blob: d512c4bc4ad7103b968d454266d1814b8d59de41 [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>
Al Viro4613ad12012-03-29 22:30:07 -040030#include "ext3.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xattr.h"
32#include "acl.h"
33
34static int ext3_writepage_trans_blocks(struct inode *inode);
Jan Karaee3e77f2011-06-03 21:58:11 +020035static int ext3_block_truncate_page(struct inode *inode, loff_t from);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37/*
38 * Test whether an inode is a fast symlink.
39 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080040static int ext3_inode_is_fast_symlink(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070041{
42 int ea_blocks = EXT3_I(inode)->i_file_acl ?
43 (inode->i_sb->s_blocksize >> 9) : 0;
44
Andrew Mortond6859bf2006-03-26 01:38:03 -080045 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046}
47
Andrew Mortond6859bf2006-03-26 01:38:03 -080048/*
49 * The ext3 forget function must perform a revoke if we are freeing data
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * which has been journaled. Metadata (eg. indirect blocks) must be
Mingming Caoae6ddcc2006-09-27 01:49:27 -070051 * revoked in all cases.
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 *
53 * "bh" may be NULL: a metadata block may have been freed from memory
54 * but there may still be a record of it in the journal, and that record
55 * still needs to be revoked.
56 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080057int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -070058 struct buffer_head *bh, ext3_fsblk_t blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
60 int err;
61
62 might_sleep();
63
Lukas Czerner785c4bc2011-05-23 18:33:01 +020064 trace_ext3_forget(inode, is_metadata, blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 BUFFER_TRACE(bh, "enter");
66
67 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
68 "data mode %lx\n",
69 bh, is_metadata, inode->i_mode,
70 test_opt(inode->i_sb, DATA_FLAGS));
71
72 /* Never use the revoke function if we are doing full data
73 * journaling: there is no need to, and a V1 superblock won't
74 * support it. Otherwise, only skip the revoke on un-journaled
75 * data blocks. */
76
77 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
78 (!is_metadata && !ext3_should_journal_data(inode))) {
79 if (bh) {
80 BUFFER_TRACE(bh, "call journal_forget");
81 return ext3_journal_forget(handle, bh);
82 }
83 return 0;
84 }
85
86 /*
87 * data!=journal && (is_metadata || should_journal_data(inode))
88 */
89 BUFFER_TRACE(bh, "call ext3_journal_revoke");
90 err = ext3_journal_revoke(handle, blocknr, bh);
91 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -070092 ext3_abort(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 "error %d when attempting revoke", err);
94 BUFFER_TRACE(bh, "exit");
95 return err;
96}
97
98/*
Andrew Mortond6859bf2006-03-26 01:38:03 -080099 * Work out how many blocks we need to proceed with the next chunk of a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * truncate transaction.
101 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700102static unsigned long blocks_for_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
104 unsigned long needed;
105
106 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
107
108 /* Give ourselves just enough room to cope with inodes in which
109 * i_blocks is corrupt: we've seen disk corruptions in the past
110 * which resulted in random data in an inode which looked enough
111 * like a regular file for ext3 to try to delete it. Things
112 * will go a bit crazy if that happens, but at least we should
113 * try not to panic the whole kernel. */
114 if (needed < 2)
115 needed = 2;
116
117 /* But we need to bound the transaction so we don't overflow the
118 * journal. */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700119 if (needed > EXT3_MAX_TRANS_DATA)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 needed = EXT3_MAX_TRANS_DATA;
121
Jan Kara1f545872005-06-23 22:01:04 -0700122 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123}
124
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700125/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 * Truncate transactions can be complex and absolutely huge. So we need to
127 * be able to restart the transaction at a conventient checkpoint to make
128 * sure we don't overflow the journal.
129 *
130 * start_transaction gets us a new handle for a truncate transaction,
131 * and extend_transaction tries to extend the existing one a bit. If
132 * extend fails, we need to propagate the failure up and restart the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700133 * transaction in the top-level truncate loop. --sct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700135static handle_t *start_transaction(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 handle_t *result;
138
139 result = ext3_journal_start(inode, blocks_for_truncate(inode));
140 if (!IS_ERR(result))
141 return result;
142
143 ext3_std_error(inode->i_sb, PTR_ERR(result));
144 return result;
145}
146
147/*
148 * Try to extend this transaction for the purposes of truncation.
149 *
150 * Returns 0 if we managed to create more room. If we can't create more
151 * room, and the transaction must be restarted we return 1.
152 */
153static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
154{
155 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
156 return 0;
157 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
158 return 0;
159 return 1;
160}
161
162/*
163 * Restart the transaction associated with *handle. This does a commit,
164 * so before we call here everything must be consistently dirtied against
165 * this transaction.
166 */
Jan Kara00171d32009-08-11 19:06:10 +0200167static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
Jan Kara00171d32009-08-11 19:06:10 +0200169 int ret;
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara00171d32009-08-11 19:06:10 +0200172 /*
173 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
174 * At this moment, get_block can be called only for blocks inside
175 * i_size since page cache has been already dropped and writes are
176 * blocked by i_mutex. So we can safely drop the truncate_mutex.
177 */
178 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
179 ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
180 mutex_lock(&EXT3_I(inode)->truncate_mutex);
181 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
184/*
Al Viroac14a952010-06-06 07:08:19 -0400185 * Called at inode eviction from icache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 */
Al Viroac14a952010-06-06 07:08:19 -0400187void ext3_evict_inode (struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Jan Karab22570d2011-07-23 01:21:38 +0200189 struct ext3_inode_info *ei = EXT3_I(inode);
Al Viroac14a952010-06-06 07:08:19 -0400190 struct ext3_block_alloc_info *rsv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 handle_t *handle;
Al Viroac14a952010-06-06 07:08:19 -0400192 int want_delete = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200194 trace_ext3_evict_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400195 if (!inode->i_nlink && !is_bad_inode(inode)) {
Christoph Hellwig871a2932010-03-03 09:05:07 -0500196 dquot_initialize(inode);
Al Viroac14a952010-06-06 07:08:19 -0400197 want_delete = 1;
198 }
Christoph Hellwig907f4552010-03-03 09:05:06 -0500199
Jan Karab22570d2011-07-23 01:21:38 +0200200 /*
201 * When journalling data dirty buffers are tracked only in the journal.
202 * So although mm thinks everything is clean and ready for reaping the
203 * inode might still have some pages to write in the running
204 * transaction or waiting to be checkpointed. Thus calling
205 * journal_invalidatepage() (via truncate_inode_pages()) to discard
206 * these buffers can cause data loss. Also even if we did not discard
207 * these buffers, we would have no way to find them after the inode
208 * is reaped and thus user could see stale data if he tries to read
209 * them before the transaction is checkpointed. So be careful and
210 * force everything to disk here... We use ei->i_datasync_tid to
211 * store the newest transaction containing inode's data.
212 *
213 * Note that directories do not have this problem because they don't
214 * use page cache.
Dan Carpenterbcdd0c12011-11-22 11:00:20 +0300215 *
216 * The s_journal check handles the case when ext3_get_journal() fails
217 * and puts the journal inode.
Jan Karab22570d2011-07-23 01:21:38 +0200218 */
219 if (inode->i_nlink && ext3_should_journal_data(inode) &&
Dan Carpenterbcdd0c12011-11-22 11:00:20 +0300220 EXT3_SB(inode->i_sb)->s_journal &&
Jan Karab22570d2011-07-23 01:21:38 +0200221 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
222 tid_t commit_tid = atomic_read(&ei->i_datasync_tid);
223 journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
224
225 log_start_commit(journal, commit_tid);
226 log_wait_commit(journal, commit_tid);
227 filemap_write_and_wait(&inode->i_data);
228 }
Mark Fashehfef26652005-09-09 13:01:31 -0700229 truncate_inode_pages(&inode->i_data, 0);
230
Al Viroac14a952010-06-06 07:08:19 -0400231 ext3_discard_reservation(inode);
Jan Karab22570d2011-07-23 01:21:38 +0200232 rsv = ei->i_block_alloc_info;
233 ei->i_block_alloc_info = NULL;
Al Viroac14a952010-06-06 07:08:19 -0400234 if (unlikely(rsv))
235 kfree(rsv);
236
237 if (!want_delete)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 goto no_delete;
239
240 handle = start_transaction(inode);
241 if (IS_ERR(handle)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800242 /*
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
245 * cleaned up.
246 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 ext3_orphan_del(NULL, inode);
248 goto no_delete;
249 }
250
251 if (IS_SYNC(inode))
252 handle->h_sync = 1;
253 inode->i_size = 0;
254 if (inode->i_blocks)
255 ext3_truncate(inode);
256 /*
Jan Kara40680f2f2011-05-24 22:24:47 +0200257 * Kill off the orphan record created when the inode lost the last
258 * link. Note that ext3_orphan_del() has to be able to cope with the
259 * deletion of a non-existent orphan - ext3_truncate() could
260 * have removed the record.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 */
262 ext3_orphan_del(handle, inode);
Jan Karab22570d2011-07-23 01:21:38 +0200263 ei->i_dtime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700265 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 * One subtle ordering requirement: if anything has gone wrong
267 * (transaction abort, IO errors, whatever), then we can still
268 * do these next steps (the fs will already have been marked as
269 * having errors), but we can't free the inode if the mark_dirty
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700270 * fails.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 */
Al Viroac14a952010-06-06 07:08:19 -0400272 if (ext3_mark_inode_dirty(handle, inode)) {
273 /* If that failed, just dquot_drop() and be done with that */
274 dquot_drop(inode);
Jan Karadbd57682012-05-03 14:48:02 +0200275 clear_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400276 } else {
277 ext3_xattr_delete_inode(handle, inode);
278 dquot_free_inode(inode);
279 dquot_drop(inode);
Jan Karadbd57682012-05-03 14:48:02 +0200280 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 ext3_free_inode(handle, inode);
Al Viroac14a952010-06-06 07:08:19 -0400282 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 ext3_journal_stop(handle);
284 return;
285no_delete:
Jan Karadbd57682012-05-03 14:48:02 +0200286 clear_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400287 dquot_drop(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290typedef struct {
291 __le32 *p;
292 __le32 key;
293 struct buffer_head *bh;
294} Indirect;
295
296static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
297{
298 p->key = *(p->p = v);
299 p->bh = bh;
300}
301
Andrew Mortond6859bf2006-03-26 01:38:03 -0800302static int verify_chain(Indirect *from, Indirect *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
304 while (from <= to && from->key == *from->p)
305 from++;
306 return (from > to);
307}
308
309/**
310 * ext3_block_to_path - parse the block number into array of offsets
311 * @inode: inode in question (we are only interested in its superblock)
312 * @i_block: block number to be parsed
313 * @offsets: array to store the offsets in
314 * @boundary: set this non-zero if the referred-to block is likely to be
315 * followed (on disk) by an indirect block.
316 *
317 * To store the locations of file's data ext3 uses a data structure common
318 * for UNIX filesystems - tree of pointers anchored in the inode, with
319 * data blocks at leaves and indirect blocks in intermediate nodes.
320 * This function translates the block number into path in that tree -
321 * return value is the path length and @offsets[n] is the offset of
322 * pointer to (n+1)th node in the nth one. If @block is out of range
323 * (negative or too large) warning is printed and zero returned.
324 *
325 * Note: function doesn't find node addresses, so no IO is needed. All
326 * we need to know is the capacity of indirect blocks (taken from the
327 * inode->i_sb).
328 */
329
330/*
331 * Portability note: the last comparison (check that we fit into triple
332 * indirect block) is spelled differently, because otherwise on an
333 * architecture with 32-bit longs and 8Kb pages we might get into trouble
334 * if our filesystem had 8Kb blocks. We might use long long, but that would
335 * kill us on x86. Oh, well, at least the sign propagation does not matter -
336 * i_block would have to be negative in the very beginning, so we would not
337 * get there at all.
338 */
339
340static int ext3_block_to_path(struct inode *inode,
341 long i_block, int offsets[4], int *boundary)
342{
343 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
344 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
345 const long direct_blocks = EXT3_NDIR_BLOCKS,
346 indirect_blocks = ptrs,
347 double_blocks = (1 << (ptrs_bits * 2));
348 int n = 0;
349 int final = 0;
350
351 if (i_block < 0) {
352 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
353 } else if (i_block < direct_blocks) {
354 offsets[n++] = i_block;
355 final = direct_blocks;
356 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
357 offsets[n++] = EXT3_IND_BLOCK;
358 offsets[n++] = i_block;
359 final = ptrs;
360 } else if ((i_block -= indirect_blocks) < double_blocks) {
361 offsets[n++] = EXT3_DIND_BLOCK;
362 offsets[n++] = i_block >> ptrs_bits;
363 offsets[n++] = i_block & (ptrs - 1);
364 final = ptrs;
365 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
366 offsets[n++] = EXT3_TIND_BLOCK;
367 offsets[n++] = i_block >> (ptrs_bits * 2);
368 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
369 offsets[n++] = i_block & (ptrs - 1);
370 final = ptrs;
371 } else {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800372 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
374 if (boundary)
Mingming Cao89747d32006-03-26 01:37:55 -0800375 *boundary = final - 1 - (i_block & (ptrs - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return n;
377}
378
379/**
380 * ext3_get_branch - read the chain of indirect blocks leading to data
381 * @inode: inode in question
382 * @depth: depth of the chain (1 - direct pointer, etc.)
383 * @offsets: offsets of pointers in inode/indirect blocks
384 * @chain: place to store the result
385 * @err: here we store the error value
386 *
387 * Function fills the array of triples <key, p, bh> and returns %NULL
388 * if everything went OK or the pointer to the last filled triple
389 * (incomplete one) otherwise. Upon the return chain[i].key contains
390 * the number of (i+1)-th block in the chain (as it is stored in memory,
391 * i.e. little-endian 32-bit), chain[i].p contains the address of that
392 * number (it points into struct inode for i==0 and into the bh->b_data
393 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
394 * block for i>0 and NULL for i==0. In other words, it holds the block
395 * numbers of the chain, addresses they were taken from (and where we can
396 * verify that chain did not change) and buffer_heads hosting these
397 * numbers.
398 *
399 * Function stops when it stumbles upon zero pointer (absent block)
400 * (pointer to last triple returned, *@err == 0)
401 * or when it gets an IO error reading an indirect block
402 * (ditto, *@err == -EIO)
403 * or when it notices that chain had been changed while it was reading
404 * (ditto, *@err == -EAGAIN)
405 * or when it reads all @depth-1 indirect blocks successfully and finds
406 * the whole chain, all way to the data (returns %NULL, *err == 0).
407 */
408static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
409 Indirect chain[4], int *err)
410{
411 struct super_block *sb = inode->i_sb;
412 Indirect *p = chain;
413 struct buffer_head *bh;
414
415 *err = 0;
416 /* i_data is not going away, no lock needed */
417 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
418 if (!p->key)
419 goto no_block;
420 while (--depth) {
421 bh = sb_bread(sb, le32_to_cpu(p->key));
422 if (!bh)
423 goto failure;
424 /* Reader: pointers */
425 if (!verify_chain(chain, p))
426 goto changed;
427 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
428 /* Reader: end */
429 if (!p->key)
430 goto no_block;
431 }
432 return NULL;
433
434changed:
435 brelse(bh);
436 *err = -EAGAIN;
437 goto no_block;
438failure:
439 *err = -EIO;
440no_block:
441 return p;
442}
443
444/**
445 * ext3_find_near - find a place for allocation with sufficient locality
446 * @inode: owner
447 * @ind: descriptor of indirect block.
448 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000449 * This function returns the preferred place for block allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 * It is used when heuristic for sequential allocation fails.
451 * Rules are:
452 * + if there is a block to the left of our position - allocate near it.
453 * + if pointer will live in indirect block - allocate near that block.
454 * + if pointer will live in inode - allocate in the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700455 * cylinder group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 *
457 * In the latter case we colour the starting block by the callers PID to
458 * prevent it from clashing with concurrent allocations for a different inode
459 * in the same block group. The PID is used here so that functionally related
460 * files will be close-by on-disk.
461 *
462 * Caller must make sure that @ind is valid and will stay that way.
463 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700464static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 struct ext3_inode_info *ei = EXT3_I(inode);
467 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
468 __le32 *p;
Mingming Cao43d23f92006-06-25 05:48:07 -0700469 ext3_fsblk_t bg_start;
470 ext3_grpblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 /* Try to find previous block */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800473 for (p = ind->p - 1; p >= start; p--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 if (*p)
475 return le32_to_cpu(*p);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
478 /* No such thing, so let's try location of indirect block */
479 if (ind->bh)
480 return ind->bh->b_blocknr;
481
482 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800483 * It is going to be referred to from the inode itself? OK, just put it
484 * into the same cylinder group then.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700486 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 colour = (current->pid % 16) *
488 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
489 return bg_start + colour;
490}
491
492/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000493 * ext3_find_goal - find a preferred place for allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 * @inode: owner
495 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000498 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800499 * returns it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 */
501
Mingming Cao43d23f92006-06-25 05:48:07 -0700502static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800503 Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Andrew Mortond6859bf2006-03-26 01:38:03 -0800505 struct ext3_block_alloc_info *block_i;
506
507 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 /*
510 * try the heuristic for sequential allocation,
511 * failing that at least try to get decent locality.
512 */
513 if (block_i && (block == block_i->last_alloc_logical_block + 1)
514 && (block_i->last_alloc_physical_block != 0)) {
Mingming Caofe55c452005-05-01 08:59:20 -0700515 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 }
517
Mingming Caofe55c452005-05-01 08:59:20 -0700518 return ext3_find_near(inode, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
Andrew Mortond6859bf2006-03-26 01:38:03 -0800520
Mingming Caob47b2472006-03-26 01:37:56 -0800521/**
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900522 * ext3_blks_to_allocate - Look up the block map and count the number
Mingming Caob47b2472006-03-26 01:37:56 -0800523 * of direct blocks need to be allocated for the given branch.
524 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700525 * @branch: chain of indirect blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800526 * @k: number of blocks need for indirect blocks
527 * @blks: number of data blocks to be mapped.
528 * @blocks_to_boundary: the offset in the indirect block
529 *
530 * return the total number of blocks to be allocate, including the
531 * direct and indirect blocks.
532 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800533static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Mingming Caob47b2472006-03-26 01:37:56 -0800534 int blocks_to_boundary)
535{
536 unsigned long count = 0;
537
538 /*
539 * Simple case, [t,d]Indirect block(s) has not allocated yet
540 * then it's clear blocks on that path have not allocated
541 */
542 if (k > 0) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800543 /* right now we don't handle cross boundary allocation */
Mingming Caob47b2472006-03-26 01:37:56 -0800544 if (blks < blocks_to_boundary + 1)
545 count += blks;
546 else
547 count += blocks_to_boundary + 1;
548 return count;
549 }
550
551 count++;
552 while (count < blks && count <= blocks_to_boundary &&
553 le32_to_cpu(*(branch[0].p + count)) == 0) {
554 count++;
555 }
556 return count;
557}
558
559/**
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900560 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
561 * @handle: handle for this transaction
562 * @inode: owner
563 * @goal: preferred place for allocation
Mingming Caob47b2472006-03-26 01:37:56 -0800564 * @indirect_blks: the number of blocks need to allocate for indirect
565 * blocks
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900566 * @blks: number of blocks need to allocated for direct blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800567 * @new_blocks: on return it will store the new block numbers for
568 * the indirect blocks(if needed) and the first direct block,
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900569 * @err: here we store the error value
570 *
571 * return the number of direct blocks allocated
Mingming Caob47b2472006-03-26 01:37:56 -0800572 */
573static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700574 ext3_fsblk_t goal, int indirect_blks, int blks,
575 ext3_fsblk_t new_blocks[4], int *err)
Mingming Caob47b2472006-03-26 01:37:56 -0800576{
577 int target, i;
578 unsigned long count = 0;
579 int index = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700580 ext3_fsblk_t current_block = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800581 int ret = 0;
582
583 /*
584 * Here we try to allocate the requested multiple blocks at once,
585 * on a best-effort basis.
586 * To build a branch, we should allocate blocks for
587 * the indirect blocks(if not allocated yet), and at least
588 * the first direct block of this branch. That's the
589 * minimum number of blocks need to allocate(required)
590 */
591 target = blks + indirect_blks;
592
593 while (1) {
594 count = target;
595 /* allocating blocks for indirect blocks and direct blocks */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800596 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
Mingming Caob47b2472006-03-26 01:37:56 -0800597 if (*err)
598 goto failed_out;
599
600 target -= count;
601 /* allocate blocks for indirect blocks */
602 while (index < indirect_blks && count) {
603 new_blocks[index++] = current_block++;
604 count--;
605 }
606
607 if (count > 0)
608 break;
609 }
610
611 /* save the new block number for the first direct block */
612 new_blocks[index] = current_block;
613
614 /* total number of blocks allocated for direct blocks */
615 ret = count;
616 *err = 0;
617 return ret;
618failed_out:
619 for (i = 0; i <index; i++)
620 ext3_free_blocks(handle, inode, new_blocks[i], 1);
621 return ret;
622}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624/**
625 * ext3_alloc_branch - allocate and set up a chain of blocks.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900626 * @handle: handle for this transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 * @inode: owner
Mingming Caob47b2472006-03-26 01:37:56 -0800628 * @indirect_blks: number of allocated indirect blocks
629 * @blks: number of allocated direct blocks
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900630 * @goal: preferred place for allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 * @offsets: offsets (in the blocks) to store the pointers to next.
632 * @branch: place to store the chain in.
633 *
Mingming Caob47b2472006-03-26 01:37:56 -0800634 * This function allocates blocks, zeroes out all but the last one,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 * links them into chain and (if we are synchronous) writes them to disk.
636 * In other words, it prepares a branch that can be spliced onto the
637 * inode. It stores the information about that chain in the branch[], in
638 * the same format as ext3_get_branch() would do. We are calling it after
639 * we had read the existing part of chain and partial points to the last
640 * triple of that (one with zero ->key). Upon the exit we have the same
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -0800641 * picture as after the successful ext3_get_block(), except that in one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 * place chain is disconnected - *branch->p is still zero (we did not
643 * set the last link), but branch->key contains the number that should
644 * be placed into *branch->p to fill that gap.
645 *
646 * If allocation fails we free all blocks we've allocated (and forget
647 * their buffer_heads) and return the error value the from failed
648 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
649 * as described above and return 0.
650 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700652 int indirect_blks, int *blks, ext3_fsblk_t goal,
Mingming Caob47b2472006-03-26 01:37:56 -0800653 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 int blocksize = inode->i_sb->s_blocksize;
Mingming Caob47b2472006-03-26 01:37:56 -0800656 int i, n = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 int err = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800658 struct buffer_head *bh;
659 int num;
Mingming Cao43d23f92006-06-25 05:48:07 -0700660 ext3_fsblk_t new_blocks[4];
661 ext3_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Mingming Caob47b2472006-03-26 01:37:56 -0800663 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
664 *blks, new_blocks, &err);
665 if (err)
666 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Mingming Caob47b2472006-03-26 01:37:56 -0800668 branch[0].key = cpu_to_le32(new_blocks[0]);
669 /*
670 * metadata blocks and data blocks are allocated.
671 */
672 for (n = 1; n <= indirect_blks; n++) {
673 /*
674 * Get buffer_head for parent block, zero it out
675 * and set the pointer to new one, then send
676 * parent to disk.
677 */
678 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
Wang Shilong61f43e62013-01-12 01:22:33 -0800679 if (unlikely(!bh)) {
680 err = -ENOMEM;
681 goto failed;
682 }
Mingming Caob47b2472006-03-26 01:37:56 -0800683 branch[n].bh = bh;
684 lock_buffer(bh);
685 BUFFER_TRACE(bh, "call get_create_access");
686 err = ext3_journal_get_create_access(handle, bh);
687 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 unlock_buffer(bh);
Mingming Caob47b2472006-03-26 01:37:56 -0800689 brelse(bh);
690 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Mingming Caob47b2472006-03-26 01:37:56 -0800693 memset(bh->b_data, 0, blocksize);
694 branch[n].p = (__le32 *) bh->b_data + offsets[n];
695 branch[n].key = cpu_to_le32(new_blocks[n]);
696 *branch[n].p = branch[n].key;
697 if ( n == indirect_blks) {
698 current_block = new_blocks[n];
699 /*
700 * End of chain, update the last new metablock of
701 * the chain to point to the new allocated
702 * data blocks numbers
703 */
704 for (i=1; i < num; i++)
705 *(branch[n].p + i) = cpu_to_le32(++current_block);
706 }
707 BUFFER_TRACE(bh, "marking uptodate");
708 set_buffer_uptodate(bh);
709 unlock_buffer(bh);
710
711 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
712 err = ext3_journal_dirty_metadata(handle, bh);
713 if (err)
714 goto failed;
715 }
716 *blks = num;
717 return err;
718failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 /* Allocation failed, free what we already allocated */
Mingming Caob47b2472006-03-26 01:37:56 -0800720 for (i = 1; i <= n ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 BUFFER_TRACE(branch[i].bh, "call journal_forget");
722 ext3_journal_forget(handle, branch[i].bh);
723 }
Wang Shilong61f43e62013-01-12 01:22:33 -0800724 for (i = 0; i < indirect_blks; i++)
Mingming Caob47b2472006-03-26 01:37:56 -0800725 ext3_free_blocks(handle, inode, new_blocks[i], 1);
726
727 ext3_free_blocks(handle, inode, new_blocks[i], num);
728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 return err;
730}
731
732/**
Andrew Mortond6859bf2006-03-26 01:38:03 -0800733 * ext3_splice_branch - splice the allocated branch onto inode.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900734 * @handle: handle for this transaction
Andrew Mortond6859bf2006-03-26 01:38:03 -0800735 * @inode: owner
736 * @block: (logical) number of block we are adding
Andrew Mortond6859bf2006-03-26 01:38:03 -0800737 * @where: location of missing link
738 * @num: number of indirect blocks we are adding
739 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800741 * This function fills the missing link and does all housekeeping needed in
742 * inode (->i_blocks, etc.). In case of success we end up with the full
743 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800745static int ext3_splice_branch(handle_t *handle, struct inode *inode,
746 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
748 int i;
749 int err = 0;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800750 struct ext3_block_alloc_info *block_i;
Mingming Cao43d23f92006-06-25 05:48:07 -0700751 ext3_fsblk_t current_block;
Jan Karafe8bc912009-10-16 19:26:15 +0200752 struct ext3_inode_info *ei = EXT3_I(inode);
Kazuya Mioac1334b2012-02-08 16:34:29 +0900753 struct timespec now;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800754
Jan Karafe8bc912009-10-16 19:26:15 +0200755 block_i = ei->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /*
757 * If we're splicing into a [td]indirect block (as opposed to the
758 * inode) then we need to get write access to the [td]indirect block
759 * before the splice.
760 */
761 if (where->bh) {
762 BUFFER_TRACE(where->bh, "get_write_access");
763 err = ext3_journal_get_write_access(handle, where->bh);
764 if (err)
765 goto err_out;
766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /* That's it */
768
769 *where->p = where->key;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800770
771 /*
772 * Update the host buffer_head or inode to point to more just allocated
773 * direct blocks blocks
774 */
Mingming Caob47b2472006-03-26 01:37:56 -0800775 if (num == 0 && blks > 1) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700776 current_block = le32_to_cpu(where->key) + 1;
Mingming Caob47b2472006-03-26 01:37:56 -0800777 for (i = 1; i < blks; i++)
778 *(where->p + i ) = cpu_to_le32(current_block++);
779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781 /*
782 * update the most recently allocated logical & physical block
783 * in i_block_alloc_info, to assist find the proper goal block for next
784 * allocation
785 */
786 if (block_i) {
Mingming Caob47b2472006-03-26 01:37:56 -0800787 block_i->last_alloc_logical_block = block + blks - 1;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800788 block_i->last_alloc_physical_block =
Mingming Cao5dea5172006-05-03 19:55:12 -0700789 le32_to_cpu(where[num].key) + blks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
791
792 /* We are done with atomic stuff, now do the rest of housekeeping */
Kazuya Mioac1334b2012-02-08 16:34:29 +0900793 now = CURRENT_TIME_SEC;
794 if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) {
795 inode->i_ctime = now;
796 ext3_mark_inode_dirty(handle, inode);
797 }
Jan Karafe8bc912009-10-16 19:26:15 +0200798 /* ext3_mark_inode_dirty already updated i_sync_tid */
799 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801 /* had we spliced it onto indirect block? */
802 if (where->bh) {
803 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800804 * If we spliced it onto an indirect block, we haven't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 * altered the inode. Note however that if it is being spliced
806 * onto an indirect block at the very end of the file (the
807 * file is growing) then we *will* alter the inode to reflect
808 * the new i_size. But that is not done here - it is done in
809 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
810 */
811 jbd_debug(5, "splicing indirect only\n");
812 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
813 err = ext3_journal_dirty_metadata(handle, where->bh);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700814 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 goto err_out;
816 } else {
817 /*
818 * OK, we spliced it into the inode itself on a direct block.
819 * Inode was dirtied above.
820 */
821 jbd_debug(5, "splicing direct\n");
822 }
823 return err;
824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825err_out:
Mingming Caob47b2472006-03-26 01:37:56 -0800826 for (i = 1; i <= num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 BUFFER_TRACE(where[i].bh, "call journal_forget");
828 ext3_journal_forget(handle, where[i].bh);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800829 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
Mingming Caob47b2472006-03-26 01:37:56 -0800831 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return err;
834}
835
836/*
837 * Allocation strategy is simple: if we have to allocate something, we will
838 * have to go the whole way to leaf. So let's do it before attaching anything
839 * to tree, set linkage between the newborn blocks, write them if sync is
840 * required, recheck the path, free and repeat if check fails, otherwise
841 * set the last missing link (that will protect us from any truncate-generated
842 * removals - all blocks on the path are immune now) and possibly force the
843 * write on the parent block.
844 * That has a nice additional property: no special recovery from the failed
845 * allocations is needed - we simply release blocks and do not touch anything
846 * reachable from inode.
847 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800848 * `handle' can be NULL if create == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 *
850 * The BKL may not be held on entry here. Be sure to take it early.
Mingming Cao89747d32006-03-26 01:37:55 -0800851 * return > 0, # of blocks mapped or allocated.
852 * return = 0, if plain lookup failed.
853 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800855int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
856 sector_t iblock, unsigned long maxblocks,
857 struct buffer_head *bh_result,
Jan Kara43237b52009-05-20 18:41:58 +0200858 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 int err = -EIO;
861 int offsets[4];
862 Indirect chain[4];
863 Indirect *partial;
Mingming Cao43d23f92006-06-25 05:48:07 -0700864 ext3_fsblk_t goal;
Mingming Caob47b2472006-03-26 01:37:56 -0800865 int indirect_blks;
Mingming Cao89747d32006-03-26 01:37:55 -0800866 int blocks_to_boundary = 0;
867 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 struct ext3_inode_info *ei = EXT3_I(inode);
Mingming Cao89747d32006-03-26 01:37:55 -0800869 int count = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700870 ext3_fsblk_t first_block = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200873 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 J_ASSERT(handle != NULL || create == 0);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800875 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
877 if (depth == 0)
878 goto out;
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
881
882 /* Simplest case - block found, no allocation needed */
883 if (!partial) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700884 first_block = le32_to_cpu(chain[depth - 1].key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 clear_buffer_new(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800886 count++;
887 /*map more blocks*/
888 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700889 ext3_fsblk_t blk;
Mingming Cao5dea5172006-05-03 19:55:12 -0700890
Jan Karae8ef7aa2009-06-17 16:26:23 -0700891 if (!verify_chain(chain, chain + depth - 1)) {
Mingming Cao89747d32006-03-26 01:37:55 -0800892 /*
893 * Indirect block might be removed by
894 * truncate while we were reading it.
895 * Handling of that case: forget what we've
896 * got now. Flag the err as EAGAIN, so it
897 * will reread.
898 */
899 err = -EAGAIN;
900 count = 0;
901 break;
902 }
Mingming Cao5dea5172006-05-03 19:55:12 -0700903 blk = le32_to_cpu(*(chain[depth-1].p + count));
904
905 if (blk == first_block + count)
Mingming Cao89747d32006-03-26 01:37:55 -0800906 count++;
907 else
908 break;
909 }
910 if (err != -EAGAIN)
911 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 }
913
914 /* Next simple case - plain lookup or failed read of indirect block */
Mingming Caofe55c452005-05-01 08:59:20 -0700915 if (!create || err == -EIO)
916 goto cleanup;
917
Jan Kara40680f2f2011-05-24 22:24:47 +0200918 /*
919 * Block out ext3_truncate while we alter the tree
920 */
Arjan van de Ven97461512006-03-23 03:00:42 -0800921 mutex_lock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700922
923 /*
924 * If the indirect block is missing while we are reading
925 * the chain(ext3_get_branch() returns -EAGAIN err), or
926 * if the chain has been changed after we grab the semaphore,
927 * (either because another process truncated this branch, or
928 * another get_block allocated this branch) re-grab the chain to see if
929 * the request block has been allocated or not.
930 *
931 * Since we already block the truncate/other get_block
932 * at this point, we will have the current copy of the chain when we
933 * splice the branch into the tree.
934 */
935 if (err == -EAGAIN || !verify_chain(chain, partial)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 while (partial > chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 brelse(partial->bh);
938 partial--;
939 }
Mingming Caofe55c452005-05-01 08:59:20 -0700940 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
941 if (!partial) {
Mingming Cao89747d32006-03-26 01:37:55 -0800942 count++;
Arjan van de Ven97461512006-03-23 03:00:42 -0800943 mutex_unlock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700944 if (err)
945 goto cleanup;
946 clear_buffer_new(bh_result);
947 goto got_it;
948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 }
950
951 /*
Mingming Caofe55c452005-05-01 08:59:20 -0700952 * Okay, we need to do block allocation. Lazily initialize the block
953 * allocation info here if necessary
954 */
955 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 ext3_init_block_alloc_info(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800958 goal = ext3_find_goal(inode, iblock, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Mingming Caob47b2472006-03-26 01:37:56 -0800960 /* the number of blocks need to allocate for [d,t]indirect blocks */
961 indirect_blks = (chain + depth) - partial - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 /*
Mingming Caob47b2472006-03-26 01:37:56 -0800964 * Next look up the indirect map to count the totoal number of
965 * direct blocks to allocate for this branch.
966 */
967 count = ext3_blks_to_allocate(partial, indirect_blks,
968 maxblocks, blocks_to_boundary);
Mingming Caob47b2472006-03-26 01:37:56 -0800969 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
Mingming Caofe55c452005-05-01 08:59:20 -0700970 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Mingming Caofe55c452005-05-01 08:59:20 -0700972 /*
973 * The ext3_splice_branch call will free and forget any buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 * on the new chain if there is a failure, but that risks using
975 * up transaction credits, especially for bitmaps where the
976 * credits cannot be returned. Can we handle this somehow? We
Mingming Caofe55c452005-05-01 08:59:20 -0700977 * may need to return -EAGAIN upwards in the worst case. --sct
978 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (!err)
Mingming Caob47b2472006-03-26 01:37:56 -0800980 err = ext3_splice_branch(handle, inode, iblock,
981 partial, indirect_blks, count);
Arjan van de Ven97461512006-03-23 03:00:42 -0800982 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 if (err)
984 goto cleanup;
985
986 set_buffer_new(bh_result);
Mingming Caofe55c452005-05-01 08:59:20 -0700987got_it:
988 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
Suparna Bhattacharya20acaa12006-09-16 12:15:58 -0700989 if (count > blocks_to_boundary)
Mingming Caofe55c452005-05-01 08:59:20 -0700990 set_buffer_boundary(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800991 err = count;
Mingming Caofe55c452005-05-01 08:59:20 -0700992 /* Clean up and exit */
993 partial = chain + depth - 1; /* the whole chain */
994cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 while (partial > chain) {
Mingming Caofe55c452005-05-01 08:59:20 -0700996 BUFFER_TRACE(partial->bh, "call brelse");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 brelse(partial->bh);
998 partial--;
999 }
Mingming Caofe55c452005-05-01 08:59:20 -07001000 BUFFER_TRACE(bh_result, "returned");
1001out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001002 trace_ext3_get_blocks_exit(inode, iblock,
1003 depth ? le32_to_cpu(chain[depth-1].key) : 0,
1004 count, err);
Mingming Caofe55c452005-05-01 08:59:20 -07001005 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
1007
Jan Karabd1939d2008-02-06 01:40:21 -08001008/* Maximum number of blocks we map for direct IO at once. */
1009#define DIO_MAX_BLOCKS 4096
1010/*
1011 * Number of credits we need for writing DIO_MAX_BLOCKS:
1012 * We need sb + group descriptor + bitmap + inode -> 4
1013 * For B blocks with A block pointers per block we need:
1014 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1015 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1016 */
1017#define DIO_CREDITS 25
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001019static int ext3_get_block(struct inode *inode, sector_t iblock,
1020 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001022 handle_t *handle = ext3_journal_current_handle();
Jan Karabd1939d2008-02-06 01:40:21 -08001023 int ret = 0, started = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -08001024 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Jan Karabd1939d2008-02-06 01:40:21 -08001026 if (create && !handle) { /* Direct IO write... */
1027 if (max_blocks > DIO_MAX_BLOCKS)
1028 max_blocks = DIO_MAX_BLOCKS;
1029 handle = ext3_journal_start(inode, DIO_CREDITS +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001030 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
Jan Karabd1939d2008-02-06 01:40:21 -08001031 if (IS_ERR(handle)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 ret = PTR_ERR(handle);
Jan Karabd1939d2008-02-06 01:40:21 -08001033 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 }
Jan Karabd1939d2008-02-06 01:40:21 -08001035 started = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037
Jan Karabd1939d2008-02-06 01:40:21 -08001038 ret = ext3_get_blocks_handle(handle, inode, iblock,
Jan Kara43237b52009-05-20 18:41:58 +02001039 max_blocks, bh_result, create);
Jan Karabd1939d2008-02-06 01:40:21 -08001040 if (ret > 0) {
1041 bh_result->b_size = (ret << inode->i_blkbits);
1042 ret = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001043 }
Jan Karabd1939d2008-02-06 01:40:21 -08001044 if (started)
1045 ext3_journal_stop(handle);
1046out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 return ret;
1048}
1049
Josef Bacik68c9d702008-10-03 17:32:43 -04001050int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1051 u64 start, u64 len)
1052{
1053 return generic_block_fiemap(inode, fieinfo, start, len,
1054 ext3_get_block);
1055}
1056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057/*
1058 * `handle' can be NULL if create is zero
1059 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001060struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1061 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
1063 struct buffer_head dummy;
1064 int fatal = 0, err;
1065
1066 J_ASSERT(handle != NULL || create == 0);
1067
1068 dummy.b_state = 0;
1069 dummy.b_blocknr = -1000;
1070 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -08001071 err = ext3_get_blocks_handle(handle, inode, block, 1,
Jan Kara43237b52009-05-20 18:41:58 +02001072 &dummy, create);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001073 /*
1074 * ext3_get_blocks_handle() returns number of blocks
1075 * mapped. 0 in case of a HOLE.
1076 */
1077 if (err > 0) {
Julia Lawall47897752012-11-03 21:30:21 +01001078 WARN_ON(err > 1);
Mingming Cao89747d32006-03-26 01:37:55 -08001079 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001080 }
1081 *errp = err;
1082 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 struct buffer_head *bh;
1084 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Wang Shilong1b7d76e2013-01-12 01:36:17 -08001085 if (unlikely(!bh)) {
Wang Shilongc04e88e2013-01-15 21:20:01 -08001086 *errp = -ENOMEM;
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001087 goto err;
1088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (buffer_new(&dummy)) {
1090 J_ASSERT(create != 0);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001091 J_ASSERT(handle != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Andrew Mortond6859bf2006-03-26 01:38:03 -08001093 /*
1094 * Now that we do not always journal data, we should
1095 * keep in mind whether this should always journal the
1096 * new buffer as metadata. For now, regular file
1097 * writes use ext3_get_block instead, so it's not a
1098 * problem.
1099 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 lock_buffer(bh);
1101 BUFFER_TRACE(bh, "call get_create_access");
1102 fatal = ext3_journal_get_create_access(handle, bh);
1103 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001104 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 set_buffer_uptodate(bh);
1106 }
1107 unlock_buffer(bh);
1108 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1109 err = ext3_journal_dirty_metadata(handle, bh);
1110 if (!fatal)
1111 fatal = err;
1112 } else {
1113 BUFFER_TRACE(bh, "not a new buffer");
1114 }
1115 if (fatal) {
1116 *errp = fatal;
1117 brelse(bh);
1118 bh = NULL;
1119 }
1120 return bh;
1121 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001122err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 return NULL;
1124}
1125
Andrew Mortond6859bf2006-03-26 01:38:03 -08001126struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 int block, int create, int *err)
1128{
1129 struct buffer_head * bh;
1130
1131 bh = ext3_getblk(handle, inode, block, create, err);
1132 if (!bh)
1133 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001134 if (bh_uptodate_or_lock(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001136 get_bh(bh);
1137 bh->b_end_io = end_buffer_read_sync;
1138 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 wait_on_buffer(bh);
1140 if (buffer_uptodate(bh))
1141 return bh;
1142 put_bh(bh);
1143 *err = -EIO;
1144 return NULL;
1145}
1146
1147static int walk_page_buffers( handle_t *handle,
1148 struct buffer_head *head,
1149 unsigned from,
1150 unsigned to,
1151 int *partial,
1152 int (*fn)( handle_t *handle,
1153 struct buffer_head *bh))
1154{
1155 struct buffer_head *bh;
1156 unsigned block_start, block_end;
1157 unsigned blocksize = head->b_size;
1158 int err, ret = 0;
1159 struct buffer_head *next;
1160
1161 for ( bh = head, block_start = 0;
1162 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001163 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 {
1165 next = bh->b_this_page;
1166 block_end = block_start + blocksize;
1167 if (block_end <= from || block_start >= to) {
1168 if (partial && !buffer_uptodate(bh))
1169 *partial = 1;
1170 continue;
1171 }
1172 err = (*fn)(handle, bh);
1173 if (!ret)
1174 ret = err;
1175 }
1176 return ret;
1177}
1178
1179/*
1180 * To preserve ordering, it is essential that the hole instantiation and
1181 * the data write be encapsulated in a single transaction. We cannot
1182 * close off a transaction and start a new one between the ext3_get_block()
1183 * and the commit_write(). So doing the journal_start at the start of
1184 * prepare_write() is the right place.
1185 *
1186 * Also, this function can nest inside ext3_writepage() ->
1187 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1188 * has generated enough buffer credits to do the whole page. So we won't
1189 * block on the journal in that case, which is good, because the caller may
1190 * be PF_MEMALLOC.
1191 *
1192 * By accident, ext3 can be reentered when a transaction is open via
1193 * quota file writes. If we were to commit the transaction while thus
1194 * reentered, there can be a deadlock - we would be holding a quota
1195 * lock, and the commit would never complete if another thread had a
1196 * transaction open and was blocking on the quota lock - a ranking
1197 * violation.
1198 *
1199 * So what we do is to rely on the fact that journal_stop/journal_start
1200 * will _not_ run commit under these circumstances because handle->h_ref
1201 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001202 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001204static int do_journal_get_write_access(handle_t *handle,
1205 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
Jan Kara5f11e6a2010-08-05 12:38:26 +02001207 int dirty = buffer_dirty(bh);
1208 int ret;
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (!buffer_mapped(bh) || buffer_freed(bh))
1211 return 0;
Jan Kara5f11e6a2010-08-05 12:38:26 +02001212 /*
1213 * __block_prepare_write() could have dirtied some buffers. Clean
1214 * the dirty bit as jbd2_journal_get_write_access() could complain
1215 * otherwise about fs integrity issues. Setting of the dirty bit
1216 * by __block_prepare_write() isn't a real problem here as we clear
1217 * the bit before releasing a page lock and thus writeback cannot
1218 * ever write the buffer.
1219 */
1220 if (dirty)
1221 clear_buffer_dirty(bh);
1222 ret = ext3_journal_get_write_access(handle, bh);
1223 if (!ret && dirty)
1224 ret = ext3_journal_dirty_metadata(handle, bh);
1225 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
1227
Jan Kara68eb3db2009-12-01 16:53:06 +01001228/*
1229 * Truncate blocks that were not used by write. We have to truncate the
1230 * pagecache as well so that corresponding buffers get properly unmapped.
1231 */
1232static void ext3_truncate_failed_write(struct inode *inode)
1233{
1234 truncate_inode_pages(inode->i_mapping, inode->i_size);
1235 ext3_truncate(inode);
1236}
1237
Jan Karaee3e77f2011-06-03 21:58:11 +02001238/*
1239 * Truncate blocks that were not used by direct IO write. We have to zero out
1240 * the last file block as well because direct IO might have written to it.
1241 */
1242static void ext3_truncate_failed_direct_write(struct inode *inode)
1243{
1244 ext3_block_truncate_page(inode, inode->i_size);
1245 ext3_truncate(inode);
1246}
1247
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001248static int ext3_write_begin(struct file *file, struct address_space *mapping,
1249 loff_t pos, unsigned len, unsigned flags,
1250 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001252 struct inode *inode = mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001253 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 handle_t *handle;
1255 int retries = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001256 struct page *page;
1257 pgoff_t index;
1258 unsigned from, to;
Jan Kara695f6ae2009-04-02 16:57:17 -07001259 /* Reserve one block more for addition to orphan list in case
1260 * we allocate blocks but write fails for some reason */
1261 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001262
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001263 trace_ext3_write_begin(inode, pos, len, flags);
1264
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001265 index = pos >> PAGE_CACHE_SHIFT;
1266 from = pos & (PAGE_CACHE_SIZE - 1);
1267 to = from + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
1269retry:
Nick Piggin54566b22009-01-04 12:00:53 -08001270 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001271 if (!page)
1272 return -ENOMEM;
1273 *pagep = page;
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 handle = ext3_journal_start(inode, needed_blocks);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001276 if (IS_ERR(handle)) {
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001277 unlock_page(page);
1278 page_cache_release(page);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001279 ret = PTR_ERR(handle);
1280 goto out;
1281 }
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001282 ret = __block_write_begin(page, pos, len, ext3_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (ret)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001284 goto write_begin_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286 if (ext3_should_journal_data(inode)) {
1287 ret = walk_page_buffers(handle, page_buffers(page),
1288 from, to, NULL, do_journal_get_write_access);
1289 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001290write_begin_failed:
1291 if (ret) {
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001292 /*
1293 * block_write_begin may have instantiated a few blocks
1294 * outside i_size. Trim these off again. Don't need
1295 * i_size_read because we hold i_mutex.
Jan Kara695f6ae2009-04-02 16:57:17 -07001296 *
1297 * Add inode to orphan list in case we crash before truncate
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001298 * finishes. Do this only if ext3_can_truncate() agrees so
1299 * that orphan processing code is happy.
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001300 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001301 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001302 ext3_orphan_add(handle, inode);
1303 ext3_journal_stop(handle);
1304 unlock_page(page);
1305 page_cache_release(page);
1306 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001307 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1310 goto retry;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001311out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 return ret;
1313}
1314
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001315
Andrew Mortond6859bf2006-03-26 01:38:03 -08001316int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317{
1318 int err = journal_dirty_data(handle, bh);
1319 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001320 ext3_journal_abort_handle(__func__, __func__,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001321 bh, handle, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 return err;
1323}
1324
Jan Kara695f6ae2009-04-02 16:57:17 -07001325/* For ordered writepage and write_end functions */
1326static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1327{
1328 /*
1329 * Write could have mapped the buffer but it didn't copy the data in
1330 * yet. So avoid filing such buffer into a transaction.
1331 */
1332 if (buffer_mapped(bh) && buffer_uptodate(bh))
1333 return ext3_journal_dirty_data(handle, bh);
1334 return 0;
1335}
1336
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001337/* For write_end() in data=journal mode */
1338static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339{
1340 if (!buffer_mapped(bh) || buffer_freed(bh))
1341 return 0;
1342 set_buffer_uptodate(bh);
1343 return ext3_journal_dirty_metadata(handle, bh);
1344}
1345
1346/*
Jan Kara695f6ae2009-04-02 16:57:17 -07001347 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1348 * for the whole page but later we failed to copy the data in. Update inode
1349 * size according to what we managed to copy. The rest is going to be
1350 * truncated in write_end function.
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001351 */
Jan Kara695f6ae2009-04-02 16:57:17 -07001352static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001353{
Jan Kara695f6ae2009-04-02 16:57:17 -07001354 /* What matters to us is i_disksize. We don't write i_size anywhere */
1355 if (pos + copied > inode->i_size)
1356 i_size_write(inode, pos + copied);
1357 if (pos + copied > EXT3_I(inode)->i_disksize) {
1358 EXT3_I(inode)->i_disksize = pos + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001359 mark_inode_dirty(inode);
1360 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001361}
1362
1363/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 * We need to pick up the new inode size which generic_commit_write gave us
1365 * `file' can be NULL - eg, when called from page_symlink().
1366 *
1367 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1368 * buffers are managed internally.
1369 */
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001370static int ext3_ordered_write_end(struct file *file,
1371 struct address_space *mapping,
1372 loff_t pos, unsigned len, unsigned copied,
1373 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
1375 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001376 struct inode *inode = file->f_mapping->host;
1377 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 int ret = 0, ret2;
1379
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001380 trace_ext3_ordered_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001381 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1382
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001383 from = pos & (PAGE_CACHE_SIZE - 1);
Jan Kara695f6ae2009-04-02 16:57:17 -07001384 to = from + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 ret = walk_page_buffers(handle, page_buffers(page),
Jan Kara695f6ae2009-04-02 16:57:17 -07001386 from, to, NULL, journal_dirty_data_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Jan Kara695f6ae2009-04-02 16:57:17 -07001388 if (ret == 0)
1389 update_file_sizes(inode, pos, copied);
1390 /*
1391 * There may be allocated blocks outside of i_size because
1392 * we failed to copy some data. Prepare for truncate.
1393 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001394 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001395 ext3_orphan_add(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 ret2 = ext3_journal_stop(handle);
1397 if (!ret)
1398 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001399 unlock_page(page);
1400 page_cache_release(page);
1401
Jan Kara695f6ae2009-04-02 16:57:17 -07001402 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001403 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001404 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405}
1406
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001407static int ext3_writeback_write_end(struct file *file,
1408 struct address_space *mapping,
1409 loff_t pos, unsigned len, unsigned copied,
1410 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
1412 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001413 struct inode *inode = file->f_mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001414 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001416 trace_ext3_writeback_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001417 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1418 update_file_sizes(inode, pos, copied);
1419 /*
1420 * There may be allocated blocks outside of i_size because
1421 * we failed to copy some data. Prepare for truncate.
1422 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001423 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001424 ext3_orphan_add(handle, inode);
1425 ret = ext3_journal_stop(handle);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001426 unlock_page(page);
1427 page_cache_release(page);
1428
Jan Kara695f6ae2009-04-02 16:57:17 -07001429 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001430 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001431 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
1433
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001434static int ext3_journalled_write_end(struct file *file,
1435 struct address_space *mapping,
1436 loff_t pos, unsigned len, unsigned copied,
1437 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438{
1439 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001440 struct inode *inode = mapping->host;
Jan Karab22570d2011-07-23 01:21:38 +02001441 struct ext3_inode_info *ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 int ret = 0, ret2;
1443 int partial = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001444 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001446 trace_ext3_journalled_write_end(inode, pos, len, copied);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001447 from = pos & (PAGE_CACHE_SIZE - 1);
1448 to = from + len;
1449
1450 if (copied < len) {
1451 if (!PageUptodate(page))
1452 copied = 0;
Jan Kara695f6ae2009-04-02 16:57:17 -07001453 page_zero_new_buffers(page, from + copied, to);
1454 to = from + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001458 to, &partial, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 if (!partial)
1460 SetPageUptodate(page);
Jan Kara695f6ae2009-04-02 16:57:17 -07001461
1462 if (pos + copied > inode->i_size)
1463 i_size_write(inode, pos + copied);
1464 /*
1465 * There may be allocated blocks outside of i_size because
1466 * we failed to copy some data. Prepare for truncate.
1467 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001468 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001469 ext3_orphan_add(handle, inode);
Jan Kara9df93932010-01-06 21:58:48 +01001470 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001471 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1472 if (inode->i_size > ei->i_disksize) {
1473 ei->i_disksize = inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001475 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 ret = ret2;
1477 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 ret2 = ext3_journal_stop(handle);
1480 if (!ret)
1481 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001482 unlock_page(page);
1483 page_cache_release(page);
1484
Jan Kara695f6ae2009-04-02 16:57:17 -07001485 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001486 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001487 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488}
1489
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001490/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 * bmap() is special. It gets used by applications such as lilo and by
1492 * the swapper to find the on-disk block of a specific piece of data.
1493 *
1494 * Naturally, this is dangerous if the block concerned is still in the
1495 * journal. If somebody makes a swapfile on an ext3 data-journaling
1496 * filesystem and enables swap, then they may get a nasty shock when the
1497 * data getting swapped to that swapfile suddenly gets overwritten by
1498 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001499 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 *
1501 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001502 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 */
1504static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1505{
1506 struct inode *inode = mapping->host;
1507 journal_t *journal;
1508 int err;
1509
Jan Kara9df93932010-01-06 21:58:48 +01001510 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001511 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 * This is a REALLY heavyweight approach, but the use of
1513 * bmap on dirty files is expected to be extremely rare:
1514 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001515 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 *
1517 * (bmap requires CAP_SYS_RAWIO so this does not
1518 * represent an unprivileged user DOS attack --- we'd be
1519 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001520 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
1522 * NB. EXT3_STATE_JDATA is not set on files other than
1523 * regular files. If somebody wants to bmap a directory
1524 * or symlink and gets confused because the buffer
1525 * hasn't yet been flushed to disk, they deserve
1526 * everything they get.
1527 */
1528
Jan Kara9df93932010-01-06 21:58:48 +01001529 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 journal = EXT3_JOURNAL(inode);
1531 journal_lock_updates(journal);
1532 err = journal_flush(journal);
1533 journal_unlock_updates(journal);
1534
1535 if (err)
1536 return 0;
1537 }
1538
1539 return generic_block_bmap(mapping,block,ext3_get_block);
1540}
1541
1542static int bget_one(handle_t *handle, struct buffer_head *bh)
1543{
1544 get_bh(bh);
1545 return 0;
1546}
1547
1548static int bput_one(handle_t *handle, struct buffer_head *bh)
1549{
1550 put_bh(bh);
1551 return 0;
1552}
1553
Jan Kara9e80d402009-03-26 13:08:04 +01001554static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1555{
1556 return !buffer_mapped(bh);
1557}
Jan Kara695f6ae2009-04-02 16:57:17 -07001558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559/*
1560 * Note that we always start a transaction even if we're not journalling
1561 * data. This is to preserve ordering: any hole instantiation within
1562 * __block_write_full_page -> ext3_get_block() should be journalled
1563 * along with the data so we don't crash and then get metadata which
1564 * refers to old data.
1565 *
1566 * In all journalling modes block_write_full_page() will start the I/O.
1567 *
1568 * Problem:
1569 *
1570 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1571 * ext3_writepage()
1572 *
1573 * Similar for:
1574 *
1575 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1576 *
1577 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001578 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 *
1580 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1581 * allocations fail.
1582 *
1583 * 16May01: If we're reentered then journal_current_handle() will be
1584 * non-zero. We simply *return*.
1585 *
1586 * 1 July 2001: @@@ FIXME:
1587 * In journalled data mode, a data buffer may be metadata against the
1588 * current transaction. But the same file is part of a shared mapping
1589 * and someone does a writepage() on it.
1590 *
1591 * We will move the buffer onto the async_data list, but *after* it has
1592 * been dirtied. So there's a small window where we have dirty data on
1593 * BJ_Metadata.
1594 *
1595 * Note that this only applies to the last partial page in the file. The
1596 * bit which block_write_full_page() uses prepare/commit for. (That's
1597 * broken code anyway: it's wrong for msync()).
1598 *
1599 * It's a rare case: affects the final partial page, for journalled data
1600 * where the file is subject to bith write() and writepage() in the same
1601 * transction. To fix it we'll need a custom block_write_full_page().
1602 * We'll probably need that anyway for journalling writepage() output.
1603 *
1604 * We don't honour synchronous mounts for writepage(). That would be
1605 * disastrous. Any write() or metadata operation will sync the fs for
1606 * us.
1607 *
1608 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1609 * we don't need to open a transaction here.
1610 */
1611static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001612 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
1614 struct inode *inode = page->mapping->host;
1615 struct buffer_head *page_bufs;
1616 handle_t *handle = NULL;
1617 int ret = 0;
1618 int err;
1619
1620 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001621 /*
1622 * We don't want to warn for emergency remount. The condition is
1623 * ordered to avoid dereferencing inode->i_sb in non-error case to
1624 * avoid slow-downs.
1625 */
1626 WARN_ON_ONCE(IS_RDONLY(inode) &&
1627 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 /*
1630 * We give up here if we're reentered, because it might be for a
1631 * different filesystem.
1632 */
1633 if (ext3_journal_current_handle())
1634 goto out_fail;
1635
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001636 trace_ext3_ordered_writepage(page);
Jan Kara9e80d402009-03-26 13:08:04 +01001637 if (!page_has_buffers(page)) {
1638 create_empty_buffers(page, inode->i_sb->s_blocksize,
1639 (1 << BH_Dirty)|(1 << BH_Uptodate));
Jan Kara430db322009-04-07 18:25:01 -04001640 page_bufs = page_buffers(page);
1641 } else {
1642 page_bufs = page_buffers(page);
1643 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1644 NULL, buffer_unmapped)) {
1645 /* Provide NULL get_block() to catch bugs if buffers
1646 * weren't really mapped */
1647 return block_write_full_page(page, NULL, wbc);
1648 }
Jan Kara9e80d402009-03-26 13:08:04 +01001649 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1651
1652 if (IS_ERR(handle)) {
1653 ret = PTR_ERR(handle);
1654 goto out_fail;
1655 }
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 walk_page_buffers(handle, page_bufs, 0,
1658 PAGE_CACHE_SIZE, NULL, bget_one);
1659
1660 ret = block_write_full_page(page, ext3_get_block, wbc);
1661
1662 /*
1663 * The page can become unlocked at any point now, and
1664 * truncate can then come in and change things. So we
1665 * can't touch *page from now on. But *page_bufs is
1666 * safe due to elevated refcount.
1667 */
1668
1669 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001670 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 * block_write_full_page() succeeded. Otherwise they are unmapped,
1672 * and generally junk.
1673 */
1674 if (ret == 0) {
1675 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1676 NULL, journal_dirty_data_fn);
1677 if (!ret)
1678 ret = err;
1679 }
1680 walk_page_buffers(handle, page_bufs, 0,
1681 PAGE_CACHE_SIZE, NULL, bput_one);
1682 err = ext3_journal_stop(handle);
1683 if (!ret)
1684 ret = err;
1685 return ret;
1686
1687out_fail:
1688 redirty_page_for_writepage(wbc, page);
1689 unlock_page(page);
1690 return ret;
1691}
1692
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693static int ext3_writeback_writepage(struct page *page,
1694 struct writeback_control *wbc)
1695{
1696 struct inode *inode = page->mapping->host;
1697 handle_t *handle = NULL;
1698 int ret = 0;
1699 int err;
1700
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001701 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001702 /*
1703 * We don't want to warn for emergency remount. The condition is
1704 * ordered to avoid dereferencing inode->i_sb in non-error case to
1705 * avoid slow-downs.
1706 */
1707 WARN_ON_ONCE(IS_RDONLY(inode) &&
1708 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 if (ext3_journal_current_handle())
1711 goto out_fail;
1712
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001713 trace_ext3_writeback_writepage(page);
Jan Kara430db322009-04-07 18:25:01 -04001714 if (page_has_buffers(page)) {
1715 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1716 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1717 /* Provide NULL get_block() to catch bugs if buffers
1718 * weren't really mapped */
1719 return block_write_full_page(page, NULL, wbc);
1720 }
1721 }
1722
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1724 if (IS_ERR(handle)) {
1725 ret = PTR_ERR(handle);
1726 goto out_fail;
1727 }
1728
Christoph Hellwig4c4d3902010-06-07 10:20:39 +02001729 ret = block_write_full_page(page, ext3_get_block, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
1731 err = ext3_journal_stop(handle);
1732 if (!ret)
1733 ret = err;
1734 return ret;
1735
1736out_fail:
1737 redirty_page_for_writepage(wbc, page);
1738 unlock_page(page);
1739 return ret;
1740}
1741
1742static int ext3_journalled_writepage(struct page *page,
1743 struct writeback_control *wbc)
1744{
1745 struct inode *inode = page->mapping->host;
1746 handle_t *handle = NULL;
1747 int ret = 0;
1748 int err;
1749
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001750 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001751 /*
1752 * We don't want to warn for emergency remount. The condition is
1753 * ordered to avoid dereferencing inode->i_sb in non-error case to
1754 * avoid slow-downs.
1755 */
1756 WARN_ON_ONCE(IS_RDONLY(inode) &&
1757 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (ext3_journal_current_handle())
1760 goto no_write;
1761
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001762 trace_ext3_journalled_writepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1764 if (IS_ERR(handle)) {
1765 ret = PTR_ERR(handle);
1766 goto no_write;
1767 }
1768
1769 if (!page_has_buffers(page) || PageChecked(page)) {
1770 /*
1771 * It's mmapped pagecache. Add buffers and journal it. There
1772 * doesn't seem much point in redirtying the page here.
1773 */
1774 ClearPageChecked(page);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001775 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1776 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001777 if (ret != 0) {
1778 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 ret = walk_page_buffers(handle, page_buffers(page), 0,
1782 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1783
1784 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001785 PAGE_CACHE_SIZE, NULL, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 if (ret == 0)
1787 ret = err;
Jan Kara9df93932010-01-06 21:58:48 +01001788 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001789 atomic_set(&EXT3_I(inode)->i_datasync_tid,
1790 handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 unlock_page(page);
1792 } else {
1793 /*
1794 * It may be a page full of checkpoint-mode buffers. We don't
1795 * really know unless we go poke around in the buffer_heads.
1796 * But block_write_full_page will do the right thing.
1797 */
1798 ret = block_write_full_page(page, ext3_get_block, wbc);
1799 }
1800 err = ext3_journal_stop(handle);
1801 if (!ret)
1802 ret = err;
1803out:
1804 return ret;
1805
1806no_write:
1807 redirty_page_for_writepage(wbc, page);
1808out_unlock:
1809 unlock_page(page);
1810 goto out;
1811}
1812
1813static int ext3_readpage(struct file *file, struct page *page)
1814{
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001815 trace_ext3_readpage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 return mpage_readpage(page, ext3_get_block);
1817}
1818
1819static int
1820ext3_readpages(struct file *file, struct address_space *mapping,
1821 struct list_head *pages, unsigned nr_pages)
1822{
1823 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1824}
1825
NeilBrown2ff28e22006-03-26 01:37:18 -08001826static void ext3_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827{
1828 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1829
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001830 trace_ext3_invalidatepage(page, offset);
1831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 /*
1833 * If it's a full truncate we just forget about the pending dirtying
1834 */
1835 if (offset == 0)
1836 ClearPageChecked(page);
1837
NeilBrown2ff28e22006-03-26 01:37:18 -08001838 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839}
1840
Al Viro27496a82005-10-21 03:20:48 -04001841static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
1843 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1844
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001845 trace_ext3_releasepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 WARN_ON(PageChecked(page));
1847 if (!page_has_buffers(page))
1848 return 0;
1849 return journal_try_to_free_buffers(journal, page, wait);
1850}
1851
1852/*
1853 * If the O_DIRECT write will extend the file then add this inode to the
1854 * orphan list. So recovery will truncate it back to the original size
1855 * if the machine crashes during the write.
1856 *
1857 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Karabd1939d2008-02-06 01:40:21 -08001858 * crashes then stale disk data _may_ be exposed inside the file. But current
1859 * VFS code falls back into buffered path in that case so we are safe.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 */
1861static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1862 const struct iovec *iov, loff_t offset,
1863 unsigned long nr_segs)
1864{
1865 struct file *file = iocb->ki_filp;
1866 struct inode *inode = file->f_mapping->host;
1867 struct ext3_inode_info *ei = EXT3_I(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001868 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 ssize_t ret;
1870 int orphan = 0;
1871 size_t count = iov_length(iov, nr_segs);
Eric Sandeenea0174a2009-10-12 21:34:27 -05001872 int retries = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001874 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 if (rw == WRITE) {
1877 loff_t final_size = offset + count;
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 if (final_size > inode->i_size) {
Jan Karabd1939d2008-02-06 01:40:21 -08001880 /* Credits for sb + inode write */
1881 handle = ext3_journal_start(inode, 2);
1882 if (IS_ERR(handle)) {
1883 ret = PTR_ERR(handle);
1884 goto out;
1885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 ret = ext3_orphan_add(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001887 if (ret) {
1888 ext3_journal_stop(handle);
1889 goto out;
1890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 orphan = 1;
1892 ei->i_disksize = inode->i_size;
Jan Karabd1939d2008-02-06 01:40:21 -08001893 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
1895 }
1896
Eric Sandeenea0174a2009-10-12 21:34:27 -05001897retry:
Christoph Hellwigaacfc19c2011-06-24 14:29:47 -04001898 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1899 ext3_get_block);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001900 /*
1901 * In case of error extending write may have instantiated a few
1902 * blocks outside i_size. Trim these off again.
1903 */
1904 if (unlikely((rw & WRITE) && ret < 0)) {
1905 loff_t isize = i_size_read(inode);
1906 loff_t end = offset + iov_length(iov, nr_segs);
1907
1908 if (end > isize)
Jan Karaee3e77f2011-06-03 21:58:11 +02001909 ext3_truncate_failed_direct_write(inode);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001910 }
Eric Sandeenea0174a2009-10-12 21:34:27 -05001911 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1912 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
Jan Karabd1939d2008-02-06 01:40:21 -08001914 if (orphan) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 int err;
1916
Jan Karabd1939d2008-02-06 01:40:21 -08001917 /* Credits for sb + inode write */
1918 handle = ext3_journal_start(inode, 2);
1919 if (IS_ERR(handle)) {
1920 /* This is really bad luck. We've written the data
Jan Kara7eb49692010-03-01 14:02:37 +01001921 * but cannot extend i_size. Truncate allocated blocks
1922 * and pretend the write failed... */
Jan Karaee3e77f2011-06-03 21:58:11 +02001923 ext3_truncate_failed_direct_write(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001924 ret = PTR_ERR(handle);
1925 goto out;
1926 }
1927 if (inode->i_nlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 ext3_orphan_del(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001929 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 loff_t end = offset + ret;
1931 if (end > inode->i_size) {
1932 ei->i_disksize = end;
1933 i_size_write(inode, end);
1934 /*
1935 * We're going to return a positive `ret'
1936 * here due to non-zero-length I/O, so there's
1937 * no way of reporting error returns from
1938 * ext3_mark_inode_dirty() to userspace. So
1939 * ignore it.
1940 */
1941 ext3_mark_inode_dirty(handle, inode);
1942 }
1943 }
1944 err = ext3_journal_stop(handle);
1945 if (ret == 0)
1946 ret = err;
1947 }
1948out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001949 trace_ext3_direct_IO_exit(inode, offset,
1950 iov_length(iov, nr_segs), rw, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 return ret;
1952}
1953
1954/*
1955 * Pages can be marked dirty completely asynchronously from ext3's journalling
1956 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1957 * much here because ->set_page_dirty is called under VFS locks. The page is
1958 * not necessarily locked.
1959 *
1960 * We cannot just dirty the page and leave attached buffers clean, because the
1961 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1962 * or jbddirty because all the journalling code will explode.
1963 *
1964 * So what we do is to mark the page "pending dirty" and next time writepage
1965 * is called, propagate that into the buffers appropriately.
1966 */
1967static int ext3_journalled_set_page_dirty(struct page *page)
1968{
1969 SetPageChecked(page);
1970 return __set_page_dirty_nobuffers(page);
1971}
1972
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001973static const struct address_space_operations ext3_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001974 .readpage = ext3_readpage,
1975 .readpages = ext3_readpages,
1976 .writepage = ext3_ordered_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001977 .write_begin = ext3_write_begin,
1978 .write_end = ext3_ordered_write_end,
1979 .bmap = ext3_bmap,
1980 .invalidatepage = ext3_invalidatepage,
1981 .releasepage = ext3_releasepage,
1982 .direct_IO = ext3_direct_IO,
1983 .migratepage = buffer_migrate_page,
1984 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001985 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986};
1987
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001988static const struct address_space_operations ext3_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001989 .readpage = ext3_readpage,
1990 .readpages = ext3_readpages,
1991 .writepage = ext3_writeback_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001992 .write_begin = ext3_write_begin,
1993 .write_end = ext3_writeback_write_end,
1994 .bmap = ext3_bmap,
1995 .invalidatepage = ext3_invalidatepage,
1996 .releasepage = ext3_releasepage,
1997 .direct_IO = ext3_direct_IO,
1998 .migratepage = buffer_migrate_page,
1999 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002000 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001};
2002
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002003static const struct address_space_operations ext3_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002004 .readpage = ext3_readpage,
2005 .readpages = ext3_readpages,
2006 .writepage = ext3_journalled_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002007 .write_begin = ext3_write_begin,
2008 .write_end = ext3_journalled_write_end,
2009 .set_page_dirty = ext3_journalled_set_page_dirty,
2010 .bmap = ext3_bmap,
2011 .invalidatepage = ext3_invalidatepage,
2012 .releasepage = ext3_releasepage,
2013 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002014 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015};
2016
2017void ext3_set_aops(struct inode *inode)
2018{
2019 if (ext3_should_order_data(inode))
2020 inode->i_mapping->a_ops = &ext3_ordered_aops;
2021 else if (ext3_should_writeback_data(inode))
2022 inode->i_mapping->a_ops = &ext3_writeback_aops;
2023 else
2024 inode->i_mapping->a_ops = &ext3_journalled_aops;
2025}
2026
2027/*
2028 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2029 * up to the end of the block which corresponds to `from'.
2030 * This required during truncate. We need to physically zero the tail end
2031 * of that block so it doesn't yield old data if the file is later grown.
2032 */
Jan Karaee3e77f2011-06-03 21:58:11 +02002033static int ext3_block_truncate_page(struct inode *inode, loff_t from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
Mingming Cao43d23f92006-06-25 05:48:07 -07002035 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Jan Karaee3e77f2011-06-03 21:58:11 +02002036 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 unsigned blocksize, iblock, length, pos;
Jan Karaee3e77f2011-06-03 21:58:11 +02002038 struct page *page;
2039 handle_t *handle = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 struct buffer_head *bh;
2041 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Jan Karaee3e77f2011-06-03 21:58:11 +02002043 /* Truncated on block boundary - nothing to do */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 blocksize = inode->i_sb->s_blocksize;
Jan Karaee3e77f2011-06-03 21:58:11 +02002045 if ((from & (blocksize - 1)) == 0)
2046 return 0;
2047
2048 page = grab_cache_page(inode->i_mapping, index);
2049 if (!page)
2050 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 length = blocksize - (offset & (blocksize - 1));
2052 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 if (!page_has_buffers(page))
2055 create_empty_buffers(page, blocksize, 0);
2056
2057 /* Find the buffer that contains "offset" */
2058 bh = page_buffers(page);
2059 pos = blocksize;
2060 while (offset >= pos) {
2061 bh = bh->b_this_page;
2062 iblock++;
2063 pos += blocksize;
2064 }
2065
2066 err = 0;
2067 if (buffer_freed(bh)) {
2068 BUFFER_TRACE(bh, "freed: skip");
2069 goto unlock;
2070 }
2071
2072 if (!buffer_mapped(bh)) {
2073 BUFFER_TRACE(bh, "unmapped");
2074 ext3_get_block(inode, iblock, bh, 0);
2075 /* unmapped? It's a hole - nothing to do */
2076 if (!buffer_mapped(bh)) {
2077 BUFFER_TRACE(bh, "still unmapped");
2078 goto unlock;
2079 }
2080 }
2081
2082 /* Ok, it's mapped. Make sure it's up-to-date */
2083 if (PageUptodate(page))
2084 set_buffer_uptodate(bh);
2085
Zheng Liud03e1292011-12-05 15:55:11 +08002086 if (!bh_uptodate_or_lock(bh)) {
2087 err = bh_submit_read(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 /* Uhhuh. Read error. Complain and punt. */
Zheng Liud03e1292011-12-05 15:55:11 +08002089 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 goto unlock;
2091 }
2092
Jan Karaee3e77f2011-06-03 21:58:11 +02002093 /* data=writeback mode doesn't need transaction to zero-out data */
2094 if (!ext3_should_writeback_data(inode)) {
2095 /* We journal at most one block */
2096 handle = ext3_journal_start(inode, 1);
2097 if (IS_ERR(handle)) {
2098 clear_highpage(page);
2099 flush_dcache_page(page);
2100 err = PTR_ERR(handle);
2101 goto unlock;
2102 }
2103 }
2104
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (ext3_should_journal_data(inode)) {
2106 BUFFER_TRACE(bh, "get write access");
2107 err = ext3_journal_get_write_access(handle, bh);
2108 if (err)
Jan Karaee3e77f2011-06-03 21:58:11 +02002109 goto stop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
2111
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002112 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 BUFFER_TRACE(bh, "zeroed end of block");
2114
2115 err = 0;
2116 if (ext3_should_journal_data(inode)) {
2117 err = ext3_journal_dirty_metadata(handle, bh);
2118 } else {
2119 if (ext3_should_order_data(inode))
2120 err = ext3_journal_dirty_data(handle, bh);
2121 mark_buffer_dirty(bh);
2122 }
Jan Karaee3e77f2011-06-03 21:58:11 +02002123stop:
2124 if (handle)
2125 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
2127unlock:
2128 unlock_page(page);
2129 page_cache_release(page);
2130 return err;
2131}
2132
2133/*
2134 * Probably it should be a library function... search for first non-zero word
2135 * or memcmp with zero_page, whatever is better for particular architecture.
2136 * Linus?
2137 */
2138static inline int all_zeroes(__le32 *p, __le32 *q)
2139{
2140 while (p < q)
2141 if (*p++)
2142 return 0;
2143 return 1;
2144}
2145
2146/**
2147 * ext3_find_shared - find the indirect blocks for partial truncation.
2148 * @inode: inode in question
2149 * @depth: depth of the affected branch
2150 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2151 * @chain: place to store the pointers to partial indirect blocks
2152 * @top: place to the (detached) top of branch
2153 *
2154 * This is a helper function used by ext3_truncate().
2155 *
2156 * When we do truncate() we may have to clean the ends of several
2157 * indirect blocks but leave the blocks themselves alive. Block is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002158 * partially truncated if some data below the new i_size is referred
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 * from it (and it is on the path to the first completely truncated
2160 * data block, indeed). We have to free the top of that path along
2161 * with everything to the right of the path. Since no allocation
2162 * past the truncation point is possible until ext3_truncate()
2163 * finishes, we may safely do the latter, but top of branch may
2164 * require special attention - pageout below the truncation point
2165 * might try to populate it.
2166 *
2167 * We atomically detach the top of branch from the tree, store the
2168 * block number of its root in *@top, pointers to buffer_heads of
2169 * partially truncated blocks - in @chain[].bh and pointers to
2170 * their last elements that should not be removed - in
2171 * @chain[].p. Return value is the pointer to last filled element
2172 * of @chain.
2173 *
2174 * The work left to caller to do the actual freeing of subtrees:
2175 * a) free the subtree starting from *@top
2176 * b) free the subtrees whose roots are stored in
2177 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2178 * c) free the subtrees growing from the inode past the @chain[0].
2179 * (no partially truncated stuff there). */
2180
Andrew Mortond6859bf2006-03-26 01:38:03 -08002181static Indirect *ext3_find_shared(struct inode *inode, int depth,
2182 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183{
2184 Indirect *partial, *p;
2185 int k, err;
2186
2187 *top = 0;
Uwe Kleine-Königbf48aab2009-10-28 20:11:03 +01002188 /* Make k index the deepest non-null offset + 1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 for (k = depth; k > 1 && !offsets[k-1]; k--)
2190 ;
2191 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2192 /* Writer: pointers */
2193 if (!partial)
2194 partial = chain + k-1;
2195 /*
2196 * If the branch acquired continuation since we've looked at it -
2197 * fine, it should all survive and (new) top doesn't belong to us.
2198 */
2199 if (!partial->key && *partial->p)
2200 /* Writer: end */
2201 goto no_top;
2202 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2203 ;
2204 /*
2205 * OK, we've found the last block that must survive. The rest of our
2206 * branch should be detached before unlocking. However, if that rest
2207 * of branch is all ours and does not grow immediately from the inode
2208 * it's easier to cheat and just decrement partial->p.
2209 */
2210 if (p == chain + k - 1 && p > chain) {
2211 p->p--;
2212 } else {
2213 *top = *p->p;
2214 /* Nope, don't do this in ext3. Must leave the tree intact */
2215#if 0
2216 *p->p = 0;
2217#endif
2218 }
2219 /* Writer: end */
2220
Andrew Mortond6859bf2006-03-26 01:38:03 -08002221 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 brelse(partial->bh);
2223 partial--;
2224 }
2225no_top:
2226 return partial;
2227}
2228
2229/*
2230 * Zero a number of block pointers in either an inode or an indirect block.
2231 * If we restart the transaction we must again get write access to the
2232 * indirect block for further modification.
2233 *
2234 * We release `count' blocks on disk, but (last - first) may be greater
2235 * than `count' because there can be holes in there.
2236 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002237static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07002238 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08002239 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240{
2241 __le32 *p;
2242 if (try_to_extend_transaction(handle, inode)) {
2243 if (bh) {
2244 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Namhyung Kim156e7432010-11-25 01:53:13 +09002245 if (ext3_journal_dirty_metadata(handle, bh))
2246 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 }
2248 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002249 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 if (bh) {
2251 BUFFER_TRACE(bh, "retaking write access");
Namhyung Kim156e7432010-11-25 01:53:13 +09002252 if (ext3_journal_get_write_access(handle, bh))
2253 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 }
2255 }
2256
2257 /*
2258 * Any buffers which are on the journal will be in memory. We find
2259 * them on the hash table so journal_revoke() will run journal_forget()
2260 * on them. We've already detached each block from the file, so
2261 * bforget() in journal_forget() should be safe.
2262 *
2263 * AKPM: turn on bforget in journal_forget()!!!
2264 */
2265 for (p = first; p < last; p++) {
2266 u32 nr = le32_to_cpu(*p);
2267 if (nr) {
2268 struct buffer_head *bh;
2269
2270 *p = 0;
2271 bh = sb_find_get_block(inode->i_sb, nr);
2272 ext3_forget(handle, 0, inode, bh, nr);
2273 }
2274 }
2275
2276 ext3_free_blocks(handle, inode, block_to_free, count);
2277}
2278
2279/**
2280 * ext3_free_data - free a list of data blocks
2281 * @handle: handle for this transaction
2282 * @inode: inode we are dealing with
2283 * @this_bh: indirect buffer_head which contains *@first and *@last
2284 * @first: array of block numbers
2285 * @last: points immediately past the end of array
2286 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002287 * We are freeing all blocks referred from that array (numbers are stored as
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2289 *
2290 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2291 * blocks are contiguous then releasing them at one time will only affect one
2292 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2293 * actually use a lot of journal space.
2294 *
2295 * @this_bh will be %NULL if @first and @last point into the inode's direct
2296 * block pointers.
2297 */
2298static void ext3_free_data(handle_t *handle, struct inode *inode,
2299 struct buffer_head *this_bh,
2300 __le32 *first, __le32 *last)
2301{
Mingming Cao43d23f92006-06-25 05:48:07 -07002302 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002303 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2305 corresponding to
2306 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002307 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 __le32 *p; /* Pointer into inode/ind
2309 for current block */
2310 int err;
2311
2312 if (this_bh) { /* For indirect block */
2313 BUFFER_TRACE(this_bh, "get_write_access");
2314 err = ext3_journal_get_write_access(handle, this_bh);
2315 /* Important: if we can't update the indirect pointers
2316 * to the blocks, we can't free them. */
2317 if (err)
2318 return;
2319 }
2320
2321 for (p = first; p < last; p++) {
2322 nr = le32_to_cpu(*p);
2323 if (nr) {
2324 /* accumulate blocks to free if they're contiguous */
2325 if (count == 0) {
2326 block_to_free = nr;
2327 block_to_free_p = p;
2328 count = 1;
2329 } else if (nr == block_to_free + count) {
2330 count++;
2331 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002332 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 block_to_free,
2334 count, block_to_free_p, p);
2335 block_to_free = nr;
2336 block_to_free_p = p;
2337 count = 1;
2338 }
2339 }
2340 }
2341
2342 if (count > 0)
2343 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2344 count, block_to_free_p, p);
2345
2346 if (this_bh) {
2347 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
Duane Griffin3ccc3162008-07-25 01:46:26 -07002348
2349 /*
2350 * The buffer head should have an attached journal head at this
2351 * point. However, if the data is corrupted and an indirect
2352 * block pointed to itself, it would have been detached when
2353 * the block was cleared. Check for this instead of OOPSing.
2354 */
2355 if (bh2jh(this_bh))
2356 ext3_journal_dirty_metadata(handle, this_bh);
2357 else
2358 ext3_error(inode->i_sb, "ext3_free_data",
2359 "circular indirect block detected, "
2360 "inode=%lu, block=%llu",
2361 inode->i_ino,
2362 (unsigned long long)this_bh->b_blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 }
2364}
2365
2366/**
2367 * ext3_free_branches - free an array of branches
2368 * @handle: JBD handle for this transaction
2369 * @inode: inode we are dealing with
2370 * @parent_bh: the buffer_head which contains *@first and *@last
2371 * @first: array of block numbers
2372 * @last: pointer immediately past the end of array
2373 * @depth: depth of the branches to free
2374 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002375 * We are freeing all blocks referred from these branches (numbers are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 * stored as little-endian 32-bit) and updating @inode->i_blocks
2377 * appropriately.
2378 */
2379static void ext3_free_branches(handle_t *handle, struct inode *inode,
2380 struct buffer_head *parent_bh,
2381 __le32 *first, __le32 *last, int depth)
2382{
Mingming Cao43d23f92006-06-25 05:48:07 -07002383 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 __le32 *p;
2385
2386 if (is_handle_aborted(handle))
2387 return;
2388
2389 if (depth--) {
2390 struct buffer_head *bh;
2391 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2392 p = last;
2393 while (--p >= first) {
2394 nr = le32_to_cpu(*p);
2395 if (!nr)
2396 continue; /* A hole */
2397
2398 /* Go read the buffer for the next level down */
2399 bh = sb_bread(inode->i_sb, nr);
2400
2401 /*
2402 * A read failure? Report error and clear slot
2403 * (should be rare).
2404 */
2405 if (!bh) {
2406 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002407 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 inode->i_ino, nr);
2409 continue;
2410 }
2411
2412 /* This zaps the entire block. Bottom up. */
2413 BUFFER_TRACE(bh, "free child branches");
2414 ext3_free_branches(handle, inode, bh,
2415 (__le32*)bh->b_data,
2416 (__le32*)bh->b_data + addr_per_block,
2417 depth);
2418
2419 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 * Everything below this this pointer has been
2421 * released. Now let this top-of-subtree go.
2422 *
2423 * We want the freeing of this indirect block to be
2424 * atomic in the journal with the updating of the
2425 * bitmap block which owns it. So make some room in
2426 * the journal.
2427 *
2428 * We zero the parent pointer *after* freeing its
2429 * pointee in the bitmaps, so if extend_transaction()
2430 * for some reason fails to put the bitmap changes and
2431 * the release into the same transaction, recovery
2432 * will merely complain about releasing a free block,
2433 * rather than leaking blocks.
2434 */
2435 if (is_handle_aborted(handle))
2436 return;
2437 if (try_to_extend_transaction(handle, inode)) {
2438 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002439 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 }
2441
Jan Karaf25f6242010-07-12 21:04:31 +02002442 /*
2443 * We've probably journalled the indirect block several
2444 * times during the truncate. But it's no longer
2445 * needed and we now drop it from the transaction via
2446 * journal_revoke().
2447 *
2448 * That's easy if it's exclusively part of this
2449 * transaction. But if it's part of the committing
2450 * transaction then journal_forget() will simply
2451 * brelse() it. That means that if the underlying
2452 * block is reallocated in ext3_get_block(),
2453 * unmap_underlying_metadata() will find this block
2454 * and will try to get rid of it. damn, damn. Thus
2455 * we don't allow a block to be reallocated until
2456 * a transaction freeing it has fully committed.
2457 *
2458 * We also have to make sure journal replay after a
2459 * crash does not overwrite non-journaled data blocks
2460 * with old metadata when the block got reallocated for
2461 * data. Thus we have to store a revoke record for a
2462 * block in the same transaction in which we free the
2463 * block.
2464 */
2465 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2466
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 ext3_free_blocks(handle, inode, nr, 1);
2468
2469 if (parent_bh) {
2470 /*
2471 * The block which we have just freed is
2472 * pointed to by an indirect block: journal it
2473 */
2474 BUFFER_TRACE(parent_bh, "get_write_access");
2475 if (!ext3_journal_get_write_access(handle,
2476 parent_bh)){
2477 *p = 0;
2478 BUFFER_TRACE(parent_bh,
2479 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002480 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 parent_bh);
2482 }
2483 }
2484 }
2485 } else {
2486 /* We have reached the bottom of the tree. */
2487 BUFFER_TRACE(parent_bh, "free data blocks");
2488 ext3_free_data(handle, inode, parent_bh, first, last);
2489 }
2490}
2491
Duane Griffinae76dd92008-07-25 01:46:23 -07002492int ext3_can_truncate(struct inode *inode)
2493{
Duane Griffinae76dd92008-07-25 01:46:23 -07002494 if (S_ISREG(inode->i_mode))
2495 return 1;
2496 if (S_ISDIR(inode->i_mode))
2497 return 1;
2498 if (S_ISLNK(inode->i_mode))
2499 return !ext3_inode_is_fast_symlink(inode);
2500 return 0;
2501}
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503/*
2504 * ext3_truncate()
2505 *
2506 * We block out ext3_get_block() block instantiations across the entire
2507 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2508 * simultaneously on behalf of the same inode.
2509 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002510 * As we work through the truncate and commit bits of it to the journal there
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 * is one core, guiding principle: the file's tree must always be consistent on
2512 * disk. We must be able to restart the truncate after a crash.
2513 *
2514 * The file's tree may be transiently inconsistent in memory (although it
2515 * probably isn't), but whenever we close off and commit a journal transaction,
2516 * the contents of (the filesystem + the journal) must be consistent and
2517 * restartable. It's pretty simple, really: bottom up, right to left (although
2518 * left-to-right works OK too).
2519 *
2520 * Note that at recovery time, journal replay occurs *before* the restart of
2521 * truncate against the orphan inode list.
2522 *
2523 * The committed inode has the new, desired i_size (which is the same as
2524 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2525 * that this inode's truncate did not complete and it will again call
2526 * ext3_truncate() to have another go. So there will be instantiated blocks
2527 * to the right of the truncation point in a crashed ext3 filesystem. But
2528 * that's fine - as long as they are linked from the inode, the post-crash
2529 * ext3_truncate() run will find them and release them.
2530 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002531void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532{
2533 handle_t *handle;
2534 struct ext3_inode_info *ei = EXT3_I(inode);
2535 __le32 *i_data = ei->i_data;
2536 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 int offsets[4];
2538 Indirect chain[4];
2539 Indirect *partial;
2540 __le32 nr = 0;
2541 int n;
2542 long last_block;
2543 unsigned blocksize = inode->i_sb->s_blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002545 trace_ext3_truncate_enter(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Duane Griffinae76dd92008-07-25 01:46:23 -07002547 if (!ext3_can_truncate(inode))
Jan Karaef436182009-06-17 16:26:24 -07002548 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002550 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
Jan Kara9df93932010-01-06 21:58:48 +01002551 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002552
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 handle = start_transaction(inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02002554 if (IS_ERR(handle))
Jan Karaef436182009-06-17 16:26:24 -07002555 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556
2557 last_block = (inode->i_size + blocksize-1)
2558 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2560 if (n == 0)
2561 goto out_stop; /* error */
2562
2563 /*
2564 * OK. This truncate is going to happen. We add the inode to the
2565 * orphan list, so that if this truncate spans multiple transactions,
2566 * and we crash, we will resume the truncate when the filesystem
2567 * recovers. It also marks the inode dirty, to catch the new size.
2568 *
2569 * Implication: the file must always be in a sane, consistent
2570 * truncatable state while each transaction commits.
2571 */
2572 if (ext3_orphan_add(handle, inode))
2573 goto out_stop;
2574
2575 /*
2576 * The orphan list entry will now protect us from any crash which
2577 * occurs before the truncate completes, so it is now safe to propagate
2578 * the new, shorter inode size (held for now in i_size) into the
2579 * on-disk inode. We do this via i_disksize, which is the value which
2580 * ext3 *really* writes onto the disk inode.
2581 */
2582 ei->i_disksize = inode->i_size;
2583
2584 /*
2585 * From here we block out all ext3_get_block() callers who want to
2586 * modify the block allocation tree.
2587 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002588 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
2590 if (n == 1) { /* direct blocks */
2591 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2592 i_data + EXT3_NDIR_BLOCKS);
2593 goto do_indirects;
2594 }
2595
2596 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2597 /* Kill the top of shared branch (not detached) */
2598 if (nr) {
2599 if (partial == chain) {
2600 /* Shared branch grows from the inode */
2601 ext3_free_branches(handle, inode, NULL,
2602 &nr, &nr+1, (chain+n-1) - partial);
2603 *partial->p = 0;
2604 /*
2605 * We mark the inode dirty prior to restart,
2606 * and prior to stop. No need for it here.
2607 */
2608 } else {
2609 /* Shared branch grows from an indirect block */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 ext3_free_branches(handle, inode, partial->bh,
2611 partial->p,
2612 partial->p+1, (chain+n-1) - partial);
2613 }
2614 }
2615 /* Clear the ends of indirect blocks on the shared branch */
2616 while (partial > chain) {
2617 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2618 (__le32*)partial->bh->b_data+addr_per_block,
2619 (chain+n-1) - partial);
2620 BUFFER_TRACE(partial->bh, "call brelse");
2621 brelse (partial->bh);
2622 partial--;
2623 }
2624do_indirects:
2625 /* Kill the remaining (whole) subtrees */
2626 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002627 default:
2628 nr = i_data[EXT3_IND_BLOCK];
2629 if (nr) {
2630 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2631 i_data[EXT3_IND_BLOCK] = 0;
2632 }
2633 case EXT3_IND_BLOCK:
2634 nr = i_data[EXT3_DIND_BLOCK];
2635 if (nr) {
2636 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2637 i_data[EXT3_DIND_BLOCK] = 0;
2638 }
2639 case EXT3_DIND_BLOCK:
2640 nr = i_data[EXT3_TIND_BLOCK];
2641 if (nr) {
2642 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2643 i_data[EXT3_TIND_BLOCK] = 0;
2644 }
2645 case EXT3_TIND_BLOCK:
2646 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 }
2648
2649 ext3_discard_reservation(inode);
2650
Arjan van de Ven97461512006-03-23 03:00:42 -08002651 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2653 ext3_mark_inode_dirty(handle, inode);
2654
Andrew Mortond6859bf2006-03-26 01:38:03 -08002655 /*
2656 * In a multi-transaction truncate, we only make the final transaction
2657 * synchronous
2658 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 if (IS_SYNC(inode))
2660 handle->h_sync = 1;
2661out_stop:
2662 /*
2663 * If this was a simple ftruncate(), and the file will remain alive
2664 * then we need to clear up the orphan record which we created above.
2665 * However, if this was a real unlink then we were called by
Al Viroac14a952010-06-06 07:08:19 -04002666 * ext3_evict_inode(), and we allow that function to clean up the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 * orphan info for us.
2668 */
2669 if (inode->i_nlink)
2670 ext3_orphan_del(handle, inode);
2671
2672 ext3_journal_stop(handle);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002673 trace_ext3_truncate_exit(inode);
Jan Karaef436182009-06-17 16:26:24 -07002674 return;
2675out_notrans:
2676 /*
2677 * Delete the inode from orphan list so that it doesn't stay there
2678 * forever and trigger assertion on umount.
2679 */
2680 if (inode->i_nlink)
2681 ext3_orphan_del(NULL, inode);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002682 trace_ext3_truncate_exit(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683}
2684
Mingming Cao43d23f92006-06-25 05:48:07 -07002685static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 unsigned long ino, struct ext3_iloc *iloc)
2687{
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002688 unsigned long block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002689 unsigned long offset;
2690 ext3_fsblk_t block;
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002691 struct ext3_group_desc *gdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Neil Brown2ccb48e2006-07-30 03:03:01 -07002693 if (!ext3_valid_inum(sb, ino)) {
2694 /*
2695 * This error is already checked for in namei.c unless we are
2696 * looking at an NFS filehandle, in which case no error
2697 * report is needed
2698 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 return 0;
2700 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002701
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002703 gdp = ext3_get_group_desc(sb, block_group, NULL);
2704 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 /*
2707 * Figure out the offset within the block group inode table
2708 */
2709 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2710 EXT3_INODE_SIZE(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002711 block = le32_to_cpu(gdp->bg_inode_table) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2713
2714 iloc->block_group = block_group;
2715 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2716 return block;
2717}
2718
2719/*
2720 * ext3_get_inode_loc returns with an extra refcount against the inode's
2721 * underlying buffer_head on success. If 'in_mem' is true, we have all
2722 * data in memory that is needed to recreate the on-disk version of this
2723 * inode.
2724 */
2725static int __ext3_get_inode_loc(struct inode *inode,
2726 struct ext3_iloc *iloc, int in_mem)
2727{
Mingming Cao43d23f92006-06-25 05:48:07 -07002728 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 struct buffer_head *bh;
2730
2731 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2732 if (!block)
2733 return -EIO;
2734
2735 bh = sb_getblk(inode->i_sb, block);
Wang Shilong1b7d76e2013-01-12 01:36:17 -08002736 if (unlikely(!bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2738 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002739 "inode=%lu, block="E3FSBLK,
2740 inode->i_ino, block);
Wang Shilongc04e88e2013-01-15 21:20:01 -08002741 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 }
2743 if (!buffer_uptodate(bh)) {
2744 lock_buffer(bh);
Hidehiro Kawai95450f52008-07-25 01:46:24 -07002745
2746 /*
2747 * If the buffer has the write error flag, we have failed
2748 * to write out another inode in the same block. In this
2749 * case, we don't have to read the block because we may
2750 * read the old inode data successfully.
2751 */
2752 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2753 set_buffer_uptodate(bh);
2754
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 if (buffer_uptodate(bh)) {
2756 /* someone brought it uptodate while we waited */
2757 unlock_buffer(bh);
2758 goto has_buffer;
2759 }
2760
2761 /*
2762 * If we have all information of the inode in memory and this
2763 * is the only valid inode in the block, we need not read the
2764 * block.
2765 */
2766 if (in_mem) {
2767 struct buffer_head *bitmap_bh;
2768 struct ext3_group_desc *desc;
2769 int inodes_per_buffer;
2770 int inode_offset, i;
2771 int block_group;
2772 int start;
2773
2774 block_group = (inode->i_ino - 1) /
2775 EXT3_INODES_PER_GROUP(inode->i_sb);
2776 inodes_per_buffer = bh->b_size /
2777 EXT3_INODE_SIZE(inode->i_sb);
2778 inode_offset = ((inode->i_ino - 1) %
2779 EXT3_INODES_PER_GROUP(inode->i_sb));
2780 start = inode_offset & ~(inodes_per_buffer - 1);
2781
2782 /* Is the inode bitmap in cache? */
2783 desc = ext3_get_group_desc(inode->i_sb,
2784 block_group, NULL);
2785 if (!desc)
2786 goto make_io;
2787
2788 bitmap_bh = sb_getblk(inode->i_sb,
2789 le32_to_cpu(desc->bg_inode_bitmap));
Wang Shilong1b7d76e2013-01-12 01:36:17 -08002790 if (unlikely(!bitmap_bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 goto make_io;
2792
2793 /*
2794 * If the inode bitmap isn't in cache then the
2795 * optimisation may end up performing two reads instead
2796 * of one, so skip it.
2797 */
2798 if (!buffer_uptodate(bitmap_bh)) {
2799 brelse(bitmap_bh);
2800 goto make_io;
2801 }
2802 for (i = start; i < start + inodes_per_buffer; i++) {
2803 if (i == inode_offset)
2804 continue;
2805 if (ext3_test_bit(i, bitmap_bh->b_data))
2806 break;
2807 }
2808 brelse(bitmap_bh);
2809 if (i == start + inodes_per_buffer) {
2810 /* all other inodes are free, so skip I/O */
2811 memset(bh->b_data, 0, bh->b_size);
2812 set_buffer_uptodate(bh);
2813 unlock_buffer(bh);
2814 goto has_buffer;
2815 }
2816 }
2817
2818make_io:
2819 /*
2820 * There are other valid inodes in the buffer, this inode
2821 * has in-inode xattrs, or we don't have this inode in memory.
2822 * Read the block from disk.
2823 */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002824 trace_ext3_load_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 get_bh(bh);
2826 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02002827 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 wait_on_buffer(bh);
2829 if (!buffer_uptodate(bh)) {
2830 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2831 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002832 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 inode->i_ino, block);
2834 brelse(bh);
2835 return -EIO;
2836 }
2837 }
2838has_buffer:
2839 iloc->bh = bh;
2840 return 0;
2841}
2842
2843int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2844{
2845 /* We have all inode data except xattrs in memory here. */
2846 return __ext3_get_inode_loc(inode, iloc,
Jan Kara9df93932010-01-06 21:58:48 +01002847 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848}
2849
2850void ext3_set_inode_flags(struct inode *inode)
2851{
2852 unsigned int flags = EXT3_I(inode)->i_flags;
2853
2854 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2855 if (flags & EXT3_SYNC_FL)
2856 inode->i_flags |= S_SYNC;
2857 if (flags & EXT3_APPEND_FL)
2858 inode->i_flags |= S_APPEND;
2859 if (flags & EXT3_IMMUTABLE_FL)
2860 inode->i_flags |= S_IMMUTABLE;
2861 if (flags & EXT3_NOATIME_FL)
2862 inode->i_flags |= S_NOATIME;
2863 if (flags & EXT3_DIRSYNC_FL)
2864 inode->i_flags |= S_DIRSYNC;
2865}
2866
Jan Kara28be5ab2007-05-08 00:30:33 -07002867/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2868void ext3_get_inode_flags(struct ext3_inode_info *ei)
2869{
2870 unsigned int flags = ei->vfs_inode.i_flags;
2871
2872 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2873 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2874 if (flags & S_SYNC)
2875 ei->i_flags |= EXT3_SYNC_FL;
2876 if (flags & S_APPEND)
2877 ei->i_flags |= EXT3_APPEND_FL;
2878 if (flags & S_IMMUTABLE)
2879 ei->i_flags |= EXT3_IMMUTABLE_FL;
2880 if (flags & S_NOATIME)
2881 ei->i_flags |= EXT3_NOATIME_FL;
2882 if (flags & S_DIRSYNC)
2883 ei->i_flags |= EXT3_DIRSYNC_FL;
2884}
2885
David Howells473043d2008-02-07 00:15:36 -08002886struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887{
2888 struct ext3_iloc iloc;
2889 struct ext3_inode *raw_inode;
David Howells473043d2008-02-07 00:15:36 -08002890 struct ext3_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 struct buffer_head *bh;
David Howells473043d2008-02-07 00:15:36 -08002892 struct inode *inode;
Jan Karafe8bc912009-10-16 19:26:15 +02002893 journal_t *journal = EXT3_SB(sb)->s_journal;
2894 transaction_t *transaction;
David Howells473043d2008-02-07 00:15:36 -08002895 long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 int block;
Eric W. Biederman15232992012-02-07 15:41:24 -08002897 uid_t i_uid;
2898 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899
David Howells473043d2008-02-07 00:15:36 -08002900 inode = iget_locked(sb, ino);
2901 if (!inode)
2902 return ERR_PTR(-ENOMEM);
2903 if (!(inode->i_state & I_NEW))
2904 return inode;
2905
2906 ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 ei->i_block_alloc_info = NULL;
2908
David Howells473043d2008-02-07 00:15:36 -08002909 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2910 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 goto bad_inode;
2912 bh = iloc.bh;
2913 raw_inode = ext3_raw_inode(&iloc);
2914 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman15232992012-02-07 15:41:24 -08002915 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2916 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 if(!(test_opt (inode->i_sb, NO_UID32))) {
Eric W. Biederman15232992012-02-07 15:41:24 -08002918 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2919 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 }
Eric W. Biederman15232992012-02-07 15:41:24 -08002921 i_uid_write(inode, i_uid);
2922 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002923 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002925 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2926 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2927 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2929
Linus Torvaldsde329822010-03-29 14:30:19 -07002930 ei->i_state_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 ei->i_dir_start_lookup = 0;
2932 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2933 /* We now have enough fields to check if the inode was active or not.
2934 * This is needed because nfsd might try to access dead inodes
2935 * the test is that same one that e2fsck uses
2936 * NeilBrown 1999oct15
2937 */
2938 if (inode->i_nlink == 0) {
2939 if (inode->i_mode == 0 ||
2940 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2941 /* this inode is deleted */
2942 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002943 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 goto bad_inode;
2945 }
2946 /* The only unlinked inodes we let through here have
2947 * valid i_mode and are being read by the orphan
2948 * recovery code: that's fine, we're about to complete
2949 * the process of deleting those. */
2950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2952 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2953#ifdef EXT3_FRAGMENTS
2954 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2955 ei->i_frag_no = raw_inode->i_frag;
2956 ei->i_frag_size = raw_inode->i_fsize;
2957#endif
2958 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2959 if (!S_ISREG(inode->i_mode)) {
2960 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2961 } else {
2962 inode->i_size |=
2963 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2964 }
2965 ei->i_disksize = inode->i_size;
2966 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2967 ei->i_block_group = iloc.block_group;
2968 /*
2969 * NOTE! The in-memory inode i_data array is in little-endian order
2970 * even on big-endian machines: we do NOT byteswap the block numbers!
2971 */
2972 for (block = 0; block < EXT3_N_BLOCKS; block++)
2973 ei->i_data[block] = raw_inode->i_block[block];
2974 INIT_LIST_HEAD(&ei->i_orphan);
2975
Jan Karafe8bc912009-10-16 19:26:15 +02002976 /*
2977 * Set transaction id's of transactions that have to be committed
2978 * to finish f[data]sync. We set them to currently running transaction
2979 * as we cannot be sure that the inode or some of its metadata isn't
2980 * part of the transaction - the inode could have been reclaimed and
2981 * now it is reread from disk.
2982 */
2983 if (journal) {
2984 tid_t tid;
2985
2986 spin_lock(&journal->j_state_lock);
2987 if (journal->j_running_transaction)
2988 transaction = journal->j_running_transaction;
2989 else
2990 transaction = journal->j_committing_transaction;
2991 if (transaction)
2992 tid = transaction->t_tid;
2993 else
2994 tid = journal->j_commit_sequence;
2995 spin_unlock(&journal->j_state_lock);
2996 atomic_set(&ei->i_sync_tid, tid);
2997 atomic_set(&ei->i_datasync_tid, tid);
2998 }
2999
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
3001 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
3002 /*
3003 * When mke2fs creates big inodes it does not zero out
3004 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
3005 * so ignore those first few inodes.
3006 */
3007 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3008 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07003009 EXT3_INODE_SIZE(inode->i_sb)) {
3010 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08003011 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 goto bad_inode;
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07003013 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 if (ei->i_extra_isize == 0) {
3015 /* The extra space is currently unused. Use it. */
3016 ei->i_extra_isize = sizeof(struct ext3_inode) -
3017 EXT3_GOOD_OLD_INODE_SIZE;
3018 } else {
3019 __le32 *magic = (void *)raw_inode +
3020 EXT3_GOOD_OLD_INODE_SIZE +
3021 ei->i_extra_isize;
3022 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
Jan Kara9df93932010-01-06 21:58:48 +01003023 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 }
3025 } else
3026 ei->i_extra_isize = 0;
3027
3028 if (S_ISREG(inode->i_mode)) {
3029 inode->i_op = &ext3_file_inode_operations;
3030 inode->i_fop = &ext3_file_operations;
3031 ext3_set_aops(inode);
3032 } else if (S_ISDIR(inode->i_mode)) {
3033 inode->i_op = &ext3_dir_inode_operations;
3034 inode->i_fop = &ext3_dir_operations;
3035 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffinb5ed3112008-12-19 20:47:14 +00003036 if (ext3_inode_is_fast_symlink(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 inode->i_op = &ext3_fast_symlink_inode_operations;
Duane Griffinb5ed3112008-12-19 20:47:14 +00003038 nd_terminate_link(ei->i_data, inode->i_size,
3039 sizeof(ei->i_data) - 1);
3040 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 inode->i_op = &ext3_symlink_inode_operations;
3042 ext3_set_aops(inode);
3043 }
3044 } else {
3045 inode->i_op = &ext3_special_inode_operations;
3046 if (raw_inode->i_block[0])
3047 init_special_inode(inode, inode->i_mode,
3048 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003049 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 init_special_inode(inode, inode->i_mode,
3051 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3052 }
3053 brelse (iloc.bh);
3054 ext3_set_inode_flags(inode);
David Howells473043d2008-02-07 00:15:36 -08003055 unlock_new_inode(inode);
3056 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
3058bad_inode:
David Howells473043d2008-02-07 00:15:36 -08003059 iget_failed(inode);
3060 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061}
3062
3063/*
3064 * Post the struct inode info into an on-disk inode location in the
3065 * buffer-cache. This gobbles the caller's reference to the
3066 * buffer_head in the inode location struct.
3067 *
3068 * The caller must have write access to iloc->bh.
3069 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003070static int ext3_do_update_inode(handle_t *handle,
3071 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 struct ext3_iloc *iloc)
3073{
3074 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3075 struct ext3_inode_info *ei = EXT3_I(inode);
3076 struct buffer_head *bh = iloc->bh;
3077 int err = 0, rc, block;
Jan Kara156bddd2012-09-03 16:50:42 +02003078 int need_datasync = 0;
3079 __le32 disksize;
Eric W. Biederman15232992012-02-07 15:41:24 -08003080 uid_t i_uid;
3081 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
Chris Mason4f003fd2009-09-08 00:22:14 +02003083again:
3084 /* we can't allow multiple procs in here at once, its a bit racey */
3085 lock_buffer(bh);
3086
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 /* For fields not not tracking in the in-memory inode,
3088 * initialise them to zero for new inodes. */
Jan Kara9df93932010-01-06 21:58:48 +01003089 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3091
Jan Kara28be5ab2007-05-08 00:30:33 -07003092 ext3_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman15232992012-02-07 15:41:24 -08003094 i_uid = i_uid_read(inode);
3095 i_gid = i_gid_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 if(!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman15232992012-02-07 15:41:24 -08003097 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3098 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099/*
3100 * Fix up interoperability with old kernels. Otherwise, old inodes get
3101 * re-used with the upper 16 bits of the uid/gid intact
3102 */
3103 if(!ei->i_dtime) {
3104 raw_inode->i_uid_high =
Eric W. Biederman15232992012-02-07 15:41:24 -08003105 cpu_to_le16(high_16_bits(i_uid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 raw_inode->i_gid_high =
Eric W. Biederman15232992012-02-07 15:41:24 -08003107 cpu_to_le16(high_16_bits(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 } else {
3109 raw_inode->i_uid_high = 0;
3110 raw_inode->i_gid_high = 0;
3111 }
3112 } else {
3113 raw_inode->i_uid_low =
Eric W. Biederman15232992012-02-07 15:41:24 -08003114 cpu_to_le16(fs_high2lowuid(i_uid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 raw_inode->i_gid_low =
Eric W. Biederman15232992012-02-07 15:41:24 -08003116 cpu_to_le16(fs_high2lowgid(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 raw_inode->i_uid_high = 0;
3118 raw_inode->i_gid_high = 0;
3119 }
3120 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Jan Kara156bddd2012-09-03 16:50:42 +02003121 disksize = cpu_to_le32(ei->i_disksize);
3122 if (disksize != raw_inode->i_size) {
3123 need_datasync = 1;
3124 raw_inode->i_size = disksize;
3125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3127 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3128 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3129 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3130 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3131 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3132#ifdef EXT3_FRAGMENTS
3133 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3134 raw_inode->i_frag = ei->i_frag_no;
3135 raw_inode->i_fsize = ei->i_frag_size;
3136#endif
3137 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3138 if (!S_ISREG(inode->i_mode)) {
3139 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3140 } else {
Jan Kara156bddd2012-09-03 16:50:42 +02003141 disksize = cpu_to_le32(ei->i_disksize >> 32);
3142 if (disksize != raw_inode->i_size_high) {
3143 raw_inode->i_size_high = disksize;
3144 need_datasync = 1;
3145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 if (ei->i_disksize > 0x7fffffffULL) {
3147 struct super_block *sb = inode->i_sb;
3148 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3149 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3150 EXT3_SB(sb)->s_es->s_rev_level ==
3151 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3152 /* If this is the first large file
3153 * created, add a flag to the superblock.
3154 */
Chris Mason4f003fd2009-09-08 00:22:14 +02003155 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 err = ext3_journal_get_write_access(handle,
3157 EXT3_SB(sb)->s_sbh);
3158 if (err)
3159 goto out_brelse;
Chris Mason4f003fd2009-09-08 00:22:14 +02003160
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 ext3_update_dynamic_rev(sb);
3162 EXT3_SET_RO_COMPAT_FEATURE(sb,
3163 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 handle->h_sync = 1;
3165 err = ext3_journal_dirty_metadata(handle,
3166 EXT3_SB(sb)->s_sbh);
Chris Mason4f003fd2009-09-08 00:22:14 +02003167 /* get our lock and start over */
3168 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 }
3170 }
3171 }
3172 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3173 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3174 if (old_valid_dev(inode->i_rdev)) {
3175 raw_inode->i_block[0] =
3176 cpu_to_le32(old_encode_dev(inode->i_rdev));
3177 raw_inode->i_block[1] = 0;
3178 } else {
3179 raw_inode->i_block[0] = 0;
3180 raw_inode->i_block[1] =
3181 cpu_to_le32(new_encode_dev(inode->i_rdev));
3182 raw_inode->i_block[2] = 0;
3183 }
3184 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3185 raw_inode->i_block[block] = ei->i_data[block];
3186
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07003187 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3189
3190 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Chris Mason4f003fd2009-09-08 00:22:14 +02003191 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 rc = ext3_journal_dirty_metadata(handle, bh);
3193 if (!err)
3194 err = rc;
Jan Kara9df93932010-01-06 21:58:48 +01003195 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196
Jan Karafe8bc912009-10-16 19:26:15 +02003197 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
Jan Kara156bddd2012-09-03 16:50:42 +02003198 if (need_datasync)
3199 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200out_brelse:
3201 brelse (bh);
3202 ext3_std_error(inode->i_sb, err);
3203 return err;
3204}
3205
3206/*
3207 * ext3_write_inode()
3208 *
3209 * We are called from a few places:
3210 *
3211 * - Within generic_file_write() for O_SYNC files.
3212 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003213 * transaction to commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 *
3215 * - Within sys_sync(), kupdate and such.
3216 * We wait on commit, if tol to.
3217 *
3218 * - Within prune_icache() (PF_MEMALLOC == true)
3219 * Here we simply return. We can't afford to block kswapd on the
3220 * journal commit.
3221 *
3222 * In all cases it is actually safe for us to return without doing anything,
3223 * because the inode has been copied into a raw inode buffer in
3224 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3225 * knfsd.
3226 *
3227 * Note that we are absolutely dependent upon all inode dirtiers doing the
3228 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3229 * which we are interested.
3230 *
3231 * It would be a bug for them to not do this. The code:
3232 *
3233 * mark_inode_dirty(inode)
3234 * stuff();
3235 * inode->i_size = expr;
3236 *
3237 * is in error because a kswapd-driven write_inode() could occur while
3238 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3239 * will no longer be on the superblock's dirty inode list.
3240 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003241int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242{
3243 if (current->flags & PF_MEMALLOC)
3244 return 0;
3245
3246 if (ext3_journal_current_handle()) {
Jose R. Santos9ad163a2007-10-18 23:39:23 -07003247 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 dump_stack();
3249 return -EIO;
3250 }
3251
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003252 if (wbc->sync_mode != WB_SYNC_ALL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 return 0;
3254
3255 return ext3_force_commit(inode->i_sb);
3256}
3257
3258/*
3259 * ext3_setattr()
3260 *
3261 * Called from notify_change.
3262 *
3263 * We want to trap VFS attempts to truncate the file as soon as
3264 * possible. In particular, we want to make sure that when the VFS
3265 * shrinks i_size, we put the inode on the orphan list and modify
3266 * i_disksize immediately, so that during the subsequent flushing of
3267 * dirty pages and freeing of disk blocks, we can guarantee that any
3268 * commit will leave the blocks being flushed in an unused state on
3269 * disk. (On recovery, the inode will get truncated and the blocks will
3270 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003271 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 *
3273 * Called with inode->sem down.
3274 */
3275int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3276{
3277 struct inode *inode = dentry->d_inode;
3278 int error, rc = 0;
3279 const unsigned int ia_valid = attr->ia_valid;
3280
3281 error = inode_change_ok(inode, attr);
3282 if (error)
3283 return error;
3284
Dmitry Monakhov12755622010-04-08 22:04:20 +04003285 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05003286 dquot_initialize(inode);
Eric W. Biederman15232992012-02-07 15:41:24 -08003287 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3288 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 handle_t *handle;
3290
3291 /* (user+group)*(old+new) structure, inode write (sb,
3292 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003293 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3294 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 if (IS_ERR(handle)) {
3296 error = PTR_ERR(handle);
3297 goto err_out;
3298 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05003299 error = dquot_transfer(inode, attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 if (error) {
3301 ext3_journal_stop(handle);
3302 return error;
3303 }
3304 /* Update corresponding info in inode so that everything is in
3305 * one transaction */
3306 if (attr->ia_valid & ATTR_UID)
3307 inode->i_uid = attr->ia_uid;
3308 if (attr->ia_valid & ATTR_GID)
3309 inode->i_gid = attr->ia_gid;
3310 error = ext3_mark_inode_dirty(handle, inode);
3311 ext3_journal_stop(handle);
3312 }
3313
Christoph Hellwig562c72a2011-06-24 14:29:45 -04003314 if (attr->ia_valid & ATTR_SIZE)
3315 inode_dio_wait(inode);
3316
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 if (S_ISREG(inode->i_mode) &&
3318 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3319 handle_t *handle;
3320
3321 handle = ext3_journal_start(inode, 3);
3322 if (IS_ERR(handle)) {
3323 error = PTR_ERR(handle);
3324 goto err_out;
3325 }
3326
3327 error = ext3_orphan_add(handle, inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02003328 if (error) {
3329 ext3_journal_stop(handle);
3330 goto err_out;
3331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 EXT3_I(inode)->i_disksize = attr->ia_size;
Jan Karaee3e77f2011-06-03 21:58:11 +02003333 error = ext3_mark_inode_dirty(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 ext3_journal_stop(handle);
Jan Karaee3e77f2011-06-03 21:58:11 +02003335 if (error) {
3336 /* Some hard fs error must have happened. Bail out. */
3337 ext3_orphan_del(NULL, inode);
3338 goto err_out;
3339 }
3340 rc = ext3_block_truncate_page(inode, attr->ia_size);
3341 if (rc) {
3342 /* Cleanup orphan list and exit */
3343 handle = ext3_journal_start(inode, 3);
3344 if (IS_ERR(handle)) {
3345 ext3_orphan_del(NULL, inode);
3346 goto err_out;
3347 }
3348 ext3_orphan_del(handle, inode);
3349 ext3_journal_stop(handle);
3350 goto err_out;
3351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 }
3353
Christoph Hellwig10257742010-06-04 11:30:02 +02003354 if ((attr->ia_valid & ATTR_SIZE) &&
3355 attr->ia_size != i_size_read(inode)) {
Jan Kara40680f2f2011-05-24 22:24:47 +02003356 truncate_setsize(inode, attr->ia_size);
3357 ext3_truncate(inode);
Christoph Hellwig10257742010-06-04 11:30:02 +02003358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359
Christoph Hellwig10257742010-06-04 11:30:02 +02003360 setattr_copy(inode, attr);
3361 mark_inode_dirty(inode);
3362
3363 if (ia_valid & ATTR_MODE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 rc = ext3_acl_chmod(inode);
3365
3366err_out:
3367 ext3_std_error(inode->i_sb, error);
3368 if (!error)
3369 error = rc;
3370 return error;
3371}
3372
3373
3374/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003375 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 *
3377 * With N blocks per page, it may be:
3378 * N data blocks
3379 * 2 indirect block
3380 * 2 dindirect
3381 * 1 tindirect
3382 * N+5 bitmap blocks (from the above)
3383 * N+5 group descriptor summary blocks
3384 * 1 inode block
3385 * 1 superblock.
3386 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3387 *
3388 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3389 *
3390 * With ordered or writeback data it's the same, less the N data blocks.
3391 *
3392 * If the inode's direct blocks can hold an integral number of pages then a
3393 * page cannot straddle two indirect blocks, and we can only touch one indirect
3394 * and dindirect block, and the "5" above becomes "3".
3395 *
3396 * This still overestimates under most circumstances. If we were to pass the
3397 * start and end offsets in here as well we could do block_to_path() on each
3398 * block and work out the exact number of indirects which are touched. Pah.
3399 */
3400
3401static int ext3_writepage_trans_blocks(struct inode *inode)
3402{
3403 int bpp = ext3_journal_blocks_per_page(inode);
3404 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3405 int ret;
3406
3407 if (ext3_should_journal_data(inode))
3408 ret = 3 * (bpp + indirects) + 2;
3409 else
Yongqiang Yang523334b2011-03-24 08:48:39 +08003410 ret = 2 * (bpp + indirects) + indirects + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411
3412#ifdef CONFIG_QUOTA
Christoph Hellwig871a2932010-03-03 09:05:07 -05003413 /* We know that structure was already allocated during dquot_initialize so
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 * we will be updating only the data blocks + inodes */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003415 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416#endif
3417
3418 return ret;
3419}
3420
3421/*
3422 * The caller must have previously called ext3_reserve_inode_write().
3423 * Give this, we know that the caller already has write access to iloc->bh.
3424 */
3425int ext3_mark_iloc_dirty(handle_t *handle,
3426 struct inode *inode, struct ext3_iloc *iloc)
3427{
3428 int err = 0;
3429
3430 /* the do_update_inode consumes one bh->b_count */
3431 get_bh(iloc->bh);
3432
3433 /* ext3_do_update_inode() does journal_dirty_metadata */
3434 err = ext3_do_update_inode(handle, inode, iloc);
3435 put_bh(iloc->bh);
3436 return err;
3437}
3438
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003439/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003441 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 */
3443
3444int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003445ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 struct ext3_iloc *iloc)
3447{
3448 int err = 0;
3449 if (handle) {
3450 err = ext3_get_inode_loc(inode, iloc);
3451 if (!err) {
3452 BUFFER_TRACE(iloc->bh, "get_write_access");
3453 err = ext3_journal_get_write_access(handle, iloc->bh);
3454 if (err) {
3455 brelse(iloc->bh);
3456 iloc->bh = NULL;
3457 }
3458 }
3459 }
3460 ext3_std_error(inode->i_sb, err);
3461 return err;
3462}
3463
3464/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003465 * What we do here is to mark the in-core inode as clean with respect to inode
3466 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467 * This means that the in-core inode may be reaped by prune_icache
3468 * without having to perform any I/O. This is a very good thing,
3469 * because *any* task may call prune_icache - even ones which
3470 * have a transaction open against a different journal.
3471 *
3472 * Is this cheating? Not really. Sure, we haven't written the
3473 * inode out, but prune_icache isn't a user-visible syncing function.
3474 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3475 * we start and wait on commits.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 */
3477int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3478{
3479 struct ext3_iloc iloc;
3480 int err;
3481
3482 might_sleep();
Lukas Czerner785c4bc2011-05-23 18:33:01 +02003483 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 err = ext3_reserve_inode_write(handle, inode, &iloc);
3485 if (!err)
3486 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3487 return err;
3488}
3489
3490/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003491 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 *
3493 * We're really interested in the case where a file is being extended.
3494 * i_size has been changed by generic_commit_write() and we thus need
3495 * to include the updated inode in the current transaction.
3496 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003497 * Also, dquot_alloc_space() will always dirty the inode when blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 * are allocated to the file.
3499 *
3500 * If the inode is marked synchronous, we don't honour that here - doing
3501 * so would cause a commit on atime updates, which we don't bother doing.
3502 * We handle synchronous inodes at the highest possible level.
3503 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04003504void ext3_dirty_inode(struct inode *inode, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505{
3506 handle_t *current_handle = ext3_journal_current_handle();
3507 handle_t *handle;
3508
3509 handle = ext3_journal_start(inode, 2);
3510 if (IS_ERR(handle))
3511 goto out;
3512 if (current_handle &&
3513 current_handle->h_transaction != handle->h_transaction) {
3514 /* This task has a transaction open against a different fs */
3515 printk(KERN_EMERG "%s: transactions do not match!\n",
Harvey Harrisone05b6b52008-04-28 02:16:15 -07003516 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 } else {
3518 jbd_debug(5, "marking dirty. outer handle=%p\n",
3519 current_handle);
3520 ext3_mark_inode_dirty(handle, inode);
3521 }
3522 ext3_journal_stop(handle);
3523out:
3524 return;
3525}
3526
Andrew Mortond6859bf2006-03-26 01:38:03 -08003527#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003528/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 * Bind an inode's backing buffer_head into this transaction, to prevent
3530 * it from being flushed to disk early. Unlike
3531 * ext3_reserve_inode_write, this leaves behind no bh reference and
3532 * returns no iloc structure, so the caller needs to repeat the iloc
3533 * lookup to mark the inode dirty later.
3534 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003535static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536{
3537 struct ext3_iloc iloc;
3538
3539 int err = 0;
3540 if (handle) {
3541 err = ext3_get_inode_loc(inode, &iloc);
3542 if (!err) {
3543 BUFFER_TRACE(iloc.bh, "get_write_access");
3544 err = journal_get_write_access(handle, iloc.bh);
3545 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003546 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 iloc.bh);
3548 brelse(iloc.bh);
3549 }
3550 }
3551 ext3_std_error(inode->i_sb, err);
3552 return err;
3553}
3554#endif
3555
3556int ext3_change_inode_journal_flag(struct inode *inode, int val)
3557{
3558 journal_t *journal;
3559 handle_t *handle;
3560 int err;
3561
3562 /*
3563 * We have to be very careful here: changing a data block's
3564 * journaling status dynamically is dangerous. If we write a
3565 * data block to the journal, change the status and then delete
3566 * that block, we risk forgetting to revoke the old log record
3567 * from the journal and so a subsequent replay can corrupt data.
3568 * So, first we make sure that the journal is empty and that
3569 * nobody is changing anything.
3570 */
3571
3572 journal = EXT3_JOURNAL(inode);
Dave Hansene3a68e32007-07-15 23:41:14 -07003573 if (is_journal_aborted(journal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 return -EROFS;
3575
3576 journal_lock_updates(journal);
3577 journal_flush(journal);
3578
3579 /*
3580 * OK, there are no updates running now, and all cached data is
3581 * synced to disk. We are now in a completely consistent state
3582 * which doesn't have anything in the journal, and we know that
3583 * no filesystem updates are running, so it is safe to modify
3584 * the inode's in-core data-journaling state flag now.
3585 */
3586
3587 if (val)
3588 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3589 else
3590 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3591 ext3_set_aops(inode);
3592
3593 journal_unlock_updates(journal);
3594
3595 /* Finally we can mark the inode as dirty. */
3596
3597 handle = ext3_journal_start(inode, 1);
3598 if (IS_ERR(handle))
3599 return PTR_ERR(handle);
3600
3601 err = ext3_mark_inode_dirty(handle, inode);
3602 handle->h_sync = 1;
3603 ext3_journal_stop(handle);
3604 ext3_std_error(inode->i_sb, err);
3605
3606 return err;
3607}