769242008-07-11 19:27:31 -04001983 if (nr_pages == 0)
1984 break;
1985 for (i = 0; i < nr_pages; i++) {
1986 struct page *page = pvec.pages[i];
1987
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001988 index = page->index;
1989 if (index > end)
1990 break;
1991 index++;
1992
1993 BUG_ON(!PageLocked(page));
1994 BUG_ON(PageWriteback(page));
1995
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001996 pages_skipped = mpd->wbc->pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001997 err = mapping->a_ops->writepage(page, mpd->wbc);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001998 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1999 /*
2000 * have successfully written the page
2001 * without skipping the same
2002 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002003 mpd->pages_written++;
Alex Tomas64769242008-07-11 19:27:31 -04002004 /*
2005 * In error case, we have to continue because
2006 * remaining pages are still locked
2007 * XXX: unlock and re-dirty them?
2008 */
2009 if (ret == 0)
2010 ret = err;
2011 }
2012 pagevec_release(&pvec);
2013 }
Alex Tomas64769242008-07-11 19:27:31 -04002014 return ret;
2015}
2016
2017/*
2018 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2019 *
2020 * @mpd->inode - inode to walk through
2021 * @exbh->b_blocknr - first block on a disk
2022 * @exbh->b_size - amount of space in bytes
2023 * @logical - first logical block to start assignment with
2024 *
2025 * the function goes through all passed space and put actual disk
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002026 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
Alex Tomas64769242008-07-11 19:27:31 -04002027 */
2028static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2029 struct buffer_head *exbh)
2030{
2031 struct inode *inode = mpd->inode;
2032 struct address_space *mapping = inode->i_mapping;
2033 int blocks = exbh->b_size >> inode->i_blkbits;
2034 sector_t pblock = exbh->b_blocknr, cur_logical;
2035 struct buffer_head *head, *bh;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002036 pgoff_t index, end;
Alex Tomas64769242008-07-11 19:27:31 -04002037 struct pagevec pvec;
2038 int nr_pages, i;
2039
2040 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2041 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2042 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2043
2044 pagevec_init(&pvec, 0);
2045
2046 while (index <= end) {
2047 /* XXX: optimize tail */
2048 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2049 if (nr_pages == 0)
2050 break;
2051 for (i = 0; i < nr_pages; i++) {
2052 struct page *page = pvec.pages[i];
2053
2054 index = page->index;
2055 if (index > end)
2056 break;
2057 index++;
2058
2059 BUG_ON(!PageLocked(page));
2060 BUG_ON(PageWriteback(page));
2061 BUG_ON(!page_has_buffers(page));
2062
2063 bh = page_buffers(page);
2064 head = bh;
2065
2066 /* skip blocks out of the range */
2067 do {
2068 if (cur_logical >= logical)
2069 break;
2070 cur_logical++;
2071 } while ((bh = bh->b_this_page) != head);
2072
2073 do {
2074 if (cur_logical >= logical + blocks)
2075 break;
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002076
2077 if (buffer_delay(bh) ||
2078 buffer_unwritten(bh)) {
2079
2080 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2081
2082 if (buffer_delay(bh)) {
2083 clear_buffer_delay(bh);
2084 bh->b_blocknr = pblock;
2085 } else {
2086 /*
2087 * unwritten already should have
2088 * blocknr assigned. Verify that
2089 */
2090 clear_buffer_unwritten(bh);
2091 BUG_ON(bh->b_blocknr != pblock);
2092 }
2093
Mingming Cao61628a32008-07-11 19:27:31 -04002094 } else if (buffer_mapped(bh))
Alex Tomas64769242008-07-11 19:27:31 -04002095 BUG_ON(bh->b_blocknr != pblock);
Alex Tomas64769242008-07-11 19:27:31 -04002096
2097 cur_logical++;
2098 pblock++;
2099 } while ((bh = bh->b_this_page) != head);
2100 }
2101 pagevec_release(&pvec);
2102 }
2103}
2104
2105
2106/*
2107 * __unmap_underlying_blocks - just a helper function to unmap
2108 * set of blocks described by @bh
2109 */
2110static inline void __unmap_underlying_blocks(struct inode *inode,
2111 struct buffer_head *bh)
2112{
2113 struct block_device *bdev = inode->i_sb->s_bdev;
2114 int blocks, i;
2115
2116 blocks = bh->b_size >> inode->i_blkbits;
2117 for (i = 0; i < blocks; i++)
2118 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2119}
2120
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002121static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2122 sector_t logical, long blk_cnt)
2123{
2124 int nr_pages, i;
2125 pgoff_t index, end;
2126 struct pagevec pvec;
2127 struct inode *inode = mpd->inode;
2128 struct address_space *mapping = inode->i_mapping;
2129
2130 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2131 end = (logical + blk_cnt - 1) >>
2132 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2133 while (index <= end) {
2134 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2135 if (nr_pages == 0)
2136 break;
2137 for (i = 0; i < nr_pages; i++) {
2138 struct page *page = pvec.pages[i];
2139 index = page->index;
2140 if (index > end)
2141 break;
2142 index++;
2143
2144 BUG_ON(!PageLocked(page));
2145 BUG_ON(PageWriteback(page));
2146 block_invalidatepage(page, 0);
2147 ClearPageUptodate(page);
2148 unlock_page(page);
2149 }
2150 }
2151 return;
2152}
2153
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002154static void ext4_print_free_blocks(struct inode *inode)
2155{
2156 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o16939182009-09-26 17:43:59 -04002157 printk(KERN_CRIT "Total free blocks count %lld\n",
2158 ext4_count_free_blocks(inode->i_sb));
2159 printk(KERN_CRIT "Free/Dirty block details\n");
2160 printk(KERN_CRIT "free_blocks=%lld\n",
2161 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2162 printk(KERN_CRIT "dirty_blocks=%lld\n",
2163 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2164 printk(KERN_CRIT "Block reservation details\n");
2165 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2166 EXT4_I(inode)->i_reserved_data_blocks);
2167 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2168 EXT4_I(inode)->i_reserved_meta_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002169 return;
2170}
2171
Theodore Ts'ob920c752009-05-14 00:54:29 -04002172/*
Alex Tomas64769242008-07-11 19:27:31 -04002173 * mpage_da_map_blocks - go through given space
2174 *
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002175 * @mpd - bh describing space
Alex Tomas64769242008-07-11 19:27:31 -04002176 *
2177 * The function skips space we know is already mapped to disk blocks.
2178 *
Alex Tomas64769242008-07-11 19:27:31 -04002179 */
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002180static int mpage_da_map_blocks(struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04002181{
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002182 int err, blks, get_blocks_flags;
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002183 struct buffer_head new;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002184 sector_t next = mpd->b_blocknr;
2185 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2186 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2187 handle_t *handle = NULL;
Alex Tomas64769242008-07-11 19:27:31 -04002188
2189 /*
2190 * We consider only non-mapped and non-allocated blocks
2191 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002192 if ((mpd->b_state & (1 << BH_Mapped)) &&
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002193 !(mpd->b_state & (1 << BH_Delay)) &&
2194 !(mpd->b_state & (1 << BH_Unwritten)))
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002195 return 0;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002196
2197 /*
2198 * If we didn't accumulate anything to write simply return
2199 */
2200 if (!mpd->b_size)
2201 return 0;
2202
2203 handle = ext4_journal_current_handle();
2204 BUG_ON(!handle);
2205
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04002206 /*
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002207 * Call ext4_get_blocks() to allocate any delayed allocation
2208 * blocks, or to convert an uninitialized extent to be
2209 * initialized (in the case where we have written into
2210 * one or more preallocated blocks).
2211 *
2212 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2213 * indicate that we are on the delayed allocation path. This
2214 * affects functions in many different parts of the allocation
2215 * call path. This flag exists primarily because we don't
2216 * want to change *many* call functions, so ext4_get_blocks()
2217 * will set the magic i_delalloc_reserved_flag once the
2218 * inode's allocation semaphore is taken.
2219 *
2220 * If the blocks in questions were delalloc blocks, set
2221 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2222 * variables are updated after the blocks have been allocated.
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04002223 */
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002224 new.b_state = 0;
2225 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2226 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2227 if (mpd->b_state & (1 << BH_Delay))
2228 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002229 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002230 &new, get_blocks_flags);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002231 if (blks < 0) {
2232 err = blks;
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002233 /*
2234 * If get block returns with error we simply
2235 * return. Later writepage will redirty the page and
2236 * writepages will find the dirty page again
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002237 */
2238 if (err == -EAGAIN)
2239 return 0;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002240
2241 if (err == -ENOSPC &&
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002242 ext4_count_free_blocks(mpd->inode->i_sb)) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002243 mpd->retval = err;
2244 return 0;
2245 }
2246
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002247 /*
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002248 * get block failure will cause us to loop in
2249 * writepages, because a_ops->writepage won't be able
2250 * to make progress. The page will be redirtied by
2251 * writepage and writepages will again try to write
2252 * the same.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002253 */
Theodore Ts'o16939182009-09-26 17:43:59 -04002254 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2255 "delayed block allocation failed for inode %lu at "
2256 "logical offset %llu with max blocks %zd with "
2257 "error %d\n", mpd->inode->i_ino,
2258 (unsigned long long) next,
2259 mpd->b_size >> mpd->inode->i_blkbits, err);
2260 printk(KERN_CRIT "This should not happen!! "
2261 "Data will be lost\n");
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002262 if (err == -ENOSPC) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002263 ext4_print_free_blocks(mpd->inode);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002264 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002265 /* invalidate all the pages */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002266 ext4_da_block_invalidatepages(mpd, next,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002267 mpd->b_size >> mpd->inode->i_blkbits);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002268 return err;
2269 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002270 BUG_ON(blks == 0);
2271
2272 new.b_size = (blks << mpd->inode->i_blkbits);
Alex Tomas64769242008-07-11 19:27:31 -04002273
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002274 if (buffer_new(&new))
2275 __unmap_underlying_blocks(mpd->inode, &new);
Alex Tomas64769242008-07-11 19:27:31 -04002276
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002277 /*
2278 * If blocks are delayed marked, we need to
2279 * put actual blocknr and drop delayed bit
2280 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002281 if ((mpd->b_state & (1 << BH_Delay)) ||
2282 (mpd->b_state & (1 << BH_Unwritten)))
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002283 mpage_put_bnr_to_bhs(mpd, next, &new);
2284
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002285 if (ext4_should_order_data(mpd->inode)) {
2286 err = ext4_jbd2_file_inode(handle, mpd->inode);
2287 if (err)
2288 return err;
2289 }
2290
2291 /*
Jan Kara03f5d8b2009-06-09 00:17:05 -04002292 * Update on-disk size along with block allocation.
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002293 */
2294 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2295 if (disksize > i_size_read(mpd->inode))
2296 disksize = i_size_read(mpd->inode);
2297 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2298 ext4_update_i_disksize(mpd->inode, disksize);
2299 return ext4_mark_inode_dirty(handle, mpd->inode);
2300 }
2301
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002302 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04002303}
2304
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04002305#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2306 (1 << BH_Delay) | (1 << BH_Unwritten))
Alex Tomas64769242008-07-11 19:27:31 -04002307
2308/*
2309 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2310 *
2311 * @mpd->lbh - extent of blocks
2312 * @logical - logical number of the block in the file
2313 * @bh - bh of the block (used to access block's state)
2314 *
2315 * the function is used to collect contig. blocks in same state
2316 */
2317static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002318 sector_t logical, size_t b_size,
2319 unsigned long b_state)
Alex Tomas64769242008-07-11 19:27:31 -04002320{
Alex Tomas64769242008-07-11 19:27:31 -04002321 sector_t next;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002322 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
Alex Tomas64769242008-07-11 19:27:31 -04002323
Mingming Cao525f4ed2008-08-19 22:15:58 -04002324 /* check if thereserved journal credits might overflow */
2325 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2326 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2327 /*
2328 * With non-extent format we are limited by the journal
2329 * credit available. Total credit needed to insert
2330 * nrblocks contiguous blocks is dependent on the
2331 * nrblocks. So limit nrblocks.
2332 */
2333 goto flush_it;
2334 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2335 EXT4_MAX_TRANS_DATA) {
2336 /*
2337 * Adding the new buffer_head would make it cross the
2338 * allowed limit for which we have journal credit
2339 * reserved. So limit the new bh->b_size
2340 */
2341 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2342 mpd->inode->i_blkbits;
2343 /* we will do mpage_da_submit_io in the next loop */
2344 }
2345 }
Alex Tomas64769242008-07-11 19:27:31 -04002346 /*
2347 * First block in the extent
2348 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002349 if (mpd->b_size == 0) {
2350 mpd->b_blocknr = logical;
2351 mpd->b_size = b_size;
2352 mpd->b_state = b_state & BH_FLAGS;
Alex Tomas64769242008-07-11 19:27:31 -04002353 return;
2354 }
2355
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002356 next = mpd->b_blocknr + nrblocks;
Alex Tomas64769242008-07-11 19:27:31 -04002357 /*
2358 * Can we merge the block to our big extent?
2359 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002360 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2361 mpd->b_size += b_size;
Alex Tomas64769242008-07-11 19:27:31 -04002362 return;
2363 }
2364
Mingming Cao525f4ed2008-08-19 22:15:58 -04002365flush_it:
Alex Tomas64769242008-07-11 19:27:31 -04002366 /*
2367 * We couldn't merge the block to our extent, so we
2368 * need to flush current extent and start new one
2369 */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002370 if (mpage_da_map_blocks(mpd) == 0)
2371 mpage_da_submit_io(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002372 mpd->io_done = 1;
2373 return;
Alex Tomas64769242008-07-11 19:27:31 -04002374}
2375
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002376static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002377{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002378 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002379}
2380
Alex Tomas64769242008-07-11 19:27:31 -04002381/*
2382 * __mpage_da_writepage - finds extent of pages and blocks
2383 *
2384 * @page: page to consider
2385 * @wbc: not used, we just follow rules
2386 * @data: context
2387 *
2388 * The function finds extents of pages and scan them for all blocks.
2389 */
2390static int __mpage_da_writepage(struct page *page,
2391 struct writeback_control *wbc, void *data)
2392{
2393 struct mpage_da_data *mpd = data;
2394 struct inode *inode = mpd->inode;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002395 struct buffer_head *bh, *head;
Alex Tomas64769242008-07-11 19:27:31 -04002396 sector_t logical;
2397
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002398 if (mpd->io_done) {
2399 /*
2400 * Rest of the page in the page_vec
2401 * redirty then and skip then. We will
Anand Gadiyarfd589a82009-07-16 17:13:03 +02002402 * try to write them again after
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002403 * starting a new transaction
2404 */
2405 redirty_page_for_writepage(wbc, page);
2406 unlock_page(page);
2407 return MPAGE_DA_EXTENT_TAIL;
2408 }
Alex Tomas64769242008-07-11 19:27:31 -04002409 /*
2410 * Can we merge this page to current extent?
2411 */
2412 if (mpd->next_page != page->index) {
2413 /*
2414 * Nope, we can't. So, we map non-allocated blocks
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002415 * and start IO on them using writepage()
Alex Tomas64769242008-07-11 19:27:31 -04002416 */
2417 if (mpd->next_page != mpd->first_page) {
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002418 if (mpage_da_map_blocks(mpd) == 0)
2419 mpage_da_submit_io(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002420 /*
2421 * skip rest of the page in the page_vec
2422 */
2423 mpd->io_done = 1;
2424 redirty_page_for_writepage(wbc, page);
2425 unlock_page(page);
2426 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002427 }
2428
2429 /*
2430 * Start next extent of pages ...
2431 */
2432 mpd->first_page = page->index;
2433
2434 /*
2435 * ... and blocks
2436 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002437 mpd->b_size = 0;
2438 mpd->b_state = 0;
2439 mpd->b_blocknr = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002440 }
2441
2442 mpd->next_page = page->index + 1;
2443 logical = (sector_t) page->index <<
2444 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2445
2446 if (!page_has_buffers(page)) {
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002447 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2448 (1 << BH_Dirty) | (1 << BH_Uptodate));
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002449 if (mpd->io_done)
2450 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002451 } else {
2452 /*
2453 * Page with regular buffer heads, just add all dirty ones
2454 */
2455 head = page_buffers(page);
2456 bh = head;
2457 do {
2458 BUG_ON(buffer_locked(bh));
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002459 /*
2460 * We need to try to allocate
2461 * unmapped blocks in the same page.
2462 * Otherwise we won't make progress
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002463 * with the page in ext4_writepage
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002464 */
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002465 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002466 mpage_add_bh_to_extent(mpd, logical,
2467 bh->b_size,
2468 bh->b_state);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002469 if (mpd->io_done)
2470 return MPAGE_DA_EXTENT_TAIL;
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002471 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2472 /*
2473 * mapped dirty buffer. We need to update
2474 * the b_state because we look at
2475 * b_state in mpage_da_map_blocks. We don't
2476 * update b_size because if we find an
2477 * unmapped buffer_head later we need to
2478 * use the b_state flag of that buffer_head.
2479 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002480 if (mpd->b_size == 0)
2481 mpd->b_state = bh->b_state & BH_FLAGS;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002482 }
Alex Tomas64769242008-07-11 19:27:31 -04002483 logical++;
2484 } while ((bh = bh->b_this_page) != head);
2485 }
2486
2487 return 0;
2488}
2489
2490/*
Theodore Ts'ob920c752009-05-14 00:54:29 -04002491 * This is a special get_blocks_t callback which is used by
2492 * ext4_da_write_begin(). It will either return mapped block or
2493 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002494 *
2495 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2496 * We also have b_blocknr = -1 and b_bdev initialized properly
2497 *
2498 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2499 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2500 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04002501 */
2502static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2503 struct buffer_head *bh_result, int create)
2504{
2505 int ret = 0;
Aneesh Kumar K.V33b98172009-05-12 14:40:37 -04002506 sector_t invalid_block = ~((sector_t) 0xffff);
2507
2508 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2509 invalid_block = ~0;
Alex Tomas64769242008-07-11 19:27:31 -04002510
2511 BUG_ON(create == 0);
2512 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2513
2514 /*
2515 * first, we need to know whether the block is allocated already
2516 * preallocated blocks are unmapped but should treated
2517 * the same as allocated blocks.
2518 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04002519 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
Mingming Caod2a17632008-07-14 17:52:37 -04002520 if ((ret == 0) && !buffer_delay(bh_result)) {
2521 /* the block isn't (pre)allocated yet, let's reserve space */
Alex Tomas64769242008-07-11 19:27:31 -04002522 /*
2523 * XXX: __block_prepare_write() unmaps passed block,
2524 * is it OK?
2525 */
Mingming Caod2a17632008-07-14 17:52:37 -04002526 ret = ext4_da_reserve_space(inode, 1);
2527 if (ret)
2528 /* not enough space to reserve */
2529 return ret;
2530
Aneesh Kumar K.V33b98172009-05-12 14:40:37 -04002531 map_bh(bh_result, inode->i_sb, invalid_block);
Alex Tomas64769242008-07-11 19:27:31 -04002532 set_buffer_new(bh_result);
2533 set_buffer_delay(bh_result);
2534 } else if (ret > 0) {
2535 bh_result->b_size = (ret << inode->i_blkbits);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002536 if (buffer_unwritten(bh_result)) {
2537 /* A delayed write to unwritten bh should
2538 * be marked new and mapped. Mapped ensures
2539 * that we don't do get_block multiple times
2540 * when we write to the same offset and new
2541 * ensures that we do proper zero out for
2542 * partial write.
2543 */
Aneesh Kumar K.V9c1ee182009-05-13 18:36:58 -04002544 set_buffer_new(bh_result);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002545 set_buffer_mapped(bh_result);
2546 }
Alex Tomas64769242008-07-11 19:27:31 -04002547 ret = 0;
2548 }
2549
2550 return ret;
2551}
Mingming Cao61628a32008-07-11 19:27:31 -04002552
Theodore Ts'ob920c752009-05-14 00:54:29 -04002553/*
2554 * This function is used as a standard get_block_t calback function
2555 * when there is no desire to allocate any blocks. It is used as a
2556 * callback function for block_prepare_write(), nobh_writepage(), and
2557 * block_write_full_page(). These functions should only try to map a
2558 * single block at a time.
2559 *
2560 * Since this function doesn't do block allocations even if the caller
2561 * requests it by passing in create=1, it is critically important that
2562 * any caller checks to make sure that any buffer heads are returned
2563 * by this function are either all already mapped or marked for
2564 * delayed allocation before calling nobh_writepage() or
2565 * block_write_full_page(). Otherwise, b_blocknr could be left
2566 * unitialized, and the page write functions will be taken by
2567 * surprise.
2568 */
2569static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002570 struct buffer_head *bh_result, int create)
2571{
2572 int ret = 0;
2573 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2574
Theodore Ts'oa2dc52b2009-05-12 13:51:29 -04002575 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2576
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002577 /*
2578 * we don't want to do block allocation in writepage
2579 * so call get_block_wrap with create = 0
2580 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04002581 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002582 if (ret > 0) {
2583 bh_result->b_size = (ret << inode->i_blkbits);
2584 ret = 0;
2585 }
2586 return ret;
Mingming Cao61628a32008-07-11 19:27:31 -04002587}
2588
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002589static int bget_one(handle_t *handle, struct buffer_head *bh)
2590{
2591 get_bh(bh);
2592 return 0;
2593}
2594
2595static int bput_one(handle_t *handle, struct buffer_head *bh)
2596{
2597 put_bh(bh);
2598 return 0;
2599}
2600
2601static int __ext4_journalled_writepage(struct page *page,
2602 struct writeback_control *wbc,
2603 unsigned int len)
2604{
2605 struct address_space *mapping = page->mapping;
2606 struct inode *inode = mapping->host;
2607 struct buffer_head *page_bufs;
2608 handle_t *handle = NULL;
2609 int ret = 0;
2610 int err;
2611
2612 page_bufs = page_buffers(page);
2613 BUG_ON(!page_bufs);
2614 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2615 /* As soon as we unlock the page, it can go away, but we have
2616 * references to buffers so we are safe */
2617 unlock_page(page);
2618
2619 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2620 if (IS_ERR(handle)) {
2621 ret = PTR_ERR(handle);
2622 goto out;
2623 }
2624
2625 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2626 do_journal_get_write_access);
2627
2628 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2629 write_end_fn);
2630 if (ret == 0)
2631 ret = err;
2632 err = ext4_journal_stop(handle);
2633 if (!ret)
2634 ret = err;
2635
2636 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2637 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2638out:
2639 return ret;
2640}
2641
Mingming Cao61628a32008-07-11 19:27:31 -04002642/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002643 * Note that we don't need to start a transaction unless we're journaling data
2644 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2645 * need to file the inode to the transaction's list in ordered mode because if
2646 * we are writing back data added by write(), the inode is already there and if
2647 * we are writing back data modified via mmap(), noone guarantees in which
2648 * transaction the data will hit the disk. In case we are journaling data, we
2649 * cannot start transaction directly because transaction start ranks above page
2650 * lock so we have to do some magic.
2651 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04002652 * This function can get called via...
2653 * - ext4_da_writepages after taking page lock (have journal handle)
2654 * - journal_submit_inode_data_buffers (no journal handle)
2655 * - shrink_page_list via pdflush (no journal handle)
2656 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002657 *
2658 * We don't do any block allocation in this function. If we have page with
2659 * multiple blocks we need to write those buffer_heads that are mapped. This
2660 * is important for mmaped based write. So if we do with blocksize 1K
2661 * truncate(f, 1024);
2662 * a = mmap(f, 0, 4096);
2663 * a[0] = 'a';
2664 * truncate(f, 4096);
2665 * we have in the page first buffer_head mapped via page_mkwrite call back
2666 * but other bufer_heads would be unmapped but dirty(dirty done via the
2667 * do_wp_page). So writepage should write the first block. If we modify
2668 * the mmap area beyond 1024 we will again get a page_fault and the
2669 * page_mkwrite callback will do the block allocation and mark the
2670 * buffer_heads mapped.
2671 *
2672 * We redirty the page if we have any buffer_heads that is either delay or
2673 * unwritten in the page.
2674 *
2675 * We can get recursively called as show below.
2676 *
2677 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2678 * ext4_writepage()
2679 *
2680 * But since we don't do any block allocation we should not deadlock.
2681 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04002682 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002683static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002684 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002685{
Alex Tomas64769242008-07-11 19:27:31 -04002686 int ret = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002687 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002688 unsigned int len;
Mingming Cao61628a32008-07-11 19:27:31 -04002689 struct buffer_head *page_bufs;
2690 struct inode *inode = page->mapping->host;
Alex Tomas64769242008-07-11 19:27:31 -04002691
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002692 trace_ext4_writepage(inode, page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002693 size = i_size_read(inode);
2694 if (page->index == size >> PAGE_CACHE_SHIFT)
2695 len = size & ~PAGE_CACHE_MASK;
2696 else
2697 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04002698
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002699 if (page_has_buffers(page)) {
Mingming Cao61628a32008-07-11 19:27:31 -04002700 page_bufs = page_buffers(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002701 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002702 ext4_bh_delay_or_unwritten)) {
Mingming Cao61628a32008-07-11 19:27:31 -04002703 /*
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002704 * We don't want to do block allocation
2705 * So redirty the page and return
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04002706 * We may reach here when we do a journal commit
2707 * via journal_submit_inode_data_buffers.
2708 * If we don't have mapping block we just ignore
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002709 * them. We can also reach here via shrink_page_list
2710 */
2711 redirty_page_for_writepage(wbc, page);
2712 unlock_page(page);
2713 return 0;
2714 }
2715 } else {
2716 /*
2717 * The test for page_has_buffers() is subtle:
2718 * We know the page is dirty but it lost buffers. That means
2719 * that at some moment in time after write_begin()/write_end()
2720 * has been called all buffers have been clean and thus they
2721 * must have been written at least once. So they are all
2722 * mapped and we can happily proceed with mapping them
2723 * and writing the page.
2724 *
2725 * Try to initialize the buffer_heads and check whether
2726 * all are mapped and non delay. We don't want to
2727 * do block allocation here.
2728 */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04002729 ret = block_prepare_write(page, 0, len,
Theodore Ts'ob920c752009-05-14 00:54:29 -04002730 noalloc_get_block_write);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002731 if (!ret) {
2732 page_bufs = page_buffers(page);
2733 /* check whether all are mapped and non delay */
2734 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002735 ext4_bh_delay_or_unwritten)) {
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002736 redirty_page_for_writepage(wbc, page);
2737 unlock_page(page);
2738 return 0;
2739 }
2740 } else {
2741 /*
2742 * We can't do block allocation here
2743 * so just redity the page and unlock
2744 * and return
Mingming Cao61628a32008-07-11 19:27:31 -04002745 */
Mingming Cao61628a32008-07-11 19:27:31 -04002746 redirty_page_for_writepage(wbc, page);
2747 unlock_page(page);
2748 return 0;
2749 }
Aneesh Kumar K.Ved9b3e32008-11-07 09:06:45 -05002750 /* now mark the buffer_heads as dirty and uptodate */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04002751 block_commit_write(page, 0, len);
Alex Tomas64769242008-07-11 19:27:31 -04002752 }
2753
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002754 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2755 /*
2756 * It's mmapped pagecache. Add buffers and journal it. There
2757 * doesn't seem much point in redirtying the page here.
2758 */
2759 ClearPageChecked(page);
2760 return __ext4_journalled_writepage(page, wbc, len);
2761 }
2762
Alex Tomas64769242008-07-11 19:27:31 -04002763 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
Theodore Ts'ob920c752009-05-14 00:54:29 -04002764 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002765 else
Theodore Ts'ob920c752009-05-14 00:54:29 -04002766 ret = block_write_full_page(page, noalloc_get_block_write,
2767 wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002768
Alex Tomas64769242008-07-11 19:27:31 -04002769 return ret;
2770}
2771
Mingming Cao61628a32008-07-11 19:27:31 -04002772/*
Mingming Cao525f4ed2008-08-19 22:15:58 -04002773 * This is called via ext4_da_writepages() to
2774 * calulate the total number of credits to reserve to fit
2775 * a single extent allocation into a single transaction,
2776 * ext4_da_writpeages() will loop calling this before
2777 * the block allocation.
Mingming Cao61628a32008-07-11 19:27:31 -04002778 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002779
2780static int ext4_da_writepages_trans_blocks(struct inode *inode)
2781{
2782 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2783
2784 /*
2785 * With non-extent format the journal credit needed to
2786 * insert nrblocks contiguous block is dependent on
2787 * number of contiguous block. So we will limit
2788 * number of contiguous block to a sane value
2789 */
2790 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2791 (max_blocks > EXT4_MAX_TRANS_DATA))
2792 max_blocks = EXT4_MAX_TRANS_DATA;
2793
2794 return ext4_chunk_trans_blocks(inode, max_blocks);
2795}
Mingming Cao61628a32008-07-11 19:27:31 -04002796
Alex Tomas64769242008-07-11 19:27:31 -04002797static int ext4_da_writepages(struct address_space *mapping,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002798 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002799{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002800 pgoff_t index;
2801 int range_whole = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002802 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002803 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002804 struct inode *inode = mapping->host;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002805 int no_nrwrite_index_update;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002806 int pages_written = 0;
2807 long pages_skipped;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002808 unsigned int max_pages;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002809 int range_cyclic, cycled = 1, io_done = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002810 int needed_blocks, ret = 0;
2811 long desired_nr_to_write, nr_to_writebump = 0;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002812 loff_t range_start = wbc->range_start;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002813 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Mingming Cao61628a32008-07-11 19:27:31 -04002814
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002815 trace_ext4_da_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002816
Mingming Cao61628a32008-07-11 19:27:31 -04002817 /*
2818 * No pages to write? This is mainly a kludge to avoid starting
2819 * a transaction for special inodes like journal inode on last iput()
2820 * because that could violate lock ordering on umount
2821 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002822 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Mingming Cao61628a32008-07-11 19:27:31 -04002823 return 0;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002824
2825 /*
2826 * If the filesystem has aborted, it is read-only, so return
2827 * right away instead of dumping stack traces later on that
2828 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002829 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002830 * the latter could be true if the filesystem is mounted
2831 * read-only, and in that case, ext4_da_writepages should
2832 * *never* be called, so if that ever happens, we would want
2833 * the stack trace.
2834 */
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002835 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002836 return -EROFS;
2837
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002838 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2839 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002840
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002841 range_cyclic = wbc->range_cyclic;
2842 if (wbc->range_cyclic) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002843 index = mapping->writeback_index;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002844 if (index)
2845 cycled = 0;
2846 wbc->range_start = index << PAGE_CACHE_SHIFT;
2847 wbc->range_end = LLONG_MAX;
2848 wbc->range_cyclic = 0;
2849 } else
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002850 index = wbc->range_start >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002851
Theodore Ts'o55138e02009-09-29 13:31:31 -04002852 /*
2853 * This works around two forms of stupidity. The first is in
2854 * the writeback code, which caps the maximum number of pages
2855 * written to be 1024 pages. This is wrong on multiple
2856 * levels; different architectues have a different page size,
2857 * which changes the maximum amount of data which gets
2858 * written. Secondly, 4 megabytes is way too small. XFS
2859 * forces this value to be 16 megabytes by multiplying
2860 * nr_to_write parameter by four, and then relies on its
2861 * allocator to allocate larger extents to make them
2862 * contiguous. Unfortunately this brings us to the second
2863 * stupidity, which is that ext4's mballoc code only allocates
2864 * at most 2048 blocks. So we force contiguous writes up to
2865 * the number of dirty blocks in the inode, or
2866 * sbi->max_writeback_mb_bump whichever is smaller.
2867 */
2868 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2869 if (!range_cyclic && range_whole)
2870 desired_nr_to_write = wbc->nr_to_write * 8;
2871 else
2872 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2873 max_pages);
2874 if (desired_nr_to_write > max_pages)
2875 desired_nr_to_write = max_pages;
2876
2877 if (wbc->nr_to_write < desired_nr_to_write) {
2878 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2879 wbc->nr_to_write = desired_nr_to_write;
2880 }
2881
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002882 mpd.wbc = wbc;
2883 mpd.inode = mapping->host;
2884
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002885 /*
2886 * we don't want write_cache_pages to update
2887 * nr_to_write and writeback_index
2888 */
2889 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2890 wbc->no_nrwrite_index_update = 1;
2891 pages_skipped = wbc->pages_skipped;
2892
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002893retry:
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002894 while (!ret && wbc->nr_to_write > 0) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002895
2896 /*
2897 * we insert one extent at a time. So we need
2898 * credit needed for single extent allocation.
2899 * journalled mode is currently not supported
2900 * by delalloc
2901 */
2902 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002903 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002904
Mingming Cao61628a32008-07-11 19:27:31 -04002905 /* start a new transaction*/
2906 handle = ext4_journal_start(inode, needed_blocks);
2907 if (IS_ERR(handle)) {
2908 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002909 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002910 "%ld pages, ino %lu; err %d\n", __func__,
2911 wbc->nr_to_write, inode->i_ino, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04002912 goto out_writepages;
2913 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002914
2915 /*
2916 * Now call __mpage_da_writepage to find the next
2917 * contiguous region of logical blocks that need
2918 * blocks to be allocated by ext4. We don't actually
2919 * submit the blocks for I/O here, even though
2920 * write_cache_pages thinks it will, and will set the
2921 * pages as clean for write before calling
2922 * __mpage_da_writepage().
2923 */
2924 mpd.b_size = 0;
2925 mpd.b_state = 0;
2926 mpd.b_blocknr = 0;
2927 mpd.first_page = 0;
2928 mpd.next_page = 0;
2929 mpd.io_done = 0;
2930 mpd.pages_written = 0;
2931 mpd.retval = 0;
2932 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2933 &mpd);
2934 /*
2935 * If we have a contigous extent of pages and we
2936 * haven't done the I/O yet, map the blocks and submit
2937 * them for I/O.
2938 */
2939 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2940 if (mpage_da_map_blocks(&mpd) == 0)
2941 mpage_da_submit_io(&mpd);
2942 mpd.io_done = 1;
2943 ret = MPAGE_DA_EXTENT_TAIL;
2944 }
Theodore Ts'ob3a3ca82009-08-31 23:13:11 -04002945 trace_ext4_da_write_pages(inode, &mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002946 wbc->nr_to_write -= mpd.pages_written;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002947
Mingming Cao61628a32008-07-11 19:27:31 -04002948 ext4_journal_stop(handle);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002949
Eric Sandeen8f64b322009-02-26 00:57:35 -05002950 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002951 /* commit the transaction which would
2952 * free blocks released in the transaction
2953 * and try again
2954 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002955 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002956 wbc->pages_skipped = pages_skipped;
2957 ret = 0;
2958 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002959 /*
2960 * got one extent now try with
2961 * rest of the pages
2962 */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002963 pages_written += mpd.pages_written;
2964 wbc->pages_skipped = pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002965 ret = 0;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002966 io_done = 1;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002967 } else if (wbc->nr_to_write)
Mingming Cao61628a32008-07-11 19:27:31 -04002968 /*
2969 * There is no more writeout needed
2970 * or we requested for a noblocking writeout
2971 * and we found the device congested
2972 */
Mingming Cao61628a32008-07-11 19:27:31 -04002973 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002974 }
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002975 if (!io_done && !cycled) {
2976 cycled = 1;
2977 index = 0;
2978 wbc->range_start = index << PAGE_CACHE_SHIFT;
2979 wbc->range_end = mapping->writeback_index - 1;
2980 goto retry;
2981 }
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002982 if (pages_skipped != wbc->pages_skipped)
Theodore Ts'o16939182009-09-26 17:43:59 -04002983 ext4_msg(inode->i_sb, KERN_CRIT,
2984 "This should not happen leaving %s "
2985 "with nr_to_write = %ld ret = %d\n",
2986 __func__, wbc->nr_to_write, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04002987
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002988 /* Update index */
2989 index += pages_written;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002990 wbc->range_cyclic = range_cyclic;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002991 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2992 /*
2993 * set the writeback_index so that range_cyclic
2994 * mode will write it back later
2995 */
2996 mapping->writeback_index = index;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002997
Mingming Cao61628a32008-07-11 19:27:31 -04002998out_writepages:
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002999 if (!no_nrwrite_index_update)
3000 wbc->no_nrwrite_index_update = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04003001 if (wbc->nr_to_write > nr_to_writebump)
3002 wbc->nr_to_write -= nr_to_writebump;
Theodore Ts'ode89de62009-08-31 17:00:59 -04003003 wbc->range_start = range_start;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003004 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
Mingming Cao61628a32008-07-11 19:27:31 -04003005 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04003006}
3007
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003008#define FALL_BACK_TO_NONDELALLOC 1
3009static int ext4_nonda_switch(struct super_block *sb)
3010{
3011 s64 free_blocks, dirty_blocks;
3012 struct ext4_sb_info *sbi = EXT4_SB(sb);
3013
3014 /*
3015 * switch to non delalloc mode if we are running low
3016 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08003017 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003018 * accumulated on each CPU without updating global counters
3019 * Delalloc need an accurate free block accounting. So switch
3020 * to non delalloc when we are near to error range.
3021 */
3022 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3023 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3024 if (2 * free_blocks < 3 * dirty_blocks ||
3025 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3026 /*
3027 * free block count is less that 150% of dirty blocks
3028 * or free blocks is less that watermark
3029 */
3030 return 1;
3031 }
3032 return 0;
3033}
3034
Alex Tomas64769242008-07-11 19:27:31 -04003035static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003036 loff_t pos, unsigned len, unsigned flags,
3037 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04003038{
Mingming Caod2a17632008-07-14 17:52:37 -04003039 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04003040 struct page *page;
3041 pgoff_t index;
3042 unsigned from, to;
3043 struct inode *inode = mapping->host;
3044 handle_t *handle;
3045
3046 index = pos >> PAGE_CACHE_SHIFT;
3047 from = pos & (PAGE_CACHE_SIZE - 1);
3048 to = from + len;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003049
3050 if (ext4_nonda_switch(inode->i_sb)) {
3051 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3052 return ext4_write_begin(file, mapping, pos,
3053 len, flags, pagep, fsdata);
3054 }
3055 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003056 trace_ext4_da_write_begin(inode, pos, len, flags);
Mingming Caod2a17632008-07-14 17:52:37 -04003057retry:
Alex Tomas64769242008-07-11 19:27:31 -04003058 /*
3059 * With delayed allocation, we don't log the i_disksize update
3060 * if there is delayed block allocation. But we still need
3061 * to journalling the i_disksize update if writes to the end
3062 * of file which has an already mapped buffer.
3063 */
3064 handle = ext4_journal_start(inode, 1);
3065 if (IS_ERR(handle)) {
3066 ret = PTR_ERR(handle);
3067 goto out;
3068 }
Jan Karaebd36102009-02-22 21:09:59 -05003069 /* We cannot recurse into the filesystem as the transaction is already
3070 * started */
3071 flags |= AOP_FLAG_NOFS;
Alex Tomas64769242008-07-11 19:27:31 -04003072
Nick Piggin54566b22009-01-04 12:00:53 -08003073 page = grab_cache_page_write_begin(mapping, index, flags);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04003074 if (!page) {
3075 ext4_journal_stop(handle);
3076 ret = -ENOMEM;
3077 goto out;
3078 }
Alex Tomas64769242008-07-11 19:27:31 -04003079 *pagep = page;
3080
3081 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
Theodore Ts'ob920c752009-05-14 00:54:29 -04003082 ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04003083 if (ret < 0) {
3084 unlock_page(page);
3085 ext4_journal_stop(handle);
3086 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04003087 /*
3088 * block_write_begin may have instantiated a few blocks
3089 * outside i_size. Trim these off again. Don't need
3090 * i_size_read because we hold i_mutex.
3091 */
3092 if (pos + len > inode->i_size)
Jan Karaffacfa72009-07-13 16:22:22 -04003093 ext4_truncate(inode);
Alex Tomas64769242008-07-11 19:27:31 -04003094 }
3095
Mingming Caod2a17632008-07-14 17:52:37 -04003096 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3097 goto retry;
Alex Tomas64769242008-07-11 19:27:31 -04003098out:
3099 return ret;
3100}
3101
Mingming Cao632eaea2008-07-11 19:27:31 -04003102/*
3103 * Check if we should update i_disksize
3104 * when write to the end of file but not require block allocation
3105 */
3106static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003107 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04003108{
3109 struct buffer_head *bh;
3110 struct inode *inode = page->mapping->host;
3111 unsigned int idx;
3112 int i;
3113
3114 bh = page_buffers(page);
3115 idx = offset >> inode->i_blkbits;
3116
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003117 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04003118 bh = bh->b_this_page;
3119
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04003120 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04003121 return 0;
3122 return 1;
3123}
3124
Alex Tomas64769242008-07-11 19:27:31 -04003125static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003126 struct address_space *mapping,
3127 loff_t pos, unsigned len, unsigned copied,
3128 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04003129{
3130 struct inode *inode = mapping->host;
3131 int ret = 0, ret2;
3132 handle_t *handle = ext4_journal_current_handle();
3133 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04003134 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003135 int write_mode = (int)(unsigned long)fsdata;
3136
3137 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3138 if (ext4_should_order_data(inode)) {
3139 return ext4_ordered_write_end(file, mapping, pos,
3140 len, copied, page, fsdata);
3141 } else if (ext4_should_writeback_data(inode)) {
3142 return ext4_writeback_write_end(file, mapping, pos,
3143 len, copied, page, fsdata);
3144 } else {
3145 BUG();
3146 }
3147 }
Mingming Cao632eaea2008-07-11 19:27:31 -04003148
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003149 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04003150 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003151 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04003152
3153 /*
3154 * generic_write_end() will run mark_inode_dirty() if i_size
3155 * changes. So let's piggyback the i_disksize mark_inode_dirty
3156 * into that.
3157 */
3158
3159 new_i_size = pos + copied;
Mingming Cao632eaea2008-07-11 19:27:31 -04003160 if (new_i_size > EXT4_I(inode)->i_disksize) {
3161 if (ext4_da_should_update_i_disksize(page, end)) {
3162 down_write(&EXT4_I(inode)->i_data_sem);
3163 if (new_i_size > EXT4_I(inode)->i_disksize) {
3164 /*
3165 * Updating i_disksize when extending file
3166 * without needing block allocation
3167 */
3168 if (ext4_should_order_data(inode))
3169 ret = ext4_jbd2_file_inode(handle,
3170 inode);
Alex Tomas64769242008-07-11 19:27:31 -04003171
Mingming Cao632eaea2008-07-11 19:27:31 -04003172 EXT4_I(inode)->i_disksize = new_i_size;
3173 }
3174 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04003175 /* We need to mark inode dirty even if
3176 * new_i_size is less that inode->i_size
3177 * bu greater than i_disksize.(hint delalloc)
3178 */
3179 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04003180 }
Mingming Cao632eaea2008-07-11 19:27:31 -04003181 }
Alex Tomas64769242008-07-11 19:27:31 -04003182 ret2 = generic_write_end(file, mapping, pos, len, copied,
3183 page, fsdata);
3184 copied = ret2;
3185 if (ret2 < 0)
3186 ret = ret2;
3187 ret2 = ext4_journal_stop(handle);
3188 if (!ret)
3189 ret = ret2;
3190
3191 return ret ? ret : copied;
3192}
3193
3194static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3195{
Alex Tomas64769242008-07-11 19:27:31 -04003196 /*
3197 * Drop reserved blocks
3198 */
3199 BUG_ON(!PageLocked(page));
3200 if (!page_has_buffers(page))
3201 goto out;
3202
Mingming Caod2a17632008-07-14 17:52:37 -04003203 ext4_da_page_release_reservation(page, offset);
Alex Tomas64769242008-07-11 19:27:31 -04003204
3205out:
3206 ext4_invalidatepage(page, offset);
3207
3208 return;
3209}
3210
Theodore Ts'occd25062009-02-26 01:04:07 -05003211/*
3212 * Force all delayed allocation blocks to be allocated for a given inode.
3213 */
3214int ext4_alloc_da_blocks(struct inode *inode)
3215{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04003216 trace_ext4_alloc_da_blocks(inode);
3217
Theodore Ts'occd25062009-02-26 01:04:07 -05003218 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3219 !EXT4_I(inode)->i_reserved_meta_blocks)
3220 return 0;
3221
3222 /*
3223 * We do something simple for now. The filemap_flush() will
3224 * also start triggering a write of the data blocks, which is
3225 * not strictly speaking necessary (and for users of
3226 * laptop_mode, not even desirable). However, to do otherwise
3227 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003228 *
Theodore Ts'occd25062009-02-26 01:04:07 -05003229 * ext4_da_writepages() ->
3230 * write_cache_pages() ---> (via passed in callback function)
3231 * __mpage_da_writepage() -->
3232 * mpage_add_bh_to_extent()
3233 * mpage_da_map_blocks()
3234 *
3235 * The problem is that write_cache_pages(), located in
3236 * mm/page-writeback.c, marks pages clean in preparation for
3237 * doing I/O, which is not desirable if we're not planning on
3238 * doing I/O at all.
3239 *
3240 * We could call write_cache_pages(), and then redirty all of
3241 * the pages by calling redirty_page_for_writeback() but that
3242 * would be ugly in the extreme. So instead we would need to
3243 * replicate parts of the code in the above functions,
3244 * simplifying them becuase we wouldn't actually intend to
3245 * write out the pages, but rather only collect contiguous
3246 * logical block extents, call the multi-block allocator, and
3247 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003248 *
Theodore Ts'occd25062009-02-26 01:04:07 -05003249 * For now, though, we'll cheat by calling filemap_flush(),
3250 * which will map the blocks, and start the I/O, but not
3251 * actually wait for the I/O to complete.
3252 */
3253 return filemap_flush(inode->i_mapping);
3254}
Alex Tomas64769242008-07-11 19:27:31 -04003255
3256/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003257 * bmap() is special. It gets used by applications such as lilo and by
3258 * the swapper to find the on-disk block of a specific piece of data.
3259 *
3260 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07003261 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003262 * filesystem and enables swap, then they may get a nasty shock when the
3263 * data getting swapped to that swapfile suddenly gets overwritten by
3264 * the original zero's written out previously to the journal and
3265 * awaiting writeback in the kernel's buffer cache.
3266 *
3267 * So, if we see any bmap calls here on a modified, data-journaled file,
3268 * take extra steps to flush any blocks which might be in the cache.
3269 */
Mingming Cao617ba132006-10-11 01:20:53 -07003270static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003271{
3272 struct inode *inode = mapping->host;
3273 journal_t *journal;
3274 int err;
3275
Alex Tomas64769242008-07-11 19:27:31 -04003276 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3277 test_opt(inode->i_sb, DELALLOC)) {
3278 /*
3279 * With delalloc we want to sync the file
3280 * so that we can make sure we allocate
3281 * blocks for file
3282 */
3283 filemap_write_and_wait(mapping);
3284 }
3285
Frank Mayhar03901312009-01-07 00:06:22 -05003286 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003287 /*
3288 * This is a REALLY heavyweight approach, but the use of
3289 * bmap on dirty files is expected to be extremely rare:
3290 * only if we run lilo or swapon on a freshly made file
3291 * do we expect this to happen.
3292 *
3293 * (bmap requires CAP_SYS_RAWIO so this does not
3294 * represent an unprivileged user DOS attack --- we'd be
3295 * in trouble if mortal users could trigger this path at
3296 * will.)
3297 *
Mingming Cao617ba132006-10-11 01:20:53 -07003298 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003299 * regular files. If somebody wants to bmap a directory
3300 * or symlink and gets confused because the buffer
3301 * hasn't yet been flushed to disk, they deserve
3302 * everything they get.
3303 */
3304
Mingming Cao617ba132006-10-11 01:20:53 -07003305 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3306 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07003307 jbd2_journal_lock_updates(journal);
3308 err = jbd2_journal_flush(journal);
3309 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003310
3311 if (err)
3312 return 0;
3313 }
3314
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003315 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003316}
3317
Mingming Cao617ba132006-10-11 01:20:53 -07003318static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003319{
Mingming Cao617ba132006-10-11 01:20:53 -07003320 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003321}
3322
3323static int
Mingming Cao617ba132006-10-11 01:20:53 -07003324ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003325 struct list_head *pages, unsigned nr_pages)
3326{
Mingming Cao617ba132006-10-11 01:20:53 -07003327 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003328}
3329
Mingming Cao617ba132006-10-11 01:20:53 -07003330static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003331{
Mingming Cao617ba132006-10-11 01:20:53 -07003332 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003333
3334 /*
3335 * If it's a full truncate we just forget about the pending dirtying
3336 */
3337 if (offset == 0)
3338 ClearPageChecked(page);
3339
Frank Mayhar03901312009-01-07 00:06:22 -05003340 if (journal)
3341 jbd2_journal_invalidatepage(journal, page, offset);
3342 else
3343 block_invalidatepage(page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003344}
3345
Mingming Cao617ba132006-10-11 01:20:53 -07003346static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003347{
Mingming Cao617ba132006-10-11 01:20:53 -07003348 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003349
3350 WARN_ON(PageChecked(page));
3351 if (!page_has_buffers(page))
3352 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05003353 if (journal)
3354 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3355 else
3356 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003357}
3358
3359/*
Mingming Cao4c0425f2009-09-28 15:48:41 -04003360 * O_DIRECT for ext3 (or indirect map) based files
3361 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003362 * If the O_DIRECT write will extend the file then add this inode to the
3363 * orphan list. So recovery will truncate it back to the original size
3364 * if the machine crashes during the write.
3365 *
3366 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Kara7fb54092008-02-10 01:08:38 -05003367 * crashes then stale disk data _may_ be exposed inside the file. But current
3368 * VFS code falls back into buffered path in that case so we are safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003369 */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003370static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003371 const struct iovec *iov, loff_t offset,
3372 unsigned long nr_segs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003373{
3374 struct file *file = iocb->ki_filp;
3375 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07003376 struct ext4_inode_info *ei = EXT4_I(inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003377 handle_t *handle;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003378 ssize_t ret;
3379 int orphan = 0;
3380 size_t count = iov_length(iov, nr_segs);
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003381 int retries = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003382
3383 if (rw == WRITE) {
3384 loff_t final_size = offset + count;
3385
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003386 if (final_size > inode->i_size) {
Jan Kara7fb54092008-02-10 01:08:38 -05003387 /* Credits for sb + inode write */
3388 handle = ext4_journal_start(inode, 2);
3389 if (IS_ERR(handle)) {
3390 ret = PTR_ERR(handle);
3391 goto out;
3392 }
Mingming Cao617ba132006-10-11 01:20:53 -07003393 ret = ext4_orphan_add(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003394 if (ret) {
3395 ext4_journal_stop(handle);
3396 goto out;
3397 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003398 orphan = 1;
3399 ei->i_disksize = inode->i_size;
Jan Kara7fb54092008-02-10 01:08:38 -05003400 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003401 }
3402 }
3403
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003404retry:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003405 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3406 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07003407 ext4_get_block, NULL);
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003408 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3409 goto retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003410
Jan Kara7fb54092008-02-10 01:08:38 -05003411 if (orphan) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003412 int err;
3413
Jan Kara7fb54092008-02-10 01:08:38 -05003414 /* Credits for sb + inode write */
3415 handle = ext4_journal_start(inode, 2);
3416 if (IS_ERR(handle)) {
3417 /* This is really bad luck. We've written the data
3418 * but cannot extend i_size. Bail out and pretend
3419 * the write failed... */
3420 ret = PTR_ERR(handle);
3421 goto out;
3422 }
3423 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003424 ext4_orphan_del(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003425 if (ret > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003426 loff_t end = offset + ret;
3427 if (end > inode->i_size) {
3428 ei->i_disksize = end;
3429 i_size_write(inode, end);
3430 /*
3431 * We're going to return a positive `ret'
3432 * here due to non-zero-length I/O, so there's
3433 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07003434 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003435 * ignore it.
3436 */
Mingming Cao617ba132006-10-11 01:20:53 -07003437 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003438 }
3439 }
Mingming Cao617ba132006-10-11 01:20:53 -07003440 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003441 if (ret == 0)
3442 ret = err;
3443 }
3444out:
3445 return ret;
3446}
3447
Mingming Cao4c0425f2009-09-28 15:48:41 -04003448/* Maximum number of blocks we map for direct IO at once. */
3449
3450static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock,
3451 struct buffer_head *bh_result, int create)
3452{
3453 handle_t *handle = NULL;
3454 int ret = 0;
3455 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3456 int dio_credits;
3457
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003458 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3459 inode->i_ino, create);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003460 /*
3461 * DIO VFS code passes create = 0 flag for write to
3462 * the middle of file. It does this to avoid block
3463 * allocation for holes, to prevent expose stale data
3464 * out when there is parallel buffered read (which does
3465 * not hold the i_mutex lock) while direct IO write has
3466 * not completed. DIO request on holes finally falls back
3467 * to buffered IO for this reason.
3468 *
3469 * For ext4 extent based file, since we support fallocate,
3470 * new allocated extent as uninitialized, for holes, we
3471 * could fallocate blocks for holes, thus parallel
3472 * buffered IO read will zero out the page when read on
3473 * a hole while parallel DIO write to the hole has not completed.
3474 *
3475 * when we come here, we know it's a direct IO write to
3476 * to the middle of file (<i_size)
3477 * so it's safe to override the create flag from VFS.
3478 */
3479 create = EXT4_GET_BLOCKS_DIO_CREATE_EXT;
3480
3481 if (max_blocks > DIO_MAX_BLOCKS)
3482 max_blocks = DIO_MAX_BLOCKS;
3483 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3484 handle = ext4_journal_start(inode, dio_credits);
3485 if (IS_ERR(handle)) {
3486 ret = PTR_ERR(handle);
3487 goto out;
3488 }
3489 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3490 create);
3491 if (ret > 0) {
3492 bh_result->b_size = (ret << inode->i_blkbits);
3493 ret = 0;
3494 }
3495 ext4_journal_stop(handle);
3496out:
3497 return ret;
3498}
3499
Mingming Cao4c0425f2009-09-28 15:48:41 -04003500static void ext4_free_io_end(ext4_io_end_t *io)
3501{
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003502 BUG_ON(!io);
3503 iput(io->inode);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003504 kfree(io);
3505}
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003506static void dump_aio_dio_list(struct inode * inode)
3507{
3508#ifdef EXT4_DEBUG
3509 struct list_head *cur, *before, *after;
3510 ext4_io_end_t *io, *io0, *io1;
3511
3512 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3513 ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino);
3514 return;
3515 }
3516
3517 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino);
3518 list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){
3519 cur = &io->list;
3520 before = cur->prev;
3521 io0 = container_of(before, ext4_io_end_t, list);
3522 after = cur->next;
3523 io1 = container_of(after, ext4_io_end_t, list);
3524
3525 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3526 io, inode->i_ino, io0, io1);
3527 }
3528#endif
3529}
Mingming Cao4c0425f2009-09-28 15:48:41 -04003530
3531/*
Mingming Cao4c0425f2009-09-28 15:48:41 -04003532 * check a range of space and convert unwritten extents to written.
3533 */
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003534static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003535{
Mingming Cao4c0425f2009-09-28 15:48:41 -04003536 struct inode *inode = io->inode;
3537 loff_t offset = io->offset;
3538 size_t size = io->size;
3539 int ret = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003540
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003541 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3542 "list->prev 0x%p\n",
3543 io, inode->i_ino, io->list.next, io->list.prev);
3544
3545 if (list_empty(&io->list))
3546 return ret;
3547
3548 if (io->flag != DIO_AIO_UNWRITTEN)
3549 return ret;
3550
Mingming Cao4c0425f2009-09-28 15:48:41 -04003551 if (offset + size <= i_size_read(inode))
3552 ret = ext4_convert_unwritten_extents(inode, offset, size);
3553
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003554 if (ret < 0) {
Mingming Cao4c0425f2009-09-28 15:48:41 -04003555 printk(KERN_EMERG "%s: failed to convert unwritten"
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003556 "extents to written extents, error is %d"
3557 " io is still on inode %lu aio dio list\n",
3558 __func__, ret, inode->i_ino);
3559 return ret;
3560 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04003561
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003562 /* clear the DIO AIO unwritten flag */
3563 io->flag = 0;
3564 return ret;
3565}
3566/*
3567 * work on completed aio dio IO, to convert unwritten extents to extents
3568 */
3569static void ext4_end_aio_dio_work(struct work_struct *work)
3570{
3571 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3572 struct inode *inode = io->inode;
3573 int ret = 0;
3574
3575 mutex_lock(&inode->i_mutex);
3576 ret = ext4_end_aio_dio_nolock(io);
3577 if (ret >= 0) {
3578 if (!list_empty(&io->list))
3579 list_del_init(&io->list);
3580 ext4_free_io_end(io);
3581 }
3582 mutex_unlock(&inode->i_mutex);
3583}
3584/*
3585 * This function is called from ext4_sync_file().
3586 *
3587 * When AIO DIO IO is completed, the work to convert unwritten
3588 * extents to written is queued on workqueue but may not get immediately
3589 * scheduled. When fsync is called, we need to ensure the
3590 * conversion is complete before fsync returns.
3591 * The inode keeps track of a list of completed AIO from DIO path
3592 * that might needs to do the conversion. This function walks through
3593 * the list and convert the related unwritten extents to written.
3594 */
3595int flush_aio_dio_completed_IO(struct inode *inode)
3596{
3597 ext4_io_end_t *io;
3598 int ret = 0;
3599 int ret2 = 0;
3600
3601 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list))
3602 return ret;
3603
3604 dump_aio_dio_list(inode);
3605 while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3606 io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next,
3607 ext4_io_end_t, list);
3608 /*
3609 * Calling ext4_end_aio_dio_nolock() to convert completed
3610 * IO to written.
3611 *
3612 * When ext4_sync_file() is called, run_queue() may already
3613 * about to flush the work corresponding to this io structure.
3614 * It will be upset if it founds the io structure related
3615 * to the work-to-be schedule is freed.
3616 *
3617 * Thus we need to keep the io structure still valid here after
3618 * convertion finished. The io structure has a flag to
3619 * avoid double converting from both fsync and background work
3620 * queue work.
3621 */
3622 ret = ext4_end_aio_dio_nolock(io);
3623 if (ret < 0)
3624 ret2 = ret;
3625 else
3626 list_del_init(&io->list);
3627 }
3628 return (ret2 < 0) ? ret2 : 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003629}
3630
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003631static ext4_io_end_t *ext4_init_io_end (struct inode *inode)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003632{
3633 ext4_io_end_t *io = NULL;
3634
3635 io = kmalloc(sizeof(*io), GFP_NOFS);
3636
3637 if (io) {
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003638 igrab(inode);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003639 io->inode = inode;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003640 io->flag = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003641 io->offset = 0;
3642 io->size = 0;
3643 io->error = 0;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003644 INIT_WORK(&io->work, ext4_end_aio_dio_work);
3645 INIT_LIST_HEAD(&io->list);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003646 }
3647
3648 return io;
3649}
3650
3651static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3652 ssize_t size, void *private)
3653{
3654 ext4_io_end_t *io_end = iocb->private;
3655 struct workqueue_struct *wq;
3656
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003657 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3658 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3659 iocb->private, io_end->inode->i_ino, iocb, offset,
3660 size);
3661 /* if not async direct IO or dio with 0 bytes write, just return */
3662 if (!io_end || !size)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003663 return;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003664
3665 /* if not aio dio with unwritten extents, just free io and return */
3666 if (io_end->flag != DIO_AIO_UNWRITTEN){
3667 ext4_free_io_end(io_end);
3668 iocb->private = NULL;
3669 return;
3670 }
3671
Mingming Cao4c0425f2009-09-28 15:48:41 -04003672 io_end->offset = offset;
3673 io_end->size = size;
3674 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3675
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003676 /* queue the work to convert unwritten extents to written */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003677 queue_work(wq, &io_end->work);
3678
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003679 /* Add the io_end to per-inode completed aio dio list*/
3680 list_add_tail(&io_end->list,
3681 &EXT4_I(io_end->inode)->i_aio_dio_complete_list);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003682 iocb->private = NULL;
3683}
3684/*
3685 * For ext4 extent files, ext4 will do direct-io write to holes,
3686 * preallocated extents, and those write extend the file, no need to
3687 * fall back to buffered IO.
3688 *
3689 * For holes, we fallocate those blocks, mark them as unintialized
3690 * If those blocks were preallocated, we mark sure they are splited, but
3691 * still keep the range to write as unintialized.
3692 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003693 * The unwrritten extents will be converted to written when DIO is completed.
3694 * For async direct IO, since the IO may still pending when return, we
3695 * set up an end_io call back function, which will do the convertion
3696 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003697 *
3698 * If the O_DIRECT write will extend the file then add this inode to the
3699 * orphan list. So recovery will truncate it back to the original size
3700 * if the machine crashes during the write.
3701 *
3702 */
3703static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3704 const struct iovec *iov, loff_t offset,
3705 unsigned long nr_segs)
3706{
3707 struct file *file = iocb->ki_filp;
3708 struct inode *inode = file->f_mapping->host;
3709 ssize_t ret;
3710 size_t count = iov_length(iov, nr_segs);
3711
3712 loff_t final_size = offset + count;
3713 if (rw == WRITE && final_size <= inode->i_size) {
3714 /*
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003715 * We could direct write to holes and fallocate.
3716 *
3717 * Allocated blocks to fill the hole are marked as uninitialized
Mingming Cao4c0425f2009-09-28 15:48:41 -04003718 * to prevent paralel buffered read to expose the stale data
3719 * before DIO complete the data IO.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003720 *
3721 * As to previously fallocated extents, ext4 get_block
Mingming Cao4c0425f2009-09-28 15:48:41 -04003722 * will just simply mark the buffer mapped but still
3723 * keep the extents uninitialized.
3724 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003725 * for non AIO case, we will convert those unwritten extents
3726 * to written after return back from blockdev_direct_IO.
3727 *
3728 * for async DIO, the conversion needs to be defered when
3729 * the IO is completed. The ext4 end_io callback function
3730 * will be called to take care of the conversion work.
3731 * Here for async case, we allocate an io_end structure to
3732 * hook to the iocb.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003733 */
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003734 iocb->private = NULL;
3735 EXT4_I(inode)->cur_aio_dio = NULL;
3736 if (!is_sync_kiocb(iocb)) {
3737 iocb->private = ext4_init_io_end(inode);
3738 if (!iocb->private)
3739 return -ENOMEM;
3740 /*
3741 * we save the io structure for current async
3742 * direct IO, so that later ext4_get_blocks()
3743 * could flag the io structure whether there
3744 * is a unwritten extents needs to be converted
3745 * when IO is completed.
3746 */
3747 EXT4_I(inode)->cur_aio_dio = iocb->private;
3748 }
3749
Mingming Cao4c0425f2009-09-28 15:48:41 -04003750 ret = blockdev_direct_IO(rw, iocb, inode,
3751 inode->i_sb->s_bdev, iov,
3752 offset, nr_segs,
3753 ext4_get_block_dio_write,
3754 ext4_end_io_dio);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003755 if (iocb->private)
3756 EXT4_I(inode)->cur_aio_dio = NULL;
3757 /*
3758 * The io_end structure takes a reference to the inode,
3759 * that structure needs to be destroyed and the
3760 * reference to the inode need to be dropped, when IO is
3761 * complete, even with 0 byte write, or failed.
3762 *
3763 * In the successful AIO DIO case, the io_end structure will be
3764 * desctroyed and the reference to the inode will be dropped
3765 * after the end_io call back function is called.
3766 *
3767 * In the case there is 0 byte write, or error case, since
3768 * VFS direct IO won't invoke the end_io call back function,
3769 * we need to free the end_io structure here.
3770 */
3771 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3772 ext4_free_io_end(iocb->private);
3773 iocb->private = NULL;
3774 } else if (ret > 0)
3775 /*
3776 * for non AIO case, since the IO is already
3777 * completed, we could do the convertion right here
3778 */
3779 ret = ext4_convert_unwritten_extents(inode,
3780 offset, ret);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003781 return ret;
3782 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003783
3784 /* for write the the end of file case, we fall back to old way */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003785 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3786}
3787
3788static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3789 const struct iovec *iov, loff_t offset,
3790 unsigned long nr_segs)
3791{
3792 struct file *file = iocb->ki_filp;
3793 struct inode *inode = file->f_mapping->host;
3794
3795 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3796 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3797
3798 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3799}
3800
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003801/*
Mingming Cao617ba132006-10-11 01:20:53 -07003802 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003803 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3804 * much here because ->set_page_dirty is called under VFS locks. The page is
3805 * not necessarily locked.
3806 *
3807 * We cannot just dirty the page and leave attached buffers clean, because the
3808 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3809 * or jbddirty because all the journalling code will explode.
3810 *
3811 * So what we do is to mark the page "pending dirty" and next time writepage
3812 * is called, propagate that into the buffers appropriately.
3813 */
Mingming Cao617ba132006-10-11 01:20:53 -07003814static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003815{
3816 SetPageChecked(page);
3817 return __set_page_dirty_nobuffers(page);
3818}
3819
Mingming Cao617ba132006-10-11 01:20:53 -07003820static const struct address_space_operations ext4_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003821 .readpage = ext4_readpage,
3822 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003823 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003824 .sync_page = block_sync_page,
3825 .write_begin = ext4_write_begin,
3826 .write_end = ext4_ordered_write_end,
3827 .bmap = ext4_bmap,
3828 .invalidatepage = ext4_invalidatepage,
3829 .releasepage = ext4_releasepage,
3830 .direct_IO = ext4_direct_IO,
3831 .migratepage = buffer_migrate_page,
3832 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003833 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003834};
3835
Mingming Cao617ba132006-10-11 01:20:53 -07003836static const struct address_space_operations ext4_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003837 .readpage = ext4_readpage,
3838 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003839 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003840 .sync_page = block_sync_page,
3841 .write_begin = ext4_write_begin,
3842 .write_end = ext4_writeback_write_end,
3843 .bmap = ext4_bmap,
3844 .invalidatepage = ext4_invalidatepage,
3845 .releasepage = ext4_releasepage,
3846 .direct_IO = ext4_direct_IO,
3847 .migratepage = buffer_migrate_page,
3848 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003849 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003850};
3851
Mingming Cao617ba132006-10-11 01:20:53 -07003852static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003853 .readpage = ext4_readpage,
3854 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003855 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003856 .sync_page = block_sync_page,
3857 .write_begin = ext4_write_begin,
3858 .write_end = ext4_journalled_write_end,
3859 .set_page_dirty = ext4_journalled_set_page_dirty,
3860 .bmap = ext4_bmap,
3861 .invalidatepage = ext4_invalidatepage,
3862 .releasepage = ext4_releasepage,
3863 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003864 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003865};
3866
Alex Tomas64769242008-07-11 19:27:31 -04003867static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003868 .readpage = ext4_readpage,
3869 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003870 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003871 .writepages = ext4_da_writepages,
3872 .sync_page = block_sync_page,
3873 .write_begin = ext4_da_write_begin,
3874 .write_end = ext4_da_write_end,
3875 .bmap = ext4_bmap,
3876 .invalidatepage = ext4_da_invalidatepage,
3877 .releasepage = ext4_releasepage,
3878 .direct_IO = ext4_direct_IO,
3879 .migratepage = buffer_migrate_page,
3880 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003881 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003882};
3883
Mingming Cao617ba132006-10-11 01:20:53 -07003884void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003885{
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04003886 if (ext4_should_order_data(inode) &&
3887 test_opt(inode->i_sb, DELALLOC))
3888 inode->i_mapping->a_ops = &ext4_da_aops;
3889 else if (ext4_should_order_data(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07003890 inode->i_mapping->a_ops = &ext4_ordered_aops;
Alex Tomas64769242008-07-11 19:27:31 -04003891 else if (ext4_should_writeback_data(inode) &&
3892 test_opt(inode->i_sb, DELALLOC))
3893 inode->i_mapping->a_ops = &ext4_da_aops;
Mingming Cao617ba132006-10-11 01:20:53 -07003894 else if (ext4_should_writeback_data(inode))
3895 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003896 else
Mingming Cao617ba132006-10-11 01:20:53 -07003897 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003898}
3899
3900/*
Mingming Cao617ba132006-10-11 01:20:53 -07003901 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003902 * up to the end of the block which corresponds to `from'.
3903 * This required during truncate. We need to physically zero the tail end
3904 * of that block so it doesn't yield old data if the file is later grown.
3905 */
Jan Karacf108bc2008-07-11 19:27:31 -04003906int ext4_block_truncate_page(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003907 struct address_space *mapping, loff_t from)
3908{
Mingming Cao617ba132006-10-11 01:20:53 -07003909 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003910 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003911 unsigned blocksize, length, pos;
3912 ext4_lblk_t iblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003913 struct inode *inode = mapping->host;
3914 struct buffer_head *bh;
Jan Karacf108bc2008-07-11 19:27:31 -04003915 struct page *page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003916 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003917
Theodore Ts'of4a01012009-07-05 22:08:16 -04003918 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3919 mapping_gfp_mask(mapping) & ~__GFP_FS);
Jan Karacf108bc2008-07-11 19:27:31 -04003920 if (!page)
3921 return -EINVAL;
3922
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003923 blocksize = inode->i_sb->s_blocksize;
3924 length = blocksize - (offset & (blocksize - 1));
3925 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3926
3927 /*
3928 * For "nobh" option, we can only work if we don't need to
3929 * read-in the page - otherwise we create buffers to do the IO.
3930 */
3931 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
Mingming Cao617ba132006-10-11 01:20:53 -07003932 ext4_should_writeback_data(inode) && PageUptodate(page)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08003933 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003934 set_page_dirty(page);
3935 goto unlock;
3936 }
3937
3938 if (!page_has_buffers(page))
3939 create_empty_buffers(page, blocksize, 0);
3940
3941 /* Find the buffer that contains "offset" */
3942 bh = page_buffers(page);
3943 pos = blocksize;
3944 while (offset >= pos) {
3945 bh = bh->b_this_page;
3946 iblock++;
3947 pos += blocksize;
3948 }
3949
3950 err = 0;
3951 if (buffer_freed(bh)) {
3952 BUFFER_TRACE(bh, "freed: skip");
3953 goto unlock;
3954 }
3955
3956 if (!buffer_mapped(bh)) {
3957 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07003958 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003959 /* unmapped? It's a hole - nothing to do */
3960 if (!buffer_mapped(bh)) {
3961 BUFFER_TRACE(bh, "still unmapped");
3962 goto unlock;
3963 }
3964 }
3965
3966 /* Ok, it's mapped. Make sure it's up-to-date */
3967 if (PageUptodate(page))
3968 set_buffer_uptodate(bh);
3969
3970 if (!buffer_uptodate(bh)) {
3971 err = -EIO;
3972 ll_rw_block(READ, 1, &bh);
3973 wait_on_buffer(bh);
3974 /* Uhhuh. Read error. Complain and punt. */
3975 if (!buffer_uptodate(bh))
3976 goto unlock;
3977 }
3978
Mingming Cao617ba132006-10-11 01:20:53 -07003979 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003980 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07003981 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003982 if (err)
3983 goto unlock;
3984 }
3985
Christoph Lametereebd2aa2008-02-04 22:28:29 -08003986 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003987
3988 BUFFER_TRACE(bh, "zeroed end of block");
3989
3990 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07003991 if (ext4_should_journal_data(inode)) {
Frank Mayhar03901312009-01-07 00:06:22 -05003992 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003993 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07003994 if (ext4_should_order_data(inode))
Jan Kara678aaf42008-07-11 19:27:31 -04003995 err = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003996 mark_buffer_dirty(bh);
3997 }
3998
3999unlock:
4000 unlock_page(page);
4001 page_cache_release(page);
4002 return err;
4003}
4004
4005/*
4006 * Probably it should be a library function... search for first non-zero word
4007 * or memcmp with zero_page, whatever is better for particular architecture.
4008 * Linus?
4009 */
4010static inline int all_zeroes(__le32 *p, __le32 *q)
4011{
4012 while (p < q)
4013 if (*p++)
4014 return 0;
4015 return 1;
4016}
4017
4018/**
Mingming Cao617ba132006-10-11 01:20:53 -07004019 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004020 * @inode: inode in question
4021 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07004022 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004023 * @chain: place to store the pointers to partial indirect blocks
4024 * @top: place to the (detached) top of branch
4025 *
Mingming Cao617ba132006-10-11 01:20:53 -07004026 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004027 *
4028 * When we do truncate() we may have to clean the ends of several
4029 * indirect blocks but leave the blocks themselves alive. Block is
4030 * partially truncated if some data below the new i_size is refered
4031 * from it (and it is on the path to the first completely truncated
4032 * data block, indeed). We have to free the top of that path along
4033 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07004034 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004035 * finishes, we may safely do the latter, but top of branch may
4036 * require special attention - pageout below the truncation point
4037 * might try to populate it.
4038 *
4039 * We atomically detach the top of branch from the tree, store the
4040 * block number of its root in *@top, pointers to buffer_heads of
4041 * partially truncated blocks - in @chain[].bh and pointers to
4042 * their last elements that should not be removed - in
4043 * @chain[].p. Return value is the pointer to last filled element
4044 * of @chain.
4045 *
4046 * The work left to caller to do the actual freeing of subtrees:
4047 * a) free the subtree starting from *@top
4048 * b) free the subtrees whose roots are stored in
4049 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4050 * c) free the subtrees growing from the inode past the @chain[0].
4051 * (no partially truncated stuff there). */
4052
Mingming Cao617ba132006-10-11 01:20:53 -07004053static Indirect *ext4_find_shared(struct inode *inode, int depth,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004054 ext4_lblk_t offsets[4], Indirect chain[4],
4055 __le32 *top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004056{
4057 Indirect *partial, *p;
4058 int k, err;
4059
4060 *top = 0;
4061 /* Make k index the deepest non-null offest + 1 */
4062 for (k = depth; k > 1 && !offsets[k-1]; k--)
4063 ;
Mingming Cao617ba132006-10-11 01:20:53 -07004064 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004065 /* Writer: pointers */
4066 if (!partial)
4067 partial = chain + k-1;
4068 /*
4069 * If the branch acquired continuation since we've looked at it -
4070 * fine, it should all survive and (new) top doesn't belong to us.
4071 */
4072 if (!partial->key && *partial->p)
4073 /* Writer: end */
4074 goto no_top;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004075 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004076 ;
4077 /*
4078 * OK, we've found the last block that must survive. The rest of our
4079 * branch should be detached before unlocking. However, if that rest
4080 * of branch is all ours and does not grow immediately from the inode
4081 * it's easier to cheat and just decrement partial->p.
4082 */
4083 if (p == chain + k - 1 && p > chain) {
4084 p->p--;
4085 } else {
4086 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07004087 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004088#if 0
4089 *p->p = 0;
4090#endif
4091 }
4092 /* Writer: end */
4093
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004094 while (partial > p) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004095 brelse(partial->bh);
4096 partial--;
4097 }
4098no_top:
4099 return partial;
4100}
4101
4102/*
4103 * Zero a number of block pointers in either an inode or an indirect block.
4104 * If we restart the transaction we must again get write access to the
4105 * indirect block for further modification.
4106 *
4107 * We release `count' blocks on disk, but (last - first) may be greater
4108 * than `count' because there can be holes in there.
4109 */
Mingming Cao617ba132006-10-11 01:20:53 -07004110static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004111 struct buffer_head *bh,
4112 ext4_fsblk_t block_to_free,
4113 unsigned long count, __le32 *first,
4114 __le32 *last)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004115{
4116 __le32 *p;
4117 if (try_to_extend_transaction(handle, inode)) {
4118 if (bh) {
Frank Mayhar03901312009-01-07 00:06:22 -05004119 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4120 ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004121 }
Mingming Cao617ba132006-10-11 01:20:53 -07004122 ext4_mark_inode_dirty(handle, inode);
Jan Kara487caee2009-08-17 22:17:20 -04004123 ext4_truncate_restart_trans(handle, inode,
4124 blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004125 if (bh) {
4126 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07004127 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004128 }
4129 }
4130
4131 /*
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004132 * Any buffers which are on the journal will be in memory. We
4133 * find them on the hash table so jbd2_journal_revoke() will
4134 * run jbd2_journal_forget() on them. We've already detached
4135 * each block from the file, so bforget() in
4136 * jbd2_journal_forget() should be safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004137 *
Mingming Caodab291a2006-10-11 01:21:01 -07004138 * AKPM: turn on bforget in jbd2_journal_forget()!!!
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004139 */
4140 for (p = first; p < last; p++) {
4141 u32 nr = le32_to_cpu(*p);
4142 if (nr) {
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004143 struct buffer_head *tbh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004144
4145 *p = 0;
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004146 tbh = sb_find_get_block(inode->i_sb, nr);
4147 ext4_forget(handle, 0, inode, tbh, nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004148 }
4149 }
4150
Alex Tomasc9de5602008-01-29 00:19:52 -05004151 ext4_free_blocks(handle, inode, block_to_free, count, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004152}
4153
4154/**
Mingming Cao617ba132006-10-11 01:20:53 -07004155 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004156 * @handle: handle for this transaction
4157 * @inode: inode we are dealing with
4158 * @this_bh: indirect buffer_head which contains *@first and *@last
4159 * @first: array of block numbers
4160 * @last: points immediately past the end of array
4161 *
4162 * We are freeing all blocks refered from that array (numbers are stored as
4163 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4164 *
4165 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4166 * blocks are contiguous then releasing them at one time will only affect one
4167 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4168 * actually use a lot of journal space.
4169 *
4170 * @this_bh will be %NULL if @first and @last point into the inode's direct
4171 * block pointers.
4172 */
Mingming Cao617ba132006-10-11 01:20:53 -07004173static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004174 struct buffer_head *this_bh,
4175 __le32 *first, __le32 *last)
4176{
Mingming Cao617ba132006-10-11 01:20:53 -07004177 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004178 unsigned long count = 0; /* Number of blocks in the run */
4179 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4180 corresponding to
4181 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07004182 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004183 __le32 *p; /* Pointer into inode/ind
4184 for current block */
4185 int err;
4186
4187 if (this_bh) { /* For indirect block */
4188 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004189 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004190 /* Important: if we can't update the indirect pointers
4191 * to the blocks, we can't free them. */
4192 if (err)
4193 return;
4194 }
4195
4196 for (p = first; p < last; p++) {
4197 nr = le32_to_cpu(*p);
4198 if (nr) {
4199 /* accumulate blocks to free if they're contiguous */
4200 if (count == 0) {
4201 block_to_free = nr;
4202 block_to_free_p = p;
4203 count = 1;
4204 } else if (nr == block_to_free + count) {
4205 count++;
4206 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004207 ext4_clear_blocks(handle, inode, this_bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004208 block_to_free,
4209 count, block_to_free_p, p);
4210 block_to_free = nr;
4211 block_to_free_p = p;
4212 count = 1;
4213 }
4214 }
4215 }
4216
4217 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07004218 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004219 count, block_to_free_p, p);
4220
4221 if (this_bh) {
Frank Mayhar03901312009-01-07 00:06:22 -05004222 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
Duane Griffin71dc8fb2008-07-11 19:27:31 -04004223
4224 /*
4225 * The buffer head should have an attached journal head at this
4226 * point. However, if the data is corrupted and an indirect
4227 * block pointed to itself, it would have been detached when
4228 * the block was cleared. Check for this instead of OOPSing.
4229 */
Theodore Ts'oe7f07962009-01-20 09:50:19 -05004230 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
Frank Mayhar03901312009-01-07 00:06:22 -05004231 ext4_handle_dirty_metadata(handle, inode, this_bh);
Duane Griffin71dc8fb2008-07-11 19:27:31 -04004232 else
4233 ext4_error(inode->i_sb, __func__,
4234 "circular indirect block detected, "
4235 "inode=%lu, block=%llu",
4236 inode->i_ino,
4237 (unsigned long long) this_bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004238 }
4239}
4240
4241/**
Mingming Cao617ba132006-10-11 01:20:53 -07004242 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004243 * @handle: JBD handle for this transaction
4244 * @inode: inode we are dealing with
4245 * @parent_bh: the buffer_head which contains *@first and *@last
4246 * @first: array of block numbers
4247 * @last: pointer immediately past the end of array
4248 * @depth: depth of the branches to free
4249 *
4250 * We are freeing all blocks refered from these branches (numbers are
4251 * stored as little-endian 32-bit) and updating @inode->i_blocks
4252 * appropriately.
4253 */
Mingming Cao617ba132006-10-11 01:20:53 -07004254static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004255 struct buffer_head *parent_bh,
4256 __le32 *first, __le32 *last, int depth)
4257{
Mingming Cao617ba132006-10-11 01:20:53 -07004258 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004259 __le32 *p;
4260
Frank Mayhar03901312009-01-07 00:06:22 -05004261 if (ext4_handle_is_aborted(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004262 return;
4263
4264 if (depth--) {
4265 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07004266 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004267 p = last;
4268 while (--p >= first) {
4269 nr = le32_to_cpu(*p);
4270 if (!nr)
4271 continue; /* A hole */
4272
4273 /* Go read the buffer for the next level down */
4274 bh = sb_bread(inode->i_sb, nr);
4275
4276 /*
4277 * A read failure? Report error and clear slot
4278 * (should be rare).
4279 */
4280 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07004281 ext4_error(inode->i_sb, "ext4_free_branches",
Mingming Cao2ae02102006-10-11 01:21:11 -07004282 "Read failure, inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004283 inode->i_ino, nr);
4284 continue;
4285 }
4286
4287 /* This zaps the entire block. Bottom up. */
4288 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07004289 ext4_free_branches(handle, inode, bh,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004290 (__le32 *) bh->b_data,
4291 (__le32 *) bh->b_data + addr_per_block,
4292 depth);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004293
4294 /*
4295 * We've probably journalled the indirect block several
4296 * times during the truncate. But it's no longer
4297 * needed and we now drop it from the transaction via
Mingming Caodab291a2006-10-11 01:21:01 -07004298 * jbd2_journal_revoke().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004299 *
4300 * That's easy if it's exclusively part of this
4301 * transaction. But if it's part of the committing
Mingming Caodab291a2006-10-11 01:21:01 -07004302 * transaction then jbd2_journal_forget() will simply
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004303 * brelse() it. That means that if the underlying
Mingming Cao617ba132006-10-11 01:20:53 -07004304 * block is reallocated in ext4_get_block(),
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004305 * unmap_underlying_metadata() will find this block
4306 * and will try to get rid of it. damn, damn.
4307 *
4308 * If this block has already been committed to the
4309 * journal, a revoke record will be written. And
4310 * revoke records must be emitted *before* clearing
4311 * this block's bit in the bitmaps.
4312 */
Mingming Cao617ba132006-10-11 01:20:53 -07004313 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004314
4315 /*
4316 * Everything below this this pointer has been
4317 * released. Now let this top-of-subtree go.
4318 *
4319 * We want the freeing of this indirect block to be
4320 * atomic in the journal with the updating of the
4321 * bitmap block which owns it. So make some room in
4322 * the journal.
4323 *
4324 * We zero the parent pointer *after* freeing its
4325 * pointee in the bitmaps, so if extend_transaction()
4326 * for some reason fails to put the bitmap changes and
4327 * the release into the same transaction, recovery
4328 * will merely complain about releasing a free block,
4329 * rather than leaking blocks.
4330 */
Frank Mayhar03901312009-01-07 00:06:22 -05004331 if (ext4_handle_is_aborted(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004332 return;
4333 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004334 ext4_mark_inode_dirty(handle, inode);
Jan Kara487caee2009-08-17 22:17:20 -04004335 ext4_truncate_restart_trans(handle, inode,
4336 blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004337 }
4338
Alex Tomasc9de5602008-01-29 00:19:52 -05004339 ext4_free_blocks(handle, inode, nr, 1, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004340
4341 if (parent_bh) {
4342 /*
4343 * The block which we have just freed is
4344 * pointed to by an indirect block: journal it
4345 */
4346 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004347 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004348 parent_bh)){
4349 *p = 0;
4350 BUFFER_TRACE(parent_bh,
Frank Mayhar03901312009-01-07 00:06:22 -05004351 "call ext4_handle_dirty_metadata");
4352 ext4_handle_dirty_metadata(handle,
4353 inode,
4354 parent_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004355 }
4356 }
4357 }
4358 } else {
4359 /* We have reached the bottom of the tree. */
4360 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07004361 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004362 }
4363}
4364
Duane Griffin91ef4ca2008-07-11 19:27:31 -04004365int ext4_can_truncate(struct inode *inode)
4366{
4367 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4368 return 0;
4369 if (S_ISREG(inode->i_mode))
4370 return 1;
4371 if (S_ISDIR(inode->i_mode))
4372 return 1;
4373 if (S_ISLNK(inode->i_mode))
4374 return !ext4_inode_is_fast_symlink(inode);
4375 return 0;
4376}
4377
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004378/*
Mingming Cao617ba132006-10-11 01:20:53 -07004379 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004380 *
Mingming Cao617ba132006-10-11 01:20:53 -07004381 * We block out ext4_get_block() block instantiations across the entire
4382 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004383 * simultaneously on behalf of the same inode.
4384 *
4385 * As we work through the truncate and commmit bits of it to the journal there
4386 * is one core, guiding principle: the file's tree must always be consistent on
4387 * disk. We must be able to restart the truncate after a crash.
4388 *
4389 * The file's tree may be transiently inconsistent in memory (although it
4390 * probably isn't), but whenever we close off and commit a journal transaction,
4391 * the contents of (the filesystem + the journal) must be consistent and
4392 * restartable. It's pretty simple, really: bottom up, right to left (although
4393 * left-to-right works OK too).
4394 *
4395 * Note that at recovery time, journal replay occurs *before* the restart of
4396 * truncate against the orphan inode list.
4397 *
4398 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07004399 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004400 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07004401 * ext4_truncate() to have another go. So there will be instantiated blocks
4402 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004403 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07004404 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004405 */
Mingming Cao617ba132006-10-11 01:20:53 -07004406void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004407{
4408 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07004409 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004410 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07004411 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004412 struct address_space *mapping = inode->i_mapping;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004413 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004414 Indirect chain[4];
4415 Indirect *partial;
4416 __le32 nr = 0;
4417 int n;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004418 ext4_lblk_t last_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004419 unsigned blocksize = inode->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004420
Duane Griffin91ef4ca2008-07-11 19:27:31 -04004421 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004422 return;
4423
Theodore Ts'o5534fb52009-09-17 09:34:16 -04004424 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05004425 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4426
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004427 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
Jan Karacf108bc2008-07-11 19:27:31 -04004428 ext4_ext_truncate(inode);
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004429 return;
4430 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004431
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004432 handle = start_transaction(inode);
Jan Karacf108bc2008-07-11 19:27:31 -04004433 if (IS_ERR(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004434 return; /* AKPM: return what? */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004435
4436 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07004437 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004438
Jan Karacf108bc2008-07-11 19:27:31 -04004439 if (inode->i_size & (blocksize - 1))
4440 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4441 goto out_stop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004442
Mingming Cao617ba132006-10-11 01:20:53 -07004443 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004444 if (n == 0)
4445 goto out_stop; /* error */
4446
4447 /*
4448 * OK. This truncate is going to happen. We add the inode to the
4449 * orphan list, so that if this truncate spans multiple transactions,
4450 * and we crash, we will resume the truncate when the filesystem
4451 * recovers. It also marks the inode dirty, to catch the new size.
4452 *
4453 * Implication: the file must always be in a sane, consistent
4454 * truncatable state while each transaction commits.
4455 */
Mingming Cao617ba132006-10-11 01:20:53 -07004456 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004457 goto out_stop;
4458
4459 /*
Mingming Cao632eaea2008-07-11 19:27:31 -04004460 * From here we block out all ext4_get_block() callers who want to
4461 * modify the block allocation tree.
4462 */
4463 down_write(&ei->i_data_sem);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04004464
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004465 ext4_discard_preallocations(inode);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04004466
Mingming Cao632eaea2008-07-11 19:27:31 -04004467 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004468 * The orphan list entry will now protect us from any crash which
4469 * occurs before the truncate completes, so it is now safe to propagate
4470 * the new, shorter inode size (held for now in i_size) into the
4471 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07004472 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004473 */
4474 ei->i_disksize = inode->i_size;
4475
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004476 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07004477 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4478 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004479 goto do_indirects;
4480 }
4481
Mingming Cao617ba132006-10-11 01:20:53 -07004482 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004483 /* Kill the top of shared branch (not detached) */
4484 if (nr) {
4485 if (partial == chain) {
4486 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07004487 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004488 &nr, &nr+1, (chain+n-1) - partial);
4489 *partial->p = 0;
4490 /*
4491 * We mark the inode dirty prior to restart,
4492 * and prior to stop. No need for it here.
4493 */
4494 } else {
4495 /* Shared branch grows from an indirect block */
4496 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004497 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004498 partial->p,
4499 partial->p+1, (chain+n-1) - partial);
4500 }
4501 }
4502 /* Clear the ends of indirect blocks on the shared branch */
4503 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07004504 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004505 (__le32*)partial->bh->b_data+addr_per_block,
4506 (chain+n-1) - partial);
4507 BUFFER_TRACE(partial->bh, "call brelse");
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004508 brelse(partial->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004509 partial--;
4510 }
4511do_indirects:
4512 /* Kill the remaining (whole) subtrees */
4513 switch (offsets[0]) {
4514 default:
Mingming Cao617ba132006-10-11 01:20:53 -07004515 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004516 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004517 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4518 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004519 }
Mingming Cao617ba132006-10-11 01:20:53 -07004520 case EXT4_IND_BLOCK:
4521 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004522 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004523 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4524 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004525 }
Mingming Cao617ba132006-10-11 01:20:53 -07004526 case EXT4_DIND_BLOCK:
4527 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004528 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004529 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4530 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004531 }
Mingming Cao617ba132006-10-11 01:20:53 -07004532 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004533 ;
4534 }
4535
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004536 up_write(&ei->i_data_sem);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004537 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07004538 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004539
4540 /*
4541 * In a multi-transaction truncate, we only make the final transaction
4542 * synchronous
4543 */
4544 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05004545 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004546out_stop:
4547 /*
4548 * If this was a simple ftruncate(), and the file will remain alive
4549 * then we need to clear up the orphan record which we created above.
4550 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07004551 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004552 * orphan info for us.
4553 */
4554 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004555 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004556
Mingming Cao617ba132006-10-11 01:20:53 -07004557 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004558}
4559
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004560/*
Mingming Cao617ba132006-10-11 01:20:53 -07004561 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004562 * underlying buffer_head on success. If 'in_mem' is true, we have all
4563 * data in memory that is needed to recreate the on-disk version of this
4564 * inode.
4565 */
Mingming Cao617ba132006-10-11 01:20:53 -07004566static int __ext4_get_inode_loc(struct inode *inode,
4567 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004568{
Theodore Ts'o240799c2008-10-09 23:53:47 -04004569 struct ext4_group_desc *gdp;
4570 struct buffer_head *bh;
4571 struct super_block *sb = inode->i_sb;
4572 ext4_fsblk_t block;
4573 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004574
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004575 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04004576 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004577 return -EIO;
4578
Theodore Ts'o240799c2008-10-09 23:53:47 -04004579 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4580 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4581 if (!gdp)
4582 return -EIO;
4583
4584 /*
4585 * Figure out the offset within the block group inode table
4586 */
4587 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4588 inode_offset = ((inode->i_ino - 1) %
4589 EXT4_INODES_PER_GROUP(sb));
4590 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4591 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4592
4593 bh = sb_getblk(sb, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004594 if (!bh) {
Theodore Ts'o240799c2008-10-09 23:53:47 -04004595 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4596 "inode block - inode=%lu, block=%llu",
4597 inode->i_ino, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004598 return -EIO;
4599 }
4600 if (!buffer_uptodate(bh)) {
4601 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04004602
4603 /*
4604 * If the buffer has the write error flag, we have failed
4605 * to write out another inode in the same block. In this
4606 * case, we don't have to read the block because we may
4607 * read the old inode data successfully.
4608 */
4609 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4610 set_buffer_uptodate(bh);
4611
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004612 if (buffer_uptodate(bh)) {
4613 /* someone brought it uptodate while we waited */
4614 unlock_buffer(bh);
4615 goto has_buffer;
4616 }
4617
4618 /*
4619 * If we have all information of the inode in memory and this
4620 * is the only valid inode in the block, we need not read the
4621 * block.
4622 */
4623 if (in_mem) {
4624 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04004625 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004626
Theodore Ts'o240799c2008-10-09 23:53:47 -04004627 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004628
4629 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04004630 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004631 if (!bitmap_bh)
4632 goto make_io;
4633
4634 /*
4635 * If the inode bitmap isn't in cache then the
4636 * optimisation may end up performing two reads instead
4637 * of one, so skip it.
4638 */
4639 if (!buffer_uptodate(bitmap_bh)) {
4640 brelse(bitmap_bh);
4641 goto make_io;
4642 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04004643 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004644 if (i == inode_offset)
4645 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07004646 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004647 break;
4648 }
4649 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04004650 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004651 /* all other inodes are free, so skip I/O */
4652 memset(bh->b_data, 0, bh->b_size);
4653 set_buffer_uptodate(bh);
4654 unlock_buffer(bh);
4655 goto has_buffer;
4656 }
4657 }
4658
4659make_io:
4660 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04004661 * If we need to do any I/O, try to pre-readahead extra
4662 * blocks from the inode table.
4663 */
4664 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4665 ext4_fsblk_t b, end, table;
4666 unsigned num;
4667
4668 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04004669 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o240799c2008-10-09 23:53:47 -04004670 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4671 if (table > b)
4672 b = table;
4673 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4674 num = EXT4_INODES_PER_GROUP(sb);
4675 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4676 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05004677 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04004678 table += num / inodes_per_block;
4679 if (end > table)
4680 end = table;
4681 while (b <= end)
4682 sb_breadahead(sb, b++);
4683 }
4684
4685 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004686 * There are other valid inodes in the buffer, this inode
4687 * has in-inode xattrs, or we don't have this inode in memory.
4688 * Read the block from disk.
4689 */
4690 get_bh(bh);
4691 bh->b_end_io = end_buffer_read_sync;
4692 submit_bh(READ_META, bh);
4693 wait_on_buffer(bh);
4694 if (!buffer_uptodate(bh)) {
Theodore Ts'o240799c2008-10-09 23:53:47 -04004695 ext4_error(sb, __func__,
4696 "unable to read inode block - inode=%lu, "
4697 "block=%llu", inode->i_ino, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004698 brelse(bh);
4699 return -EIO;
4700 }
4701 }
4702has_buffer:
4703 iloc->bh = bh;
4704 return 0;
4705}
4706
Mingming Cao617ba132006-10-11 01:20:53 -07004707int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004708{
4709 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07004710 return __ext4_get_inode_loc(inode, iloc,
4711 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004712}
4713
Mingming Cao617ba132006-10-11 01:20:53 -07004714void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004715{
Mingming Cao617ba132006-10-11 01:20:53 -07004716 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004717
4718 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07004719 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004720 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07004721 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004722 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07004723 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004724 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07004725 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004726 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07004727 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004728 inode->i_flags |= S_DIRSYNC;
4729}
4730
Jan Karaff9ddf72007-07-18 09:24:20 -04004731/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4732void ext4_get_inode_flags(struct ext4_inode_info *ei)
4733{
4734 unsigned int flags = ei->vfs_inode.i_flags;
4735
4736 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4737 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4738 if (flags & S_SYNC)
4739 ei->i_flags |= EXT4_SYNC_FL;
4740 if (flags & S_APPEND)
4741 ei->i_flags |= EXT4_APPEND_FL;
4742 if (flags & S_IMMUTABLE)
4743 ei->i_flags |= EXT4_IMMUTABLE_FL;
4744 if (flags & S_NOATIME)
4745 ei->i_flags |= EXT4_NOATIME_FL;
4746 if (flags & S_DIRSYNC)
4747 ei->i_flags |= EXT4_DIRSYNC_FL;
4748}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004749
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004750static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004751 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004752{
4753 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004754 struct inode *inode = &(ei->vfs_inode);
4755 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004756
4757 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4758 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4759 /* we are using combined 48 bit field */
4760 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4761 le32_to_cpu(raw_inode->i_blocks_lo);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004762 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4763 /* i_blocks represent file system block size */
4764 return i_blocks << (inode->i_blkbits - 9);
4765 } else {
4766 return i_blocks;
4767 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004768 } else {
4769 return le32_to_cpu(raw_inode->i_blocks_lo);
4770 }
4771}
Jan Karaff9ddf72007-07-18 09:24:20 -04004772
David Howells1d1fe1e2008-02-07 00:15:37 -08004773struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004774{
Mingming Cao617ba132006-10-11 01:20:53 -07004775 struct ext4_iloc iloc;
4776 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004777 struct ext4_inode_info *ei;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004778 struct buffer_head *bh;
David Howells1d1fe1e2008-02-07 00:15:37 -08004779 struct inode *inode;
4780 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004781 int block;
4782
David Howells1d1fe1e2008-02-07 00:15:37 -08004783 inode = iget_locked(sb, ino);
4784 if (!inode)
4785 return ERR_PTR(-ENOMEM);
4786 if (!(inode->i_state & I_NEW))
4787 return inode;
4788
4789 ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004790
David Howells1d1fe1e2008-02-07 00:15:37 -08004791 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4792 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004793 goto bad_inode;
4794 bh = iloc.bh;
Mingming Cao617ba132006-10-11 01:20:53 -07004795 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004796 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4797 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4798 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004799 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004800 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4801 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4802 }
4803 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004804
4805 ei->i_state = 0;
4806 ei->i_dir_start_lookup = 0;
4807 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4808 /* We now have enough fields to check if the inode was active or not.
4809 * This is needed because nfsd might try to access dead inodes
4810 * the test is that same one that e2fsck uses
4811 * NeilBrown 1999oct15
4812 */
4813 if (inode->i_nlink == 0) {
4814 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07004815 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004816 /* this inode is deleted */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004817 brelse(bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004818 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004819 goto bad_inode;
4820 }
4821 /* The only unlinked inodes we let through here have
4822 * valid i_mode and are being read by the orphan
4823 * recovery code: that's fine, we're about to complete
4824 * the process of deleting those. */
4825 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004826 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004827 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004828 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04004829 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004830 ei->i_file_acl |=
4831 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004832 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004833 ei->i_disksize = inode->i_size;
4834 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4835 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04004836 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004837 /*
4838 * NOTE! The in-memory inode i_data array is in little-endian order
4839 * even on big-endian machines: we do NOT byteswap the block numbers!
4840 */
Mingming Cao617ba132006-10-11 01:20:53 -07004841 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004842 ei->i_data[block] = raw_inode->i_block[block];
4843 INIT_LIST_HEAD(&ei->i_orphan);
4844
Eric Sandeen0040d982008-02-05 22:36:43 -05004845 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004846 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07004847 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07004848 EXT4_INODE_SIZE(inode->i_sb)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004849 brelse(bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004850 ret = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004851 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07004852 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004853 if (ei->i_extra_isize == 0) {
4854 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07004855 ei->i_extra_isize = sizeof(struct ext4_inode) -
4856 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004857 } else {
4858 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07004859 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004860 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07004861 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004862 ei->i_state |= EXT4_STATE_XATTR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004863 }
4864 } else
4865 ei->i_extra_isize = 0;
4866
Kalpak Shahef7f3832007-07-18 09:15:20 -04004867 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4868 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4869 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4870 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4871
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004872 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4873 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4874 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4875 inode->i_version |=
4876 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4877 }
4878
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004879 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004880 if (ei->i_file_acl &&
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004881 ((ei->i_file_acl <
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004882 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4883 EXT4_SB(sb)->s_gdb_count)) ||
4884 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4885 ext4_error(sb, __func__,
4886 "bad extended attribute block %llu in inode #%lu",
4887 ei->i_file_acl, inode->i_ino);
4888 ret = -EIO;
4889 goto bad_inode;
4890 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004891 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4892 (S_ISLNK(inode->i_mode) &&
4893 !ext4_inode_is_fast_symlink(inode)))
4894 /* Validate extent which is part of inode */
4895 ret = ext4_ext_check_inode(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004896 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004897 (S_ISLNK(inode->i_mode) &&
4898 !ext4_inode_is_fast_symlink(inode))) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004899 /* Validate block references which are part of inode */
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004900 ret = ext4_check_inode_blockref(inode);
4901 }
4902 if (ret) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004903 brelse(bh);
4904 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04004905 }
4906
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004907 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004908 inode->i_op = &ext4_file_inode_operations;
4909 inode->i_fop = &ext4_file_operations;
4910 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004911 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004912 inode->i_op = &ext4_dir_inode_operations;
4913 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004914 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00004915 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004916 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00004917 nd_terminate_link(ei->i_data, inode->i_size,
4918 sizeof(ei->i_data) - 1);
4919 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004920 inode->i_op = &ext4_symlink_inode_operations;
4921 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004922 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004923 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4924 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004925 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004926 if (raw_inode->i_block[0])
4927 init_special_inode(inode, inode->i_mode,
4928 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4929 else
4930 init_special_inode(inode, inode->i_mode,
4931 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004932 } else {
4933 brelse(bh);
4934 ret = -EIO;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004935 ext4_error(inode->i_sb, __func__,
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004936 "bogus i_mode (%o) for inode=%lu",
4937 inode->i_mode, inode->i_ino);
4938 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004939 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004940 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004941 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004942 unlock_new_inode(inode);
4943 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004944
4945bad_inode:
David Howells1d1fe1e2008-02-07 00:15:37 -08004946 iget_failed(inode);
4947 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004948}
4949
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004950static int ext4_inode_blocks_set(handle_t *handle,
4951 struct ext4_inode *raw_inode,
4952 struct ext4_inode_info *ei)
4953{
4954 struct inode *inode = &(ei->vfs_inode);
4955 u64 i_blocks = inode->i_blocks;
4956 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004957
4958 if (i_blocks <= ~0U) {
4959 /*
4960 * i_blocks can be represnted in a 32 bit variable
4961 * as multiple of 512 bytes
4962 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004963 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004964 raw_inode->i_blocks_high = 0;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004965 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004966 return 0;
4967 }
4968 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4969 return -EFBIG;
4970
4971 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004972 /*
4973 * i_blocks can be represented in a 48 bit variable
4974 * as multiple of 512 bytes
4975 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004976 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004977 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004978 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004979 } else {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004980 ei->i_flags |= EXT4_HUGE_FILE_FL;
4981 /* i_block is stored in file system block size */
4982 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4983 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4984 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004985 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004986 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004987}
4988
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004989/*
4990 * Post the struct inode info into an on-disk inode location in the
4991 * buffer-cache. This gobbles the caller's reference to the
4992 * buffer_head in the inode location struct.
4993 *
4994 * The caller must have write access to iloc->bh.
4995 */
Mingming Cao617ba132006-10-11 01:20:53 -07004996static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004997 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04004998 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004999{
Mingming Cao617ba132006-10-11 01:20:53 -07005000 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5001 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005002 struct buffer_head *bh = iloc->bh;
5003 int err = 0, rc, block;
5004
5005 /* For fields not not tracking in the in-memory inode,
5006 * initialise them to zero for new inodes. */
Mingming Cao617ba132006-10-11 01:20:53 -07005007 if (ei->i_state & EXT4_STATE_NEW)
5008 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005009
Jan Karaff9ddf72007-07-18 09:24:20 -04005010 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005011 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005012 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005013 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5014 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5015/*
5016 * Fix up interoperability with old kernels. Otherwise, old inodes get
5017 * re-used with the upper 16 bits of the uid/gid intact
5018 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005019 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005020 raw_inode->i_uid_high =
5021 cpu_to_le16(high_16_bits(inode->i_uid));
5022 raw_inode->i_gid_high =
5023 cpu_to_le16(high_16_bits(inode->i_gid));
5024 } else {
5025 raw_inode->i_uid_high = 0;
5026 raw_inode->i_gid_high = 0;
5027 }
5028 } else {
5029 raw_inode->i_uid_low =
5030 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5031 raw_inode->i_gid_low =
5032 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5033 raw_inode->i_uid_high = 0;
5034 raw_inode->i_gid_high = 0;
5035 }
5036 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04005037
5038 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5039 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5040 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5041 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5042
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005043 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5044 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005045 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o1b9c12f2009-09-17 08:32:22 -04005046 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07005047 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5048 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07005049 raw_inode->i_file_acl_high =
5050 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05005051 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005052 ext4_isize_set(raw_inode, ei->i_disksize);
5053 if (ei->i_disksize > 0x7fffffffULL) {
5054 struct super_block *sb = inode->i_sb;
5055 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5056 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5057 EXT4_SB(sb)->s_es->s_rev_level ==
5058 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5059 /* If this is the first large file
5060 * created, add a flag to the superblock.
5061 */
5062 err = ext4_journal_get_write_access(handle,
5063 EXT4_SB(sb)->s_sbh);
5064 if (err)
5065 goto out_brelse;
5066 ext4_update_dynamic_rev(sb);
5067 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07005068 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005069 sb->s_dirt = 1;
Frank Mayhar03901312009-01-07 00:06:22 -05005070 ext4_handle_sync(handle);
5071 err = ext4_handle_dirty_metadata(handle, inode,
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005072 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005073 }
5074 }
5075 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5076 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5077 if (old_valid_dev(inode->i_rdev)) {
5078 raw_inode->i_block[0] =
5079 cpu_to_le32(old_encode_dev(inode->i_rdev));
5080 raw_inode->i_block[1] = 0;
5081 } else {
5082 raw_inode->i_block[0] = 0;
5083 raw_inode->i_block[1] =
5084 cpu_to_le32(new_encode_dev(inode->i_rdev));
5085 raw_inode->i_block[2] = 0;
5086 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005087 } else
5088 for (block = 0; block < EXT4_N_BLOCKS; block++)
5089 raw_inode->i_block[block] = ei->i_data[block];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005090
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005091 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5092 if (ei->i_extra_isize) {
5093 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5094 raw_inode->i_version_hi =
5095 cpu_to_le32(inode->i_version >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005096 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005097 }
5098
Frank Mayhar830156c2009-09-29 10:07:47 -04005099 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5100 rc = ext4_handle_dirty_metadata(handle, inode, bh);
5101 if (!err)
5102 err = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07005103 ei->i_state &= ~EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005104
5105out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005106 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07005107 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005108 return err;
5109}
5110
5111/*
Mingming Cao617ba132006-10-11 01:20:53 -07005112 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005113 *
5114 * We are called from a few places:
5115 *
5116 * - Within generic_file_write() for O_SYNC files.
5117 * Here, there will be no transaction running. We wait for any running
5118 * trasnaction to commit.
5119 *
5120 * - Within sys_sync(), kupdate and such.
5121 * We wait on commit, if tol to.
5122 *
5123 * - Within prune_icache() (PF_MEMALLOC == true)
5124 * Here we simply return. We can't afford to block kswapd on the
5125 * journal commit.
5126 *
5127 * In all cases it is actually safe for us to return without doing anything,
5128 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07005129 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005130 * knfsd.
5131 *
5132 * Note that we are absolutely dependent upon all inode dirtiers doing the
5133 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5134 * which we are interested.
5135 *
5136 * It would be a bug for them to not do this. The code:
5137 *
5138 * mark_inode_dirty(inode)
5139 * stuff();
5140 * inode->i_size = expr;
5141 *
5142 * is in error because a kswapd-driven write_inode() could occur while
5143 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5144 * will no longer be on the superblock's dirty inode list.
5145 */
Mingming Cao617ba132006-10-11 01:20:53 -07005146int ext4_write_inode(struct inode *inode, int wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005147{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005148 int err;
5149
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005150 if (current->flags & PF_MEMALLOC)
5151 return 0;
5152
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005153 if (EXT4_SB(inode->i_sb)->s_journal) {
5154 if (ext4_journal_current_handle()) {
5155 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5156 dump_stack();
5157 return -EIO;
5158 }
5159
5160 if (!wait)
5161 return 0;
5162
5163 err = ext4_force_commit(inode->i_sb);
5164 } else {
5165 struct ext4_iloc iloc;
5166
5167 err = ext4_get_inode_loc(inode, &iloc);
5168 if (err)
5169 return err;
Frank Mayhar830156c2009-09-29 10:07:47 -04005170 if (wait)
5171 sync_dirty_buffer(iloc.bh);
5172 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5173 ext4_error(inode->i_sb, __func__,
5174 "IO error syncing inode, "
5175 "inode=%lu, block=%llu",
5176 inode->i_ino,
5177 (unsigned long long)iloc.bh->b_blocknr);
5178 err = -EIO;
5179 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005180 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005181 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005182}
5183
5184/*
Mingming Cao617ba132006-10-11 01:20:53 -07005185 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005186 *
5187 * Called from notify_change.
5188 *
5189 * We want to trap VFS attempts to truncate the file as soon as
5190 * possible. In particular, we want to make sure that when the VFS
5191 * shrinks i_size, we put the inode on the orphan list and modify
5192 * i_disksize immediately, so that during the subsequent flushing of
5193 * dirty pages and freeing of disk blocks, we can guarantee that any
5194 * commit will leave the blocks being flushed in an unused state on
5195 * disk. (On recovery, the inode will get truncated and the blocks will
5196 * be freed, so we have a strong guarantee that no future commit will
5197 * leave these blocks visible to the user.)
5198 *
Jan Kara678aaf42008-07-11 19:27:31 -04005199 * Another thing we have to assure is that if we are in ordered mode
5200 * and inode is still attached to the committing transaction, we must
5201 * we start writeout of all the dirty pages which are being truncated.
5202 * This way we are sure that all the data written in the previous
5203 * transaction are already on disk (truncate waits for pages under
5204 * writeback).
5205 *
5206 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005207 */
Mingming Cao617ba132006-10-11 01:20:53 -07005208int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005209{
5210 struct inode *inode = dentry->d_inode;
5211 int error, rc = 0;
5212 const unsigned int ia_valid = attr->ia_valid;
5213
5214 error = inode_change_ok(inode, attr);
5215 if (error)
5216 return error;
5217
5218 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5219 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5220 handle_t *handle;
5221
5222 /* (user+group)*(old+new) structure, inode write (sb,
5223 * inode block, ? - but truncate inode update has it) */
Mingming Cao617ba132006-10-11 01:20:53 -07005224 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
5225 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005226 if (IS_ERR(handle)) {
5227 error = PTR_ERR(handle);
5228 goto err_out;
5229 }
Jan Karaa269eb12009-01-26 17:04:39 +01005230 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005231 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07005232 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005233 return error;
5234 }
5235 /* Update corresponding info in inode so that everything is in
5236 * one transaction */
5237 if (attr->ia_valid & ATTR_UID)
5238 inode->i_uid = attr->ia_uid;
5239 if (attr->ia_valid & ATTR_GID)
5240 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07005241 error = ext4_mark_inode_dirty(handle, inode);
5242 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005243 }
5244
Eric Sandeene2b46572008-01-28 23:58:27 -05005245 if (attr->ia_valid & ATTR_SIZE) {
5246 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5247 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5248
5249 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5250 error = -EFBIG;
5251 goto err_out;
5252 }
5253 }
5254 }
5255
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005256 if (S_ISREG(inode->i_mode) &&
5257 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
5258 handle_t *handle;
5259
Mingming Cao617ba132006-10-11 01:20:53 -07005260 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005261 if (IS_ERR(handle)) {
5262 error = PTR_ERR(handle);
5263 goto err_out;
5264 }
5265
Mingming Cao617ba132006-10-11 01:20:53 -07005266 error = ext4_orphan_add(handle, inode);
5267 EXT4_I(inode)->i_disksize = attr->ia_size;
5268 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005269 if (!error)
5270 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07005271 ext4_journal_stop(handle);
Jan Kara678aaf42008-07-11 19:27:31 -04005272
5273 if (ext4_should_order_data(inode)) {
5274 error = ext4_begin_ordered_truncate(inode,
5275 attr->ia_size);
5276 if (error) {
5277 /* Do as much error cleanup as possible */
5278 handle = ext4_journal_start(inode, 3);
5279 if (IS_ERR(handle)) {
5280 ext4_orphan_del(NULL, inode);
5281 goto err_out;
5282 }
5283 ext4_orphan_del(handle, inode);
5284 ext4_journal_stop(handle);
5285 goto err_out;
5286 }
5287 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005288 }
5289
5290 rc = inode_setattr(inode, attr);
5291
Mingming Cao617ba132006-10-11 01:20:53 -07005292 /* If inode_setattr's call to ext4_truncate failed to get a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005293 * transaction handle at all, we need to clean up the in-core
5294 * orphan list manually. */
5295 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07005296 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005297
5298 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07005299 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005300
5301err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07005302 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005303 if (!error)
5304 error = rc;
5305 return error;
5306}
5307
Mingming Cao3e3398a2008-07-11 19:27:31 -04005308int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5309 struct kstat *stat)
5310{
5311 struct inode *inode;
5312 unsigned long delalloc_blocks;
5313
5314 inode = dentry->d_inode;
5315 generic_fillattr(inode, stat);
5316
5317 /*
5318 * We can't update i_blocks if the block allocation is delayed
5319 * otherwise in the case of system crash before the real block
5320 * allocation is done, we will have i_blocks inconsistent with
5321 * on-disk file blocks.
5322 * We always keep i_blocks updated together with real
5323 * allocation. But to not confuse with user, stat
5324 * will return the blocks that include the delayed allocation
5325 * blocks for this file.
5326 */
5327 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5328 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5329 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5330
5331 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5332 return 0;
5333}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005334
Mingming Caoa02908f2008-08-19 22:16:07 -04005335static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5336 int chunk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005337{
Mingming Caoa02908f2008-08-19 22:16:07 -04005338 int indirects;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005339
Mingming Caoa02908f2008-08-19 22:16:07 -04005340 /* if nrblocks are contiguous */
5341 if (chunk) {
5342 /*
5343 * With N contiguous data blocks, it need at most
5344 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5345 * 2 dindirect blocks
5346 * 1 tindirect block
5347 */
5348 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5349 return indirects + 3;
5350 }
5351 /*
5352 * if nrblocks are not contiguous, worse case, each block touch
5353 * a indirect block, and each indirect block touch a double indirect
5354 * block, plus a triple indirect block
5355 */
5356 indirects = nrblocks * 2 + 1;
5357 return indirects;
5358}
Alex Tomasa86c6182006-10-11 01:21:03 -07005359
Mingming Caoa02908f2008-08-19 22:16:07 -04005360static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5361{
5362 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
Theodore Ts'oac51d832008-11-06 16:49:36 -05005363 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5364 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
Mingming Caoa02908f2008-08-19 22:16:07 -04005365}
Theodore Ts'oac51d832008-11-06 16:49:36 -05005366
Mingming Caoa02908f2008-08-19 22:16:07 -04005367/*
5368 * Account for index blocks, block groups bitmaps and block group
5369 * descriptor blocks if modify datablocks and index blocks
5370 * worse case, the indexs blocks spread over different block groups
5371 *
5372 * If datablocks are discontiguous, they are possible to spread over
5373 * different block groups too. If they are contiugous, with flexbg,
5374 * they could still across block group boundary.
5375 *
5376 * Also account for superblock, inode, quota and xattr blocks
5377 */
5378int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5379{
Theodore Ts'o8df96752009-05-01 08:50:38 -04005380 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5381 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04005382 int idxblocks;
5383 int ret = 0;
5384
5385 /*
5386 * How many index blocks need to touch to modify nrblocks?
5387 * The "Chunk" flag indicating whether the nrblocks is
5388 * physically contiguous on disk
5389 *
5390 * For Direct IO and fallocate, they calls get_block to allocate
5391 * one single extent at a time, so they could set the "Chunk" flag
5392 */
5393 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5394
5395 ret = idxblocks;
5396
5397 /*
5398 * Now let's see how many group bitmaps and group descriptors need
5399 * to account
5400 */
5401 groups = idxblocks;
5402 if (chunk)
5403 groups += 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005404 else
Mingming Caoa02908f2008-08-19 22:16:07 -04005405 groups += nrblocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005406
Mingming Caoa02908f2008-08-19 22:16:07 -04005407 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04005408 if (groups > ngroups)
5409 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04005410 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5411 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5412
5413 /* bitmaps and block group descriptor blocks */
5414 ret += groups + gdpblocks;
5415
5416 /* Blocks for super block, inode, quota and xattr blocks */
5417 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005418
5419 return ret;
5420}
5421
5422/*
Mingming Caoa02908f2008-08-19 22:16:07 -04005423 * Calulate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04005424 * the modification of a single pages into a single transaction,
5425 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04005426 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04005427 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04005428 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04005429 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04005430 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04005431 */
5432int ext4_writepage_trans_blocks(struct inode *inode)
5433{
5434 int bpp = ext4_journal_blocks_per_page(inode);
5435 int ret;
5436
5437 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5438
5439 /* Account for data blocks for journalled mode */
5440 if (ext4_should_journal_data(inode))
5441 ret += bpp;
5442 return ret;
5443}
Mingming Caof3bd1f32008-08-19 22:16:03 -04005444
5445/*
5446 * Calculate the journal credits for a chunk of data modification.
5447 *
5448 * This is called from DIO, fallocate or whoever calling
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04005449 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04005450 *
5451 * journal buffers for data blocks are not included here, as DIO
5452 * and fallocate do no need to journal data buffers.
5453 */
5454int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5455{
5456 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5457}
5458
Mingming Caoa02908f2008-08-19 22:16:07 -04005459/*
Mingming Cao617ba132006-10-11 01:20:53 -07005460 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005461 * Give this, we know that the caller already has write access to iloc->bh.
5462 */
Mingming Cao617ba132006-10-11 01:20:53 -07005463int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005464 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005465{
5466 int err = 0;
5467
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005468 if (test_opt(inode->i_sb, I_VERSION))
5469 inode_inc_iversion(inode);
5470
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005471 /* the do_update_inode consumes one bh->b_count */
5472 get_bh(iloc->bh);
5473
Mingming Caodab291a2006-10-11 01:21:01 -07005474 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04005475 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005476 put_bh(iloc->bh);
5477 return err;
5478}
5479
5480/*
5481 * On success, We end up with an outstanding reference count against
5482 * iloc->bh. This _must_ be cleaned up later.
5483 */
5484
5485int
Mingming Cao617ba132006-10-11 01:20:53 -07005486ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5487 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005488{
Frank Mayhar03901312009-01-07 00:06:22 -05005489 int err;
5490
5491 err = ext4_get_inode_loc(inode, iloc);
5492 if (!err) {
5493 BUFFER_TRACE(iloc->bh, "get_write_access");
5494 err = ext4_journal_get_write_access(handle, iloc->bh);
5495 if (err) {
5496 brelse(iloc->bh);
5497 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005498 }
5499 }
Mingming Cao617ba132006-10-11 01:20:53 -07005500 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005501 return err;
5502}
5503
5504/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005505 * Expand an inode by new_extra_isize bytes.
5506 * Returns 0 on success or negative error number on failure.
5507 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05005508static int ext4_expand_extra_isize(struct inode *inode,
5509 unsigned int new_extra_isize,
5510 struct ext4_iloc iloc,
5511 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005512{
5513 struct ext4_inode *raw_inode;
5514 struct ext4_xattr_ibody_header *header;
5515 struct ext4_xattr_entry *entry;
5516
5517 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5518 return 0;
5519
5520 raw_inode = ext4_raw_inode(&iloc);
5521
5522 header = IHDR(inode, raw_inode);
5523 entry = IFIRST(header);
5524
5525 /* No extended attributes present */
5526 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5527 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5528 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5529 new_extra_isize);
5530 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5531 return 0;
5532 }
5533
5534 /* try to expand with EAs present */
5535 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5536 raw_inode, handle);
5537}
5538
5539/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005540 * What we do here is to mark the in-core inode as clean with respect to inode
5541 * dirtiness (it may still be data-dirty).
5542 * This means that the in-core inode may be reaped by prune_icache
5543 * without having to perform any I/O. This is a very good thing,
5544 * because *any* task may call prune_icache - even ones which
5545 * have a transaction open against a different journal.
5546 *
5547 * Is this cheating? Not really. Sure, we haven't written the
5548 * inode out, but prune_icache isn't a user-visible syncing function.
5549 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5550 * we start and wait on commits.
5551 *
5552 * Is this efficient/effective? Well, we're being nice to the system
5553 * by cleaning up our inodes proactively so they can be reaped
5554 * without I/O. But we are potentially leaving up to five seconds'
5555 * worth of inodes floating about which prune_icache wants us to
5556 * write out. One way to fix that would be to get prune_icache()
5557 * to do a write_super() to free up some memory. It has the desired
5558 * effect.
5559 */
Mingming Cao617ba132006-10-11 01:20:53 -07005560int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005561{
Mingming Cao617ba132006-10-11 01:20:53 -07005562 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005563 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5564 static unsigned int mnt_count;
5565 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005566
5567 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07005568 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05005569 if (ext4_handle_valid(handle) &&
5570 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005571 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5572 /*
5573 * We need extra buffer credits since we may write into EA block
5574 * with this same handle. If journal_extend fails, then it will
5575 * only result in a minor loss of functionality for that inode.
5576 * If this is felt to be critical, then e2fsck should be run to
5577 * force a large enough s_min_extra_isize.
5578 */
5579 if ((jbd2_journal_extend(handle,
5580 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5581 ret = ext4_expand_extra_isize(inode,
5582 sbi->s_want_extra_isize,
5583 iloc, handle);
5584 if (ret) {
5585 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005586 if (mnt_count !=
5587 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04005588 ext4_warning(inode->i_sb, __func__,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005589 "Unable to expand inode %lu. Delete"
5590 " some EAs or run e2fsck.",
5591 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005592 mnt_count =
5593 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005594 }
5595 }
5596 }
5597 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005598 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07005599 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005600 return err;
5601}
5602
5603/*
Mingming Cao617ba132006-10-11 01:20:53 -07005604 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005605 *
5606 * We're really interested in the case where a file is being extended.
5607 * i_size has been changed by generic_commit_write() and we thus need
5608 * to include the updated inode in the current transaction.
5609 *
Jan Karaa269eb12009-01-26 17:04:39 +01005610 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005611 * are allocated to the file.
5612 *
5613 * If the inode is marked synchronous, we don't honour that here - doing
5614 * so would cause a commit on atime updates, which we don't bother doing.
5615 * We handle synchronous inodes at the highest possible level.
5616 */
Mingming Cao617ba132006-10-11 01:20:53 -07005617void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005618{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005619 handle_t *handle;
5620
Mingming Cao617ba132006-10-11 01:20:53 -07005621 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005622 if (IS_ERR(handle))
5623 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005624
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005625 ext4_mark_inode_dirty(handle, inode);
5626
Mingming Cao617ba132006-10-11 01:20:53 -07005627 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005628out:
5629 return;
5630}
5631
5632#if 0
5633/*
5634 * Bind an inode's backing buffer_head into this transaction, to prevent
5635 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07005636 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005637 * returns no iloc structure, so the caller needs to repeat the iloc
5638 * lookup to mark the inode dirty later.
5639 */
Mingming Cao617ba132006-10-11 01:20:53 -07005640static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005641{
Mingming Cao617ba132006-10-11 01:20:53 -07005642 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005643
5644 int err = 0;
5645 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07005646 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005647 if (!err) {
5648 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07005649 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005650 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05005651 err = ext4_handle_dirty_metadata(handle,
5652 inode,
5653 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005654 brelse(iloc.bh);
5655 }
5656 }
Mingming Cao617ba132006-10-11 01:20:53 -07005657 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005658 return err;
5659}
5660#endif
5661
Mingming Cao617ba132006-10-11 01:20:53 -07005662int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005663{
5664 journal_t *journal;
5665 handle_t *handle;
5666 int err;
5667
5668 /*
5669 * We have to be very careful here: changing a data block's
5670 * journaling status dynamically is dangerous. If we write a
5671 * data block to the journal, change the status and then delete
5672 * that block, we risk forgetting to revoke the old log record
5673 * from the journal and so a subsequent replay can corrupt data.
5674 * So, first we make sure that the journal is empty and that
5675 * nobody is changing anything.
5676 */
5677
Mingming Cao617ba132006-10-11 01:20:53 -07005678 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005679 if (!journal)
5680 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04005681 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005682 return -EROFS;
5683
Mingming Caodab291a2006-10-11 01:21:01 -07005684 jbd2_journal_lock_updates(journal);
5685 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005686
5687 /*
5688 * OK, there are no updates running now, and all cached data is
5689 * synced to disk. We are now in a completely consistent state
5690 * which doesn't have anything in the journal, and we know that
5691 * no filesystem updates are running, so it is safe to modify
5692 * the inode's in-core data-journaling state flag now.
5693 */
5694
5695 if (val)
Mingming Cao617ba132006-10-11 01:20:53 -07005696 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005697 else
Mingming Cao617ba132006-10-11 01:20:53 -07005698 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5699 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005700
Mingming Caodab291a2006-10-11 01:21:01 -07005701 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005702
5703 /* Finally we can mark the inode as dirty. */
5704
Mingming Cao617ba132006-10-11 01:20:53 -07005705 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005706 if (IS_ERR(handle))
5707 return PTR_ERR(handle);
5708
Mingming Cao617ba132006-10-11 01:20:53 -07005709 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005710 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07005711 ext4_journal_stop(handle);
5712 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005713
5714 return err;
5715}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005716
5717static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5718{
5719 return !buffer_mapped(bh);
5720}
5721
Nick Pigginc2ec1752009-03-31 15:23:21 -07005722int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005723{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005724 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005725 loff_t size;
5726 unsigned long len;
5727 int ret = -EINVAL;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005728 void *fsdata;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005729 struct file *file = vma->vm_file;
5730 struct inode *inode = file->f_path.dentry->d_inode;
5731 struct address_space *mapping = inode->i_mapping;
5732
5733 /*
5734 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5735 * get i_mutex because we are already holding mmap_sem.
5736 */
5737 down_read(&inode->i_alloc_sem);
5738 size = i_size_read(inode);
5739 if (page->mapping != mapping || size <= page_offset(page)
5740 || !PageUptodate(page)) {
5741 /* page got truncated from under us? */
5742 goto out_unlock;
5743 }
5744 ret = 0;
5745 if (PageMappedToDisk(page))
5746 goto out_unlock;
5747
5748 if (page->index == size >> PAGE_CACHE_SHIFT)
5749 len = size & ~PAGE_CACHE_MASK;
5750 else
5751 len = PAGE_CACHE_SIZE;
5752
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005753 lock_page(page);
5754 /*
5755 * return if we have all the buffers mapped. This avoid
5756 * the need to call write_begin/write_end which does a
5757 * journal_start/journal_stop which can block and take
5758 * long time
5759 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005760 if (page_has_buffers(page)) {
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005761 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005762 ext4_bh_unmapped)) {
5763 unlock_page(page);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005764 goto out_unlock;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005765 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005766 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005767 unlock_page(page);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005768 /*
5769 * OK, we need to fill the hole... Do write_begin write_end
5770 * to do block allocation/reservation.We are not holding
5771 * inode.i__mutex here. That allow * parallel write_begin,
5772 * write_end call. lock_page prevent this from happening
5773 * on the same page though
5774 */
5775 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005776 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005777 if (ret < 0)
5778 goto out_unlock;
5779 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005780 len, len, page, fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005781 if (ret < 0)
5782 goto out_unlock;
5783 ret = 0;
5784out_unlock:
Nick Pigginc2ec1752009-03-31 15:23:21 -07005785 if (ret)
5786 ret = VM_FAULT_SIGBUS;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005787 up_read(&inode->i_alloc_sem);
5788 return ret;
5789}