blob: b176d425354428043ef92c4e680353d90a28c760 [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 Boissinot1cc8dcf2008-04-21 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 Boissinot1cc8dcf2008-04-21 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 Boissinot1cc8dcf2008-04-21 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]);
679 branch[n].bh = bh;
680 lock_buffer(bh);
681 BUFFER_TRACE(bh, "call get_create_access");
682 err = ext3_journal_get_create_access(handle, bh);
683 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 unlock_buffer(bh);
Mingming Caob47b2472006-03-26 01:37:56 -0800685 brelse(bh);
686 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Mingming Caob47b2472006-03-26 01:37:56 -0800689 memset(bh->b_data, 0, blocksize);
690 branch[n].p = (__le32 *) bh->b_data + offsets[n];
691 branch[n].key = cpu_to_le32(new_blocks[n]);
692 *branch[n].p = branch[n].key;
693 if ( n == indirect_blks) {
694 current_block = new_blocks[n];
695 /*
696 * End of chain, update the last new metablock of
697 * the chain to point to the new allocated
698 * data blocks numbers
699 */
700 for (i=1; i < num; i++)
701 *(branch[n].p + i) = cpu_to_le32(++current_block);
702 }
703 BUFFER_TRACE(bh, "marking uptodate");
704 set_buffer_uptodate(bh);
705 unlock_buffer(bh);
706
707 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
708 err = ext3_journal_dirty_metadata(handle, bh);
709 if (err)
710 goto failed;
711 }
712 *blks = num;
713 return err;
714failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 /* Allocation failed, free what we already allocated */
Mingming Caob47b2472006-03-26 01:37:56 -0800716 for (i = 1; i <= n ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 BUFFER_TRACE(branch[i].bh, "call journal_forget");
718 ext3_journal_forget(handle, branch[i].bh);
719 }
Mingming Caob47b2472006-03-26 01:37:56 -0800720 for (i = 0; i <indirect_blks; i++)
721 ext3_free_blocks(handle, inode, new_blocks[i], 1);
722
723 ext3_free_blocks(handle, inode, new_blocks[i], num);
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return err;
726}
727
728/**
Andrew Mortond6859bf2006-03-26 01:38:03 -0800729 * ext3_splice_branch - splice the allocated branch onto inode.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900730 * @handle: handle for this transaction
Andrew Mortond6859bf2006-03-26 01:38:03 -0800731 * @inode: owner
732 * @block: (logical) number of block we are adding
Andrew Mortond6859bf2006-03-26 01:38:03 -0800733 * @where: location of missing link
734 * @num: number of indirect blocks we are adding
735 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800737 * This function fills the missing link and does all housekeeping needed in
738 * inode (->i_blocks, etc.). In case of success we end up with the full
739 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800741static int ext3_splice_branch(handle_t *handle, struct inode *inode,
742 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
744 int i;
745 int err = 0;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800746 struct ext3_block_alloc_info *block_i;
Mingming Cao43d23f92006-06-25 05:48:07 -0700747 ext3_fsblk_t current_block;
Jan Karafe8bc912009-10-16 19:26:15 +0200748 struct ext3_inode_info *ei = EXT3_I(inode);
Kazuya Mioac1334b2012-02-08 16:34:29 +0900749 struct timespec now;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800750
Jan Karafe8bc912009-10-16 19:26:15 +0200751 block_i = ei->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 /*
753 * If we're splicing into a [td]indirect block (as opposed to the
754 * inode) then we need to get write access to the [td]indirect block
755 * before the splice.
756 */
757 if (where->bh) {
758 BUFFER_TRACE(where->bh, "get_write_access");
759 err = ext3_journal_get_write_access(handle, where->bh);
760 if (err)
761 goto err_out;
762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 /* That's it */
764
765 *where->p = where->key;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800766
767 /*
768 * Update the host buffer_head or inode to point to more just allocated
769 * direct blocks blocks
770 */
Mingming Caob47b2472006-03-26 01:37:56 -0800771 if (num == 0 && blks > 1) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700772 current_block = le32_to_cpu(where->key) + 1;
Mingming Caob47b2472006-03-26 01:37:56 -0800773 for (i = 1; i < blks; i++)
774 *(where->p + i ) = cpu_to_le32(current_block++);
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 /*
778 * update the most recently allocated logical & physical block
779 * in i_block_alloc_info, to assist find the proper goal block for next
780 * allocation
781 */
782 if (block_i) {
Mingming Caob47b2472006-03-26 01:37:56 -0800783 block_i->last_alloc_logical_block = block + blks - 1;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800784 block_i->last_alloc_physical_block =
Mingming Cao5dea5172006-05-03 19:55:12 -0700785 le32_to_cpu(where[num].key) + blks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
787
788 /* We are done with atomic stuff, now do the rest of housekeeping */
Kazuya Mioac1334b2012-02-08 16:34:29 +0900789 now = CURRENT_TIME_SEC;
790 if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) {
791 inode->i_ctime = now;
792 ext3_mark_inode_dirty(handle, inode);
793 }
Jan Karafe8bc912009-10-16 19:26:15 +0200794 /* ext3_mark_inode_dirty already updated i_sync_tid */
795 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
797 /* had we spliced it onto indirect block? */
798 if (where->bh) {
799 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800800 * If we spliced it onto an indirect block, we haven't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 * altered the inode. Note however that if it is being spliced
802 * onto an indirect block at the very end of the file (the
803 * file is growing) then we *will* alter the inode to reflect
804 * the new i_size. But that is not done here - it is done in
805 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
806 */
807 jbd_debug(5, "splicing indirect only\n");
808 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
809 err = ext3_journal_dirty_metadata(handle, where->bh);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700810 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 goto err_out;
812 } else {
813 /*
814 * OK, we spliced it into the inode itself on a direct block.
815 * Inode was dirtied above.
816 */
817 jbd_debug(5, "splicing direct\n");
818 }
819 return err;
820
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821err_out:
Mingming Caob47b2472006-03-26 01:37:56 -0800822 for (i = 1; i <= num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 BUFFER_TRACE(where[i].bh, "call journal_forget");
824 ext3_journal_forget(handle, where[i].bh);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800825 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
Mingming Caob47b2472006-03-26 01:37:56 -0800827 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 return err;
830}
831
832/*
833 * Allocation strategy is simple: if we have to allocate something, we will
834 * have to go the whole way to leaf. So let's do it before attaching anything
835 * to tree, set linkage between the newborn blocks, write them if sync is
836 * required, recheck the path, free and repeat if check fails, otherwise
837 * set the last missing link (that will protect us from any truncate-generated
838 * removals - all blocks on the path are immune now) and possibly force the
839 * write on the parent block.
840 * That has a nice additional property: no special recovery from the failed
841 * allocations is needed - we simply release blocks and do not touch anything
842 * reachable from inode.
843 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800844 * `handle' can be NULL if create == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 *
846 * The BKL may not be held on entry here. Be sure to take it early.
Mingming Cao89747d32006-03-26 01:37:55 -0800847 * return > 0, # of blocks mapped or allocated.
848 * return = 0, if plain lookup failed.
849 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800851int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
852 sector_t iblock, unsigned long maxblocks,
853 struct buffer_head *bh_result,
Jan Kara43237b52009-05-20 18:41:58 +0200854 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855{
856 int err = -EIO;
857 int offsets[4];
858 Indirect chain[4];
859 Indirect *partial;
Mingming Cao43d23f92006-06-25 05:48:07 -0700860 ext3_fsblk_t goal;
Mingming Caob47b2472006-03-26 01:37:56 -0800861 int indirect_blks;
Mingming Cao89747d32006-03-26 01:37:55 -0800862 int blocks_to_boundary = 0;
863 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 struct ext3_inode_info *ei = EXT3_I(inode);
Mingming Cao89747d32006-03-26 01:37:55 -0800865 int count = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700866 ext3_fsblk_t first_block = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200869 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 J_ASSERT(handle != NULL || create == 0);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800871 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873 if (depth == 0)
874 goto out;
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
877
878 /* Simplest case - block found, no allocation needed */
879 if (!partial) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700880 first_block = le32_to_cpu(chain[depth - 1].key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 clear_buffer_new(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800882 count++;
883 /*map more blocks*/
884 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700885 ext3_fsblk_t blk;
Mingming Cao5dea5172006-05-03 19:55:12 -0700886
Jan Karae8ef7aa2009-06-17 16:26:23 -0700887 if (!verify_chain(chain, chain + depth - 1)) {
Mingming Cao89747d32006-03-26 01:37:55 -0800888 /*
889 * Indirect block might be removed by
890 * truncate while we were reading it.
891 * Handling of that case: forget what we've
892 * got now. Flag the err as EAGAIN, so it
893 * will reread.
894 */
895 err = -EAGAIN;
896 count = 0;
897 break;
898 }
Mingming Cao5dea5172006-05-03 19:55:12 -0700899 blk = le32_to_cpu(*(chain[depth-1].p + count));
900
901 if (blk == first_block + count)
Mingming Cao89747d32006-03-26 01:37:55 -0800902 count++;
903 else
904 break;
905 }
906 if (err != -EAGAIN)
907 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 }
909
910 /* Next simple case - plain lookup or failed read of indirect block */
Mingming Caofe55c452005-05-01 08:59:20 -0700911 if (!create || err == -EIO)
912 goto cleanup;
913
Jan Kara40680f2f2011-05-24 22:24:47 +0200914 /*
915 * Block out ext3_truncate while we alter the tree
916 */
Arjan van de Ven97461512006-03-23 03:00:42 -0800917 mutex_lock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700918
919 /*
920 * If the indirect block is missing while we are reading
921 * the chain(ext3_get_branch() returns -EAGAIN err), or
922 * if the chain has been changed after we grab the semaphore,
923 * (either because another process truncated this branch, or
924 * another get_block allocated this branch) re-grab the chain to see if
925 * the request block has been allocated or not.
926 *
927 * Since we already block the truncate/other get_block
928 * at this point, we will have the current copy of the chain when we
929 * splice the branch into the tree.
930 */
931 if (err == -EAGAIN || !verify_chain(chain, partial)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 while (partial > chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 brelse(partial->bh);
934 partial--;
935 }
Mingming Caofe55c452005-05-01 08:59:20 -0700936 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
937 if (!partial) {
Mingming Cao89747d32006-03-26 01:37:55 -0800938 count++;
Arjan van de Ven97461512006-03-23 03:00:42 -0800939 mutex_unlock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700940 if (err)
941 goto cleanup;
942 clear_buffer_new(bh_result);
943 goto got_it;
944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 }
946
947 /*
Mingming Caofe55c452005-05-01 08:59:20 -0700948 * Okay, we need to do block allocation. Lazily initialize the block
949 * allocation info here if necessary
950 */
951 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 ext3_init_block_alloc_info(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800954 goal = ext3_find_goal(inode, iblock, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Mingming Caob47b2472006-03-26 01:37:56 -0800956 /* the number of blocks need to allocate for [d,t]indirect blocks */
957 indirect_blks = (chain + depth) - partial - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
959 /*
Mingming Caob47b2472006-03-26 01:37:56 -0800960 * Next look up the indirect map to count the totoal number of
961 * direct blocks to allocate for this branch.
962 */
963 count = ext3_blks_to_allocate(partial, indirect_blks,
964 maxblocks, blocks_to_boundary);
Mingming Caob47b2472006-03-26 01:37:56 -0800965 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
Mingming Caofe55c452005-05-01 08:59:20 -0700966 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Mingming Caofe55c452005-05-01 08:59:20 -0700968 /*
969 * The ext3_splice_branch call will free and forget any buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 * on the new chain if there is a failure, but that risks using
971 * up transaction credits, especially for bitmaps where the
972 * credits cannot be returned. Can we handle this somehow? We
Mingming Caofe55c452005-05-01 08:59:20 -0700973 * may need to return -EAGAIN upwards in the worst case. --sct
974 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (!err)
Mingming Caob47b2472006-03-26 01:37:56 -0800976 err = ext3_splice_branch(handle, inode, iblock,
977 partial, indirect_blks, count);
Arjan van de Ven97461512006-03-23 03:00:42 -0800978 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (err)
980 goto cleanup;
981
982 set_buffer_new(bh_result);
Mingming Caofe55c452005-05-01 08:59:20 -0700983got_it:
984 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
Suparna Bhattacharya20acaa12006-09-16 12:15:58 -0700985 if (count > blocks_to_boundary)
Mingming Caofe55c452005-05-01 08:59:20 -0700986 set_buffer_boundary(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800987 err = count;
Mingming Caofe55c452005-05-01 08:59:20 -0700988 /* Clean up and exit */
989 partial = chain + depth - 1; /* the whole chain */
990cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 while (partial > chain) {
Mingming Caofe55c452005-05-01 08:59:20 -0700992 BUFFER_TRACE(partial->bh, "call brelse");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 brelse(partial->bh);
994 partial--;
995 }
Mingming Caofe55c452005-05-01 08:59:20 -0700996 BUFFER_TRACE(bh_result, "returned");
997out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200998 trace_ext3_get_blocks_exit(inode, iblock,
999 depth ? le32_to_cpu(chain[depth-1].key) : 0,
1000 count, err);
Mingming Caofe55c452005-05-01 08:59:20 -07001001 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002}
1003
Jan Karabd1939d2008-02-06 01:40:21 -08001004/* Maximum number of blocks we map for direct IO at once. */
1005#define DIO_MAX_BLOCKS 4096
1006/*
1007 * Number of credits we need for writing DIO_MAX_BLOCKS:
1008 * We need sb + group descriptor + bitmap + inode -> 4
1009 * For B blocks with A block pointers per block we need:
1010 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1011 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1012 */
1013#define DIO_CREDITS 25
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001015static int ext3_get_block(struct inode *inode, sector_t iblock,
1016 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001018 handle_t *handle = ext3_journal_current_handle();
Jan Karabd1939d2008-02-06 01:40:21 -08001019 int ret = 0, started = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -08001020 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Jan Karabd1939d2008-02-06 01:40:21 -08001022 if (create && !handle) { /* Direct IO write... */
1023 if (max_blocks > DIO_MAX_BLOCKS)
1024 max_blocks = DIO_MAX_BLOCKS;
1025 handle = ext3_journal_start(inode, DIO_CREDITS +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001026 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
Jan Karabd1939d2008-02-06 01:40:21 -08001027 if (IS_ERR(handle)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 ret = PTR_ERR(handle);
Jan Karabd1939d2008-02-06 01:40:21 -08001029 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 }
Jan Karabd1939d2008-02-06 01:40:21 -08001031 started = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
1033
Jan Karabd1939d2008-02-06 01:40:21 -08001034 ret = ext3_get_blocks_handle(handle, inode, iblock,
Jan Kara43237b52009-05-20 18:41:58 +02001035 max_blocks, bh_result, create);
Jan Karabd1939d2008-02-06 01:40:21 -08001036 if (ret > 0) {
1037 bh_result->b_size = (ret << inode->i_blkbits);
1038 ret = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001039 }
Jan Karabd1939d2008-02-06 01:40:21 -08001040 if (started)
1041 ext3_journal_stop(handle);
1042out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 return ret;
1044}
1045
Josef Bacik68c9d702008-10-03 17:32:43 -04001046int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1047 u64 start, u64 len)
1048{
1049 return generic_block_fiemap(inode, fieinfo, start, len,
1050 ext3_get_block);
1051}
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053/*
1054 * `handle' can be NULL if create is zero
1055 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001056struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1057 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
1059 struct buffer_head dummy;
1060 int fatal = 0, err;
1061
1062 J_ASSERT(handle != NULL || create == 0);
1063
1064 dummy.b_state = 0;
1065 dummy.b_blocknr = -1000;
1066 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -08001067 err = ext3_get_blocks_handle(handle, inode, block, 1,
Jan Kara43237b52009-05-20 18:41:58 +02001068 &dummy, create);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001069 /*
1070 * ext3_get_blocks_handle() returns number of blocks
1071 * mapped. 0 in case of a HOLE.
1072 */
1073 if (err > 0) {
Julia Lawall47897752012-11-03 21:30:21 +01001074 WARN_ON(err > 1);
Mingming Cao89747d32006-03-26 01:37:55 -08001075 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001076 }
1077 *errp = err;
1078 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 struct buffer_head *bh;
1080 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001081 if (!bh) {
1082 *errp = -EIO;
1083 goto err;
1084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 if (buffer_new(&dummy)) {
1086 J_ASSERT(create != 0);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001087 J_ASSERT(handle != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Andrew Mortond6859bf2006-03-26 01:38:03 -08001089 /*
1090 * Now that we do not always journal data, we should
1091 * keep in mind whether this should always journal the
1092 * new buffer as metadata. For now, regular file
1093 * writes use ext3_get_block instead, so it's not a
1094 * problem.
1095 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 lock_buffer(bh);
1097 BUFFER_TRACE(bh, "call get_create_access");
1098 fatal = ext3_journal_get_create_access(handle, bh);
1099 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001100 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 set_buffer_uptodate(bh);
1102 }
1103 unlock_buffer(bh);
1104 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1105 err = ext3_journal_dirty_metadata(handle, bh);
1106 if (!fatal)
1107 fatal = err;
1108 } else {
1109 BUFFER_TRACE(bh, "not a new buffer");
1110 }
1111 if (fatal) {
1112 *errp = fatal;
1113 brelse(bh);
1114 bh = NULL;
1115 }
1116 return bh;
1117 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001118err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 return NULL;
1120}
1121
Andrew Mortond6859bf2006-03-26 01:38:03 -08001122struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 int block, int create, int *err)
1124{
1125 struct buffer_head * bh;
1126
1127 bh = ext3_getblk(handle, inode, block, create, err);
1128 if (!bh)
1129 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001130 if (bh_uptodate_or_lock(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001132 get_bh(bh);
1133 bh->b_end_io = end_buffer_read_sync;
1134 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 wait_on_buffer(bh);
1136 if (buffer_uptodate(bh))
1137 return bh;
1138 put_bh(bh);
1139 *err = -EIO;
1140 return NULL;
1141}
1142
1143static int walk_page_buffers( handle_t *handle,
1144 struct buffer_head *head,
1145 unsigned from,
1146 unsigned to,
1147 int *partial,
1148 int (*fn)( handle_t *handle,
1149 struct buffer_head *bh))
1150{
1151 struct buffer_head *bh;
1152 unsigned block_start, block_end;
1153 unsigned blocksize = head->b_size;
1154 int err, ret = 0;
1155 struct buffer_head *next;
1156
1157 for ( bh = head, block_start = 0;
1158 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001159 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 {
1161 next = bh->b_this_page;
1162 block_end = block_start + blocksize;
1163 if (block_end <= from || block_start >= to) {
1164 if (partial && !buffer_uptodate(bh))
1165 *partial = 1;
1166 continue;
1167 }
1168 err = (*fn)(handle, bh);
1169 if (!ret)
1170 ret = err;
1171 }
1172 return ret;
1173}
1174
1175/*
1176 * To preserve ordering, it is essential that the hole instantiation and
1177 * the data write be encapsulated in a single transaction. We cannot
1178 * close off a transaction and start a new one between the ext3_get_block()
1179 * and the commit_write(). So doing the journal_start at the start of
1180 * prepare_write() is the right place.
1181 *
1182 * Also, this function can nest inside ext3_writepage() ->
1183 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1184 * has generated enough buffer credits to do the whole page. So we won't
1185 * block on the journal in that case, which is good, because the caller may
1186 * be PF_MEMALLOC.
1187 *
1188 * By accident, ext3 can be reentered when a transaction is open via
1189 * quota file writes. If we were to commit the transaction while thus
1190 * reentered, there can be a deadlock - we would be holding a quota
1191 * lock, and the commit would never complete if another thread had a
1192 * transaction open and was blocking on the quota lock - a ranking
1193 * violation.
1194 *
1195 * So what we do is to rely on the fact that journal_stop/journal_start
1196 * will _not_ run commit under these circumstances because handle->h_ref
1197 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001198 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001200static int do_journal_get_write_access(handle_t *handle,
1201 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Jan Kara5f11e6a2010-08-05 12:38:26 +02001203 int dirty = buffer_dirty(bh);
1204 int ret;
1205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 if (!buffer_mapped(bh) || buffer_freed(bh))
1207 return 0;
Jan Kara5f11e6a2010-08-05 12:38:26 +02001208 /*
1209 * __block_prepare_write() could have dirtied some buffers. Clean
1210 * the dirty bit as jbd2_journal_get_write_access() could complain
1211 * otherwise about fs integrity issues. Setting of the dirty bit
1212 * by __block_prepare_write() isn't a real problem here as we clear
1213 * the bit before releasing a page lock and thus writeback cannot
1214 * ever write the buffer.
1215 */
1216 if (dirty)
1217 clear_buffer_dirty(bh);
1218 ret = ext3_journal_get_write_access(handle, bh);
1219 if (!ret && dirty)
1220 ret = ext3_journal_dirty_metadata(handle, bh);
1221 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
Jan Kara68eb3db2009-12-01 16:53:06 +01001224/*
1225 * Truncate blocks that were not used by write. We have to truncate the
1226 * pagecache as well so that corresponding buffers get properly unmapped.
1227 */
1228static void ext3_truncate_failed_write(struct inode *inode)
1229{
1230 truncate_inode_pages(inode->i_mapping, inode->i_size);
1231 ext3_truncate(inode);
1232}
1233
Jan Karaee3e77f2011-06-03 21:58:11 +02001234/*
1235 * Truncate blocks that were not used by direct IO write. We have to zero out
1236 * the last file block as well because direct IO might have written to it.
1237 */
1238static void ext3_truncate_failed_direct_write(struct inode *inode)
1239{
1240 ext3_block_truncate_page(inode, inode->i_size);
1241 ext3_truncate(inode);
1242}
1243
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001244static int ext3_write_begin(struct file *file, struct address_space *mapping,
1245 loff_t pos, unsigned len, unsigned flags,
1246 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247{
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001248 struct inode *inode = mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001249 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 handle_t *handle;
1251 int retries = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001252 struct page *page;
1253 pgoff_t index;
1254 unsigned from, to;
Jan Kara695f6ae2009-04-02 16:57:17 -07001255 /* Reserve one block more for addition to orphan list in case
1256 * we allocate blocks but write fails for some reason */
1257 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001258
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001259 trace_ext3_write_begin(inode, pos, len, flags);
1260
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001261 index = pos >> PAGE_CACHE_SHIFT;
1262 from = pos & (PAGE_CACHE_SIZE - 1);
1263 to = from + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
1265retry:
Nick Piggin54566b22009-01-04 12:00:53 -08001266 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001267 if (!page)
1268 return -ENOMEM;
1269 *pagep = page;
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 handle = ext3_journal_start(inode, needed_blocks);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001272 if (IS_ERR(handle)) {
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001273 unlock_page(page);
1274 page_cache_release(page);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001275 ret = PTR_ERR(handle);
1276 goto out;
1277 }
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001278 ret = __block_write_begin(page, pos, len, ext3_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (ret)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001280 goto write_begin_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
1282 if (ext3_should_journal_data(inode)) {
1283 ret = walk_page_buffers(handle, page_buffers(page),
1284 from, to, NULL, do_journal_get_write_access);
1285 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001286write_begin_failed:
1287 if (ret) {
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001288 /*
1289 * block_write_begin may have instantiated a few blocks
1290 * outside i_size. Trim these off again. Don't need
1291 * i_size_read because we hold i_mutex.
Jan Kara695f6ae2009-04-02 16:57:17 -07001292 *
1293 * Add inode to orphan list in case we crash before truncate
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001294 * finishes. Do this only if ext3_can_truncate() agrees so
1295 * that orphan processing code is happy.
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001296 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001297 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001298 ext3_orphan_add(handle, inode);
1299 ext3_journal_stop(handle);
1300 unlock_page(page);
1301 page_cache_release(page);
1302 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001303 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1306 goto retry;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001307out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 return ret;
1309}
1310
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001311
Andrew Mortond6859bf2006-03-26 01:38:03 -08001312int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
1314 int err = journal_dirty_data(handle, bh);
1315 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001316 ext3_journal_abort_handle(__func__, __func__,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001317 bh, handle, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 return err;
1319}
1320
Jan Kara695f6ae2009-04-02 16:57:17 -07001321/* For ordered writepage and write_end functions */
1322static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1323{
1324 /*
1325 * Write could have mapped the buffer but it didn't copy the data in
1326 * yet. So avoid filing such buffer into a transaction.
1327 */
1328 if (buffer_mapped(bh) && buffer_uptodate(bh))
1329 return ext3_journal_dirty_data(handle, bh);
1330 return 0;
1331}
1332
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001333/* For write_end() in data=journal mode */
1334static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
1336 if (!buffer_mapped(bh) || buffer_freed(bh))
1337 return 0;
1338 set_buffer_uptodate(bh);
1339 return ext3_journal_dirty_metadata(handle, bh);
1340}
1341
1342/*
Jan Kara695f6ae2009-04-02 16:57:17 -07001343 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1344 * for the whole page but later we failed to copy the data in. Update inode
1345 * size according to what we managed to copy. The rest is going to be
1346 * truncated in write_end function.
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001347 */
Jan Kara695f6ae2009-04-02 16:57:17 -07001348static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001349{
Jan Kara695f6ae2009-04-02 16:57:17 -07001350 /* What matters to us is i_disksize. We don't write i_size anywhere */
1351 if (pos + copied > inode->i_size)
1352 i_size_write(inode, pos + copied);
1353 if (pos + copied > EXT3_I(inode)->i_disksize) {
1354 EXT3_I(inode)->i_disksize = pos + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001355 mark_inode_dirty(inode);
1356 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001357}
1358
1359/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 * We need to pick up the new inode size which generic_commit_write gave us
1361 * `file' can be NULL - eg, when called from page_symlink().
1362 *
1363 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1364 * buffers are managed internally.
1365 */
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001366static int ext3_ordered_write_end(struct file *file,
1367 struct address_space *mapping,
1368 loff_t pos, unsigned len, unsigned copied,
1369 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370{
1371 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001372 struct inode *inode = file->f_mapping->host;
1373 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 int ret = 0, ret2;
1375
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001376 trace_ext3_ordered_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001377 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1378
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001379 from = pos & (PAGE_CACHE_SIZE - 1);
Jan Kara695f6ae2009-04-02 16:57:17 -07001380 to = from + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 ret = walk_page_buffers(handle, page_buffers(page),
Jan Kara695f6ae2009-04-02 16:57:17 -07001382 from, to, NULL, journal_dirty_data_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Jan Kara695f6ae2009-04-02 16:57:17 -07001384 if (ret == 0)
1385 update_file_sizes(inode, pos, copied);
1386 /*
1387 * There may be allocated blocks outside of i_size because
1388 * we failed to copy some data. Prepare for truncate.
1389 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001390 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001391 ext3_orphan_add(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 ret2 = ext3_journal_stop(handle);
1393 if (!ret)
1394 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001395 unlock_page(page);
1396 page_cache_release(page);
1397
Jan Kara695f6ae2009-04-02 16:57:17 -07001398 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001399 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001400 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001403static int ext3_writeback_write_end(struct file *file,
1404 struct address_space *mapping,
1405 loff_t pos, unsigned len, unsigned copied,
1406 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
1408 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001409 struct inode *inode = file->f_mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001410 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001412 trace_ext3_writeback_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001413 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1414 update_file_sizes(inode, pos, copied);
1415 /*
1416 * There may be allocated blocks outside of i_size because
1417 * we failed to copy some data. Prepare for truncate.
1418 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001419 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001420 ext3_orphan_add(handle, inode);
1421 ret = ext3_journal_stop(handle);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001422 unlock_page(page);
1423 page_cache_release(page);
1424
Jan Kara695f6ae2009-04-02 16:57:17 -07001425 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001426 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001427 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001430static int ext3_journalled_write_end(struct file *file,
1431 struct address_space *mapping,
1432 loff_t pos, unsigned len, unsigned copied,
1433 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434{
1435 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001436 struct inode *inode = mapping->host;
Jan Karab22570d2011-07-23 01:21:38 +02001437 struct ext3_inode_info *ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 int ret = 0, ret2;
1439 int partial = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001440 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001442 trace_ext3_journalled_write_end(inode, pos, len, copied);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001443 from = pos & (PAGE_CACHE_SIZE - 1);
1444 to = from + len;
1445
1446 if (copied < len) {
1447 if (!PageUptodate(page))
1448 copied = 0;
Jan Kara695f6ae2009-04-02 16:57:17 -07001449 page_zero_new_buffers(page, from + copied, to);
1450 to = from + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001454 to, &partial, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (!partial)
1456 SetPageUptodate(page);
Jan Kara695f6ae2009-04-02 16:57:17 -07001457
1458 if (pos + copied > inode->i_size)
1459 i_size_write(inode, pos + copied);
1460 /*
1461 * There may be allocated blocks outside of i_size because
1462 * we failed to copy some data. Prepare for truncate.
1463 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001464 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001465 ext3_orphan_add(handle, inode);
Jan Kara9df93932010-01-06 21:58:48 +01001466 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001467 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1468 if (inode->i_size > ei->i_disksize) {
1469 ei->i_disksize = inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001471 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 ret = ret2;
1473 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 ret2 = ext3_journal_stop(handle);
1476 if (!ret)
1477 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001478 unlock_page(page);
1479 page_cache_release(page);
1480
Jan Kara695f6ae2009-04-02 16:57:17 -07001481 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001482 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001483 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001486/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 * bmap() is special. It gets used by applications such as lilo and by
1488 * the swapper to find the on-disk block of a specific piece of data.
1489 *
1490 * Naturally, this is dangerous if the block concerned is still in the
1491 * journal. If somebody makes a swapfile on an ext3 data-journaling
1492 * filesystem and enables swap, then they may get a nasty shock when the
1493 * data getting swapped to that swapfile suddenly gets overwritten by
1494 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001495 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 *
1497 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001498 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 */
1500static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1501{
1502 struct inode *inode = mapping->host;
1503 journal_t *journal;
1504 int err;
1505
Jan Kara9df93932010-01-06 21:58:48 +01001506 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001507 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 * This is a REALLY heavyweight approach, but the use of
1509 * bmap on dirty files is expected to be extremely rare:
1510 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001511 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 *
1513 * (bmap requires CAP_SYS_RAWIO so this does not
1514 * represent an unprivileged user DOS attack --- we'd be
1515 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001516 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 *
1518 * NB. EXT3_STATE_JDATA is not set on files other than
1519 * regular files. If somebody wants to bmap a directory
1520 * or symlink and gets confused because the buffer
1521 * hasn't yet been flushed to disk, they deserve
1522 * everything they get.
1523 */
1524
Jan Kara9df93932010-01-06 21:58:48 +01001525 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 journal = EXT3_JOURNAL(inode);
1527 journal_lock_updates(journal);
1528 err = journal_flush(journal);
1529 journal_unlock_updates(journal);
1530
1531 if (err)
1532 return 0;
1533 }
1534
1535 return generic_block_bmap(mapping,block,ext3_get_block);
1536}
1537
1538static int bget_one(handle_t *handle, struct buffer_head *bh)
1539{
1540 get_bh(bh);
1541 return 0;
1542}
1543
1544static int bput_one(handle_t *handle, struct buffer_head *bh)
1545{
1546 put_bh(bh);
1547 return 0;
1548}
1549
Jan Kara9e80d402009-03-26 13:08:04 +01001550static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1551{
1552 return !buffer_mapped(bh);
1553}
Jan Kara695f6ae2009-04-02 16:57:17 -07001554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555/*
1556 * Note that we always start a transaction even if we're not journalling
1557 * data. This is to preserve ordering: any hole instantiation within
1558 * __block_write_full_page -> ext3_get_block() should be journalled
1559 * along with the data so we don't crash and then get metadata which
1560 * refers to old data.
1561 *
1562 * In all journalling modes block_write_full_page() will start the I/O.
1563 *
1564 * Problem:
1565 *
1566 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1567 * ext3_writepage()
1568 *
1569 * Similar for:
1570 *
1571 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1572 *
1573 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001574 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 *
1576 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1577 * allocations fail.
1578 *
1579 * 16May01: If we're reentered then journal_current_handle() will be
1580 * non-zero. We simply *return*.
1581 *
1582 * 1 July 2001: @@@ FIXME:
1583 * In journalled data mode, a data buffer may be metadata against the
1584 * current transaction. But the same file is part of a shared mapping
1585 * and someone does a writepage() on it.
1586 *
1587 * We will move the buffer onto the async_data list, but *after* it has
1588 * been dirtied. So there's a small window where we have dirty data on
1589 * BJ_Metadata.
1590 *
1591 * Note that this only applies to the last partial page in the file. The
1592 * bit which block_write_full_page() uses prepare/commit for. (That's
1593 * broken code anyway: it's wrong for msync()).
1594 *
1595 * It's a rare case: affects the final partial page, for journalled data
1596 * where the file is subject to bith write() and writepage() in the same
1597 * transction. To fix it we'll need a custom block_write_full_page().
1598 * We'll probably need that anyway for journalling writepage() output.
1599 *
1600 * We don't honour synchronous mounts for writepage(). That would be
1601 * disastrous. Any write() or metadata operation will sync the fs for
1602 * us.
1603 *
1604 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1605 * we don't need to open a transaction here.
1606 */
1607static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001608 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609{
1610 struct inode *inode = page->mapping->host;
1611 struct buffer_head *page_bufs;
1612 handle_t *handle = NULL;
1613 int ret = 0;
1614 int err;
1615
1616 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001617 /*
1618 * We don't want to warn for emergency remount. The condition is
1619 * ordered to avoid dereferencing inode->i_sb in non-error case to
1620 * avoid slow-downs.
1621 */
1622 WARN_ON_ONCE(IS_RDONLY(inode) &&
1623 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 /*
1626 * We give up here if we're reentered, because it might be for a
1627 * different filesystem.
1628 */
1629 if (ext3_journal_current_handle())
1630 goto out_fail;
1631
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001632 trace_ext3_ordered_writepage(page);
Jan Kara9e80d402009-03-26 13:08:04 +01001633 if (!page_has_buffers(page)) {
1634 create_empty_buffers(page, inode->i_sb->s_blocksize,
1635 (1 << BH_Dirty)|(1 << BH_Uptodate));
Jan Kara430db322009-04-07 18:25:01 -04001636 page_bufs = page_buffers(page);
1637 } else {
1638 page_bufs = page_buffers(page);
1639 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1640 NULL, buffer_unmapped)) {
1641 /* Provide NULL get_block() to catch bugs if buffers
1642 * weren't really mapped */
1643 return block_write_full_page(page, NULL, wbc);
1644 }
Jan Kara9e80d402009-03-26 13:08:04 +01001645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1647
1648 if (IS_ERR(handle)) {
1649 ret = PTR_ERR(handle);
1650 goto out_fail;
1651 }
1652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 walk_page_buffers(handle, page_bufs, 0,
1654 PAGE_CACHE_SIZE, NULL, bget_one);
1655
1656 ret = block_write_full_page(page, ext3_get_block, wbc);
1657
1658 /*
1659 * The page can become unlocked at any point now, and
1660 * truncate can then come in and change things. So we
1661 * can't touch *page from now on. But *page_bufs is
1662 * safe due to elevated refcount.
1663 */
1664
1665 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001666 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 * block_write_full_page() succeeded. Otherwise they are unmapped,
1668 * and generally junk.
1669 */
1670 if (ret == 0) {
1671 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1672 NULL, journal_dirty_data_fn);
1673 if (!ret)
1674 ret = err;
1675 }
1676 walk_page_buffers(handle, page_bufs, 0,
1677 PAGE_CACHE_SIZE, NULL, bput_one);
1678 err = ext3_journal_stop(handle);
1679 if (!ret)
1680 ret = err;
1681 return ret;
1682
1683out_fail:
1684 redirty_page_for_writepage(wbc, page);
1685 unlock_page(page);
1686 return ret;
1687}
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689static int ext3_writeback_writepage(struct page *page,
1690 struct writeback_control *wbc)
1691{
1692 struct inode *inode = page->mapping->host;
1693 handle_t *handle = NULL;
1694 int ret = 0;
1695 int err;
1696
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001697 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001698 /*
1699 * We don't want to warn for emergency remount. The condition is
1700 * ordered to avoid dereferencing inode->i_sb in non-error case to
1701 * avoid slow-downs.
1702 */
1703 WARN_ON_ONCE(IS_RDONLY(inode) &&
1704 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 if (ext3_journal_current_handle())
1707 goto out_fail;
1708
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001709 trace_ext3_writeback_writepage(page);
Jan Kara430db322009-04-07 18:25:01 -04001710 if (page_has_buffers(page)) {
1711 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1712 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1713 /* Provide NULL get_block() to catch bugs if buffers
1714 * weren't really mapped */
1715 return block_write_full_page(page, NULL, wbc);
1716 }
1717 }
1718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1720 if (IS_ERR(handle)) {
1721 ret = PTR_ERR(handle);
1722 goto out_fail;
1723 }
1724
Christoph Hellwig4c4d3902010-06-07 10:20:39 +02001725 ret = block_write_full_page(page, ext3_get_block, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
1727 err = ext3_journal_stop(handle);
1728 if (!ret)
1729 ret = err;
1730 return ret;
1731
1732out_fail:
1733 redirty_page_for_writepage(wbc, page);
1734 unlock_page(page);
1735 return ret;
1736}
1737
1738static int ext3_journalled_writepage(struct page *page,
1739 struct writeback_control *wbc)
1740{
1741 struct inode *inode = page->mapping->host;
1742 handle_t *handle = NULL;
1743 int ret = 0;
1744 int err;
1745
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001746 J_ASSERT(PageLocked(page));
Jan Kara33c104d2011-12-22 16:49:05 +01001747 /*
1748 * We don't want to warn for emergency remount. The condition is
1749 * ordered to avoid dereferencing inode->i_sb in non-error case to
1750 * avoid slow-downs.
1751 */
1752 WARN_ON_ONCE(IS_RDONLY(inode) &&
1753 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 if (ext3_journal_current_handle())
1756 goto no_write;
1757
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001758 trace_ext3_journalled_writepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1760 if (IS_ERR(handle)) {
1761 ret = PTR_ERR(handle);
1762 goto no_write;
1763 }
1764
1765 if (!page_has_buffers(page) || PageChecked(page)) {
1766 /*
1767 * It's mmapped pagecache. Add buffers and journal it. There
1768 * doesn't seem much point in redirtying the page here.
1769 */
1770 ClearPageChecked(page);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001771 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1772 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001773 if (ret != 0) {
1774 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001776 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 ret = walk_page_buffers(handle, page_buffers(page), 0,
1778 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1779
1780 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001781 PAGE_CACHE_SIZE, NULL, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 if (ret == 0)
1783 ret = err;
Jan Kara9df93932010-01-06 21:58:48 +01001784 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001785 atomic_set(&EXT3_I(inode)->i_datasync_tid,
1786 handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 unlock_page(page);
1788 } else {
1789 /*
1790 * It may be a page full of checkpoint-mode buffers. We don't
1791 * really know unless we go poke around in the buffer_heads.
1792 * But block_write_full_page will do the right thing.
1793 */
1794 ret = block_write_full_page(page, ext3_get_block, wbc);
1795 }
1796 err = ext3_journal_stop(handle);
1797 if (!ret)
1798 ret = err;
1799out:
1800 return ret;
1801
1802no_write:
1803 redirty_page_for_writepage(wbc, page);
1804out_unlock:
1805 unlock_page(page);
1806 goto out;
1807}
1808
1809static int ext3_readpage(struct file *file, struct page *page)
1810{
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001811 trace_ext3_readpage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 return mpage_readpage(page, ext3_get_block);
1813}
1814
1815static int
1816ext3_readpages(struct file *file, struct address_space *mapping,
1817 struct list_head *pages, unsigned nr_pages)
1818{
1819 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1820}
1821
NeilBrown2ff28e22006-03-26 01:37:18 -08001822static void ext3_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823{
1824 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1825
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001826 trace_ext3_invalidatepage(page, offset);
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 /*
1829 * If it's a full truncate we just forget about the pending dirtying
1830 */
1831 if (offset == 0)
1832 ClearPageChecked(page);
1833
NeilBrown2ff28e22006-03-26 01:37:18 -08001834 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835}
1836
Al Viro27496a82005-10-21 03:20:48 -04001837static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838{
1839 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1840
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001841 trace_ext3_releasepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 WARN_ON(PageChecked(page));
1843 if (!page_has_buffers(page))
1844 return 0;
1845 return journal_try_to_free_buffers(journal, page, wait);
1846}
1847
1848/*
1849 * If the O_DIRECT write will extend the file then add this inode to the
1850 * orphan list. So recovery will truncate it back to the original size
1851 * if the machine crashes during the write.
1852 *
1853 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Karabd1939d2008-02-06 01:40:21 -08001854 * crashes then stale disk data _may_ be exposed inside the file. But current
1855 * VFS code falls back into buffered path in that case so we are safe.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 */
1857static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1858 const struct iovec *iov, loff_t offset,
1859 unsigned long nr_segs)
1860{
1861 struct file *file = iocb->ki_filp;
1862 struct inode *inode = file->f_mapping->host;
1863 struct ext3_inode_info *ei = EXT3_I(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001864 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 ssize_t ret;
1866 int orphan = 0;
1867 size_t count = iov_length(iov, nr_segs);
Eric Sandeenea0174a2009-10-12 21:34:27 -05001868 int retries = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001870 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 if (rw == WRITE) {
1873 loff_t final_size = offset + count;
1874
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if (final_size > inode->i_size) {
Jan Karabd1939d2008-02-06 01:40:21 -08001876 /* Credits for sb + inode write */
1877 handle = ext3_journal_start(inode, 2);
1878 if (IS_ERR(handle)) {
1879 ret = PTR_ERR(handle);
1880 goto out;
1881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 ret = ext3_orphan_add(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001883 if (ret) {
1884 ext3_journal_stop(handle);
1885 goto out;
1886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 orphan = 1;
1888 ei->i_disksize = inode->i_size;
Jan Karabd1939d2008-02-06 01:40:21 -08001889 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 }
1891 }
1892
Eric Sandeenea0174a2009-10-12 21:34:27 -05001893retry:
Christoph Hellwigaacfc19c2011-06-24 14:29:47 -04001894 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1895 ext3_get_block);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001896 /*
1897 * In case of error extending write may have instantiated a few
1898 * blocks outside i_size. Trim these off again.
1899 */
1900 if (unlikely((rw & WRITE) && ret < 0)) {
1901 loff_t isize = i_size_read(inode);
1902 loff_t end = offset + iov_length(iov, nr_segs);
1903
1904 if (end > isize)
Jan Karaee3e77f2011-06-03 21:58:11 +02001905 ext3_truncate_failed_direct_write(inode);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001906 }
Eric Sandeenea0174a2009-10-12 21:34:27 -05001907 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1908 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
Jan Karabd1939d2008-02-06 01:40:21 -08001910 if (orphan) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 int err;
1912
Jan Karabd1939d2008-02-06 01:40:21 -08001913 /* Credits for sb + inode write */
1914 handle = ext3_journal_start(inode, 2);
1915 if (IS_ERR(handle)) {
1916 /* This is really bad luck. We've written the data
Jan Kara7eb49692010-03-01 14:02:37 +01001917 * but cannot extend i_size. Truncate allocated blocks
1918 * and pretend the write failed... */
Jan Karaee3e77f2011-06-03 21:58:11 +02001919 ext3_truncate_failed_direct_write(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001920 ret = PTR_ERR(handle);
1921 goto out;
1922 }
1923 if (inode->i_nlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 ext3_orphan_del(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001925 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 loff_t end = offset + ret;
1927 if (end > inode->i_size) {
1928 ei->i_disksize = end;
1929 i_size_write(inode, end);
1930 /*
1931 * We're going to return a positive `ret'
1932 * here due to non-zero-length I/O, so there's
1933 * no way of reporting error returns from
1934 * ext3_mark_inode_dirty() to userspace. So
1935 * ignore it.
1936 */
1937 ext3_mark_inode_dirty(handle, inode);
1938 }
1939 }
1940 err = ext3_journal_stop(handle);
1941 if (ret == 0)
1942 ret = err;
1943 }
1944out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001945 trace_ext3_direct_IO_exit(inode, offset,
1946 iov_length(iov, nr_segs), rw, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 return ret;
1948}
1949
1950/*
1951 * Pages can be marked dirty completely asynchronously from ext3's journalling
1952 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1953 * much here because ->set_page_dirty is called under VFS locks. The page is
1954 * not necessarily locked.
1955 *
1956 * We cannot just dirty the page and leave attached buffers clean, because the
1957 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1958 * or jbddirty because all the journalling code will explode.
1959 *
1960 * So what we do is to mark the page "pending dirty" and next time writepage
1961 * is called, propagate that into the buffers appropriately.
1962 */
1963static int ext3_journalled_set_page_dirty(struct page *page)
1964{
1965 SetPageChecked(page);
1966 return __set_page_dirty_nobuffers(page);
1967}
1968
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001969static const struct address_space_operations ext3_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001970 .readpage = ext3_readpage,
1971 .readpages = ext3_readpages,
1972 .writepage = ext3_ordered_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001973 .write_begin = ext3_write_begin,
1974 .write_end = ext3_ordered_write_end,
1975 .bmap = ext3_bmap,
1976 .invalidatepage = ext3_invalidatepage,
1977 .releasepage = ext3_releasepage,
1978 .direct_IO = ext3_direct_IO,
1979 .migratepage = buffer_migrate_page,
1980 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001981 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982};
1983
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001984static const struct address_space_operations ext3_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001985 .readpage = ext3_readpage,
1986 .readpages = ext3_readpages,
1987 .writepage = ext3_writeback_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001988 .write_begin = ext3_write_begin,
1989 .write_end = ext3_writeback_write_end,
1990 .bmap = ext3_bmap,
1991 .invalidatepage = ext3_invalidatepage,
1992 .releasepage = ext3_releasepage,
1993 .direct_IO = ext3_direct_IO,
1994 .migratepage = buffer_migrate_page,
1995 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001996 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997};
1998
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001999static const struct address_space_operations ext3_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002000 .readpage = ext3_readpage,
2001 .readpages = ext3_readpages,
2002 .writepage = ext3_journalled_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002003 .write_begin = ext3_write_begin,
2004 .write_end = ext3_journalled_write_end,
2005 .set_page_dirty = ext3_journalled_set_page_dirty,
2006 .bmap = ext3_bmap,
2007 .invalidatepage = ext3_invalidatepage,
2008 .releasepage = ext3_releasepage,
2009 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002010 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011};
2012
2013void ext3_set_aops(struct inode *inode)
2014{
2015 if (ext3_should_order_data(inode))
2016 inode->i_mapping->a_ops = &ext3_ordered_aops;
2017 else if (ext3_should_writeback_data(inode))
2018 inode->i_mapping->a_ops = &ext3_writeback_aops;
2019 else
2020 inode->i_mapping->a_ops = &ext3_journalled_aops;
2021}
2022
2023/*
2024 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2025 * up to the end of the block which corresponds to `from'.
2026 * This required during truncate. We need to physically zero the tail end
2027 * of that block so it doesn't yield old data if the file is later grown.
2028 */
Jan Karaee3e77f2011-06-03 21:58:11 +02002029static int ext3_block_truncate_page(struct inode *inode, loff_t from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030{
Mingming Cao43d23f92006-06-25 05:48:07 -07002031 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Jan Karaee3e77f2011-06-03 21:58:11 +02002032 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 unsigned blocksize, iblock, length, pos;
Jan Karaee3e77f2011-06-03 21:58:11 +02002034 struct page *page;
2035 handle_t *handle = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 struct buffer_head *bh;
2037 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Jan Karaee3e77f2011-06-03 21:58:11 +02002039 /* Truncated on block boundary - nothing to do */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 blocksize = inode->i_sb->s_blocksize;
Jan Karaee3e77f2011-06-03 21:58:11 +02002041 if ((from & (blocksize - 1)) == 0)
2042 return 0;
2043
2044 page = grab_cache_page(inode->i_mapping, index);
2045 if (!page)
2046 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 length = blocksize - (offset & (blocksize - 1));
2048 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2049
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (!page_has_buffers(page))
2051 create_empty_buffers(page, blocksize, 0);
2052
2053 /* Find the buffer that contains "offset" */
2054 bh = page_buffers(page);
2055 pos = blocksize;
2056 while (offset >= pos) {
2057 bh = bh->b_this_page;
2058 iblock++;
2059 pos += blocksize;
2060 }
2061
2062 err = 0;
2063 if (buffer_freed(bh)) {
2064 BUFFER_TRACE(bh, "freed: skip");
2065 goto unlock;
2066 }
2067
2068 if (!buffer_mapped(bh)) {
2069 BUFFER_TRACE(bh, "unmapped");
2070 ext3_get_block(inode, iblock, bh, 0);
2071 /* unmapped? It's a hole - nothing to do */
2072 if (!buffer_mapped(bh)) {
2073 BUFFER_TRACE(bh, "still unmapped");
2074 goto unlock;
2075 }
2076 }
2077
2078 /* Ok, it's mapped. Make sure it's up-to-date */
2079 if (PageUptodate(page))
2080 set_buffer_uptodate(bh);
2081
Zheng Liud03e1292011-12-05 15:55:11 +08002082 if (!bh_uptodate_or_lock(bh)) {
2083 err = bh_submit_read(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 /* Uhhuh. Read error. Complain and punt. */
Zheng Liud03e1292011-12-05 15:55:11 +08002085 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 goto unlock;
2087 }
2088
Jan Karaee3e77f2011-06-03 21:58:11 +02002089 /* data=writeback mode doesn't need transaction to zero-out data */
2090 if (!ext3_should_writeback_data(inode)) {
2091 /* We journal at most one block */
2092 handle = ext3_journal_start(inode, 1);
2093 if (IS_ERR(handle)) {
2094 clear_highpage(page);
2095 flush_dcache_page(page);
2096 err = PTR_ERR(handle);
2097 goto unlock;
2098 }
2099 }
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 if (ext3_should_journal_data(inode)) {
2102 BUFFER_TRACE(bh, "get write access");
2103 err = ext3_journal_get_write_access(handle, bh);
2104 if (err)
Jan Karaee3e77f2011-06-03 21:58:11 +02002105 goto stop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 }
2107
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002108 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 BUFFER_TRACE(bh, "zeroed end of block");
2110
2111 err = 0;
2112 if (ext3_should_journal_data(inode)) {
2113 err = ext3_journal_dirty_metadata(handle, bh);
2114 } else {
2115 if (ext3_should_order_data(inode))
2116 err = ext3_journal_dirty_data(handle, bh);
2117 mark_buffer_dirty(bh);
2118 }
Jan Karaee3e77f2011-06-03 21:58:11 +02002119stop:
2120 if (handle)
2121 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122
2123unlock:
2124 unlock_page(page);
2125 page_cache_release(page);
2126 return err;
2127}
2128
2129/*
2130 * Probably it should be a library function... search for first non-zero word
2131 * or memcmp with zero_page, whatever is better for particular architecture.
2132 * Linus?
2133 */
2134static inline int all_zeroes(__le32 *p, __le32 *q)
2135{
2136 while (p < q)
2137 if (*p++)
2138 return 0;
2139 return 1;
2140}
2141
2142/**
2143 * ext3_find_shared - find the indirect blocks for partial truncation.
2144 * @inode: inode in question
2145 * @depth: depth of the affected branch
2146 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2147 * @chain: place to store the pointers to partial indirect blocks
2148 * @top: place to the (detached) top of branch
2149 *
2150 * This is a helper function used by ext3_truncate().
2151 *
2152 * When we do truncate() we may have to clean the ends of several
2153 * indirect blocks but leave the blocks themselves alive. Block is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002154 * partially truncated if some data below the new i_size is referred
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 * from it (and it is on the path to the first completely truncated
2156 * data block, indeed). We have to free the top of that path along
2157 * with everything to the right of the path. Since no allocation
2158 * past the truncation point is possible until ext3_truncate()
2159 * finishes, we may safely do the latter, but top of branch may
2160 * require special attention - pageout below the truncation point
2161 * might try to populate it.
2162 *
2163 * We atomically detach the top of branch from the tree, store the
2164 * block number of its root in *@top, pointers to buffer_heads of
2165 * partially truncated blocks - in @chain[].bh and pointers to
2166 * their last elements that should not be removed - in
2167 * @chain[].p. Return value is the pointer to last filled element
2168 * of @chain.
2169 *
2170 * The work left to caller to do the actual freeing of subtrees:
2171 * a) free the subtree starting from *@top
2172 * b) free the subtrees whose roots are stored in
2173 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2174 * c) free the subtrees growing from the inode past the @chain[0].
2175 * (no partially truncated stuff there). */
2176
Andrew Mortond6859bf2006-03-26 01:38:03 -08002177static Indirect *ext3_find_shared(struct inode *inode, int depth,
2178 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
2180 Indirect *partial, *p;
2181 int k, err;
2182
2183 *top = 0;
Uwe Kleine-Königbf48aab2009-10-28 20:11:03 +01002184 /* Make k index the deepest non-null offset + 1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 for (k = depth; k > 1 && !offsets[k-1]; k--)
2186 ;
2187 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2188 /* Writer: pointers */
2189 if (!partial)
2190 partial = chain + k-1;
2191 /*
2192 * If the branch acquired continuation since we've looked at it -
2193 * fine, it should all survive and (new) top doesn't belong to us.
2194 */
2195 if (!partial->key && *partial->p)
2196 /* Writer: end */
2197 goto no_top;
2198 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2199 ;
2200 /*
2201 * OK, we've found the last block that must survive. The rest of our
2202 * branch should be detached before unlocking. However, if that rest
2203 * of branch is all ours and does not grow immediately from the inode
2204 * it's easier to cheat and just decrement partial->p.
2205 */
2206 if (p == chain + k - 1 && p > chain) {
2207 p->p--;
2208 } else {
2209 *top = *p->p;
2210 /* Nope, don't do this in ext3. Must leave the tree intact */
2211#if 0
2212 *p->p = 0;
2213#endif
2214 }
2215 /* Writer: end */
2216
Andrew Mortond6859bf2006-03-26 01:38:03 -08002217 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 brelse(partial->bh);
2219 partial--;
2220 }
2221no_top:
2222 return partial;
2223}
2224
2225/*
2226 * Zero a number of block pointers in either an inode or an indirect block.
2227 * If we restart the transaction we must again get write access to the
2228 * indirect block for further modification.
2229 *
2230 * We release `count' blocks on disk, but (last - first) may be greater
2231 * than `count' because there can be holes in there.
2232 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002233static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07002234 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08002235 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236{
2237 __le32 *p;
2238 if (try_to_extend_transaction(handle, inode)) {
2239 if (bh) {
2240 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Namhyung Kim156e7432010-11-25 01:53:13 +09002241 if (ext3_journal_dirty_metadata(handle, bh))
2242 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 }
2244 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002245 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 if (bh) {
2247 BUFFER_TRACE(bh, "retaking write access");
Namhyung Kim156e7432010-11-25 01:53:13 +09002248 if (ext3_journal_get_write_access(handle, bh))
2249 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 }
2251 }
2252
2253 /*
2254 * Any buffers which are on the journal will be in memory. We find
2255 * them on the hash table so journal_revoke() will run journal_forget()
2256 * on them. We've already detached each block from the file, so
2257 * bforget() in journal_forget() should be safe.
2258 *
2259 * AKPM: turn on bforget in journal_forget()!!!
2260 */
2261 for (p = first; p < last; p++) {
2262 u32 nr = le32_to_cpu(*p);
2263 if (nr) {
2264 struct buffer_head *bh;
2265
2266 *p = 0;
2267 bh = sb_find_get_block(inode->i_sb, nr);
2268 ext3_forget(handle, 0, inode, bh, nr);
2269 }
2270 }
2271
2272 ext3_free_blocks(handle, inode, block_to_free, count);
2273}
2274
2275/**
2276 * ext3_free_data - free a list of data blocks
2277 * @handle: handle for this transaction
2278 * @inode: inode we are dealing with
2279 * @this_bh: indirect buffer_head which contains *@first and *@last
2280 * @first: array of block numbers
2281 * @last: points immediately past the end of array
2282 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002283 * We are freeing all blocks referred from that array (numbers are stored as
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2285 *
2286 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2287 * blocks are contiguous then releasing them at one time will only affect one
2288 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2289 * actually use a lot of journal space.
2290 *
2291 * @this_bh will be %NULL if @first and @last point into the inode's direct
2292 * block pointers.
2293 */
2294static void ext3_free_data(handle_t *handle, struct inode *inode,
2295 struct buffer_head *this_bh,
2296 __le32 *first, __le32 *last)
2297{
Mingming Cao43d23f92006-06-25 05:48:07 -07002298 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002299 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2301 corresponding to
2302 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002303 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 __le32 *p; /* Pointer into inode/ind
2305 for current block */
2306 int err;
2307
2308 if (this_bh) { /* For indirect block */
2309 BUFFER_TRACE(this_bh, "get_write_access");
2310 err = ext3_journal_get_write_access(handle, this_bh);
2311 /* Important: if we can't update the indirect pointers
2312 * to the blocks, we can't free them. */
2313 if (err)
2314 return;
2315 }
2316
2317 for (p = first; p < last; p++) {
2318 nr = le32_to_cpu(*p);
2319 if (nr) {
2320 /* accumulate blocks to free if they're contiguous */
2321 if (count == 0) {
2322 block_to_free = nr;
2323 block_to_free_p = p;
2324 count = 1;
2325 } else if (nr == block_to_free + count) {
2326 count++;
2327 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002328 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 block_to_free,
2330 count, block_to_free_p, p);
2331 block_to_free = nr;
2332 block_to_free_p = p;
2333 count = 1;
2334 }
2335 }
2336 }
2337
2338 if (count > 0)
2339 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2340 count, block_to_free_p, p);
2341
2342 if (this_bh) {
2343 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
Duane Griffin3ccc3162008-07-25 01:46:26 -07002344
2345 /*
2346 * The buffer head should have an attached journal head at this
2347 * point. However, if the data is corrupted and an indirect
2348 * block pointed to itself, it would have been detached when
2349 * the block was cleared. Check for this instead of OOPSing.
2350 */
2351 if (bh2jh(this_bh))
2352 ext3_journal_dirty_metadata(handle, this_bh);
2353 else
2354 ext3_error(inode->i_sb, "ext3_free_data",
2355 "circular indirect block detected, "
2356 "inode=%lu, block=%llu",
2357 inode->i_ino,
2358 (unsigned long long)this_bh->b_blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 }
2360}
2361
2362/**
2363 * ext3_free_branches - free an array of branches
2364 * @handle: JBD handle for this transaction
2365 * @inode: inode we are dealing with
2366 * @parent_bh: the buffer_head which contains *@first and *@last
2367 * @first: array of block numbers
2368 * @last: pointer immediately past the end of array
2369 * @depth: depth of the branches to free
2370 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002371 * We are freeing all blocks referred from these branches (numbers are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 * stored as little-endian 32-bit) and updating @inode->i_blocks
2373 * appropriately.
2374 */
2375static void ext3_free_branches(handle_t *handle, struct inode *inode,
2376 struct buffer_head *parent_bh,
2377 __le32 *first, __le32 *last, int depth)
2378{
Mingming Cao43d23f92006-06-25 05:48:07 -07002379 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 __le32 *p;
2381
2382 if (is_handle_aborted(handle))
2383 return;
2384
2385 if (depth--) {
2386 struct buffer_head *bh;
2387 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2388 p = last;
2389 while (--p >= first) {
2390 nr = le32_to_cpu(*p);
2391 if (!nr)
2392 continue; /* A hole */
2393
2394 /* Go read the buffer for the next level down */
2395 bh = sb_bread(inode->i_sb, nr);
2396
2397 /*
2398 * A read failure? Report error and clear slot
2399 * (should be rare).
2400 */
2401 if (!bh) {
2402 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002403 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 inode->i_ino, nr);
2405 continue;
2406 }
2407
2408 /* This zaps the entire block. Bottom up. */
2409 BUFFER_TRACE(bh, "free child branches");
2410 ext3_free_branches(handle, inode, bh,
2411 (__le32*)bh->b_data,
2412 (__le32*)bh->b_data + addr_per_block,
2413 depth);
2414
2415 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 * Everything below this this pointer has been
2417 * released. Now let this top-of-subtree go.
2418 *
2419 * We want the freeing of this indirect block to be
2420 * atomic in the journal with the updating of the
2421 * bitmap block which owns it. So make some room in
2422 * the journal.
2423 *
2424 * We zero the parent pointer *after* freeing its
2425 * pointee in the bitmaps, so if extend_transaction()
2426 * for some reason fails to put the bitmap changes and
2427 * the release into the same transaction, recovery
2428 * will merely complain about releasing a free block,
2429 * rather than leaking blocks.
2430 */
2431 if (is_handle_aborted(handle))
2432 return;
2433 if (try_to_extend_transaction(handle, inode)) {
2434 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002435 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 }
2437
Jan Karaf25f6242010-07-12 21:04:31 +02002438 /*
2439 * We've probably journalled the indirect block several
2440 * times during the truncate. But it's no longer
2441 * needed and we now drop it from the transaction via
2442 * journal_revoke().
2443 *
2444 * That's easy if it's exclusively part of this
2445 * transaction. But if it's part of the committing
2446 * transaction then journal_forget() will simply
2447 * brelse() it. That means that if the underlying
2448 * block is reallocated in ext3_get_block(),
2449 * unmap_underlying_metadata() will find this block
2450 * and will try to get rid of it. damn, damn. Thus
2451 * we don't allow a block to be reallocated until
2452 * a transaction freeing it has fully committed.
2453 *
2454 * We also have to make sure journal replay after a
2455 * crash does not overwrite non-journaled data blocks
2456 * with old metadata when the block got reallocated for
2457 * data. Thus we have to store a revoke record for a
2458 * block in the same transaction in which we free the
2459 * block.
2460 */
2461 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2462
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 ext3_free_blocks(handle, inode, nr, 1);
2464
2465 if (parent_bh) {
2466 /*
2467 * The block which we have just freed is
2468 * pointed to by an indirect block: journal it
2469 */
2470 BUFFER_TRACE(parent_bh, "get_write_access");
2471 if (!ext3_journal_get_write_access(handle,
2472 parent_bh)){
2473 *p = 0;
2474 BUFFER_TRACE(parent_bh,
2475 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002476 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 parent_bh);
2478 }
2479 }
2480 }
2481 } else {
2482 /* We have reached the bottom of the tree. */
2483 BUFFER_TRACE(parent_bh, "free data blocks");
2484 ext3_free_data(handle, inode, parent_bh, first, last);
2485 }
2486}
2487
Duane Griffinae76dd92008-07-25 01:46:23 -07002488int ext3_can_truncate(struct inode *inode)
2489{
Duane Griffinae76dd92008-07-25 01:46:23 -07002490 if (S_ISREG(inode->i_mode))
2491 return 1;
2492 if (S_ISDIR(inode->i_mode))
2493 return 1;
2494 if (S_ISLNK(inode->i_mode))
2495 return !ext3_inode_is_fast_symlink(inode);
2496 return 0;
2497}
2498
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499/*
2500 * ext3_truncate()
2501 *
2502 * We block out ext3_get_block() block instantiations across the entire
2503 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2504 * simultaneously on behalf of the same inode.
2505 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002506 * As we work through the truncate and commit bits of it to the journal there
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 * is one core, guiding principle: the file's tree must always be consistent on
2508 * disk. We must be able to restart the truncate after a crash.
2509 *
2510 * The file's tree may be transiently inconsistent in memory (although it
2511 * probably isn't), but whenever we close off and commit a journal transaction,
2512 * the contents of (the filesystem + the journal) must be consistent and
2513 * restartable. It's pretty simple, really: bottom up, right to left (although
2514 * left-to-right works OK too).
2515 *
2516 * Note that at recovery time, journal replay occurs *before* the restart of
2517 * truncate against the orphan inode list.
2518 *
2519 * The committed inode has the new, desired i_size (which is the same as
2520 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2521 * that this inode's truncate did not complete and it will again call
2522 * ext3_truncate() to have another go. So there will be instantiated blocks
2523 * to the right of the truncation point in a crashed ext3 filesystem. But
2524 * that's fine - as long as they are linked from the inode, the post-crash
2525 * ext3_truncate() run will find them and release them.
2526 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002527void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528{
2529 handle_t *handle;
2530 struct ext3_inode_info *ei = EXT3_I(inode);
2531 __le32 *i_data = ei->i_data;
2532 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 int offsets[4];
2534 Indirect chain[4];
2535 Indirect *partial;
2536 __le32 nr = 0;
2537 int n;
2538 long last_block;
2539 unsigned blocksize = inode->i_sb->s_blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002541 trace_ext3_truncate_enter(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Duane Griffinae76dd92008-07-25 01:46:23 -07002543 if (!ext3_can_truncate(inode))
Jan Karaef436182009-06-17 16:26:24 -07002544 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002546 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
Jan Kara9df93932010-01-06 21:58:48 +01002547 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002548
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 handle = start_transaction(inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02002550 if (IS_ERR(handle))
Jan Karaef436182009-06-17 16:26:24 -07002551 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
2553 last_block = (inode->i_size + blocksize-1)
2554 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2556 if (n == 0)
2557 goto out_stop; /* error */
2558
2559 /*
2560 * OK. This truncate is going to happen. We add the inode to the
2561 * orphan list, so that if this truncate spans multiple transactions,
2562 * and we crash, we will resume the truncate when the filesystem
2563 * recovers. It also marks the inode dirty, to catch the new size.
2564 *
2565 * Implication: the file must always be in a sane, consistent
2566 * truncatable state while each transaction commits.
2567 */
2568 if (ext3_orphan_add(handle, inode))
2569 goto out_stop;
2570
2571 /*
2572 * The orphan list entry will now protect us from any crash which
2573 * occurs before the truncate completes, so it is now safe to propagate
2574 * the new, shorter inode size (held for now in i_size) into the
2575 * on-disk inode. We do this via i_disksize, which is the value which
2576 * ext3 *really* writes onto the disk inode.
2577 */
2578 ei->i_disksize = inode->i_size;
2579
2580 /*
2581 * From here we block out all ext3_get_block() callers who want to
2582 * modify the block allocation tree.
2583 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002584 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
2586 if (n == 1) { /* direct blocks */
2587 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2588 i_data + EXT3_NDIR_BLOCKS);
2589 goto do_indirects;
2590 }
2591
2592 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2593 /* Kill the top of shared branch (not detached) */
2594 if (nr) {
2595 if (partial == chain) {
2596 /* Shared branch grows from the inode */
2597 ext3_free_branches(handle, inode, NULL,
2598 &nr, &nr+1, (chain+n-1) - partial);
2599 *partial->p = 0;
2600 /*
2601 * We mark the inode dirty prior to restart,
2602 * and prior to stop. No need for it here.
2603 */
2604 } else {
2605 /* Shared branch grows from an indirect block */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 ext3_free_branches(handle, inode, partial->bh,
2607 partial->p,
2608 partial->p+1, (chain+n-1) - partial);
2609 }
2610 }
2611 /* Clear the ends of indirect blocks on the shared branch */
2612 while (partial > chain) {
2613 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2614 (__le32*)partial->bh->b_data+addr_per_block,
2615 (chain+n-1) - partial);
2616 BUFFER_TRACE(partial->bh, "call brelse");
2617 brelse (partial->bh);
2618 partial--;
2619 }
2620do_indirects:
2621 /* Kill the remaining (whole) subtrees */
2622 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002623 default:
2624 nr = i_data[EXT3_IND_BLOCK];
2625 if (nr) {
2626 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2627 i_data[EXT3_IND_BLOCK] = 0;
2628 }
2629 case EXT3_IND_BLOCK:
2630 nr = i_data[EXT3_DIND_BLOCK];
2631 if (nr) {
2632 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2633 i_data[EXT3_DIND_BLOCK] = 0;
2634 }
2635 case EXT3_DIND_BLOCK:
2636 nr = i_data[EXT3_TIND_BLOCK];
2637 if (nr) {
2638 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2639 i_data[EXT3_TIND_BLOCK] = 0;
2640 }
2641 case EXT3_TIND_BLOCK:
2642 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 }
2644
2645 ext3_discard_reservation(inode);
2646
Arjan van de Ven97461512006-03-23 03:00:42 -08002647 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2649 ext3_mark_inode_dirty(handle, inode);
2650
Andrew Mortond6859bf2006-03-26 01:38:03 -08002651 /*
2652 * In a multi-transaction truncate, we only make the final transaction
2653 * synchronous
2654 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if (IS_SYNC(inode))
2656 handle->h_sync = 1;
2657out_stop:
2658 /*
2659 * If this was a simple ftruncate(), and the file will remain alive
2660 * then we need to clear up the orphan record which we created above.
2661 * However, if this was a real unlink then we were called by
Al Viroac14a952010-06-06 07:08:19 -04002662 * ext3_evict_inode(), and we allow that function to clean up the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 * orphan info for us.
2664 */
2665 if (inode->i_nlink)
2666 ext3_orphan_del(handle, inode);
2667
2668 ext3_journal_stop(handle);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002669 trace_ext3_truncate_exit(inode);
Jan Karaef436182009-06-17 16:26:24 -07002670 return;
2671out_notrans:
2672 /*
2673 * Delete the inode from orphan list so that it doesn't stay there
2674 * forever and trigger assertion on umount.
2675 */
2676 if (inode->i_nlink)
2677 ext3_orphan_del(NULL, inode);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002678 trace_ext3_truncate_exit(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679}
2680
Mingming Cao43d23f92006-06-25 05:48:07 -07002681static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 unsigned long ino, struct ext3_iloc *iloc)
2683{
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002684 unsigned long block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002685 unsigned long offset;
2686 ext3_fsblk_t block;
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002687 struct ext3_group_desc *gdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
Neil Brown2ccb48e2006-07-30 03:03:01 -07002689 if (!ext3_valid_inum(sb, ino)) {
2690 /*
2691 * This error is already checked for in namei.c unless we are
2692 * looking at an NFS filehandle, in which case no error
2693 * report is needed
2694 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 return 0;
2696 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002697
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002699 gdp = ext3_get_group_desc(sb, block_group, NULL);
2700 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 /*
2703 * Figure out the offset within the block group inode table
2704 */
2705 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2706 EXT3_INODE_SIZE(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002707 block = le32_to_cpu(gdp->bg_inode_table) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2709
2710 iloc->block_group = block_group;
2711 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2712 return block;
2713}
2714
2715/*
2716 * ext3_get_inode_loc returns with an extra refcount against the inode's
2717 * underlying buffer_head on success. If 'in_mem' is true, we have all
2718 * data in memory that is needed to recreate the on-disk version of this
2719 * inode.
2720 */
2721static int __ext3_get_inode_loc(struct inode *inode,
2722 struct ext3_iloc *iloc, int in_mem)
2723{
Mingming Cao43d23f92006-06-25 05:48:07 -07002724 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 struct buffer_head *bh;
2726
2727 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2728 if (!block)
2729 return -EIO;
2730
2731 bh = sb_getblk(inode->i_sb, block);
2732 if (!bh) {
2733 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2734 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002735 "inode=%lu, block="E3FSBLK,
2736 inode->i_ino, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 return -EIO;
2738 }
2739 if (!buffer_uptodate(bh)) {
2740 lock_buffer(bh);
Hidehiro Kawai95450f52008-07-25 01:46:24 -07002741
2742 /*
2743 * If the buffer has the write error flag, we have failed
2744 * to write out another inode in the same block. In this
2745 * case, we don't have to read the block because we may
2746 * read the old inode data successfully.
2747 */
2748 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2749 set_buffer_uptodate(bh);
2750
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 if (buffer_uptodate(bh)) {
2752 /* someone brought it uptodate while we waited */
2753 unlock_buffer(bh);
2754 goto has_buffer;
2755 }
2756
2757 /*
2758 * If we have all information of the inode in memory and this
2759 * is the only valid inode in the block, we need not read the
2760 * block.
2761 */
2762 if (in_mem) {
2763 struct buffer_head *bitmap_bh;
2764 struct ext3_group_desc *desc;
2765 int inodes_per_buffer;
2766 int inode_offset, i;
2767 int block_group;
2768 int start;
2769
2770 block_group = (inode->i_ino - 1) /
2771 EXT3_INODES_PER_GROUP(inode->i_sb);
2772 inodes_per_buffer = bh->b_size /
2773 EXT3_INODE_SIZE(inode->i_sb);
2774 inode_offset = ((inode->i_ino - 1) %
2775 EXT3_INODES_PER_GROUP(inode->i_sb));
2776 start = inode_offset & ~(inodes_per_buffer - 1);
2777
2778 /* Is the inode bitmap in cache? */
2779 desc = ext3_get_group_desc(inode->i_sb,
2780 block_group, NULL);
2781 if (!desc)
2782 goto make_io;
2783
2784 bitmap_bh = sb_getblk(inode->i_sb,
2785 le32_to_cpu(desc->bg_inode_bitmap));
2786 if (!bitmap_bh)
2787 goto make_io;
2788
2789 /*
2790 * If the inode bitmap isn't in cache then the
2791 * optimisation may end up performing two reads instead
2792 * of one, so skip it.
2793 */
2794 if (!buffer_uptodate(bitmap_bh)) {
2795 brelse(bitmap_bh);
2796 goto make_io;
2797 }
2798 for (i = start; i < start + inodes_per_buffer; i++) {
2799 if (i == inode_offset)
2800 continue;
2801 if (ext3_test_bit(i, bitmap_bh->b_data))
2802 break;
2803 }
2804 brelse(bitmap_bh);
2805 if (i == start + inodes_per_buffer) {
2806 /* all other inodes are free, so skip I/O */
2807 memset(bh->b_data, 0, bh->b_size);
2808 set_buffer_uptodate(bh);
2809 unlock_buffer(bh);
2810 goto has_buffer;
2811 }
2812 }
2813
2814make_io:
2815 /*
2816 * There are other valid inodes in the buffer, this inode
2817 * has in-inode xattrs, or we don't have this inode in memory.
2818 * Read the block from disk.
2819 */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002820 trace_ext3_load_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 get_bh(bh);
2822 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02002823 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 wait_on_buffer(bh);
2825 if (!buffer_uptodate(bh)) {
2826 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2827 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002828 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 inode->i_ino, block);
2830 brelse(bh);
2831 return -EIO;
2832 }
2833 }
2834has_buffer:
2835 iloc->bh = bh;
2836 return 0;
2837}
2838
2839int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2840{
2841 /* We have all inode data except xattrs in memory here. */
2842 return __ext3_get_inode_loc(inode, iloc,
Jan Kara9df93932010-01-06 21:58:48 +01002843 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844}
2845
2846void ext3_set_inode_flags(struct inode *inode)
2847{
2848 unsigned int flags = EXT3_I(inode)->i_flags;
2849
2850 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2851 if (flags & EXT3_SYNC_FL)
2852 inode->i_flags |= S_SYNC;
2853 if (flags & EXT3_APPEND_FL)
2854 inode->i_flags |= S_APPEND;
2855 if (flags & EXT3_IMMUTABLE_FL)
2856 inode->i_flags |= S_IMMUTABLE;
2857 if (flags & EXT3_NOATIME_FL)
2858 inode->i_flags |= S_NOATIME;
2859 if (flags & EXT3_DIRSYNC_FL)
2860 inode->i_flags |= S_DIRSYNC;
2861}
2862
Jan Kara28be5ab2007-05-08 00:30:33 -07002863/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2864void ext3_get_inode_flags(struct ext3_inode_info *ei)
2865{
2866 unsigned int flags = ei->vfs_inode.i_flags;
2867
2868 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2869 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2870 if (flags & S_SYNC)
2871 ei->i_flags |= EXT3_SYNC_FL;
2872 if (flags & S_APPEND)
2873 ei->i_flags |= EXT3_APPEND_FL;
2874 if (flags & S_IMMUTABLE)
2875 ei->i_flags |= EXT3_IMMUTABLE_FL;
2876 if (flags & S_NOATIME)
2877 ei->i_flags |= EXT3_NOATIME_FL;
2878 if (flags & S_DIRSYNC)
2879 ei->i_flags |= EXT3_DIRSYNC_FL;
2880}
2881
David Howells473043d2008-02-07 00:15:36 -08002882struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883{
2884 struct ext3_iloc iloc;
2885 struct ext3_inode *raw_inode;
David Howells473043d2008-02-07 00:15:36 -08002886 struct ext3_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 struct buffer_head *bh;
David Howells473043d2008-02-07 00:15:36 -08002888 struct inode *inode;
Jan Karafe8bc912009-10-16 19:26:15 +02002889 journal_t *journal = EXT3_SB(sb)->s_journal;
2890 transaction_t *transaction;
David Howells473043d2008-02-07 00:15:36 -08002891 long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 int block;
Eric W. Biederman15232992012-02-07 15:41:24 -08002893 uid_t i_uid;
2894 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
David Howells473043d2008-02-07 00:15:36 -08002896 inode = iget_locked(sb, ino);
2897 if (!inode)
2898 return ERR_PTR(-ENOMEM);
2899 if (!(inode->i_state & I_NEW))
2900 return inode;
2901
2902 ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 ei->i_block_alloc_info = NULL;
2904
David Howells473043d2008-02-07 00:15:36 -08002905 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2906 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 goto bad_inode;
2908 bh = iloc.bh;
2909 raw_inode = ext3_raw_inode(&iloc);
2910 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman15232992012-02-07 15:41:24 -08002911 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2912 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 if(!(test_opt (inode->i_sb, NO_UID32))) {
Eric W. Biederman15232992012-02-07 15:41:24 -08002914 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2915 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 }
Eric W. Biederman15232992012-02-07 15:41:24 -08002917 i_uid_write(inode, i_uid);
2918 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002919 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002921 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2922 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2923 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2925
Linus Torvaldsde329822010-03-29 14:30:19 -07002926 ei->i_state_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 ei->i_dir_start_lookup = 0;
2928 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2929 /* We now have enough fields to check if the inode was active or not.
2930 * This is needed because nfsd might try to access dead inodes
2931 * the test is that same one that e2fsck uses
2932 * NeilBrown 1999oct15
2933 */
2934 if (inode->i_nlink == 0) {
2935 if (inode->i_mode == 0 ||
2936 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2937 /* this inode is deleted */
2938 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002939 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 goto bad_inode;
2941 }
2942 /* The only unlinked inodes we let through here have
2943 * valid i_mode and are being read by the orphan
2944 * recovery code: that's fine, we're about to complete
2945 * the process of deleting those. */
2946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2948 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2949#ifdef EXT3_FRAGMENTS
2950 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2951 ei->i_frag_no = raw_inode->i_frag;
2952 ei->i_frag_size = raw_inode->i_fsize;
2953#endif
2954 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2955 if (!S_ISREG(inode->i_mode)) {
2956 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2957 } else {
2958 inode->i_size |=
2959 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2960 }
2961 ei->i_disksize = inode->i_size;
2962 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2963 ei->i_block_group = iloc.block_group;
2964 /*
2965 * NOTE! The in-memory inode i_data array is in little-endian order
2966 * even on big-endian machines: we do NOT byteswap the block numbers!
2967 */
2968 for (block = 0; block < EXT3_N_BLOCKS; block++)
2969 ei->i_data[block] = raw_inode->i_block[block];
2970 INIT_LIST_HEAD(&ei->i_orphan);
2971
Jan Karafe8bc912009-10-16 19:26:15 +02002972 /*
2973 * Set transaction id's of transactions that have to be committed
2974 * to finish f[data]sync. We set them to currently running transaction
2975 * as we cannot be sure that the inode or some of its metadata isn't
2976 * part of the transaction - the inode could have been reclaimed and
2977 * now it is reread from disk.
2978 */
2979 if (journal) {
2980 tid_t tid;
2981
2982 spin_lock(&journal->j_state_lock);
2983 if (journal->j_running_transaction)
2984 transaction = journal->j_running_transaction;
2985 else
2986 transaction = journal->j_committing_transaction;
2987 if (transaction)
2988 tid = transaction->t_tid;
2989 else
2990 tid = journal->j_commit_sequence;
2991 spin_unlock(&journal->j_state_lock);
2992 atomic_set(&ei->i_sync_tid, tid);
2993 atomic_set(&ei->i_datasync_tid, tid);
2994 }
2995
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2997 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2998 /*
2999 * When mke2fs creates big inodes it does not zero out
3000 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
3001 * so ignore those first few inodes.
3002 */
3003 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3004 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07003005 EXT3_INODE_SIZE(inode->i_sb)) {
3006 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08003007 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 goto bad_inode;
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07003009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 if (ei->i_extra_isize == 0) {
3011 /* The extra space is currently unused. Use it. */
3012 ei->i_extra_isize = sizeof(struct ext3_inode) -
3013 EXT3_GOOD_OLD_INODE_SIZE;
3014 } else {
3015 __le32 *magic = (void *)raw_inode +
3016 EXT3_GOOD_OLD_INODE_SIZE +
3017 ei->i_extra_isize;
3018 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
Jan Kara9df93932010-01-06 21:58:48 +01003019 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 }
3021 } else
3022 ei->i_extra_isize = 0;
3023
3024 if (S_ISREG(inode->i_mode)) {
3025 inode->i_op = &ext3_file_inode_operations;
3026 inode->i_fop = &ext3_file_operations;
3027 ext3_set_aops(inode);
3028 } else if (S_ISDIR(inode->i_mode)) {
3029 inode->i_op = &ext3_dir_inode_operations;
3030 inode->i_fop = &ext3_dir_operations;
3031 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffinb5ed3112008-12-19 20:47:14 +00003032 if (ext3_inode_is_fast_symlink(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 inode->i_op = &ext3_fast_symlink_inode_operations;
Duane Griffinb5ed3112008-12-19 20:47:14 +00003034 nd_terminate_link(ei->i_data, inode->i_size,
3035 sizeof(ei->i_data) - 1);
3036 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 inode->i_op = &ext3_symlink_inode_operations;
3038 ext3_set_aops(inode);
3039 }
3040 } else {
3041 inode->i_op = &ext3_special_inode_operations;
3042 if (raw_inode->i_block[0])
3043 init_special_inode(inode, inode->i_mode,
3044 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003045 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 init_special_inode(inode, inode->i_mode,
3047 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3048 }
3049 brelse (iloc.bh);
3050 ext3_set_inode_flags(inode);
David Howells473043d2008-02-07 00:15:36 -08003051 unlock_new_inode(inode);
3052 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
3054bad_inode:
David Howells473043d2008-02-07 00:15:36 -08003055 iget_failed(inode);
3056 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057}
3058
3059/*
3060 * Post the struct inode info into an on-disk inode location in the
3061 * buffer-cache. This gobbles the caller's reference to the
3062 * buffer_head in the inode location struct.
3063 *
3064 * The caller must have write access to iloc->bh.
3065 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003066static int ext3_do_update_inode(handle_t *handle,
3067 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 struct ext3_iloc *iloc)
3069{
3070 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3071 struct ext3_inode_info *ei = EXT3_I(inode);
3072 struct buffer_head *bh = iloc->bh;
3073 int err = 0, rc, block;
Jan Kara156bddd2012-09-03 16:50:42 +02003074 int need_datasync = 0;
3075 __le32 disksize;
Eric W. Biederman15232992012-02-07 15:41:24 -08003076 uid_t i_uid;
3077 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078
Chris Mason4f003fd2009-09-08 00:22:14 +02003079again:
3080 /* we can't allow multiple procs in here at once, its a bit racey */
3081 lock_buffer(bh);
3082
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 /* For fields not not tracking in the in-memory inode,
3084 * initialise them to zero for new inodes. */
Jan Kara9df93932010-01-06 21:58:48 +01003085 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3087
Jan Kara28be5ab2007-05-08 00:30:33 -07003088 ext3_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman15232992012-02-07 15:41:24 -08003090 i_uid = i_uid_read(inode);
3091 i_gid = i_gid_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 if(!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman15232992012-02-07 15:41:24 -08003093 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3094 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095/*
3096 * Fix up interoperability with old kernels. Otherwise, old inodes get
3097 * re-used with the upper 16 bits of the uid/gid intact
3098 */
3099 if(!ei->i_dtime) {
3100 raw_inode->i_uid_high =
Eric W. Biederman15232992012-02-07 15:41:24 -08003101 cpu_to_le16(high_16_bits(i_uid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 raw_inode->i_gid_high =
Eric W. Biederman15232992012-02-07 15:41:24 -08003103 cpu_to_le16(high_16_bits(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 } else {
3105 raw_inode->i_uid_high = 0;
3106 raw_inode->i_gid_high = 0;
3107 }
3108 } else {
3109 raw_inode->i_uid_low =
Eric W. Biederman15232992012-02-07 15:41:24 -08003110 cpu_to_le16(fs_high2lowuid(i_uid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 raw_inode->i_gid_low =
Eric W. Biederman15232992012-02-07 15:41:24 -08003112 cpu_to_le16(fs_high2lowgid(i_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 raw_inode->i_uid_high = 0;
3114 raw_inode->i_gid_high = 0;
3115 }
3116 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Jan Kara156bddd2012-09-03 16:50:42 +02003117 disksize = cpu_to_le32(ei->i_disksize);
3118 if (disksize != raw_inode->i_size) {
3119 need_datasync = 1;
3120 raw_inode->i_size = disksize;
3121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3123 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3124 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3125 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3126 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3127 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3128#ifdef EXT3_FRAGMENTS
3129 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3130 raw_inode->i_frag = ei->i_frag_no;
3131 raw_inode->i_fsize = ei->i_frag_size;
3132#endif
3133 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3134 if (!S_ISREG(inode->i_mode)) {
3135 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3136 } else {
Jan Kara156bddd2012-09-03 16:50:42 +02003137 disksize = cpu_to_le32(ei->i_disksize >> 32);
3138 if (disksize != raw_inode->i_size_high) {
3139 raw_inode->i_size_high = disksize;
3140 need_datasync = 1;
3141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 if (ei->i_disksize > 0x7fffffffULL) {
3143 struct super_block *sb = inode->i_sb;
3144 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3145 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3146 EXT3_SB(sb)->s_es->s_rev_level ==
3147 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3148 /* If this is the first large file
3149 * created, add a flag to the superblock.
3150 */
Chris Mason4f003fd2009-09-08 00:22:14 +02003151 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 err = ext3_journal_get_write_access(handle,
3153 EXT3_SB(sb)->s_sbh);
3154 if (err)
3155 goto out_brelse;
Chris Mason4f003fd2009-09-08 00:22:14 +02003156
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 ext3_update_dynamic_rev(sb);
3158 EXT3_SET_RO_COMPAT_FEATURE(sb,
3159 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160 handle->h_sync = 1;
3161 err = ext3_journal_dirty_metadata(handle,
3162 EXT3_SB(sb)->s_sbh);
Chris Mason4f003fd2009-09-08 00:22:14 +02003163 /* get our lock and start over */
3164 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 }
3166 }
3167 }
3168 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3169 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3170 if (old_valid_dev(inode->i_rdev)) {
3171 raw_inode->i_block[0] =
3172 cpu_to_le32(old_encode_dev(inode->i_rdev));
3173 raw_inode->i_block[1] = 0;
3174 } else {
3175 raw_inode->i_block[0] = 0;
3176 raw_inode->i_block[1] =
3177 cpu_to_le32(new_encode_dev(inode->i_rdev));
3178 raw_inode->i_block[2] = 0;
3179 }
3180 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3181 raw_inode->i_block[block] = ei->i_data[block];
3182
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07003183 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3185
3186 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Chris Mason4f003fd2009-09-08 00:22:14 +02003187 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 rc = ext3_journal_dirty_metadata(handle, bh);
3189 if (!err)
3190 err = rc;
Jan Kara9df93932010-01-06 21:58:48 +01003191 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192
Jan Karafe8bc912009-10-16 19:26:15 +02003193 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
Jan Kara156bddd2012-09-03 16:50:42 +02003194 if (need_datasync)
3195 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196out_brelse:
3197 brelse (bh);
3198 ext3_std_error(inode->i_sb, err);
3199 return err;
3200}
3201
3202/*
3203 * ext3_write_inode()
3204 *
3205 * We are called from a few places:
3206 *
3207 * - Within generic_file_write() for O_SYNC files.
3208 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003209 * transaction to commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 *
3211 * - Within sys_sync(), kupdate and such.
3212 * We wait on commit, if tol to.
3213 *
3214 * - Within prune_icache() (PF_MEMALLOC == true)
3215 * Here we simply return. We can't afford to block kswapd on the
3216 * journal commit.
3217 *
3218 * In all cases it is actually safe for us to return without doing anything,
3219 * because the inode has been copied into a raw inode buffer in
3220 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3221 * knfsd.
3222 *
3223 * Note that we are absolutely dependent upon all inode dirtiers doing the
3224 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3225 * which we are interested.
3226 *
3227 * It would be a bug for them to not do this. The code:
3228 *
3229 * mark_inode_dirty(inode)
3230 * stuff();
3231 * inode->i_size = expr;
3232 *
3233 * is in error because a kswapd-driven write_inode() could occur while
3234 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3235 * will no longer be on the superblock's dirty inode list.
3236 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003237int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238{
3239 if (current->flags & PF_MEMALLOC)
3240 return 0;
3241
3242 if (ext3_journal_current_handle()) {
Jose R. Santos9ad163a2007-10-18 23:39:23 -07003243 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 dump_stack();
3245 return -EIO;
3246 }
3247
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003248 if (wbc->sync_mode != WB_SYNC_ALL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 return 0;
3250
3251 return ext3_force_commit(inode->i_sb);
3252}
3253
3254/*
3255 * ext3_setattr()
3256 *
3257 * Called from notify_change.
3258 *
3259 * We want to trap VFS attempts to truncate the file as soon as
3260 * possible. In particular, we want to make sure that when the VFS
3261 * shrinks i_size, we put the inode on the orphan list and modify
3262 * i_disksize immediately, so that during the subsequent flushing of
3263 * dirty pages and freeing of disk blocks, we can guarantee that any
3264 * commit will leave the blocks being flushed in an unused state on
3265 * disk. (On recovery, the inode will get truncated and the blocks will
3266 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003267 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 *
3269 * Called with inode->sem down.
3270 */
3271int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3272{
3273 struct inode *inode = dentry->d_inode;
3274 int error, rc = 0;
3275 const unsigned int ia_valid = attr->ia_valid;
3276
3277 error = inode_change_ok(inode, attr);
3278 if (error)
3279 return error;
3280
Dmitry Monakhov12755622010-04-08 22:04:20 +04003281 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05003282 dquot_initialize(inode);
Eric W. Biederman15232992012-02-07 15:41:24 -08003283 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3284 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 handle_t *handle;
3286
3287 /* (user+group)*(old+new) structure, inode write (sb,
3288 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003289 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3290 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 if (IS_ERR(handle)) {
3292 error = PTR_ERR(handle);
3293 goto err_out;
3294 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05003295 error = dquot_transfer(inode, attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 if (error) {
3297 ext3_journal_stop(handle);
3298 return error;
3299 }
3300 /* Update corresponding info in inode so that everything is in
3301 * one transaction */
3302 if (attr->ia_valid & ATTR_UID)
3303 inode->i_uid = attr->ia_uid;
3304 if (attr->ia_valid & ATTR_GID)
3305 inode->i_gid = attr->ia_gid;
3306 error = ext3_mark_inode_dirty(handle, inode);
3307 ext3_journal_stop(handle);
3308 }
3309
Christoph Hellwig562c72a2011-06-24 14:29:45 -04003310 if (attr->ia_valid & ATTR_SIZE)
3311 inode_dio_wait(inode);
3312
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 if (S_ISREG(inode->i_mode) &&
3314 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3315 handle_t *handle;
3316
3317 handle = ext3_journal_start(inode, 3);
3318 if (IS_ERR(handle)) {
3319 error = PTR_ERR(handle);
3320 goto err_out;
3321 }
3322
3323 error = ext3_orphan_add(handle, inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02003324 if (error) {
3325 ext3_journal_stop(handle);
3326 goto err_out;
3327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 EXT3_I(inode)->i_disksize = attr->ia_size;
Jan Karaee3e77f2011-06-03 21:58:11 +02003329 error = ext3_mark_inode_dirty(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 ext3_journal_stop(handle);
Jan Karaee3e77f2011-06-03 21:58:11 +02003331 if (error) {
3332 /* Some hard fs error must have happened. Bail out. */
3333 ext3_orphan_del(NULL, inode);
3334 goto err_out;
3335 }
3336 rc = ext3_block_truncate_page(inode, attr->ia_size);
3337 if (rc) {
3338 /* Cleanup orphan list and exit */
3339 handle = ext3_journal_start(inode, 3);
3340 if (IS_ERR(handle)) {
3341 ext3_orphan_del(NULL, inode);
3342 goto err_out;
3343 }
3344 ext3_orphan_del(handle, inode);
3345 ext3_journal_stop(handle);
3346 goto err_out;
3347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 }
3349
Christoph Hellwig10257742010-06-04 11:30:02 +02003350 if ((attr->ia_valid & ATTR_SIZE) &&
3351 attr->ia_size != i_size_read(inode)) {
Jan Kara40680f2f2011-05-24 22:24:47 +02003352 truncate_setsize(inode, attr->ia_size);
3353 ext3_truncate(inode);
Christoph Hellwig10257742010-06-04 11:30:02 +02003354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355
Christoph Hellwig10257742010-06-04 11:30:02 +02003356 setattr_copy(inode, attr);
3357 mark_inode_dirty(inode);
3358
3359 if (ia_valid & ATTR_MODE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 rc = ext3_acl_chmod(inode);
3361
3362err_out:
3363 ext3_std_error(inode->i_sb, error);
3364 if (!error)
3365 error = rc;
3366 return error;
3367}
3368
3369
3370/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003371 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 *
3373 * With N blocks per page, it may be:
3374 * N data blocks
3375 * 2 indirect block
3376 * 2 dindirect
3377 * 1 tindirect
3378 * N+5 bitmap blocks (from the above)
3379 * N+5 group descriptor summary blocks
3380 * 1 inode block
3381 * 1 superblock.
3382 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3383 *
3384 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3385 *
3386 * With ordered or writeback data it's the same, less the N data blocks.
3387 *
3388 * If the inode's direct blocks can hold an integral number of pages then a
3389 * page cannot straddle two indirect blocks, and we can only touch one indirect
3390 * and dindirect block, and the "5" above becomes "3".
3391 *
3392 * This still overestimates under most circumstances. If we were to pass the
3393 * start and end offsets in here as well we could do block_to_path() on each
3394 * block and work out the exact number of indirects which are touched. Pah.
3395 */
3396
3397static int ext3_writepage_trans_blocks(struct inode *inode)
3398{
3399 int bpp = ext3_journal_blocks_per_page(inode);
3400 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3401 int ret;
3402
3403 if (ext3_should_journal_data(inode))
3404 ret = 3 * (bpp + indirects) + 2;
3405 else
Yongqiang Yang523334b2011-03-24 08:48:39 +08003406 ret = 2 * (bpp + indirects) + indirects + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407
3408#ifdef CONFIG_QUOTA
Christoph Hellwig871a2932010-03-03 09:05:07 -05003409 /* We know that structure was already allocated during dquot_initialize so
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410 * we will be updating only the data blocks + inodes */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003411 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412#endif
3413
3414 return ret;
3415}
3416
3417/*
3418 * The caller must have previously called ext3_reserve_inode_write().
3419 * Give this, we know that the caller already has write access to iloc->bh.
3420 */
3421int ext3_mark_iloc_dirty(handle_t *handle,
3422 struct inode *inode, struct ext3_iloc *iloc)
3423{
3424 int err = 0;
3425
3426 /* the do_update_inode consumes one bh->b_count */
3427 get_bh(iloc->bh);
3428
3429 /* ext3_do_update_inode() does journal_dirty_metadata */
3430 err = ext3_do_update_inode(handle, inode, iloc);
3431 put_bh(iloc->bh);
3432 return err;
3433}
3434
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003435/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003437 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 */
3439
3440int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003441ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 struct ext3_iloc *iloc)
3443{
3444 int err = 0;
3445 if (handle) {
3446 err = ext3_get_inode_loc(inode, iloc);
3447 if (!err) {
3448 BUFFER_TRACE(iloc->bh, "get_write_access");
3449 err = ext3_journal_get_write_access(handle, iloc->bh);
3450 if (err) {
3451 brelse(iloc->bh);
3452 iloc->bh = NULL;
3453 }
3454 }
3455 }
3456 ext3_std_error(inode->i_sb, err);
3457 return err;
3458}
3459
3460/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003461 * What we do here is to mark the in-core inode as clean with respect to inode
3462 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 * This means that the in-core inode may be reaped by prune_icache
3464 * without having to perform any I/O. This is a very good thing,
3465 * because *any* task may call prune_icache - even ones which
3466 * have a transaction open against a different journal.
3467 *
3468 * Is this cheating? Not really. Sure, we haven't written the
3469 * inode out, but prune_icache isn't a user-visible syncing function.
3470 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3471 * we start and wait on commits.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 */
3473int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3474{
3475 struct ext3_iloc iloc;
3476 int err;
3477
3478 might_sleep();
Lukas Czerner785c4bc2011-05-23 18:33:01 +02003479 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 err = ext3_reserve_inode_write(handle, inode, &iloc);
3481 if (!err)
3482 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3483 return err;
3484}
3485
3486/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003487 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 *
3489 * We're really interested in the case where a file is being extended.
3490 * i_size has been changed by generic_commit_write() and we thus need
3491 * to include the updated inode in the current transaction.
3492 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003493 * Also, dquot_alloc_space() will always dirty the inode when blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 * are allocated to the file.
3495 *
3496 * If the inode is marked synchronous, we don't honour that here - doing
3497 * so would cause a commit on atime updates, which we don't bother doing.
3498 * We handle synchronous inodes at the highest possible level.
3499 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04003500void ext3_dirty_inode(struct inode *inode, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501{
3502 handle_t *current_handle = ext3_journal_current_handle();
3503 handle_t *handle;
3504
3505 handle = ext3_journal_start(inode, 2);
3506 if (IS_ERR(handle))
3507 goto out;
3508 if (current_handle &&
3509 current_handle->h_transaction != handle->h_transaction) {
3510 /* This task has a transaction open against a different fs */
3511 printk(KERN_EMERG "%s: transactions do not match!\n",
Harvey Harrisone05b6b52008-04-28 02:16:15 -07003512 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 } else {
3514 jbd_debug(5, "marking dirty. outer handle=%p\n",
3515 current_handle);
3516 ext3_mark_inode_dirty(handle, inode);
3517 }
3518 ext3_journal_stop(handle);
3519out:
3520 return;
3521}
3522
Andrew Mortond6859bf2006-03-26 01:38:03 -08003523#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003524/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 * Bind an inode's backing buffer_head into this transaction, to prevent
3526 * it from being flushed to disk early. Unlike
3527 * ext3_reserve_inode_write, this leaves behind no bh reference and
3528 * returns no iloc structure, so the caller needs to repeat the iloc
3529 * lookup to mark the inode dirty later.
3530 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003531static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532{
3533 struct ext3_iloc iloc;
3534
3535 int err = 0;
3536 if (handle) {
3537 err = ext3_get_inode_loc(inode, &iloc);
3538 if (!err) {
3539 BUFFER_TRACE(iloc.bh, "get_write_access");
3540 err = journal_get_write_access(handle, iloc.bh);
3541 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003542 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543 iloc.bh);
3544 brelse(iloc.bh);
3545 }
3546 }
3547 ext3_std_error(inode->i_sb, err);
3548 return err;
3549}
3550#endif
3551
3552int ext3_change_inode_journal_flag(struct inode *inode, int val)
3553{
3554 journal_t *journal;
3555 handle_t *handle;
3556 int err;
3557
3558 /*
3559 * We have to be very careful here: changing a data block's
3560 * journaling status dynamically is dangerous. If we write a
3561 * data block to the journal, change the status and then delete
3562 * that block, we risk forgetting to revoke the old log record
3563 * from the journal and so a subsequent replay can corrupt data.
3564 * So, first we make sure that the journal is empty and that
3565 * nobody is changing anything.
3566 */
3567
3568 journal = EXT3_JOURNAL(inode);
Dave Hansene3a68e32007-07-15 23:41:14 -07003569 if (is_journal_aborted(journal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 return -EROFS;
3571
3572 journal_lock_updates(journal);
3573 journal_flush(journal);
3574
3575 /*
3576 * OK, there are no updates running now, and all cached data is
3577 * synced to disk. We are now in a completely consistent state
3578 * which doesn't have anything in the journal, and we know that
3579 * no filesystem updates are running, so it is safe to modify
3580 * the inode's in-core data-journaling state flag now.
3581 */
3582
3583 if (val)
3584 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3585 else
3586 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3587 ext3_set_aops(inode);
3588
3589 journal_unlock_updates(journal);
3590
3591 /* Finally we can mark the inode as dirty. */
3592
3593 handle = ext3_journal_start(inode, 1);
3594 if (IS_ERR(handle))
3595 return PTR_ERR(handle);
3596
3597 err = ext3_mark_inode_dirty(handle, inode);
3598 handle->h_sync = 1;
3599 ext3_journal_stop(handle);
3600 ext3_std_error(inode->i_sb, err);
3601
3602 return err;
3603}