blob: b6e5934f5dd68f6fea324f182c76086486c8edb3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext3/inode.c
3 *
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
Dave Kleikampe9ad5622006-09-27 01:49:35 -070016 * (sct@redhat.com), 1993, 1998
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 * 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
Dave Kleikampe9ad5622006-09-27 01:49:35 -070020 * (jj@sunsite.ms.mff.cuni.cz)
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 *
22 * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
23 */
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/highuid.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/quotaops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/writeback.h>
28#include <linux/mpage.h>
Duane Griffinb5ed3112008-12-19 20:47:14 +000029#include <linux/namei.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070030#include <linux/aio.h>
Al Viro4613ad12012-03-29 22:30:07 -040031#include "ext3.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xattr.h"
33#include "acl.h"
34
35static int ext3_writepage_trans_blocks(struct inode *inode);
Jan Karaee3e77f2011-06-03 21:58:11 +020036static int ext3_block_truncate_page(struct inode *inode, loff_t from);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38/*
39 * Test whether an inode is a fast symlink.
40 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080041static int ext3_inode_is_fast_symlink(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070042{
43 int ea_blocks = EXT3_I(inode)->i_file_acl ?
44 (inode->i_sb->s_blocksize >> 9) : 0;
45
Andrew Mortond6859bf2006-03-26 01:38:03 -080046 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047}
48
Andrew Mortond6859bf2006-03-26 01:38:03 -080049/*
50 * The ext3 forget function must perform a revoke if we are freeing data
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 * which has been journaled. Metadata (eg. indirect blocks) must be
Mingming Caoae6ddcc2006-09-27 01:49:27 -070052 * revoked in all cases.
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 *
54 * "bh" may be NULL: a metadata block may have been freed from memory
55 * but there may still be a record of it in the journal, and that record
56 * still needs to be revoked.
57 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080058int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -070059 struct buffer_head *bh, ext3_fsblk_t blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
61 int err;
62
63 might_sleep();
64
Lukas Czerner785c4bc2011-05-23 18:33:01 +020065 trace_ext3_forget(inode, is_metadata, blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 BUFFER_TRACE(bh, "enter");
67
68 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
69 "data mode %lx\n",
70 bh, is_metadata, inode->i_mode,
71 test_opt(inode->i_sb, DATA_FLAGS));
72
73 /* Never use the revoke function if we are doing full data
74 * journaling: there is no need to, and a V1 superblock won't
75 * support it. Otherwise, only skip the revoke on un-journaled
76 * data blocks. */
77
78 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
79 (!is_metadata && !ext3_should_journal_data(inode))) {
80 if (bh) {
81 BUFFER_TRACE(bh, "call journal_forget");
82 return ext3_journal_forget(handle, bh);
83 }
84 return 0;
85 }
86
87 /*
88 * data!=journal && (is_metadata || should_journal_data(inode))
89 */
90 BUFFER_TRACE(bh, "call ext3_journal_revoke");
91 err = ext3_journal_revoke(handle, blocknr, bh);
92 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -070093 ext3_abort(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 "error %d when attempting revoke", err);
95 BUFFER_TRACE(bh, "exit");
96 return err;
97}
98
99/*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800100 * Work out how many blocks we need to proceed with the next chunk of a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 * truncate transaction.
102 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700103static unsigned long blocks_for_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
105 unsigned long needed;
106
107 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
108
109 /* Give ourselves just enough room to cope with inodes in which
110 * i_blocks is corrupt: we've seen disk corruptions in the past
111 * which resulted in random data in an inode which looked enough
112 * like a regular file for ext3 to try to delete it. Things
113 * will go a bit crazy if that happens, but at least we should
114 * try not to panic the whole kernel. */
115 if (needed < 2)
116 needed = 2;
117
118 /* But we need to bound the transaction so we don't overflow the
119 * journal. */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700120 if (needed > EXT3_MAX_TRANS_DATA)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 needed = EXT3_MAX_TRANS_DATA;
122
Jan Kara1f545872005-06-23 22:01:04 -0700123 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
125
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700126/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 * Truncate transactions can be complex and absolutely huge. So we need to
128 * be able to restart the transaction at a conventient checkpoint to make
129 * sure we don't overflow the journal.
130 *
131 * start_transaction gets us a new handle for a truncate transaction,
132 * and extend_transaction tries to extend the existing one a bit. If
133 * extend fails, we need to propagate the failure up and restart the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700134 * transaction in the top-level truncate loop. --sct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700136static handle_t *start_transaction(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
138 handle_t *result;
139
140 result = ext3_journal_start(inode, blocks_for_truncate(inode));
141 if (!IS_ERR(result))
142 return result;
143
144 ext3_std_error(inode->i_sb, PTR_ERR(result));
145 return result;
146}
147
148/*
149 * Try to extend this transaction for the purposes of truncation.
150 *
151 * Returns 0 if we managed to create more room. If we can't create more
152 * room, and the transaction must be restarted we return 1.
153 */
154static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
155{
156 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
157 return 0;
158 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
159 return 0;
160 return 1;
161}
162
163/*
164 * Restart the transaction associated with *handle. This does a commit,
165 * so before we call here everything must be consistently dirtied against
166 * this transaction.
167 */
Jan Kara00171d32009-08-11 19:06:10 +0200168static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Jan Kara00171d32009-08-11 19:06:10 +0200170 int ret;
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara00171d32009-08-11 19:06:10 +0200173 /*
174 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
175 * At this moment, get_block can be called only for blocks inside
176 * i_size since page cache has been already dropped and writes are
177 * blocked by i_mutex. So we can safely drop the truncate_mutex.
178 */
179 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
180 ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
181 mutex_lock(&EXT3_I(inode)->truncate_mutex);
182 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185/*
Al Viroac14a952010-06-06 07:08:19 -0400186 * Called at inode eviction from icache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 */
Al Viroac14a952010-06-06 07:08:19 -0400188void ext3_evict_inode (struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
Jan Karab22570d2011-07-23 01:21:38 +0200190 struct ext3_inode_info *ei = EXT3_I(inode);
Al Viroac14a952010-06-06 07:08:19 -0400191 struct ext3_block_alloc_info *rsv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 handle_t *handle;
Al Viroac14a952010-06-06 07:08:19 -0400193 int want_delete = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200195 trace_ext3_evict_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400196 if (!inode->i_nlink && !is_bad_inode(inode)) {
Christoph Hellwig871a2932010-03-03 09:05:07 -0500197 dquot_initialize(inode);
Al Viroac14a952010-06-06 07:08:19 -0400198 want_delete = 1;
199 }
Christoph Hellwig907f4552010-03-03 09:05:06 -0500200
Jan Karab22570d2011-07-23 01:21:38 +0200201 /*
202 * When journalling data dirty buffers are tracked only in the journal.
203 * So although mm thinks everything is clean and ready for reaping the
204 * inode might still have some pages to write in the running
205 * transaction or waiting to be checkpointed. Thus calling
206 * journal_invalidatepage() (via truncate_inode_pages()) to discard
207 * these buffers can cause data loss. Also even if we did not discard
208 * these buffers, we would have no way to find them after the inode
209 * is reaped and thus user could see stale data if he tries to read
210 * them before the transaction is checkpointed. So be careful and
211 * force everything to disk here... We use ei->i_datasync_tid to
212 * store the newest transaction containing inode's data.
213 *
214 * Note that directories do not have this problem because they don't
215 * use page cache.
Dan Carpenterbcdd0c12011-11-22 11:00:20 +0300216 *
217 * The s_journal check handles the case when ext3_get_journal() fails
218 * and puts the journal inode.
Jan Karab22570d2011-07-23 01:21:38 +0200219 */
220 if (inode->i_nlink && ext3_should_journal_data(inode) &&
Dan Carpenterbcdd0c12011-11-22 11:00:20 +0300221 EXT3_SB(inode->i_sb)->s_journal &&
Jan Karae6436922013-03-20 14:39:05 +0100222 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
223 inode->i_ino != EXT3_JOURNAL_INO) {
Jan Karab22570d2011-07-23 01:21:38 +0200224 tid_t commit_tid = atomic_read(&ei->i_datasync_tid);
225 journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
226
227 log_start_commit(journal, commit_tid);
228 log_wait_commit(journal, commit_tid);
229 filemap_write_and_wait(&inode->i_data);
230 }
Mark Fashehfef26652005-09-09 13:01:31 -0700231 truncate_inode_pages(&inode->i_data, 0);
232
Al Viroac14a952010-06-06 07:08:19 -0400233 ext3_discard_reservation(inode);
Jan Karab22570d2011-07-23 01:21:38 +0200234 rsv = ei->i_block_alloc_info;
235 ei->i_block_alloc_info = NULL;
Al Viroac14a952010-06-06 07:08:19 -0400236 if (unlikely(rsv))
237 kfree(rsv);
238
239 if (!want_delete)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 goto no_delete;
241
242 handle = start_transaction(inode);
243 if (IS_ERR(handle)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800244 /*
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
247 * cleaned up.
248 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 ext3_orphan_del(NULL, inode);
250 goto no_delete;
251 }
252
253 if (IS_SYNC(inode))
254 handle->h_sync = 1;
255 inode->i_size = 0;
256 if (inode->i_blocks)
257 ext3_truncate(inode);
258 /*
Jan Kara40680f2f2011-05-24 22:24:47 +0200259 * Kill off the orphan record created when the inode lost the last
260 * link. Note that ext3_orphan_del() has to be able to cope with the
261 * deletion of a non-existent orphan - ext3_truncate() could
262 * have removed the record.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
264 ext3_orphan_del(handle, inode);
Jan Karab22570d2011-07-23 01:21:38 +0200265 ei->i_dtime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700267 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 * One subtle ordering requirement: if anything has gone wrong
269 * (transaction abort, IO errors, whatever), then we can still
270 * do these next steps (the fs will already have been marked as
271 * having errors), but we can't free the inode if the mark_dirty
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700272 * fails.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 */
Al Viroac14a952010-06-06 07:08:19 -0400274 if (ext3_mark_inode_dirty(handle, inode)) {
275 /* If that failed, just dquot_drop() and be done with that */
276 dquot_drop(inode);
Jan Karadbd57682012-05-03 14:48:02 +0200277 clear_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400278 } else {
279 ext3_xattr_delete_inode(handle, inode);
280 dquot_free_inode(inode);
281 dquot_drop(inode);
Jan Karadbd57682012-05-03 14:48:02 +0200282 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 ext3_free_inode(handle, inode);
Al Viroac14a952010-06-06 07:08:19 -0400284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 ext3_journal_stop(handle);
286 return;
287no_delete:
Jan Karadbd57682012-05-03 14:48:02 +0200288 clear_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400289 dquot_drop(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292typedef struct {
293 __le32 *p;
294 __le32 key;
295 struct buffer_head *bh;
296} Indirect;
297
298static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
299{
300 p->key = *(p->p = v);
301 p->bh = bh;
302}
303
Andrew Mortond6859bf2006-03-26 01:38:03 -0800304static int verify_chain(Indirect *from, Indirect *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 while (from <= to && from->key == *from->p)
307 from++;
308 return (from > to);
309}
310
311/**
312 * ext3_block_to_path - parse the block number into array of offsets
313 * @inode: inode in question (we are only interested in its superblock)
314 * @i_block: block number to be parsed
315 * @offsets: array to store the offsets in
316 * @boundary: set this non-zero if the referred-to block is likely to be
317 * followed (on disk) by an indirect block.
318 *
319 * To store the locations of file's data ext3 uses a data structure common
320 * for UNIX filesystems - tree of pointers anchored in the inode, with
321 * data blocks at leaves and indirect blocks in intermediate nodes.
322 * This function translates the block number into path in that tree -
323 * return value is the path length and @offsets[n] is the offset of
324 * pointer to (n+1)th node in the nth one. If @block is out of range
325 * (negative or too large) warning is printed and zero returned.
326 *
327 * Note: function doesn't find node addresses, so no IO is needed. All
328 * we need to know is the capacity of indirect blocks (taken from the
329 * inode->i_sb).
330 */
331
332/*
333 * Portability note: the last comparison (check that we fit into triple
334 * indirect block) is spelled differently, because otherwise on an
335 * architecture with 32-bit longs and 8Kb pages we might get into trouble
336 * if our filesystem had 8Kb blocks. We might use long long, but that would
337 * kill us on x86. Oh, well, at least the sign propagation does not matter -
338 * i_block would have to be negative in the very beginning, so we would not
339 * get there at all.
340 */
341
342static int ext3_block_to_path(struct inode *inode,
343 long i_block, int offsets[4], int *boundary)
344{
345 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
346 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
347 const long direct_blocks = EXT3_NDIR_BLOCKS,
348 indirect_blocks = ptrs,
349 double_blocks = (1 << (ptrs_bits * 2));
350 int n = 0;
351 int final = 0;
352
353 if (i_block < 0) {
354 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
355 } else if (i_block < direct_blocks) {
356 offsets[n++] = i_block;
357 final = direct_blocks;
358 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
359 offsets[n++] = EXT3_IND_BLOCK;
360 offsets[n++] = i_block;
361 final = ptrs;
362 } else if ((i_block -= indirect_blocks) < double_blocks) {
363 offsets[n++] = EXT3_DIND_BLOCK;
364 offsets[n++] = i_block >> ptrs_bits;
365 offsets[n++] = i_block & (ptrs - 1);
366 final = ptrs;
367 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
368 offsets[n++] = EXT3_TIND_BLOCK;
369 offsets[n++] = i_block >> (ptrs_bits * 2);
370 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
371 offsets[n++] = i_block & (ptrs - 1);
372 final = ptrs;
373 } else {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800374 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 }
376 if (boundary)
Mingming Cao89747d32006-03-26 01:37:55 -0800377 *boundary = final - 1 - (i_block & (ptrs - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return n;
379}
380
381/**
382 * ext3_get_branch - read the chain of indirect blocks leading to data
383 * @inode: inode in question
384 * @depth: depth of the chain (1 - direct pointer, etc.)
385 * @offsets: offsets of pointers in inode/indirect blocks
386 * @chain: place to store the result
387 * @err: here we store the error value
388 *
389 * Function fills the array of triples <key, p, bh> and returns %NULL
390 * if everything went OK or the pointer to the last filled triple
391 * (incomplete one) otherwise. Upon the return chain[i].key contains
392 * the number of (i+1)-th block in the chain (as it is stored in memory,
393 * i.e. little-endian 32-bit), chain[i].p contains the address of that
394 * number (it points into struct inode for i==0 and into the bh->b_data
395 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
396 * block for i>0 and NULL for i==0. In other words, it holds the block
397 * numbers of the chain, addresses they were taken from (and where we can
398 * verify that chain did not change) and buffer_heads hosting these
399 * numbers.
400 *
401 * Function stops when it stumbles upon zero pointer (absent block)
402 * (pointer to last triple returned, *@err == 0)
403 * or when it gets an IO error reading an indirect block
404 * (ditto, *@err == -EIO)
405 * or when it notices that chain had been changed while it was reading
406 * (ditto, *@err == -EAGAIN)
407 * or when it reads all @depth-1 indirect blocks successfully and finds
408 * the whole chain, all way to the data (returns %NULL, *err == 0).
409 */
410static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
411 Indirect chain[4], int *err)
412{
413 struct super_block *sb = inode->i_sb;
414 Indirect *p = chain;
415 struct buffer_head *bh;
416
417 *err = 0;
418 /* i_data is not going away, no lock needed */
419 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
420 if (!p->key)
421 goto no_block;
422 while (--depth) {
423 bh = sb_bread(sb, le32_to_cpu(p->key));
424 if (!bh)
425 goto failure;
426 /* Reader: pointers */
427 if (!verify_chain(chain, p))
428 goto changed;
429 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
430 /* Reader: end */
431 if (!p->key)
432 goto no_block;
433 }
434 return NULL;
435
436changed:
437 brelse(bh);
438 *err = -EAGAIN;
439 goto no_block;
440failure:
441 *err = -EIO;
442no_block:
443 return p;
444}
445
446/**
447 * ext3_find_near - find a place for allocation with sufficient locality
448 * @inode: owner
449 * @ind: descriptor of indirect block.
450 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000451 * This function returns the preferred place for block allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 * It is used when heuristic for sequential allocation fails.
453 * Rules are:
454 * + if there is a block to the left of our position - allocate near it.
455 * + if pointer will live in indirect block - allocate near that block.
456 * + if pointer will live in inode - allocate in the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700457 * cylinder group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 *
459 * In the latter case we colour the starting block by the callers PID to
460 * prevent it from clashing with concurrent allocations for a different inode
461 * in the same block group. The PID is used here so that functionally related
462 * files will be close-by on-disk.
463 *
464 * Caller must make sure that @ind is valid and will stay that way.
465 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700466static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct ext3_inode_info *ei = EXT3_I(inode);
469 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
470 __le32 *p;
Mingming Cao43d23f92006-06-25 05:48:07 -0700471 ext3_fsblk_t bg_start;
472 ext3_grpblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 /* Try to find previous block */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800475 for (p = ind->p - 1; p >= start; p--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 if (*p)
477 return le32_to_cpu(*p);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800478 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480 /* No such thing, so let's try location of indirect block */
481 if (ind->bh)
482 return ind->bh->b_blocknr;
483
484 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800485 * It is going to be referred to from the inode itself? OK, just put it
486 * into the same cylinder group then.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700488 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 colour = (current->pid % 16) *
490 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
491 return bg_start + colour;
492}
493
494/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000495 * ext3_find_goal - find a preferred place for allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 * @inode: owner
497 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000500 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800501 * returns it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 */
503
Mingming Cao43d23f92006-06-25 05:48:07 -0700504static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800505 Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
Andrew Mortond6859bf2006-03-26 01:38:03 -0800507 struct ext3_block_alloc_info *block_i;
508
509 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511 /*
512 * try the heuristic for sequential allocation,
513 * failing that at least try to get decent locality.
514 */
515 if (block_i && (block == block_i->last_alloc_logical_block + 1)
516 && (block_i->last_alloc_physical_block != 0)) {
Mingming Caofe55c452005-05-01 08:59:20 -0700517 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 }
519
Mingming Caofe55c452005-05-01 08:59:20 -0700520 return ext3_find_near(inode, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521}
Andrew Mortond6859bf2006-03-26 01:38:03 -0800522
Mingming Caob47b2472006-03-26 01:37:56 -0800523/**
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900524 * ext3_blks_to_allocate - Look up the block map and count the number
Mingming Caob47b2472006-03-26 01:37:56 -0800525 * of direct blocks need to be allocated for the given branch.
526 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700527 * @branch: chain of indirect blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800528 * @k: number of blocks need for indirect blocks
529 * @blks: number of data blocks to be mapped.
530 * @blocks_to_boundary: the offset in the indirect block
531 *
532 * return the total number of blocks to be allocate, including the
533 * direct and indirect blocks.
534 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800535static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Mingming Caob47b2472006-03-26 01:37:56 -0800536 int blocks_to_boundary)
537{
538 unsigned long count = 0;
539
540 /*
541 * Simple case, [t,d]Indirect block(s) has not allocated yet
542 * then it's clear blocks on that path have not allocated
543 */
544 if (k > 0) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800545 /* right now we don't handle cross boundary allocation */
Mingming Caob47b2472006-03-26 01:37:56 -0800546 if (blks < blocks_to_boundary + 1)
547 count += blks;
548 else
549 count += blocks_to_boundary + 1;
550 return count;
551 }
552
553 count++;
554 while (count < blks && count <= blocks_to_boundary &&
555 le32_to_cpu(*(branch[0].p + count)) == 0) {
556 count++;
557 }
558 return count;
559}
560
561/**
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900562 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
563 * @handle: handle for this transaction
564 * @inode: owner
565 * @goal: preferred place for allocation
Mingming Caob47b2472006-03-26 01:37:56 -0800566 * @indirect_blks: the number of blocks need to allocate for indirect
567 * blocks
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900568 * @blks: number of blocks need to allocated for direct blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800569 * @new_blocks: on return it will store the new block numbers for
570 * the indirect blocks(if needed) and the first direct block,
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900571 * @err: here we store the error value
572 *
573 * return the number of direct blocks allocated
Mingming Caob47b2472006-03-26 01:37:56 -0800574 */
575static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700576 ext3_fsblk_t goal, int indirect_blks, int blks,
577 ext3_fsblk_t new_blocks[4], int *err)
Mingming Caob47b2472006-03-26 01:37:56 -0800578{
579 int target, i;
580 unsigned long count = 0;
581 int index = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700582 ext3_fsblk_t current_block = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800583 int ret = 0;
584
585 /*
586 * Here we try to allocate the requested multiple blocks at once,
587 * on a best-effort basis.
588 * To build a branch, we should allocate blocks for
589 * the indirect blocks(if not allocated yet), and at least
590 * the first direct block of this branch. That's the
591 * minimum number of blocks need to allocate(required)
592 */
593 target = blks + indirect_blks;
594
595 while (1) {
596 count = target;
597 /* allocating blocks for indirect blocks and direct blocks */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800598 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
Mingming Caob47b2472006-03-26 01:37:56 -0800599 if (*err)
600 goto failed_out;
601
602 target -= count;
603 /* allocate blocks for indirect blocks */
604 while (index < indirect_blks && count) {
605 new_blocks[index++] = current_block++;
606 count--;
607 }
608
609 if (count > 0)
610 break;
611 }
612
613 /* save the new block number for the first direct block */
614 new_blocks[index] = current_block;
615
616 /* total number of blocks allocated for direct blocks */
617 ret = count;
618 *err = 0;
619 return ret;
620failed_out:
621 for (i = 0; i <index; i++)
622 ext3_free_blocks(handle, inode, new_blocks[i], 1);
623 return ret;
624}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
626/**
627 * ext3_alloc_branch - allocate and set up a chain of blocks.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900628 * @handle: handle for this transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 * @inode: owner
Mingming Caob47b2472006-03-26 01:37:56 -0800630 * @indirect_blks: number of allocated indirect blocks
631 * @blks: number of allocated direct blocks
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900632 * @goal: preferred place for allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 * @offsets: offsets (in the blocks) to store the pointers to next.
634 * @branch: place to store the chain in.
635 *
Mingming Caob47b2472006-03-26 01:37:56 -0800636 * This function allocates blocks, zeroes out all but the last one,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 * links them into chain and (if we are synchronous) writes them to disk.
638 * In other words, it prepares a branch that can be spliced onto the
639 * inode. It stores the information about that chain in the branch[], in
640 * the same format as ext3_get_branch() would do. We are calling it after
641 * we had read the existing part of chain and partial points to the last
642 * triple of that (one with zero ->key). Upon the exit we have the same
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -0800643 * picture as after the successful ext3_get_block(), except that in one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 * place chain is disconnected - *branch->p is still zero (we did not
645 * set the last link), but branch->key contains the number that should
646 * be placed into *branch->p to fill that gap.
647 *
648 * If allocation fails we free all blocks we've allocated (and forget
649 * their buffer_heads) and return the error value the from failed
650 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
651 * as described above and return 0.
652 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700654 int indirect_blks, int *blks, ext3_fsblk_t goal,
Mingming Caob47b2472006-03-26 01:37:56 -0800655 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 int blocksize = inode->i_sb->s_blocksize;
Mingming Caob47b2472006-03-26 01:37:56 -0800658 int i, n = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 int err = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800660 struct buffer_head *bh;
661 int num;
Mingming Cao43d23f92006-06-25 05:48:07 -0700662 ext3_fsblk_t new_blocks[4];
663 ext3_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Mingming Caob47b2472006-03-26 01:37:56 -0800665 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
666 *blks, new_blocks, &err);
667 if (err)
668 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Mingming Caob47b2472006-03-26 01:37:56 -0800670 branch[0].key = cpu_to_le32(new_blocks[0]);
671 /*
672 * metadata blocks and data blocks are allocated.
673 */
674 for (n = 1; n <= indirect_blks; n++) {
675 /*
676 * Get buffer_head for parent block, zero it out
677 * and set the pointer to new one, then send
678 * parent to disk.
679 */
680 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
Wang Shilong61f43e62013-01-12 01:22:33 -0800681 if (unlikely(!bh)) {
682 err = -ENOMEM;
683 goto failed;
684 }
Mingming Caob47b2472006-03-26 01:37:56 -0800685 branch[n].bh = bh;
686 lock_buffer(bh);
687 BUFFER_TRACE(bh, "call get_create_access");
688 err = ext3_journal_get_create_access(handle, bh);
689 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 unlock_buffer(bh);
Mingming Caob47b2472006-03-26 01:37:56 -0800691 brelse(bh);
692 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Mingming Caob47b2472006-03-26 01:37:56 -0800695 memset(bh->b_data, 0, blocksize);
696 branch[n].p = (__le32 *) bh->b_data + offsets[n];
697 branch[n].key = cpu_to_le32(new_blocks[n]);
698 *branch[n].p = branch[n].key;
699 if ( n == indirect_blks) {
700 current_block = new_blocks[n];
701 /*
702 * End of chain, update the last new metablock of
703 * the chain to point to the new allocated
704 * data blocks numbers
705 */
706 for (i=1; i < num; i++)
707 *(branch[n].p + i) = cpu_to_le32(++current_block);
708 }
709 BUFFER_TRACE(bh, "marking uptodate");
710 set_buffer_uptodate(bh);
711 unlock_buffer(bh);
712
713 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
714 err = ext3_journal_dirty_metadata(handle, bh);
715 if (err)
716 goto failed;
717 }
718 *blks = num;
719 return err;
720failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 /* Allocation failed, free what we already allocated */
Mingming Caob47b2472006-03-26 01:37:56 -0800722 for (i = 1; i <= n ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 BUFFER_TRACE(branch[i].bh, "call journal_forget");
724 ext3_journal_forget(handle, branch[i].bh);
725 }
Wang Shilong61f43e62013-01-12 01:22:33 -0800726 for (i = 0; i < indirect_blks; i++)
Mingming Caob47b2472006-03-26 01:37:56 -0800727 ext3_free_blocks(handle, inode, new_blocks[i], 1);
728
729 ext3_free_blocks(handle, inode, new_blocks[i], num);
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return err;
732}
733
734/**
Andrew Mortond6859bf2006-03-26 01:38:03 -0800735 * ext3_splice_branch - splice the allocated branch onto inode.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900736 * @handle: handle for this transaction
Andrew Mortond6859bf2006-03-26 01:38:03 -0800737 * @inode: owner
738 * @block: (logical) number of block we are adding
Andrew Mortond6859bf2006-03-26 01:38:03 -0800739 * @where: location of missing link
740 * @num: number of indirect blocks we are adding
741 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800743 * This function fills the missing link and does all housekeeping needed in
744 * inode (->i_blocks, etc.). In case of success we end up with the full
745 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800747static int ext3_splice_branch(handle_t *handle, struct inode *inode,
748 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
750 int i;
751 int err = 0;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800752 struct ext3_block_alloc_info *block_i;
Mingming Cao43d23f92006-06-25 05:48:07 -0700753 ext3_fsblk_t current_block;
Jan Karafe8bc912009-10-16 19:26:15 +0200754 struct ext3_inode_info *ei = EXT3_I(inode);
Kazuya Mioac1334b2012-02-08 16:34:29 +0900755 struct timespec now;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800756
Jan Karafe8bc912009-10-16 19:26:15 +0200757 block_i = ei->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 /*
759 * If we're splicing into a [td]indirect block (as opposed to the
760 * inode) then we need to get write access to the [td]indirect block
761 * before the splice.
762 */
763 if (where->bh) {
764 BUFFER_TRACE(where->bh, "get_write_access");
765 err = ext3_journal_get_write_access(handle, where->bh);
766 if (err)
767 goto err_out;
768 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 /* That's it */
770
771 *where->p = where->key;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800772
773 /*
774 * Update the host buffer_head or inode to point to more just allocated
775 * direct blocks blocks
776 */
Mingming Caob47b2472006-03-26 01:37:56 -0800777 if (num == 0 && blks > 1) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700778 current_block = le32_to_cpu(where->key) + 1;
Mingming Caob47b2472006-03-26 01:37:56 -0800779 for (i = 1; i < blks; i++)
780 *(where->p + i ) = cpu_to_le32(current_block++);
781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 /*
784 * update the most recently allocated logical & physical block
785 * in i_block_alloc_info, to assist find the proper goal block for next
786 * allocation
787 */
788 if (block_i) {
Mingming Caob47b2472006-03-26 01:37:56 -0800789 block_i->last_alloc_logical_block = block + blks - 1;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800790 block_i->last_alloc_physical_block =
Mingming Cao5dea5172006-05-03 19:55:12 -0700791 le32_to_cpu(where[num].key) + blks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
793
794 /* We are done with atomic stuff, now do the rest of housekeeping */
Kazuya Mioac1334b2012-02-08 16:34:29 +0900795 now = CURRENT_TIME_SEC;
796 if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) {
797 inode->i_ctime = now;
798 ext3_mark_inode_dirty(handle, inode);
799 }
Jan Karafe8bc912009-10-16 19:26:15 +0200800 /* ext3_mark_inode_dirty already updated i_sync_tid */
801 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
803 /* had we spliced it onto indirect block? */
804 if (where->bh) {
805 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800806 * If we spliced it onto an indirect block, we haven't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * altered the inode. Note however that if it is being spliced
808 * onto an indirect block at the very end of the file (the
809 * file is growing) then we *will* alter the inode to reflect
810 * the new i_size. But that is not done here - it is done in
811 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
812 */
813 jbd_debug(5, "splicing indirect only\n");
814 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
815 err = ext3_journal_dirty_metadata(handle, where->bh);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700816 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 goto err_out;
818 } else {
819 /*
820 * OK, we spliced it into the inode itself on a direct block.
821 * Inode was dirtied above.
822 */
823 jbd_debug(5, "splicing direct\n");
824 }
825 return err;
826
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827err_out:
Mingming Caob47b2472006-03-26 01:37:56 -0800828 for (i = 1; i <= num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 BUFFER_TRACE(where[i].bh, "call journal_forget");
830 ext3_journal_forget(handle, where[i].bh);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800831 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 }
Mingming Caob47b2472006-03-26 01:37:56 -0800833 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 return err;
836}
837
838/*
839 * Allocation strategy is simple: if we have to allocate something, we will
840 * have to go the whole way to leaf. So let's do it before attaching anything
841 * to tree, set linkage between the newborn blocks, write them if sync is
842 * required, recheck the path, free and repeat if check fails, otherwise
843 * set the last missing link (that will protect us from any truncate-generated
844 * removals - all blocks on the path are immune now) and possibly force the
845 * write on the parent block.
846 * That has a nice additional property: no special recovery from the failed
847 * allocations is needed - we simply release blocks and do not touch anything
848 * reachable from inode.
849 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800850 * `handle' can be NULL if create == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 *
852 * The BKL may not be held on entry here. Be sure to take it early.
Mingming Cao89747d32006-03-26 01:37:55 -0800853 * return > 0, # of blocks mapped or allocated.
854 * return = 0, if plain lookup failed.
855 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800857int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
858 sector_t iblock, unsigned long maxblocks,
859 struct buffer_head *bh_result,
Jan Kara43237b52009-05-20 18:41:58 +0200860 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861{
862 int err = -EIO;
863 int offsets[4];
864 Indirect chain[4];
865 Indirect *partial;
Mingming Cao43d23f92006-06-25 05:48:07 -0700866 ext3_fsblk_t goal;
Mingming Caob47b2472006-03-26 01:37:56 -0800867 int indirect_blks;
Mingming Cao89747d32006-03-26 01:37:55 -0800868 int blocks_to_boundary = 0;
869 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 struct ext3_inode_info *ei = EXT3_I(inode);
Mingming Cao89747d32006-03-26 01:37:55 -0800871 int count = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700872 ext3_fsblk_t first_block = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200875 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 J_ASSERT(handle != NULL || create == 0);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800877 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
879 if (depth == 0)
880 goto out;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
883
884 /* Simplest case - block found, no allocation needed */
885 if (!partial) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700886 first_block = le32_to_cpu(chain[depth - 1].key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 clear_buffer_new(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800888 count++;
889 /*map more blocks*/
890 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700891 ext3_fsblk_t blk;
Mingming Cao5dea5172006-05-03 19:55:12 -0700892
Jan Karae8ef7aa2009-06-17 16:26:23 -0700893 if (!verify_chain(chain, chain + depth - 1)) {
Mingming Cao89747d32006-03-26 01:37:55 -0800894 /*
895 * Indirect block might be removed by
896 * truncate while we were reading it.
897 * Handling of that case: forget what we've
898 * got now. Flag the err as EAGAIN, so it
899 * will reread.
900 */
901 err = -EAGAIN;
902 count = 0;
903 break;
904 }
Mingming Cao5dea5172006-05-03 19:55:12 -0700905 blk = le32_to_cpu(*(chain[depth-1].p + count));
906
907 if (blk == first_block + count)
Mingming Cao89747d32006-03-26 01:37:55 -0800908 count++;
909 else
910 break;
911 }
912 if (err != -EAGAIN)
913 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 }
915
916 /* Next simple case - plain lookup or failed read of indirect block */
Mingming Caofe55c452005-05-01 08:59:20 -0700917 if (!create || err == -EIO)
918 goto cleanup;
919
Jan Kara40680f2f2011-05-24 22:24:47 +0200920 /*
921 * Block out ext3_truncate while we alter the tree
922 */
Arjan van de Ven97461512006-03-23 03:00:42 -0800923 mutex_lock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700924
925 /*
926 * If the indirect block is missing while we are reading
927 * the chain(ext3_get_branch() returns -EAGAIN err), or
928 * if the chain has been changed after we grab the semaphore,
929 * (either because another process truncated this branch, or
930 * another get_block allocated this branch) re-grab the chain to see if
931 * the request block has been allocated or not.
932 *
933 * Since we already block the truncate/other get_block
934 * at this point, we will have the current copy of the chain when we
935 * splice the branch into the tree.
936 */
937 if (err == -EAGAIN || !verify_chain(chain, partial)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 while (partial > chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 brelse(partial->bh);
940 partial--;
941 }
Mingming Caofe55c452005-05-01 08:59:20 -0700942 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
943 if (!partial) {
Mingming Cao89747d32006-03-26 01:37:55 -0800944 count++;
Arjan van de Ven97461512006-03-23 03:00:42 -0800945 mutex_unlock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700946 if (err)
947 goto cleanup;
948 clear_buffer_new(bh_result);
949 goto got_it;
950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 }
952
953 /*
Mingming Caofe55c452005-05-01 08:59:20 -0700954 * Okay, we need to do block allocation. Lazily initialize the block
955 * allocation info here if necessary
956 */
957 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 ext3_init_block_alloc_info(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800960 goal = ext3_find_goal(inode, iblock, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Mingming Caob47b2472006-03-26 01:37:56 -0800962 /* the number of blocks need to allocate for [d,t]indirect blocks */
963 indirect_blks = (chain + depth) - partial - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
965 /*
Mingming Caob47b2472006-03-26 01:37:56 -0800966 * Next look up the indirect map to count the totoal number of
967 * direct blocks to allocate for this branch.
968 */
969 count = ext3_blks_to_allocate(partial, indirect_blks,
970 maxblocks, blocks_to_boundary);
Mingming Caob47b2472006-03-26 01:37:56 -0800971 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
Mingming Caofe55c452005-05-01 08:59:20 -0700972 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Mingming Caofe55c452005-05-01 08:59:20 -0700974 /*
975 * The ext3_splice_branch call will free and forget any buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 * on the new chain if there is a failure, but that risks using
977 * up transaction credits, especially for bitmaps where the
978 * credits cannot be returned. Can we handle this somehow? We
Mingming Caofe55c452005-05-01 08:59:20 -0700979 * may need to return -EAGAIN upwards in the worst case. --sct
980 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 if (!err)
Mingming Caob47b2472006-03-26 01:37:56 -0800982 err = ext3_splice_branch(handle, inode, iblock,
983 partial, indirect_blks, count);
Arjan van de Ven97461512006-03-23 03:00:42 -0800984 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 if (err)
986 goto cleanup;
987
988 set_buffer_new(bh_result);
Mingming Caofe55c452005-05-01 08:59:20 -0700989got_it:
990 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
Suparna Bhattacharya20acaa12006-09-16 12:15:58 -0700991 if (count > blocks_to_boundary)
Mingming Caofe55c452005-05-01 08:59:20 -0700992 set_buffer_boundary(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800993 err = count;
Mingming Caofe55c452005-05-01 08:59:20 -0700994 /* Clean up and exit */
995 partial = chain + depth - 1; /* the whole chain */
996cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 while (partial > chain) {
Mingming Caofe55c452005-05-01 08:59:20 -0700998 BUFFER_TRACE(partial->bh, "call brelse");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 brelse(partial->bh);
1000 partial--;
1001 }
Mingming Caofe55c452005-05-01 08:59:20 -07001002 BUFFER_TRACE(bh_result, "returned");
1003out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001004 trace_ext3_get_blocks_exit(inode, iblock,
1005 depth ? le32_to_cpu(chain[depth-1].key) : 0,
1006 count, err);
Mingming Caofe55c452005-05-01 08:59:20 -07001007 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008}
1009
Jan Karabd1939d2008-02-06 01:40:21 -08001010/* Maximum number of blocks we map for direct IO at once. */
1011#define DIO_MAX_BLOCKS 4096
1012/*
1013 * Number of credits we need for writing DIO_MAX_BLOCKS:
1014 * We need sb + group descriptor + bitmap + inode -> 4
1015 * For B blocks with A block pointers per block we need:
1016 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1017 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1018 */
1019#define DIO_CREDITS 25
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001021static int ext3_get_block(struct inode *inode, sector_t iblock,
1022 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001024 handle_t *handle = ext3_journal_current_handle();
Jan Karabd1939d2008-02-06 01:40:21 -08001025 int ret = 0, started = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -08001026 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Jan Karabd1939d2008-02-06 01:40:21 -08001028 if (create && !handle) { /* Direct IO write... */
1029 if (max_blocks > DIO_MAX_BLOCKS)
1030 max_blocks = DIO_MAX_BLOCKS;
1031 handle = ext3_journal_start(inode, DIO_CREDITS +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001032 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
Jan Karabd1939d2008-02-06 01:40:21 -08001033 if (IS_ERR(handle)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 ret = PTR_ERR(handle);
Jan Karabd1939d2008-02-06 01:40:21 -08001035 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
Jan Karabd1939d2008-02-06 01:40:21 -08001037 started = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
1039
Jan Karabd1939d2008-02-06 01:40:21 -08001040 ret = ext3_get_blocks_handle(handle, inode, iblock,
Jan Kara43237b52009-05-20 18:41:58 +02001041 max_blocks, bh_result, create);
Jan Karabd1939d2008-02-06 01:40:21 -08001042 if (ret > 0) {
1043 bh_result->b_size = (ret << inode->i_blkbits);
1044 ret = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001045 }
Jan Karabd1939d2008-02-06 01:40:21 -08001046 if (started)
1047 ext3_journal_stop(handle);
1048out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 return ret;
1050}
1051
Josef Bacik68c9d702008-10-03 17:32:43 -04001052int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1053 u64 start, u64 len)
1054{
1055 return generic_block_fiemap(inode, fieinfo, start, len,
1056 ext3_get_block);
1057}
1058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059/*
1060 * `handle' can be NULL if create is zero
1061 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001062struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1063 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
1065 struct buffer_head dummy;
1066 int fatal = 0, err;
1067
1068 J_ASSERT(handle != NULL || create == 0);
1069
1070 dummy.b_state = 0;
1071 dummy.b_blocknr = -1000;
1072 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -08001073 err = ext3_get_blocks_handle(handle, inode, block, 1,
Jan Kara43237b52009-05-20 18:41:58 +02001074 &dummy, create);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001075 /*
1076 * ext3_get_blocks_handle() returns number of blocks
1077 * mapped. 0 in case of a HOLE.
1078 */
1079 if (err > 0) {
Julia Lawall47897752012-11-03 21:30:21 +01001080 WARN_ON(err > 1);
Mingming Cao89747d32006-03-26 01:37:55 -08001081 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001082 }
1083 *errp = err;
1084 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 struct buffer_head *bh;
1086 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Wang Shilong1b7d76e2013-01-12 01:36:17 -08001087 if (unlikely(!bh)) {
Wang Shilongc04e88e2013-01-15 21:20:01 -08001088 *errp = -ENOMEM;
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001089 goto err;
1090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (buffer_new(&dummy)) {
1092 J_ASSERT(create != 0);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001093 J_ASSERT(handle != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Andrew Mortond6859bf2006-03-26 01:38:03 -08001095 /*
1096 * Now that we do not always journal data, we should
1097 * keep in mind whether this should always journal the
1098 * new buffer as metadata. For now, regular file
1099 * writes use ext3_get_block instead, so it's not a
1100 * problem.
1101 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 lock_buffer(bh);
1103 BUFFER_TRACE(bh, "call get_create_access");
1104 fatal = ext3_journal_get_create_access(handle, bh);
1105 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001106 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 set_buffer_uptodate(bh);
1108 }
1109 unlock_buffer(bh);
1110 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1111 err = ext3_journal_dirty_metadata(handle, bh);
1112 if (!fatal)
1113 fatal = err;
1114 } else {
1115 BUFFER_TRACE(bh, "not a new buffer");
1116 }
1117 if (fatal) {
1118 *errp = fatal;
1119 brelse(bh);
1120 bh = NULL;
1121 }
1122 return bh;
1123 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001124err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return NULL;
1126}
1127
Andrew Mortond6859bf2006-03-26 01:38:03 -08001128struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 int block, int create, int *err)
1130{
1131 struct buffer_head * bh;
1132
1133 bh = ext3_getblk(handle, inode, block, create, err);
1134 if (!bh)
1135 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001136 if (bh_uptodate_or_lock(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001138 get_bh(bh);
1139 bh->b_end_io = end_buffer_read_sync;
1140 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 wait_on_buffer(bh);
1142 if (buffer_uptodate(bh))
1143 return bh;
1144 put_bh(bh);
1145 *err = -EIO;
1146 return NULL;
1147}
1148
1149static int walk_page_buffers( handle_t *handle,
1150 struct buffer_head *head,
1151 unsigned from,
1152 unsigned to,
1153 int *partial,
1154 int (*fn)( handle_t *handle,
1155 struct buffer_head *bh))
1156{
1157 struct buffer_head *bh;
1158 unsigned block_start, block_end;
1159 unsigned blocksize = head->b_size;
1160 int err, ret = 0;
1161 struct buffer_head *next;
1162
1163 for ( bh = head, block_start = 0;
1164 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001165 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 {
1167 next = bh->b_this_page;
1168 block_end = block_start + blocksize;
1169 if (block_end <= from || block_start >= to) {
1170 if (partial && !buffer_uptodate(bh))
1171 *partial = 1;
1172 continue;
1173 }
1174 err = (*fn)(handle, bh);
1175 if (!ret)
1176 ret = err;
1177 }
1178 return ret;
1179}
1180
1181/*
1182 * To preserve ordering, it is essential that the hole instantiation and
1183 * the data write be encapsulated in a single transaction. We cannot
1184 * close off a transaction and start a new one between the ext3_get_block()
1185 * and the commit_write(). So doing the journal_start at the start of
1186 * prepare_write() is the right place.
1187 *
1188 * Also, this function can nest inside ext3_writepage() ->
1189 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1190 * has generated enough buffer credits to do the whole page. So we won't
1191 * block on the journal in that case, which is good, because the caller may
1192 * be PF_MEMALLOC.
1193 *
1194 * By accident, ext3 can be reentered when a transaction is open via
1195 * quota file writes. If we were to commit the transaction while thus
1196 * reentered, there can be a deadlock - we would be holding a quota
1197 * lock, and the commit would never complete if another thread had a
1198 * transaction open and was blocking on the quota lock - a ranking
1199 * violation.
1200 *
1201 * So what we do is to rely on the fact that journal_stop/journal_start
1202 * will _not_ run commit under these circumstances because handle->h_ref
1203 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001204 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001206static int do_journal_get_write_access(handle_t *handle,
1207 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
Jan Kara5f11e6a2010-08-05 12:38:26 +02001209 int dirty = buffer_dirty(bh);
1210 int ret;
1211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (!buffer_mapped(bh) || buffer_freed(bh))
1213 return 0;
Jan Kara5f11e6a2010-08-05 12:38:26 +02001214 /*
1215 * __block_prepare_write() could have dirtied some buffers. Clean
1216 * the dirty bit as jbd2_journal_get_write_access() could complain
1217 * otherwise about fs integrity issues. Setting of the dirty bit
1218 * by __block_prepare_write() isn't a real problem here as we clear
1219 * the bit before releasing a page lock and thus writeback cannot
1220 * ever write the buffer.
1221 */
1222 if (dirty)
1223 clear_buffer_dirty(bh);
1224 ret = ext3_journal_get_write_access(handle, bh);
1225 if (!ret && dirty)
1226 ret = ext3_journal_dirty_metadata(handle, bh);
1227 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228}
1229
Jan Kara68eb3db2009-12-01 16:53:06 +01001230/*
1231 * Truncate blocks that were not used by write. We have to truncate the
1232 * pagecache as well so that corresponding buffers get properly unmapped.
1233 */
1234static void ext3_truncate_failed_write(struct inode *inode)
1235{
1236 truncate_inode_pages(inode->i_mapping, inode->i_size);
1237 ext3_truncate(inode);
1238}
1239
Jan Karaee3e77f2011-06-03 21:58:11 +02001240/*
1241 * Truncate blocks that were not used by direct IO write. We have to zero out
1242 * the last file block as well because direct IO might have written to it.
1243 */
1244static void ext3_truncate_failed_direct_write(struct inode *inode)
1245{
1246 ext3_block_truncate_page(inode, inode->i_size);
1247 ext3_truncate(inode);
1248}
1249
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001250static int ext3_write_begin(struct file *file, struct address_space *mapping,
1251 loff_t pos, unsigned len, unsigned flags,
1252 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253{
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001254 struct inode *inode = mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001255 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 handle_t *handle;
1257 int retries = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001258 struct page *page;
1259 pgoff_t index;
1260 unsigned from, to;
Jan Kara695f6ae2009-04-02 16:57:17 -07001261 /* Reserve one block more for addition to orphan list in case
1262 * we allocate blocks but write fails for some reason */
1263 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001264
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001265 trace_ext3_write_begin(inode, pos, len, flags);
1266
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001267 index = pos >> PAGE_CACHE_SHIFT;
1268 from = pos & (PAGE_CACHE_SIZE - 1);
1269 to = from + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271retry:
Nick Piggin54566b22009-01-04 12:00:53 -08001272 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001273 if (!page)
1274 return -ENOMEM;
1275 *pagep = page;
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 handle = ext3_journal_start(inode, needed_blocks);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001278 if (IS_ERR(handle)) {
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001279 unlock_page(page);
1280 page_cache_release(page);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001281 ret = PTR_ERR(handle);
1282 goto out;
1283 }
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001284 ret = __block_write_begin(page, pos, len, ext3_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 if (ret)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001286 goto write_begin_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 if (ext3_should_journal_data(inode)) {
1289 ret = walk_page_buffers(handle, page_buffers(page),
1290 from, to, NULL, do_journal_get_write_access);
1291 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001292write_begin_failed:
1293 if (ret) {
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001294 /*
1295 * block_write_begin may have instantiated a few blocks
1296 * outside i_size. Trim these off again. Don't need
1297 * i_size_read because we hold i_mutex.
Jan Kara695f6ae2009-04-02 16:57:17 -07001298 *
1299 * Add inode to orphan list in case we crash before truncate
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001300 * finishes. Do this only if ext3_can_truncate() agrees so
1301 * that orphan processing code is happy.
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001302 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001303 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001304 ext3_orphan_add(handle, inode);
1305 ext3_journal_stop(handle);
1306 unlock_page(page);
1307 page_cache_release(page);
1308 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001309 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1312 goto retry;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001313out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return ret;
1315}
1316
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001317
Andrew Mortond6859bf2006-03-26 01:38:03 -08001318int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319{
1320 int err = journal_dirty_data(handle, bh);
1321 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001322 ext3_journal_abort_handle(__func__, __func__,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001323 bh, handle, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 return err;
1325}
1326
Jan Kara695f6ae2009-04-02 16:57:17 -07001327/* For ordered writepage and write_end functions */
1328static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1329{
1330 /*
1331 * Write could have mapped the buffer but it didn't copy the data in
1332 * yet. So avoid filing such buffer into a transaction.
1333 */
1334 if (buffer_mapped(bh) && buffer_uptodate(bh))
1335 return ext3_journal_dirty_data(handle, bh);
1336 return 0;
1337}
1338
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001339/* For write_end() in data=journal mode */
1340static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 if (!buffer_mapped(bh) || buffer_freed(bh))
1343 return 0;
1344 set_buffer_uptodate(bh);
1345 return ext3_journal_dirty_metadata(handle, bh);
1346}
1347
1348/*
Jan Kara695f6ae2009-04-02 16:57:17 -07001349 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1350 * for the whole page but later we failed to copy the data in. Update inode
1351 * size according to what we managed to copy. The rest is going to be
1352 * truncated in write_end function.
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001353 */
Jan Kara695f6ae2009-04-02 16:57:17 -07001354static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001355{
Jan Kara695f6ae2009-04-02 16:57:17 -07001356 /* What matters to us is i_disksize. We don't write i_size anywhere */
1357 if (pos + copied > inode->i_size)
1358 i_size_write(inode, pos + copied);
1359 if (pos + copied > EXT3_I(inode)->i_disksize) {
1360 EXT3_I(inode)->i_disksize = pos + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001361 mark_inode_dirty(inode);
1362 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001363}
1364
1365/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 * We need to pick up the new inode size which generic_commit_write gave us
1367 * `file' can be NULL - eg, when called from page_symlink().
1368 *
1369 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1370 * buffers are managed internally.
1371 */
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001372static int ext3_ordered_write_end(struct file *file,
1373 struct address_space *mapping,
1374 loff_t pos, unsigned len, unsigned copied,
1375 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
1377 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001378 struct inode *inode = file->f_mapping->host;
1379 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 int ret = 0, ret2;
1381
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001382 trace_ext3_ordered_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001383 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1384
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001385 from = pos & (PAGE_CACHE_SIZE - 1);
Jan Kara695f6ae2009-04-02 16:57:17 -07001386 to = from + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 ret = walk_page_buffers(handle, page_buffers(page),
Jan Kara695f6ae2009-04-02 16:57:17 -07001388 from, to, NULL, journal_dirty_data_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Jan Kara695f6ae2009-04-02 16:57:17 -07001390 if (ret == 0)
1391 update_file_sizes(inode, pos, copied);
1392 /*
1393 * There may be allocated blocks outside of i_size because
1394 * we failed to copy some data. Prepare for truncate.
1395 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001396 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001397 ext3_orphan_add(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 ret2 = ext3_journal_stop(handle);
1399 if (!ret)
1400 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001401 unlock_page(page);
1402 page_cache_release(page);
1403
Jan Kara695f6ae2009-04-02 16:57:17 -07001404 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001405 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001406 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001409static int ext3_writeback_write_end(struct file *file,
1410 struct address_space *mapping,
1411 loff_t pos, unsigned len, unsigned copied,
1412 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
1414 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001415 struct inode *inode = file->f_mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001416 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001418 trace_ext3_writeback_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001419 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1420 update_file_sizes(inode, pos, copied);
1421 /*
1422 * There may be allocated blocks outside of i_size because
1423 * we failed to copy some data. Prepare for truncate.
1424 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001425 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001426 ext3_orphan_add(handle, inode);
1427 ret = ext3_journal_stop(handle);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001428 unlock_page(page);
1429 page_cache_release(page);
1430
Jan Kara695f6ae2009-04-02 16:57:17 -07001431 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001432 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001433 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434}
1435
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001436static int ext3_journalled_write_end(struct file *file,
1437 struct address_space *mapping,
1438 loff_t pos, unsigned len, unsigned copied,
1439 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440{
1441 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001442 struct inode *inode = mapping->host;
Jan Karab22570d2011-07-23 01:21:38 +02001443 struct ext3_inode_info *ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 int ret = 0, ret2;
1445 int partial = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001446 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001448 trace_ext3_journalled_write_end(inode, pos, len, copied);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001449 from = pos & (PAGE_CACHE_SIZE - 1);
1450 to = from + len;
1451
1452 if (copied < len) {
1453 if (!PageUptodate(page))
1454 copied = 0;
Jan Kara695f6ae2009-04-02 16:57:17 -07001455 page_zero_new_buffers(page, from + copied, to);
1456 to = from + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001457 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001460 to, &partial, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 if (!partial)
1462 SetPageUptodate(page);
Jan Kara695f6ae2009-04-02 16:57:17 -07001463
1464 if (pos + copied > inode->i_size)
1465 i_size_write(inode, pos + copied);
1466 /*
1467 * There may be allocated blocks outside of i_size because
1468 * we failed to copy some data. Prepare for truncate.
1469 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001470 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001471 ext3_orphan_add(handle, inode);
Jan Kara9df93932010-01-06 21:58:48 +01001472 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001473 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1474 if (inode->i_size > ei->i_disksize) {
1475 ei->i_disksize = inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001477 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 ret = ret2;
1479 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 ret2 = ext3_journal_stop(handle);
1482 if (!ret)
1483 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001484 unlock_page(page);
1485 page_cache_release(page);
1486
Jan Kara695f6ae2009-04-02 16:57:17 -07001487 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001488 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001489 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490}
1491
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001492/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 * bmap() is special. It gets used by applications such as lilo and by
1494 * the swapper to find the on-disk block of a specific piece of data.
1495 *
1496 * Naturally, this is dangerous if the block concerned is still in the
1497 * journal. If somebody makes a swapfile on an ext3 data-journaling
1498 * filesystem and enables swap, then they may get a nasty shock when the
1499 * data getting swapped to that swapfile suddenly gets overwritten by
1500 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001501 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 *
1503 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001504 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 */
1506static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1507{
1508 struct inode *inode = mapping->host;
1509 journal_t *journal;
1510 int err;
1511
Jan Kara9df93932010-01-06 21:58:48 +01001512 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001513 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 * This is a REALLY heavyweight approach, but the use of
1515 * bmap on dirty files is expected to be extremely rare:
1516 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001517 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 *
1519 * (bmap requires CAP_SYS_RAWIO so this does not
1520 * represent an unprivileged user DOS attack --- we'd be
1521 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001522 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 *
1524 * NB. EXT3_STATE_JDATA is not set on files other than
1525 * regular files. If somebody wants to bmap a directory
1526 * or symlink and gets confused because the buffer
1527 * hasn't yet been flushed to disk, they deserve
1528 * everything they get.
1529 */
1530
Jan Kara9df93932010-01-06 21:58:48 +01001531 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 journal = EXT3_JOURNAL(inode);
1533 journal_lock_updates(journal);
1534 err = journal_flush(journal);
1535 journal_unlock_updates(journal);
1536
1537 if (err)
1538 return 0;
1539 }
1540
1541 return generic_block_bmap(mapping,block,ext3_get_block);
1542}
1543
1544static int bget_one(handle_t *handle, struct buffer_head *bh)
1545{
1546 get_bh(bh);
1547 return 0;
1548}
1549
1550static int bput_one(handle_t *handle, struct buffer_head *bh)
1551{
1552 put_bh(bh);
1553 return 0;
1554}
1555
Jan Kara9e80d402009-03-26 13:08:04 +01001556static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1557{
1558 return !buffer_mapped(bh);
1559}
Jan Kara695f6ae2009-04-02 16:57:17 -07001560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561/*
1562 * Note that we always start a transaction even if we're not journalling
1563 * data. This is to preserve ordering: any hole instantiation within
1564 * __block_write_full_page -> ext3_get_block() should be journalled
1565 * along with the data so we don't crash and then get metadata which
1566 * refers to old data.
1567 *
1568 * In all journalling modes block_write_full_page() will start the I/O.
1569 *
1570 * Problem:
1571 *
1572 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1573 * ext3_writepage()
1574 *
1575 * Similar for:
1576 *
1577 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1578 *
1579 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001580 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 *
1582 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1583 * allocations fail.
1584 *
1585 * 16May01: If we're reentered then journal_current_handle() will be
1586 * non-zero. We simply *return*.
1587 *
1588 * 1 July 2001: @@@ FIXME:
1589 * In journalled data mode, a data buffer may be metadata against the
1590 * current transaction. But the same file is part of a shared mapping
1591 * and someone does a writepage() on it.
1592 *
1593 * We will move the buffer onto the async_data list, but *after* it has
1594 * been dirtied. So there's a small window where we have dirty data on
1595 * BJ_Metadata.
1596 *
1597 * Note that this only applies to the last partial page in the file. The
1598 * bit which block_write_full_page() uses prepare/commit for. (That's
1599 * broken code anyway: it's wrong for msync()).
1600 *
1601 * It's a rare case: affects the final partial page, for journalled data
1602 * where the file is subject to bith write() and writepage() in the same
1603 * transction. To fix it we'll need a custom block_write_full_page().
1604 * We'll probably need that anyway for journalling writepage() output.
1605 *
1606 * We don't honour synchronous mounts for writepage(). That would be
1607 * disastrous. Any write() or metadata operation will sync the fs for
1608 * us.
1609 *
1610 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1611 * we don't need to open a transaction here.
1612 */
1613static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001614 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
1616 struct inode *inode = page->mapping->host;
1617 struct buffer_head *page_bufs;
1618 handle_t *handle = NULL;
1619 int ret = 0;
1620 int err;
1621
1622 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001623 /*
1624 * We don't want to warn for emergency remount. The condition is
1625 * ordered to avoid dereferencing inode->i_sb in non-error case to
1626 * avoid slow-downs.
1627 */
1628 WARN_ON_ONCE(IS_RDONLY(inode) &&
1629 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 /*
1632 * We give up here if we're reentered, because it might be for a
1633 * different filesystem.
1634 */
1635 if (ext3_journal_current_handle())
1636 goto out_fail;
1637
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001638 trace_ext3_ordered_writepage(page);
Jan Kara9e80d402009-03-26 13:08:04 +01001639 if (!page_has_buffers(page)) {
1640 create_empty_buffers(page, inode->i_sb->s_blocksize,
1641 (1 << BH_Dirty)|(1 << BH_Uptodate));
Jan Kara430db322009-04-07 18:25:01 -04001642 page_bufs = page_buffers(page);
1643 } else {
1644 page_bufs = page_buffers(page);
1645 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1646 NULL, buffer_unmapped)) {
1647 /* Provide NULL get_block() to catch bugs if buffers
1648 * weren't really mapped */
1649 return block_write_full_page(page, NULL, wbc);
1650 }
Jan Kara9e80d402009-03-26 13:08:04 +01001651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1653
1654 if (IS_ERR(handle)) {
1655 ret = PTR_ERR(handle);
1656 goto out_fail;
1657 }
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 walk_page_buffers(handle, page_bufs, 0,
1660 PAGE_CACHE_SIZE, NULL, bget_one);
1661
1662 ret = block_write_full_page(page, ext3_get_block, wbc);
1663
1664 /*
1665 * The page can become unlocked at any point now, and
1666 * truncate can then come in and change things. So we
1667 * can't touch *page from now on. But *page_bufs is
1668 * safe due to elevated refcount.
1669 */
1670
1671 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001672 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 * block_write_full_page() succeeded. Otherwise they are unmapped,
1674 * and generally junk.
1675 */
1676 if (ret == 0) {
1677 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1678 NULL, journal_dirty_data_fn);
1679 if (!ret)
1680 ret = err;
1681 }
1682 walk_page_buffers(handle, page_bufs, 0,
1683 PAGE_CACHE_SIZE, NULL, bput_one);
1684 err = ext3_journal_stop(handle);
1685 if (!ret)
1686 ret = err;
1687 return ret;
1688
1689out_fail:
1690 redirty_page_for_writepage(wbc, page);
1691 unlock_page(page);
1692 return ret;
1693}
1694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695static int ext3_writeback_writepage(struct page *page,
1696 struct writeback_control *wbc)
1697{
1698 struct inode *inode = page->mapping->host;
1699 handle_t *handle = NULL;
1700 int ret = 0;
1701 int err;
1702
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001703 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001704 /*
1705 * We don't want to warn for emergency remount. The condition is
1706 * ordered to avoid dereferencing inode->i_sb in non-error case to
1707 * avoid slow-downs.
1708 */
1709 WARN_ON_ONCE(IS_RDONLY(inode) &&
1710 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (ext3_journal_current_handle())
1713 goto out_fail;
1714
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001715 trace_ext3_writeback_writepage(page);
Jan Kara430db322009-04-07 18:25:01 -04001716 if (page_has_buffers(page)) {
1717 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1718 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1719 /* Provide NULL get_block() to catch bugs if buffers
1720 * weren't really mapped */
1721 return block_write_full_page(page, NULL, wbc);
1722 }
1723 }
1724
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1726 if (IS_ERR(handle)) {
1727 ret = PTR_ERR(handle);
1728 goto out_fail;
1729 }
1730
Christoph Hellwig4c4d3902010-06-07 10:20:39 +02001731 ret = block_write_full_page(page, ext3_get_block, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 err = ext3_journal_stop(handle);
1734 if (!ret)
1735 ret = err;
1736 return ret;
1737
1738out_fail:
1739 redirty_page_for_writepage(wbc, page);
1740 unlock_page(page);
1741 return ret;
1742}
1743
1744static int ext3_journalled_writepage(struct page *page,
1745 struct writeback_control *wbc)
1746{
1747 struct inode *inode = page->mapping->host;
1748 handle_t *handle = NULL;
1749 int ret = 0;
1750 int err;
1751
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001752 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001753 /*
1754 * We don't want to warn for emergency remount. The condition is
1755 * ordered to avoid dereferencing inode->i_sb in non-error case to
1756 * avoid slow-downs.
1757 */
1758 WARN_ON_ONCE(IS_RDONLY(inode) &&
1759 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 if (ext3_journal_current_handle())
1762 goto no_write;
1763
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001764 trace_ext3_journalled_writepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1766 if (IS_ERR(handle)) {
1767 ret = PTR_ERR(handle);
1768 goto no_write;
1769 }
1770
1771 if (!page_has_buffers(page) || PageChecked(page)) {
1772 /*
1773 * It's mmapped pagecache. Add buffers and journal it. There
1774 * doesn't seem much point in redirtying the page here.
1775 */
1776 ClearPageChecked(page);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001777 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1778 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001779 if (ret != 0) {
1780 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001782 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 ret = walk_page_buffers(handle, page_buffers(page), 0,
1784 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1785
1786 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001787 PAGE_CACHE_SIZE, NULL, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 if (ret == 0)
1789 ret = err;
Jan Kara9df93932010-01-06 21:58:48 +01001790 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001791 atomic_set(&EXT3_I(inode)->i_datasync_tid,
1792 handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 unlock_page(page);
1794 } else {
1795 /*
1796 * It may be a page full of checkpoint-mode buffers. We don't
1797 * really know unless we go poke around in the buffer_heads.
1798 * But block_write_full_page will do the right thing.
1799 */
1800 ret = block_write_full_page(page, ext3_get_block, wbc);
1801 }
1802 err = ext3_journal_stop(handle);
1803 if (!ret)
1804 ret = err;
1805out:
1806 return ret;
1807
1808no_write:
1809 redirty_page_for_writepage(wbc, page);
1810out_unlock:
1811 unlock_page(page);
1812 goto out;
1813}
1814
1815static int ext3_readpage(struct file *file, struct page *page)
1816{
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001817 trace_ext3_readpage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 return mpage_readpage(page, ext3_get_block);
1819}
1820
1821static int
1822ext3_readpages(struct file *file, struct address_space *mapping,
1823 struct list_head *pages, unsigned nr_pages)
1824{
1825 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1826}
1827
Lukas Czernerd47992f2013-05-21 23:17:23 -04001828static void ext3_invalidatepage(struct page *page, unsigned int offset,
1829 unsigned int length)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830{
1831 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1832
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001833 trace_ext3_invalidatepage(page, offset);
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 /*
1836 * If it's a full truncate we just forget about the pending dirtying
1837 */
1838 if (offset == 0)
1839 ClearPageChecked(page);
1840
NeilBrown2ff28e22006-03-26 01:37:18 -08001841 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842}
1843
Al Viro27496a82005-10-21 03:20:48 -04001844static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845{
1846 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1847
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001848 trace_ext3_releasepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 WARN_ON(PageChecked(page));
1850 if (!page_has_buffers(page))
1851 return 0;
1852 return journal_try_to_free_buffers(journal, page, wait);
1853}
1854
1855/*
1856 * If the O_DIRECT write will extend the file then add this inode to the
1857 * orphan list. So recovery will truncate it back to the original size
1858 * if the machine crashes during the write.
1859 *
1860 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Karabd1939d2008-02-06 01:40:21 -08001861 * crashes then stale disk data _may_ be exposed inside the file. But current
1862 * VFS code falls back into buffered path in that case so we are safe.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 */
1864static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1865 const struct iovec *iov, loff_t offset,
1866 unsigned long nr_segs)
1867{
1868 struct file *file = iocb->ki_filp;
1869 struct inode *inode = file->f_mapping->host;
1870 struct ext3_inode_info *ei = EXT3_I(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001871 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 ssize_t ret;
1873 int orphan = 0;
1874 size_t count = iov_length(iov, nr_segs);
Eric Sandeenea0174a2009-10-12 21:34:27 -05001875 int retries = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001877 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 if (rw == WRITE) {
1880 loff_t final_size = offset + count;
1881
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 if (final_size > inode->i_size) {
Jan Karabd1939d2008-02-06 01:40:21 -08001883 /* Credits for sb + inode write */
1884 handle = ext3_journal_start(inode, 2);
1885 if (IS_ERR(handle)) {
1886 ret = PTR_ERR(handle);
1887 goto out;
1888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 ret = ext3_orphan_add(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001890 if (ret) {
1891 ext3_journal_stop(handle);
1892 goto out;
1893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 orphan = 1;
1895 ei->i_disksize = inode->i_size;
Jan Karabd1939d2008-02-06 01:40:21 -08001896 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 }
1898 }
1899
Eric Sandeenea0174a2009-10-12 21:34:27 -05001900retry:
Christoph Hellwigaacfc19c2011-06-24 14:29:47 -04001901 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1902 ext3_get_block);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001903 /*
1904 * In case of error extending write may have instantiated a few
1905 * blocks outside i_size. Trim these off again.
1906 */
1907 if (unlikely((rw & WRITE) && ret < 0)) {
1908 loff_t isize = i_size_read(inode);
1909 loff_t end = offset + iov_length(iov, nr_segs);
1910
1911 if (end > isize)
Jan Karaee3e77f2011-06-03 21:58:11 +02001912 ext3_truncate_failed_direct_write(inode);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001913 }
Eric Sandeenea0174a2009-10-12 21:34:27 -05001914 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1915 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Jan Karabd1939d2008-02-06 01:40:21 -08001917 if (orphan) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 int err;
1919
Jan Karabd1939d2008-02-06 01:40:21 -08001920 /* Credits for sb + inode write */
1921 handle = ext3_journal_start(inode, 2);
1922 if (IS_ERR(handle)) {
1923 /* This is really bad luck. We've written the data
Jan Kara7eb49692010-03-01 14:02:37 +01001924 * but cannot extend i_size. Truncate allocated blocks
1925 * and pretend the write failed... */
Jan Karaee3e77f2011-06-03 21:58:11 +02001926 ext3_truncate_failed_direct_write(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001927 ret = PTR_ERR(handle);
1928 goto out;
1929 }
1930 if (inode->i_nlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 ext3_orphan_del(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001932 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 loff_t end = offset + ret;
1934 if (end > inode->i_size) {
1935 ei->i_disksize = end;
1936 i_size_write(inode, end);
1937 /*
1938 * We're going to return a positive `ret'
1939 * here due to non-zero-length I/O, so there's
1940 * no way of reporting error returns from
1941 * ext3_mark_inode_dirty() to userspace. So
1942 * ignore it.
1943 */
1944 ext3_mark_inode_dirty(handle, inode);
1945 }
1946 }
1947 err = ext3_journal_stop(handle);
1948 if (ret == 0)
1949 ret = err;
1950 }
1951out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001952 trace_ext3_direct_IO_exit(inode, offset,
1953 iov_length(iov, nr_segs), rw, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return ret;
1955}
1956
1957/*
1958 * Pages can be marked dirty completely asynchronously from ext3's journalling
1959 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1960 * much here because ->set_page_dirty is called under VFS locks. The page is
1961 * not necessarily locked.
1962 *
1963 * We cannot just dirty the page and leave attached buffers clean, because the
1964 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1965 * or jbddirty because all the journalling code will explode.
1966 *
1967 * So what we do is to mark the page "pending dirty" and next time writepage
1968 * is called, propagate that into the buffers appropriately.
1969 */
1970static int ext3_journalled_set_page_dirty(struct page *page)
1971{
1972 SetPageChecked(page);
1973 return __set_page_dirty_nobuffers(page);
1974}
1975
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001976static const struct address_space_operations ext3_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001977 .readpage = ext3_readpage,
1978 .readpages = ext3_readpages,
1979 .writepage = ext3_ordered_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001980 .write_begin = ext3_write_begin,
1981 .write_end = ext3_ordered_write_end,
1982 .bmap = ext3_bmap,
1983 .invalidatepage = ext3_invalidatepage,
1984 .releasepage = ext3_releasepage,
1985 .direct_IO = ext3_direct_IO,
1986 .migratepage = buffer_migrate_page,
1987 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001988 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989};
1990
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001991static const struct address_space_operations ext3_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001992 .readpage = ext3_readpage,
1993 .readpages = ext3_readpages,
1994 .writepage = ext3_writeback_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001995 .write_begin = ext3_write_begin,
1996 .write_end = ext3_writeback_write_end,
1997 .bmap = ext3_bmap,
1998 .invalidatepage = ext3_invalidatepage,
1999 .releasepage = ext3_releasepage,
2000 .direct_IO = ext3_direct_IO,
2001 .migratepage = buffer_migrate_page,
2002 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002003 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004};
2005
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002006static const struct address_space_operations ext3_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002007 .readpage = ext3_readpage,
2008 .readpages = ext3_readpages,
2009 .writepage = ext3_journalled_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002010 .write_begin = ext3_write_begin,
2011 .write_end = ext3_journalled_write_end,
2012 .set_page_dirty = ext3_journalled_set_page_dirty,
2013 .bmap = ext3_bmap,
2014 .invalidatepage = ext3_invalidatepage,
2015 .releasepage = ext3_releasepage,
2016 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002017 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018};
2019
2020void ext3_set_aops(struct inode *inode)
2021{
2022 if (ext3_should_order_data(inode))
2023 inode->i_mapping->a_ops = &ext3_ordered_aops;
2024 else if (ext3_should_writeback_data(inode))
2025 inode->i_mapping->a_ops = &ext3_writeback_aops;
2026 else
2027 inode->i_mapping->a_ops = &ext3_journalled_aops;
2028}
2029
2030/*
2031 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2032 * up to the end of the block which corresponds to `from'.
2033 * This required during truncate. We need to physically zero the tail end
2034 * of that block so it doesn't yield old data if the file is later grown.
2035 */
Jan Karaee3e77f2011-06-03 21:58:11 +02002036static int ext3_block_truncate_page(struct inode *inode, loff_t from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037{
Mingming Cao43d23f92006-06-25 05:48:07 -07002038 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Jan Karaee3e77f2011-06-03 21:58:11 +02002039 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 unsigned blocksize, iblock, length, pos;
Jan Karaee3e77f2011-06-03 21:58:11 +02002041 struct page *page;
2042 handle_t *handle = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 struct buffer_head *bh;
2044 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Jan Karaee3e77f2011-06-03 21:58:11 +02002046 /* Truncated on block boundary - nothing to do */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 blocksize = inode->i_sb->s_blocksize;
Jan Karaee3e77f2011-06-03 21:58:11 +02002048 if ((from & (blocksize - 1)) == 0)
2049 return 0;
2050
2051 page = grab_cache_page(inode->i_mapping, index);
2052 if (!page)
2053 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 length = blocksize - (offset & (blocksize - 1));
2055 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 if (!page_has_buffers(page))
2058 create_empty_buffers(page, blocksize, 0);
2059
2060 /* Find the buffer that contains "offset" */
2061 bh = page_buffers(page);
2062 pos = blocksize;
2063 while (offset >= pos) {
2064 bh = bh->b_this_page;
2065 iblock++;
2066 pos += blocksize;
2067 }
2068
2069 err = 0;
2070 if (buffer_freed(bh)) {
2071 BUFFER_TRACE(bh, "freed: skip");
2072 goto unlock;
2073 }
2074
2075 if (!buffer_mapped(bh)) {
2076 BUFFER_TRACE(bh, "unmapped");
2077 ext3_get_block(inode, iblock, bh, 0);
2078 /* unmapped? It's a hole - nothing to do */
2079 if (!buffer_mapped(bh)) {
2080 BUFFER_TRACE(bh, "still unmapped");
2081 goto unlock;
2082 }
2083 }
2084
2085 /* Ok, it's mapped. Make sure it's up-to-date */
2086 if (PageUptodate(page))
2087 set_buffer_uptodate(bh);
2088
Zheng Liud03e1292011-12-05 15:55:11 +08002089 if (!bh_uptodate_or_lock(bh)) {
2090 err = bh_submit_read(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 /* Uhhuh. Read error. Complain and punt. */
Zheng Liud03e1292011-12-05 15:55:11 +08002092 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 goto unlock;
2094 }
2095
Jan Karaee3e77f2011-06-03 21:58:11 +02002096 /* data=writeback mode doesn't need transaction to zero-out data */
2097 if (!ext3_should_writeback_data(inode)) {
2098 /* We journal at most one block */
2099 handle = ext3_journal_start(inode, 1);
2100 if (IS_ERR(handle)) {
2101 clear_highpage(page);
2102 flush_dcache_page(page);
2103 err = PTR_ERR(handle);
2104 goto unlock;
2105 }
2106 }
2107
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 if (ext3_should_journal_data(inode)) {
2109 BUFFER_TRACE(bh, "get write access");
2110 err = ext3_journal_get_write_access(handle, bh);
2111 if (err)
Jan Karaee3e77f2011-06-03 21:58:11 +02002112 goto stop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 }
2114
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002115 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 BUFFER_TRACE(bh, "zeroed end of block");
2117
2118 err = 0;
2119 if (ext3_should_journal_data(inode)) {
2120 err = ext3_journal_dirty_metadata(handle, bh);
2121 } else {
2122 if (ext3_should_order_data(inode))
2123 err = ext3_journal_dirty_data(handle, bh);
2124 mark_buffer_dirty(bh);
2125 }
Jan Karaee3e77f2011-06-03 21:58:11 +02002126stop:
2127 if (handle)
2128 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
2130unlock:
2131 unlock_page(page);
2132 page_cache_release(page);
2133 return err;
2134}
2135
2136/*
2137 * Probably it should be a library function... search for first non-zero word
2138 * or memcmp with zero_page, whatever is better for particular architecture.
2139 * Linus?
2140 */
2141static inline int all_zeroes(__le32 *p, __le32 *q)
2142{
2143 while (p < q)
2144 if (*p++)
2145 return 0;
2146 return 1;
2147}
2148
2149/**
2150 * ext3_find_shared - find the indirect blocks for partial truncation.
2151 * @inode: inode in question
2152 * @depth: depth of the affected branch
2153 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2154 * @chain: place to store the pointers to partial indirect blocks
2155 * @top: place to the (detached) top of branch
2156 *
2157 * This is a helper function used by ext3_truncate().
2158 *
2159 * When we do truncate() we may have to clean the ends of several
2160 * indirect blocks but leave the blocks themselves alive. Block is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002161 * partially truncated if some data below the new i_size is referred
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 * from it (and it is on the path to the first completely truncated
2163 * data block, indeed). We have to free the top of that path along
2164 * with everything to the right of the path. Since no allocation
2165 * past the truncation point is possible until ext3_truncate()
2166 * finishes, we may safely do the latter, but top of branch may
2167 * require special attention - pageout below the truncation point
2168 * might try to populate it.
2169 *
2170 * We atomically detach the top of branch from the tree, store the
2171 * block number of its root in *@top, pointers to buffer_heads of
2172 * partially truncated blocks - in @chain[].bh and pointers to
2173 * their last elements that should not be removed - in
2174 * @chain[].p. Return value is the pointer to last filled element
2175 * of @chain.
2176 *
2177 * The work left to caller to do the actual freeing of subtrees:
2178 * a) free the subtree starting from *@top
2179 * b) free the subtrees whose roots are stored in
2180 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2181 * c) free the subtrees growing from the inode past the @chain[0].
2182 * (no partially truncated stuff there). */
2183
Andrew Mortond6859bf2006-03-26 01:38:03 -08002184static Indirect *ext3_find_shared(struct inode *inode, int depth,
2185 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186{
2187 Indirect *partial, *p;
2188 int k, err;
2189
2190 *top = 0;
Uwe Kleine-Königbf48aab2009-10-28 20:11:03 +01002191 /* Make k index the deepest non-null offset + 1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 for (k = depth; k > 1 && !offsets[k-1]; k--)
2193 ;
2194 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2195 /* Writer: pointers */
2196 if (!partial)
2197 partial = chain + k-1;
2198 /*
2199 * If the branch acquired continuation since we've looked at it -
2200 * fine, it should all survive and (new) top doesn't belong to us.
2201 */
2202 if (!partial->key && *partial->p)
2203 /* Writer: end */
2204 goto no_top;
2205 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2206 ;
2207 /*
2208 * OK, we've found the last block that must survive. The rest of our
2209 * branch should be detached before unlocking. However, if that rest
2210 * of branch is all ours and does not grow immediately from the inode
2211 * it's easier to cheat and just decrement partial->p.
2212 */
2213 if (p == chain + k - 1 && p > chain) {
2214 p->p--;
2215 } else {
2216 *top = *p->p;
2217 /* Nope, don't do this in ext3. Must leave the tree intact */
2218#if 0
2219 *p->p = 0;
2220#endif
2221 }
2222 /* Writer: end */
2223
Andrew Mortond6859bf2006-03-26 01:38:03 -08002224 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 brelse(partial->bh);
2226 partial--;
2227 }
2228no_top:
2229 return partial;
2230}
2231
2232/*
2233 * Zero a number of block pointers in either an inode or an indirect block.
2234 * If we restart the transaction we must again get write access to the
2235 * indirect block for further modification.
2236 *
2237 * We release `count' blocks on disk, but (last - first) may be greater
2238 * than `count' because there can be holes in there.
2239 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002240static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07002241 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08002242 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243{
2244 __le32 *p;
2245 if (try_to_extend_transaction(handle, inode)) {
2246 if (bh) {
2247 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Namhyung Kim156e7432010-11-25 01:53:13 +09002248 if (ext3_journal_dirty_metadata(handle, bh))
2249 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 }
2251 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002252 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (bh) {
2254 BUFFER_TRACE(bh, "retaking write access");
Namhyung Kim156e7432010-11-25 01:53:13 +09002255 if (ext3_journal_get_write_access(handle, bh))
2256 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 }
2258 }
2259
2260 /*
2261 * Any buffers which are on the journal will be in memory. We find
2262 * them on the hash table so journal_revoke() will run journal_forget()
2263 * on them. We've already detached each block from the file, so
2264 * bforget() in journal_forget() should be safe.
2265 *
2266 * AKPM: turn on bforget in journal_forget()!!!
2267 */
2268 for (p = first; p < last; p++) {
2269 u32 nr = le32_to_cpu(*p);
2270 if (nr) {
2271 struct buffer_head *bh;
2272
2273 *p = 0;
2274 bh = sb_find_get_block(inode->i_sb, nr);
2275 ext3_forget(handle, 0, inode, bh, nr);
2276 }
2277 }
2278
2279 ext3_free_blocks(handle, inode, block_to_free, count);
2280}
2281
2282/**
2283 * ext3_free_data - free a list of data blocks
2284 * @handle: handle for this transaction
2285 * @inode: inode we are dealing with
2286 * @this_bh: indirect buffer_head which contains *@first and *@last
2287 * @first: array of block numbers
2288 * @last: points immediately past the end of array
2289 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002290 * We are freeing all blocks referred from that array (numbers are stored as
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2292 *
2293 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2294 * blocks are contiguous then releasing them at one time will only affect one
2295 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2296 * actually use a lot of journal space.
2297 *
2298 * @this_bh will be %NULL if @first and @last point into the inode's direct
2299 * block pointers.
2300 */
2301static void ext3_free_data(handle_t *handle, struct inode *inode,
2302 struct buffer_head *this_bh,
2303 __le32 *first, __le32 *last)
2304{
Mingming Cao43d23f92006-06-25 05:48:07 -07002305 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002306 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2308 corresponding to
2309 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002310 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 __le32 *p; /* Pointer into inode/ind
2312 for current block */
2313 int err;
2314
2315 if (this_bh) { /* For indirect block */
2316 BUFFER_TRACE(this_bh, "get_write_access");
2317 err = ext3_journal_get_write_access(handle, this_bh);
2318 /* Important: if we can't update the indirect pointers
2319 * to the blocks, we can't free them. */
2320 if (err)
2321 return;
2322 }
2323
2324 for (p = first; p < last; p++) {
2325 nr = le32_to_cpu(*p);
2326 if (nr) {
2327 /* accumulate blocks to free if they're contiguous */
2328 if (count == 0) {
2329 block_to_free = nr;
2330 block_to_free_p = p;
2331 count = 1;
2332 } else if (nr == block_to_free + count) {
2333 count++;
2334 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002335 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 block_to_free,
2337 count, block_to_free_p, p);
2338 block_to_free = nr;
2339 block_to_free_p = p;
2340 count = 1;
2341 }
2342 }
2343 }
2344
2345 if (count > 0)
2346 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2347 count, block_to_free_p, p);
2348
2349 if (this_bh) {
2350 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
Duane Griffin3ccc3162008-07-25 01:46:26 -07002351
2352 /*
2353 * The buffer head should have an attached journal head at this
2354 * point. However, if the data is corrupted and an indirect
2355 * block pointed to itself, it would have been detached when
2356 * the block was cleared. Check for this instead of OOPSing.
2357 */
2358 if (bh2jh(this_bh))
2359 ext3_journal_dirty_metadata(handle, this_bh);
2360 else
2361 ext3_error(inode->i_sb, "ext3_free_data",
2362 "circular indirect block detected, "
2363 "inode=%lu, block=%llu",
2364 inode->i_ino,
2365 (unsigned long long)this_bh->b_blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 }
2367}
2368
2369/**
2370 * ext3_free_branches - free an array of branches
2371 * @handle: JBD handle for this transaction
2372 * @inode: inode we are dealing with
2373 * @parent_bh: the buffer_head which contains *@first and *@last
2374 * @first: array of block numbers
2375 * @last: pointer immediately past the end of array
2376 * @depth: depth of the branches to free
2377 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002378 * We are freeing all blocks referred from these branches (numbers are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 * stored as little-endian 32-bit) and updating @inode->i_blocks
2380 * appropriately.
2381 */
2382static void ext3_free_branches(handle_t *handle, struct inode *inode,
2383 struct buffer_head *parent_bh,
2384 __le32 *first, __le32 *last, int depth)
2385{
Mingming Cao43d23f92006-06-25 05:48:07 -07002386 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 __le32 *p;
2388
2389 if (is_handle_aborted(handle))
2390 return;
2391
2392 if (depth--) {
2393 struct buffer_head *bh;
2394 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2395 p = last;
2396 while (--p >= first) {
2397 nr = le32_to_cpu(*p);
2398 if (!nr)
2399 continue; /* A hole */
2400
2401 /* Go read the buffer for the next level down */
2402 bh = sb_bread(inode->i_sb, nr);
2403
2404 /*
2405 * A read failure? Report error and clear slot
2406 * (should be rare).
2407 */
2408 if (!bh) {
2409 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002410 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 inode->i_ino, nr);
2412 continue;
2413 }
2414
2415 /* This zaps the entire block. Bottom up. */
2416 BUFFER_TRACE(bh, "free child branches");
2417 ext3_free_branches(handle, inode, bh,
2418 (__le32*)bh->b_data,
2419 (__le32*)bh->b_data + addr_per_block,
2420 depth);
2421
2422 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 * Everything below this this pointer has been
2424 * released. Now let this top-of-subtree go.
2425 *
2426 * We want the freeing of this indirect block to be
2427 * atomic in the journal with the updating of the
2428 * bitmap block which owns it. So make some room in
2429 * the journal.
2430 *
2431 * We zero the parent pointer *after* freeing its
2432 * pointee in the bitmaps, so if extend_transaction()
2433 * for some reason fails to put the bitmap changes and
2434 * the release into the same transaction, recovery
2435 * will merely complain about releasing a free block,
2436 * rather than leaking blocks.
2437 */
2438 if (is_handle_aborted(handle))
2439 return;
2440 if (try_to_extend_transaction(handle, inode)) {
2441 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002442 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 }
2444
Jan Karaf25f6242010-07-12 21:04:31 +02002445 /*
2446 * We've probably journalled the indirect block several
2447 * times during the truncate. But it's no longer
2448 * needed and we now drop it from the transaction via
2449 * journal_revoke().
2450 *
2451 * That's easy if it's exclusively part of this
2452 * transaction. But if it's part of the committing
2453 * transaction then journal_forget() will simply
2454 * brelse() it. That means that if the underlying
2455 * block is reallocated in ext3_get_block(),
2456 * unmap_underlying_metadata() will find this block
2457 * and will try to get rid of it. damn, damn. Thus
2458 * we don't allow a block to be reallocated until
2459 * a transaction freeing it has fully committed.
2460 *
2461 * We also have to make sure journal replay after a
2462 * crash does not overwrite non-journaled data blocks
2463 * with old metadata when the block got reallocated for
2464 * data. Thus we have to store a revoke record for a
2465 * block in the same transaction in which we free the
2466 * block.
2467 */
2468 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2469
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 ext3_free_blocks(handle, inode, nr, 1);
2471
2472 if (parent_bh) {
2473 /*
2474 * The block which we have just freed is
2475 * pointed to by an indirect block: journal it
2476 */
2477 BUFFER_TRACE(parent_bh, "get_write_access");
2478 if (!ext3_journal_get_write_access(handle,
2479 parent_bh)){
2480 *p = 0;
2481 BUFFER_TRACE(parent_bh,
2482 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002483 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 parent_bh);
2485 }
2486 }
2487 }
2488 } else {
2489 /* We have reached the bottom of the tree. */
2490 BUFFER_TRACE(parent_bh, "free data blocks");
2491 ext3_free_data(handle, inode, parent_bh, first, last);
2492 }
2493}
2494
Duane Griffinae76dd92008-07-25 01:46:23 -07002495int ext3_can_truncate(struct inode *inode)
2496{
Duane Griffinae76dd92008-07-25 01:46:23 -07002497 if (S_ISREG(inode->i_mode))
2498 return 1;
2499 if (S_ISDIR(inode->i_mode))
2500 return 1;
2501 if (S_ISLNK(inode->i_mode))
2502 return !ext3_inode_is_fast_symlink(inode);
2503 return 0;
2504}
2505
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506/*
2507 * ext3_truncate()
2508 *
2509 * We block out ext3_get_block() block instantiations across the entire
2510 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2511 * simultaneously on behalf of the same inode.
2512 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002513 * As we work through the truncate and commit bits of it to the journal there
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 * is one core, guiding principle: the file's tree must always be consistent on
2515 * disk. We must be able to restart the truncate after a crash.
2516 *
2517 * The file's tree may be transiently inconsistent in memory (although it
2518 * probably isn't), but whenever we close off and commit a journal transaction,
2519 * the contents of (the filesystem + the journal) must be consistent and
2520 * restartable. It's pretty simple, really: bottom up, right to left (although
2521 * left-to-right works OK too).
2522 *
2523 * Note that at recovery time, journal replay occurs *before* the restart of
2524 * truncate against the orphan inode list.
2525 *
2526 * The committed inode has the new, desired i_size (which is the same as
2527 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2528 * that this inode's truncate did not complete and it will again call
2529 * ext3_truncate() to have another go. So there will be instantiated blocks
2530 * to the right of the truncation point in a crashed ext3 filesystem. But
2531 * that's fine - as long as they are linked from the inode, the post-crash
2532 * ext3_truncate() run will find them and release them.
2533 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002534void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535{
2536 handle_t *handle;
2537 struct ext3_inode_info *ei = EXT3_I(inode);
2538 __le32 *i_data = ei->i_data;
2539 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 int offsets[4];
2541 Indirect chain[4];
2542 Indirect *partial;
2543 __le32 nr = 0;
2544 int n;
2545 long last_block;
2546 unsigned blocksize = inode->i_sb->s_blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002548 trace_ext3_truncate_enter(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Duane Griffinae76dd92008-07-25 01:46:23 -07002550 if (!ext3_can_truncate(inode))
Jan Karaef436182009-06-17 16:26:24 -07002551 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002553 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
Jan Kara9df93932010-01-06 21:58:48 +01002554 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002555
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 handle = start_transaction(inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02002557 if (IS_ERR(handle))
Jan Karaef436182009-06-17 16:26:24 -07002558 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
2560 last_block = (inode->i_size + blocksize-1)
2561 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2563 if (n == 0)
2564 goto out_stop; /* error */
2565
2566 /*
2567 * OK. This truncate is going to happen. We add the inode to the
2568 * orphan list, so that if this truncate spans multiple transactions,
2569 * and we crash, we will resume the truncate when the filesystem
2570 * recovers. It also marks the inode dirty, to catch the new size.
2571 *
2572 * Implication: the file must always be in a sane, consistent
2573 * truncatable state while each transaction commits.
2574 */
2575 if (ext3_orphan_add(handle, inode))
2576 goto out_stop;
2577
2578 /*
2579 * The orphan list entry will now protect us from any crash which
2580 * occurs before the truncate completes, so it is now safe to propagate
2581 * the new, shorter inode size (held for now in i_size) into the
2582 * on-disk inode. We do this via i_disksize, which is the value which
2583 * ext3 *really* writes onto the disk inode.
2584 */
2585 ei->i_disksize = inode->i_size;
2586
2587 /*
2588 * From here we block out all ext3_get_block() callers who want to
2589 * modify the block allocation tree.
2590 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002591 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
2593 if (n == 1) { /* direct blocks */
2594 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2595 i_data + EXT3_NDIR_BLOCKS);
2596 goto do_indirects;
2597 }
2598
2599 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2600 /* Kill the top of shared branch (not detached) */
2601 if (nr) {
2602 if (partial == chain) {
2603 /* Shared branch grows from the inode */
2604 ext3_free_branches(handle, inode, NULL,
2605 &nr, &nr+1, (chain+n-1) - partial);
2606 *partial->p = 0;
2607 /*
2608 * We mark the inode dirty prior to restart,
2609 * and prior to stop. No need for it here.
2610 */
2611 } else {
2612 /* Shared branch grows from an indirect block */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 ext3_free_branches(handle, inode, partial->bh,
2614 partial->p,
2615 partial->p+1, (chain+n-1) - partial);
2616 }
2617 }
2618 /* Clear the ends of indirect blocks on the shared branch */
2619 while (partial > chain) {
2620 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2621 (__le32*)partial->bh->b_data+addr_per_block,
2622 (chain+n-1) - partial);
2623 BUFFER_TRACE(partial->bh, "call brelse");
2624 brelse (partial->bh);
2625 partial--;
2626 }
2627do_indirects:
2628 /* Kill the remaining (whole) subtrees */
2629 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002630 default:
2631 nr = i_data[EXT3_IND_BLOCK];
2632 if (nr) {
2633 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2634 i_data[EXT3_IND_BLOCK] = 0;
2635 }
2636 case EXT3_IND_BLOCK:
2637 nr = i_data[EXT3_DIND_BLOCK];
2638 if (nr) {
2639 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2640 i_data[EXT3_DIND_BLOCK] = 0;
2641 }
2642 case EXT3_DIND_BLOCK:
2643 nr = i_data[EXT3_TIND_BLOCK];
2644 if (nr) {
2645 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2646 i_data[EXT3_TIND_BLOCK] = 0;
2647 }
2648 case EXT3_TIND_BLOCK:
2649 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 }
2651
2652 ext3_discard_reservation(inode);
2653
Arjan van de Ven97461512006-03-23 03:00:42 -08002654 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2656 ext3_mark_inode_dirty(handle, inode);
2657
Andrew Mortond6859bf2006-03-26 01:38:03 -08002658 /*
2659 * In a multi-transaction truncate, we only make the final transaction
2660 * synchronous
2661 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (IS_SYNC(inode))
2663 handle->h_sync = 1;
2664out_stop:
2665 /*
2666 * If this was a simple ftruncate(), and the file will remain alive
2667 * then we need to clear up the orphan record which we created above.
2668 * However, if this was a real unlink then we were called by
Al Viroac14a952010-06-06 07:08:19 -04002669 * ext3_evict_inode(), and we allow that function to clean up the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 * orphan info for us.
2671 */
2672 if (inode->i_nlink)
2673 ext3_orphan_del(handle, inode);
2674
2675 ext3_journal_stop(handle);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002676 trace_ext3_truncate_exit(inode);
Jan Karaef436182009-06-17 16:26:24 -07002677 return;
2678out_notrans:
2679 /*
2680 * Delete the inode from orphan list so that it doesn't stay there
2681 * forever and trigger assertion on umount.
2682 */
2683 if (inode->i_nlink)
2684 ext3_orphan_del(NULL, inode);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002685 trace_ext3_truncate_exit(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686}
2687
Mingming Cao43d23f92006-06-25 05:48:07 -07002688static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 unsigned long ino, struct ext3_iloc *iloc)
2690{
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002691 unsigned long block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002692 unsigned long offset;
2693 ext3_fsblk_t block;
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002694 struct ext3_group_desc *gdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
Neil Brown2ccb48e2006-07-30 03:03:01 -07002696 if (!ext3_valid_inum(sb, ino)) {
2697 /*
2698 * This error is already checked for in namei.c unless we are
2699 * looking at an NFS filehandle, in which case no error
2700 * report is needed
2701 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 return 0;
2703 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002704
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002706 gdp = ext3_get_group_desc(sb, block_group, NULL);
2707 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 /*
2710 * Figure out the offset within the block group inode table
2711 */
2712 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2713 EXT3_INODE_SIZE(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002714 block = le32_to_cpu(gdp->bg_inode_table) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2716
2717 iloc->block_group = block_group;
2718 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2719 return block;
2720}
2721
2722/*
2723 * ext3_get_inode_loc returns with an extra refcount against the inode's
2724 * underlying buffer_head on success. If 'in_mem' is true, we have all
2725 * data in memory that is needed to recreate the on-disk version of this
2726 * inode.
2727 */
2728static int __ext3_get_inode_loc(struct inode *inode,
2729 struct ext3_iloc *iloc, int in_mem)
2730{
Mingming Cao43d23f92006-06-25 05:48:07 -07002731 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 struct buffer_head *bh;
2733
2734 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2735 if (!block)
2736 return -EIO;
2737
2738 bh = sb_getblk(inode->i_sb, block);
Wang Shilong1b7d76e2013-01-12 01:36:17 -08002739 if (unlikely(!bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2741 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002742 "inode=%lu, block="E3FSBLK,
2743 inode->i_ino, block);
Wang Shilongc04e88e2013-01-15 21:20:01 -08002744 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 }
2746 if (!buffer_uptodate(bh)) {
2747 lock_buffer(bh);
Hidehiro Kawai95450f52008-07-25 01:46:24 -07002748
2749 /*
2750 * If the buffer has the write error flag, we have failed
2751 * to write out another inode in the same block. In this
2752 * case, we don't have to read the block because we may
2753 * read the old inode data successfully.
2754 */
2755 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2756 set_buffer_uptodate(bh);
2757
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 if (buffer_uptodate(bh)) {
2759 /* someone brought it uptodate while we waited */
2760 unlock_buffer(bh);
2761 goto has_buffer;
2762 }
2763
2764 /*
2765 * If we have all information of the inode in memory and this
2766 * is the only valid inode in the block, we need not read the
2767 * block.
2768 */
2769 if (in_mem) {
2770 struct buffer_head *bitmap_bh;
2771 struct ext3_group_desc *desc;
2772 int inodes_per_buffer;
2773 int inode_offset, i;
2774 int block_group;
2775 int start;
2776
2777 block_group = (inode->i_ino - 1) /
2778 EXT3_INODES_PER_GROUP(inode->i_sb);
2779 inodes_per_buffer = bh->b_size /
2780 EXT3_INODE_SIZE(inode->i_sb);
2781 inode_offset = ((inode->i_ino - 1) %
2782 EXT3_INODES_PER_GROUP(inode->i_sb));
2783 start = inode_offset & ~(inodes_per_buffer - 1);
2784
2785 /* Is the inode bitmap in cache? */
2786 desc = ext3_get_group_desc(inode->i_sb,
2787 block_group, NULL);
2788 if (!desc)
2789 goto make_io;
2790
2791 bitmap_bh = sb_getblk(inode->i_sb,
2792 le32_to_cpu(desc->bg_inode_bitmap));
Wang Shilong1b7d76e2013-01-12 01:36:17 -08002793 if (unlikely(!bitmap_bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 goto make_io;
2795
2796 /*
2797 * If the inode bitmap isn't in cache then the
2798 * optimisation may end up performing two reads instead
2799 * of one, so skip it.
2800 */
2801 if (!buffer_uptodate(bitmap_bh)) {
2802 brelse(bitmap_bh);
2803 goto make_io;
2804 }
2805 for (i = start; i < start + inodes_per_buffer; i++) {
2806 if (i == inode_offset)
2807 continue;
2808 if (ext3_test_bit(i, bitmap_bh->b_data))
2809 break;
2810 }
2811 brelse(bitmap_bh);
2812 if (i == start + inodes_per_buffer) {
2813 /* all other inodes are free, so skip I/O */
2814 memset(bh->b_data, 0, bh->b_size);
2815 set_buffer_uptodate(bh);
2816 unlock_buffer(bh);
2817 goto has_buffer;
2818 }
2819 }
2820
2821make_io:
2822 /*
2823 * There are other valid inodes in the buffer, this inode
2824 * has in-inode xattrs, or we don't have this inode in memory.
2825 * Read the block from disk.
2826 */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002827 trace_ext3_load_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 get_bh(bh);
2829 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02002830 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 wait_on_buffer(bh);
2832 if (!buffer_uptodate(bh)) {
2833 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2834 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002835 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 inode->i_ino, block);
2837 brelse(bh);
2838 return -EIO;
2839 }
2840 }
2841has_buffer:
2842 iloc->bh = bh;
2843 return 0;
2844}
2845
2846int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2847{
2848 /* We have all inode data except xattrs in memory here. */
2849 return __ext3_get_inode_loc(inode, iloc,
Jan Kara9df93932010-01-06 21:58:48 +01002850 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851}
2852
2853void ext3_set_inode_flags(struct inode *inode)
2854{
2855 unsigned int flags = EXT3_I(inode)->i_flags;
2856
2857 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2858 if (flags & EXT3_SYNC_FL)
2859 inode->i_flags |= S_SYNC;
2860 if (flags & EXT3_APPEND_FL)
2861 inode->i_flags |= S_APPEND;
2862 if (flags & EXT3_IMMUTABLE_FL)
2863 inode->i_flags |= S_IMMUTABLE;
2864 if (flags & EXT3_NOATIME_FL)
2865 inode->i_flags |= S_NOATIME;
2866 if (flags & EXT3_DIRSYNC_FL)
2867 inode->i_flags |= S_DIRSYNC;
2868}
2869
Jan Kara28be5ab2007-05-08 00:30:33 -07002870/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2871void ext3_get_inode_flags(struct ext3_inode_info *ei)
2872{
2873 unsigned int flags = ei->vfs_inode.i_flags;
2874
2875 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2876 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2877 if (flags & S_SYNC)
2878 ei->i_flags |= EXT3_SYNC_FL;
2879 if (flags & S_APPEND)
2880 ei->i_flags |= EXT3_APPEND_FL;
2881 if (flags & S_IMMUTABLE)
2882 ei->i_flags |= EXT3_IMMUTABLE_FL;
2883 if (flags & S_NOATIME)
2884 ei->i_flags |= EXT3_NOATIME_FL;
2885 if (flags & S_DIRSYNC)
2886 ei->i_flags |= EXT3_DIRSYNC_FL;
2887}
2888
David Howells473043d2008-02-07 00:15:36 -08002889struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890{
2891 struct ext3_iloc iloc;
2892 struct ext3_inode *raw_inode;
David Howells473043d2008-02-07 00:15:36 -08002893 struct ext3_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 struct buffer_head *bh;
David Howells473043d2008-02-07 00:15:36 -08002895 struct inode *inode;
Jan Karafe8bc912009-10-16 19:26:15 +02002896 journal_t *journal = EXT3_SB(sb)->s_journal;
2897 transaction_t *transaction;
David Howells473043d2008-02-07 00:15:36 -08002898 long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 int block;
Eric W. Biederman15232992012-02-07 15:41:24 -08002900 uid_t i_uid;
2901 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
David Howells473043d2008-02-07 00:15:36 -08002903 inode = iget_locked(sb, ino);
2904 if (!inode)
2905 return ERR_PTR(-ENOMEM);
2906 if (!(inode->i_state & I_NEW))
2907 return inode;
2908
2909 ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 ei->i_block_alloc_info = NULL;
2911
David Howells473043d2008-02-07 00:15:36 -08002912 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2913 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 goto bad_inode;
2915 bh = iloc.bh;
2916 raw_inode = ext3_raw_inode(&iloc);
2917 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman15232992012-02-07 15:41:24 -08002918 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2919 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 if(!(test_opt (inode->i_sb, NO_UID32))) {
Eric W. Biederman15232992012-02-07 15:41:24 -08002921 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2922 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 }
Eric W. Biederman15232992012-02-07 15:41:24 -08002924 i_uid_write(inode, i_uid);
2925 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002926 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002928 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2929 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2930 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2932
Linus Torvaldsde329822010-03-29 14:30:19 -07002933 ei->i_state_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 ei->i_dir_start_lookup = 0;
2935 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2936 /* We now have enough fields to check if the inode was active or not.
2937 * This is needed because nfsd might try to access dead inodes
2938 * the test is that same one that e2fsck uses
2939 * NeilBrown 1999oct15
2940 */
2941 if (inode->i_nlink == 0) {
2942 if (inode->i_mode == 0 ||
2943 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2944 /* this inode is deleted */
2945 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002946 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 goto bad_inode;
2948 }
2949 /* The only unlinked inodes we let through here have
2950 * valid i_mode and are being read by the orphan
2951 * recovery code: that's fine, we're about to complete
2952 * the process of deleting those. */
2953 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2955 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2956#ifdef EXT3_FRAGMENTS
2957 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2958 ei->i_frag_no = raw_inode->i_frag;
2959 ei->i_frag_size = raw_inode->i_fsize;
2960#endif
2961 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2962 if (!S_ISREG(inode->i_mode)) {
2963 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2964 } else {
2965 inode->i_size |=
2966 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2967 }
2968 ei->i_disksize = inode->i_size;
2969 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2970 ei->i_block_group = iloc.block_group;
2971 /*
2972 * NOTE! The in-memory inode i_data array is in little-endian order
2973 * even on big-endian machines: we do NOT byteswap the block numbers!
2974 */
2975 for (block = 0; block < EXT3_N_BLOCKS; block++)
2976 ei->i_data[block] = raw_inode->i_block[block];
2977 INIT_LIST_HEAD(&ei->i_orphan);
2978
Jan Karafe8bc912009-10-16 19:26:15 +02002979 /*
2980 * Set transaction id's of transactions that have to be committed
2981 * to finish f[data]sync. We set them to currently running transaction
2982 * as we cannot be sure that the inode or some of its metadata isn't
2983 * part of the transaction - the inode could have been reclaimed and
2984 * now it is reread from disk.
2985 */
2986 if (journal) {
2987 tid_t tid;
2988
2989 spin_lock(&journal->j_state_lock);
2990 if (journal->j_running_transaction)
2991 transaction = journal->j_running_transaction;
2992 else
2993 transaction = journal->j_committing_transaction;
2994 if (transaction)
2995 tid = transaction->t_tid;
2996 else
2997 tid = journal->j_commit_sequence;
2998 spin_unlock(&journal->j_state_lock);
2999 atomic_set(&ei->i_sync_tid, tid);
3000 atomic_set(&ei->i_datasync_tid, tid);
3001 }
3002
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
3004 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
3005 /*
3006 * When mke2fs creates big inodes it does not zero out
3007 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
3008 * so ignore those first few inodes.
3009 */
3010 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3011 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07003012 EXT3_INODE_SIZE(inode->i_sb)) {
3013 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08003014 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 goto bad_inode;
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07003016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 if (ei->i_extra_isize == 0) {
3018 /* The extra space is currently unused. Use it. */
3019 ei->i_extra_isize = sizeof(struct ext3_inode) -
3020 EXT3_GOOD_OLD_INODE_SIZE;
3021 } else {
3022 __le32 *magic = (void *)raw_inode +
3023 EXT3_GOOD_OLD_INODE_SIZE +
3024 ei->i_extra_isize;
3025 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
Jan Kara9df93932010-01-06 21:58:48 +01003026 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 }
3028 } else
3029 ei->i_extra_isize = 0;
3030
3031 if (S_ISREG(inode->i_mode)) {
3032 inode->i_op = &ext3_file_inode_operations;
3033 inode->i_fop = &ext3_file_operations;
3034 ext3_set_aops(inode);
3035 } else if (S_ISDIR(inode->i_mode)) {
3036 inode->i_op = &ext3_dir_inode_operations;
3037 inode->i_fop = &ext3_dir_operations;
3038 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffinb5ed3112008-12-19 20:47:14 +00003039 if (ext3_inode_is_fast_symlink(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 inode->i_op = &ext3_fast_symlink_inode_operations;
Duane Griffinb5ed3112008-12-19 20:47:14 +00003041 nd_terminate_link(ei->i_data, inode->i_size,
3042 sizeof(ei->i_data) - 1);
3043 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 inode->i_op = &ext3_symlink_inode_operations;
3045 ext3_set_aops(inode);
3046 }
3047 } else {
3048 inode->i_op = &ext3_special_inode_operations;
3049 if (raw_inode->i_block[0])
3050 init_special_inode(inode, inode->i_mode,
3051 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003052 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 init_special_inode(inode, inode->i_mode,
3054 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3055 }
3056 brelse (iloc.bh);
3057 ext3_set_inode_flags(inode);
David Howells473043d2008-02-07 00:15:36 -08003058 unlock_new_inode(inode);
3059 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
3061bad_inode:
David Howells473043d2008-02-07 00:15:36 -08003062 iget_failed(inode);
3063 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064}
3065
3066/*
3067 * Post the struct inode info into an on-disk inode location in the
3068 * buffer-cache. This gobbles the caller's reference to the
3069 * buffer_head in the inode location struct.
3070 *
3071 * The caller must have write access to iloc->bh.
3072 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003073static int ext3_do_update_inode(handle_t *handle,
3074 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 struct ext3_iloc *iloc)
3076{
3077 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3078 struct ext3_inode_info *ei = EXT3_I(inode);
3079 struct buffer_head *bh = iloc->bh;
3080 int err = 0, rc, block;
Jan Kara156bddd2012-09-03 16:50:42 +02003081 int need_datasync = 0;
3082 __le32 disksize;
Eric W. Biederman15232992012-02-07 15:41:24 -08003083 uid_t i_uid;
3084 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085
Chris Mason4f003fd2009-09-08 00:22:14 +02003086again:
3087 /* we can't allow multiple procs in here at once, its a bit racey */
3088 lock_buffer(bh);
3089
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 /* For fields not not tracking in the in-memory inode,
3091 * initialise them to zero for new inodes. */
Jan Kara9df93932010-01-06 21:58:48 +01003092 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3094
Jan Kara28be5ab2007-05-08 00:30:33 -07003095 ext3_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman15232992012-02-07 15:41:24 -08003097 i_uid = i_uid_read(inode);
3098 i_gid = i_gid_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 if(!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman15232992012-02-07 15:41:24 -08003100 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3101 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102/*
3103 * Fix up interoperability with old kernels. Otherwise, old inodes get
3104 * re-used with the upper 16 bits of the uid/gid intact
3105 */
3106 if(!ei->i_dtime) {
3107 raw_inode->i_uid_high =
Eric W. Biederman15232992012-02-07 15:41:24 -08003108 cpu_to_le16(high_16_bits(i_uid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 raw_inode->i_gid_high =
Eric W. Biederman15232992012-02-07 15:41:24 -08003110 cpu_to_le16(high_16_bits(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 } else {
3112 raw_inode->i_uid_high = 0;
3113 raw_inode->i_gid_high = 0;
3114 }
3115 } else {
3116 raw_inode->i_uid_low =
Eric W. Biederman15232992012-02-07 15:41:24 -08003117 cpu_to_le16(fs_high2lowuid(i_uid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 raw_inode->i_gid_low =
Eric W. Biederman15232992012-02-07 15:41:24 -08003119 cpu_to_le16(fs_high2lowgid(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 raw_inode->i_uid_high = 0;
3121 raw_inode->i_gid_high = 0;
3122 }
3123 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Jan Kara156bddd2012-09-03 16:50:42 +02003124 disksize = cpu_to_le32(ei->i_disksize);
3125 if (disksize != raw_inode->i_size) {
3126 need_datasync = 1;
3127 raw_inode->i_size = disksize;
3128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3130 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3131 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3132 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3133 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3134 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3135#ifdef EXT3_FRAGMENTS
3136 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3137 raw_inode->i_frag = ei->i_frag_no;
3138 raw_inode->i_fsize = ei->i_frag_size;
3139#endif
3140 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3141 if (!S_ISREG(inode->i_mode)) {
3142 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3143 } else {
Jan Kara156bddd2012-09-03 16:50:42 +02003144 disksize = cpu_to_le32(ei->i_disksize >> 32);
3145 if (disksize != raw_inode->i_size_high) {
3146 raw_inode->i_size_high = disksize;
3147 need_datasync = 1;
3148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 if (ei->i_disksize > 0x7fffffffULL) {
3150 struct super_block *sb = inode->i_sb;
3151 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3152 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3153 EXT3_SB(sb)->s_es->s_rev_level ==
3154 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3155 /* If this is the first large file
3156 * created, add a flag to the superblock.
3157 */
Chris Mason4f003fd2009-09-08 00:22:14 +02003158 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 err = ext3_journal_get_write_access(handle,
3160 EXT3_SB(sb)->s_sbh);
3161 if (err)
3162 goto out_brelse;
Chris Mason4f003fd2009-09-08 00:22:14 +02003163
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 ext3_update_dynamic_rev(sb);
3165 EXT3_SET_RO_COMPAT_FEATURE(sb,
3166 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 handle->h_sync = 1;
3168 err = ext3_journal_dirty_metadata(handle,
3169 EXT3_SB(sb)->s_sbh);
Chris Mason4f003fd2009-09-08 00:22:14 +02003170 /* get our lock and start over */
3171 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 }
3173 }
3174 }
3175 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3176 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3177 if (old_valid_dev(inode->i_rdev)) {
3178 raw_inode->i_block[0] =
3179 cpu_to_le32(old_encode_dev(inode->i_rdev));
3180 raw_inode->i_block[1] = 0;
3181 } else {
3182 raw_inode->i_block[0] = 0;
3183 raw_inode->i_block[1] =
3184 cpu_to_le32(new_encode_dev(inode->i_rdev));
3185 raw_inode->i_block[2] = 0;
3186 }
3187 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3188 raw_inode->i_block[block] = ei->i_data[block];
3189
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07003190 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3192
3193 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Chris Mason4f003fd2009-09-08 00:22:14 +02003194 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 rc = ext3_journal_dirty_metadata(handle, bh);
3196 if (!err)
3197 err = rc;
Jan Kara9df93932010-01-06 21:58:48 +01003198 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199
Jan Karafe8bc912009-10-16 19:26:15 +02003200 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
Jan Kara156bddd2012-09-03 16:50:42 +02003201 if (need_datasync)
3202 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203out_brelse:
3204 brelse (bh);
3205 ext3_std_error(inode->i_sb, err);
3206 return err;
3207}
3208
3209/*
3210 * ext3_write_inode()
3211 *
3212 * We are called from a few places:
3213 *
3214 * - Within generic_file_write() for O_SYNC files.
3215 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003216 * transaction to commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217 *
3218 * - Within sys_sync(), kupdate and such.
3219 * We wait on commit, if tol to.
3220 *
3221 * - Within prune_icache() (PF_MEMALLOC == true)
3222 * Here we simply return. We can't afford to block kswapd on the
3223 * journal commit.
3224 *
3225 * In all cases it is actually safe for us to return without doing anything,
3226 * because the inode has been copied into a raw inode buffer in
3227 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3228 * knfsd.
3229 *
3230 * Note that we are absolutely dependent upon all inode dirtiers doing the
3231 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3232 * which we are interested.
3233 *
3234 * It would be a bug for them to not do this. The code:
3235 *
3236 * mark_inode_dirty(inode)
3237 * stuff();
3238 * inode->i_size = expr;
3239 *
3240 * is in error because a kswapd-driven write_inode() could occur while
3241 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3242 * will no longer be on the superblock's dirty inode list.
3243 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003244int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245{
3246 if (current->flags & PF_MEMALLOC)
3247 return 0;
3248
3249 if (ext3_journal_current_handle()) {
Jose R. Santos9ad163a2007-10-18 23:39:23 -07003250 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 dump_stack();
3252 return -EIO;
3253 }
3254
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003255 if (wbc->sync_mode != WB_SYNC_ALL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 return 0;
3257
3258 return ext3_force_commit(inode->i_sb);
3259}
3260
3261/*
3262 * ext3_setattr()
3263 *
3264 * Called from notify_change.
3265 *
3266 * We want to trap VFS attempts to truncate the file as soon as
3267 * possible. In particular, we want to make sure that when the VFS
3268 * shrinks i_size, we put the inode on the orphan list and modify
3269 * i_disksize immediately, so that during the subsequent flushing of
3270 * dirty pages and freeing of disk blocks, we can guarantee that any
3271 * commit will leave the blocks being flushed in an unused state on
3272 * disk. (On recovery, the inode will get truncated and the blocks will
3273 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003274 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 *
3276 * Called with inode->sem down.
3277 */
3278int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3279{
3280 struct inode *inode = dentry->d_inode;
3281 int error, rc = 0;
3282 const unsigned int ia_valid = attr->ia_valid;
3283
3284 error = inode_change_ok(inode, attr);
3285 if (error)
3286 return error;
3287
Dmitry Monakhov12755622010-04-08 22:04:20 +04003288 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05003289 dquot_initialize(inode);
Eric W. Biederman15232992012-02-07 15:41:24 -08003290 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3291 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 handle_t *handle;
3293
3294 /* (user+group)*(old+new) structure, inode write (sb,
3295 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003296 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3297 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 if (IS_ERR(handle)) {
3299 error = PTR_ERR(handle);
3300 goto err_out;
3301 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05003302 error = dquot_transfer(inode, attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 if (error) {
3304 ext3_journal_stop(handle);
3305 return error;
3306 }
3307 /* Update corresponding info in inode so that everything is in
3308 * one transaction */
3309 if (attr->ia_valid & ATTR_UID)
3310 inode->i_uid = attr->ia_uid;
3311 if (attr->ia_valid & ATTR_GID)
3312 inode->i_gid = attr->ia_gid;
3313 error = ext3_mark_inode_dirty(handle, inode);
3314 ext3_journal_stop(handle);
3315 }
3316
Christoph Hellwig562c72a2011-06-24 14:29:45 -04003317 if (attr->ia_valid & ATTR_SIZE)
3318 inode_dio_wait(inode);
3319
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 if (S_ISREG(inode->i_mode) &&
3321 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3322 handle_t *handle;
3323
3324 handle = ext3_journal_start(inode, 3);
3325 if (IS_ERR(handle)) {
3326 error = PTR_ERR(handle);
3327 goto err_out;
3328 }
3329
3330 error = ext3_orphan_add(handle, inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02003331 if (error) {
3332 ext3_journal_stop(handle);
3333 goto err_out;
3334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 EXT3_I(inode)->i_disksize = attr->ia_size;
Jan Karaee3e77f2011-06-03 21:58:11 +02003336 error = ext3_mark_inode_dirty(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 ext3_journal_stop(handle);
Jan Karaee3e77f2011-06-03 21:58:11 +02003338 if (error) {
3339 /* Some hard fs error must have happened. Bail out. */
3340 ext3_orphan_del(NULL, inode);
3341 goto err_out;
3342 }
3343 rc = ext3_block_truncate_page(inode, attr->ia_size);
3344 if (rc) {
3345 /* Cleanup orphan list and exit */
3346 handle = ext3_journal_start(inode, 3);
3347 if (IS_ERR(handle)) {
3348 ext3_orphan_del(NULL, inode);
3349 goto err_out;
3350 }
3351 ext3_orphan_del(handle, inode);
3352 ext3_journal_stop(handle);
3353 goto err_out;
3354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 }
3356
Christoph Hellwig10257742010-06-04 11:30:02 +02003357 if ((attr->ia_valid & ATTR_SIZE) &&
3358 attr->ia_size != i_size_read(inode)) {
Jan Kara40680f2f2011-05-24 22:24:47 +02003359 truncate_setsize(inode, attr->ia_size);
3360 ext3_truncate(inode);
Christoph Hellwig10257742010-06-04 11:30:02 +02003361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362
Christoph Hellwig10257742010-06-04 11:30:02 +02003363 setattr_copy(inode, attr);
3364 mark_inode_dirty(inode);
3365
3366 if (ia_valid & ATTR_MODE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 rc = ext3_acl_chmod(inode);
3368
3369err_out:
3370 ext3_std_error(inode->i_sb, error);
3371 if (!error)
3372 error = rc;
3373 return error;
3374}
3375
3376
3377/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003378 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 *
3380 * With N blocks per page, it may be:
3381 * N data blocks
3382 * 2 indirect block
3383 * 2 dindirect
3384 * 1 tindirect
3385 * N+5 bitmap blocks (from the above)
3386 * N+5 group descriptor summary blocks
3387 * 1 inode block
3388 * 1 superblock.
3389 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3390 *
3391 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3392 *
3393 * With ordered or writeback data it's the same, less the N data blocks.
3394 *
3395 * If the inode's direct blocks can hold an integral number of pages then a
3396 * page cannot straddle two indirect blocks, and we can only touch one indirect
3397 * and dindirect block, and the "5" above becomes "3".
3398 *
3399 * This still overestimates under most circumstances. If we were to pass the
3400 * start and end offsets in here as well we could do block_to_path() on each
3401 * block and work out the exact number of indirects which are touched. Pah.
3402 */
3403
3404static int ext3_writepage_trans_blocks(struct inode *inode)
3405{
3406 int bpp = ext3_journal_blocks_per_page(inode);
3407 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3408 int ret;
3409
3410 if (ext3_should_journal_data(inode))
3411 ret = 3 * (bpp + indirects) + 2;
3412 else
Yongqiang Yang523334b2011-03-24 08:48:39 +08003413 ret = 2 * (bpp + indirects) + indirects + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414
3415#ifdef CONFIG_QUOTA
Christoph Hellwig871a2932010-03-03 09:05:07 -05003416 /* We know that structure was already allocated during dquot_initialize so
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 * we will be updating only the data blocks + inodes */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003418 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419#endif
3420
3421 return ret;
3422}
3423
3424/*
3425 * The caller must have previously called ext3_reserve_inode_write().
3426 * Give this, we know that the caller already has write access to iloc->bh.
3427 */
3428int ext3_mark_iloc_dirty(handle_t *handle,
3429 struct inode *inode, struct ext3_iloc *iloc)
3430{
3431 int err = 0;
3432
3433 /* the do_update_inode consumes one bh->b_count */
3434 get_bh(iloc->bh);
3435
3436 /* ext3_do_update_inode() does journal_dirty_metadata */
3437 err = ext3_do_update_inode(handle, inode, iloc);
3438 put_bh(iloc->bh);
3439 return err;
3440}
3441
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003442/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003444 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 */
3446
3447int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003448ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 struct ext3_iloc *iloc)
3450{
3451 int err = 0;
3452 if (handle) {
3453 err = ext3_get_inode_loc(inode, iloc);
3454 if (!err) {
3455 BUFFER_TRACE(iloc->bh, "get_write_access");
3456 err = ext3_journal_get_write_access(handle, iloc->bh);
3457 if (err) {
3458 brelse(iloc->bh);
3459 iloc->bh = NULL;
3460 }
3461 }
3462 }
3463 ext3_std_error(inode->i_sb, err);
3464 return err;
3465}
3466
3467/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003468 * What we do here is to mark the in-core inode as clean with respect to inode
3469 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470 * This means that the in-core inode may be reaped by prune_icache
3471 * without having to perform any I/O. This is a very good thing,
3472 * because *any* task may call prune_icache - even ones which
3473 * have a transaction open against a different journal.
3474 *
3475 * Is this cheating? Not really. Sure, we haven't written the
3476 * inode out, but prune_icache isn't a user-visible syncing function.
3477 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3478 * we start and wait on commits.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 */
3480int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3481{
3482 struct ext3_iloc iloc;
3483 int err;
3484
3485 might_sleep();
Lukas Czerner785c4bc2011-05-23 18:33:01 +02003486 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 err = ext3_reserve_inode_write(handle, inode, &iloc);
3488 if (!err)
3489 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3490 return err;
3491}
3492
3493/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003494 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 *
3496 * We're really interested in the case where a file is being extended.
3497 * i_size has been changed by generic_commit_write() and we thus need
3498 * to include the updated inode in the current transaction.
3499 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003500 * Also, dquot_alloc_space() will always dirty the inode when blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 * are allocated to the file.
3502 *
3503 * If the inode is marked synchronous, we don't honour that here - doing
3504 * so would cause a commit on atime updates, which we don't bother doing.
3505 * We handle synchronous inodes at the highest possible level.
3506 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04003507void ext3_dirty_inode(struct inode *inode, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508{
3509 handle_t *current_handle = ext3_journal_current_handle();
3510 handle_t *handle;
3511
3512 handle = ext3_journal_start(inode, 2);
3513 if (IS_ERR(handle))
3514 goto out;
3515 if (current_handle &&
3516 current_handle->h_transaction != handle->h_transaction) {
3517 /* This task has a transaction open against a different fs */
3518 printk(KERN_EMERG "%s: transactions do not match!\n",
Harvey Harrisone05b6b52008-04-28 02:16:15 -07003519 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520 } else {
3521 jbd_debug(5, "marking dirty. outer handle=%p\n",
3522 current_handle);
3523 ext3_mark_inode_dirty(handle, inode);
3524 }
3525 ext3_journal_stop(handle);
3526out:
3527 return;
3528}
3529
Andrew Mortond6859bf2006-03-26 01:38:03 -08003530#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003531/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 * Bind an inode's backing buffer_head into this transaction, to prevent
3533 * it from being flushed to disk early. Unlike
3534 * ext3_reserve_inode_write, this leaves behind no bh reference and
3535 * returns no iloc structure, so the caller needs to repeat the iloc
3536 * lookup to mark the inode dirty later.
3537 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003538static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539{
3540 struct ext3_iloc iloc;
3541
3542 int err = 0;
3543 if (handle) {
3544 err = ext3_get_inode_loc(inode, &iloc);
3545 if (!err) {
3546 BUFFER_TRACE(iloc.bh, "get_write_access");
3547 err = journal_get_write_access(handle, iloc.bh);
3548 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003549 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 iloc.bh);
3551 brelse(iloc.bh);
3552 }
3553 }
3554 ext3_std_error(inode->i_sb, err);
3555 return err;
3556}
3557#endif
3558
3559int ext3_change_inode_journal_flag(struct inode *inode, int val)
3560{
3561 journal_t *journal;
3562 handle_t *handle;
3563 int err;
3564
3565 /*
3566 * We have to be very careful here: changing a data block's
3567 * journaling status dynamically is dangerous. If we write a
3568 * data block to the journal, change the status and then delete
3569 * that block, we risk forgetting to revoke the old log record
3570 * from the journal and so a subsequent replay can corrupt data.
3571 * So, first we make sure that the journal is empty and that
3572 * nobody is changing anything.
3573 */
3574
3575 journal = EXT3_JOURNAL(inode);
Dave Hansene3a68e32007-07-15 23:41:14 -07003576 if (is_journal_aborted(journal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577 return -EROFS;
3578
3579 journal_lock_updates(journal);
3580 journal_flush(journal);
3581
3582 /*
3583 * OK, there are no updates running now, and all cached data is
3584 * synced to disk. We are now in a completely consistent state
3585 * which doesn't have anything in the journal, and we know that
3586 * no filesystem updates are running, so it is safe to modify
3587 * the inode's in-core data-journaling state flag now.
3588 */
3589
3590 if (val)
3591 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3592 else
3593 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3594 ext3_set_aops(inode);
3595
3596 journal_unlock_updates(journal);
3597
3598 /* Finally we can mark the inode as dirty. */
3599
3600 handle = ext3_journal_start(inode, 1);
3601 if (IS_ERR(handle))
3602 return PTR_ERR(handle);
3603
3604 err = ext3_mark_inode_dirty(handle, inode);
3605 handle->h_sync = 1;
3606 ext3_journal_stop(handle);
3607 ext3_std_error(inode->i_sb, err);
3608
3609 return err;
3610}