blob: a8d3217c5cfe45a142f6d15acddff25828786a35 [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
25#include <linux/module.h>
26#include <linux/fs.h>
27#include <linux/time.h>
28#include <linux/ext3_jbd.h>
29#include <linux/jbd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/highuid.h>
31#include <linux/pagemap.h>
32#include <linux/quotaops.h>
33#include <linux/string.h>
34#include <linux/buffer_head.h>
35#include <linux/writeback.h>
36#include <linux/mpage.h>
37#include <linux/uio.h>
Jens Axboecaa38fb2006-07-23 01:41:26 +020038#include <linux/bio.h>
Josef Bacik68c9d702008-10-03 17:32:43 -040039#include <linux/fiemap.h>
Duane Griffinb5ed3112008-12-19 20:47:14 +000040#include <linux/namei.h>
Lukas Czerner785c4bc2011-05-23 18:33:01 +020041#include <trace/events/ext3.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xattr.h"
43#include "acl.h"
44
45static int ext3_writepage_trans_blocks(struct inode *inode);
Jan Karaee3e77f2011-06-03 21:58:11 +020046static int ext3_block_truncate_page(struct inode *inode, loff_t from);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
49 * Test whether an inode is a fast symlink.
50 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080051static int ext3_inode_is_fast_symlink(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 int ea_blocks = EXT3_I(inode)->i_file_acl ?
54 (inode->i_sb->s_blocksize >> 9) : 0;
55
Andrew Mortond6859bf2006-03-26 01:38:03 -080056 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057}
58
Andrew Mortond6859bf2006-03-26 01:38:03 -080059/*
60 * The ext3 forget function must perform a revoke if we are freeing data
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * which has been journaled. Metadata (eg. indirect blocks) must be
Mingming Caoae6ddcc2006-09-27 01:49:27 -070062 * revoked in all cases.
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 *
64 * "bh" may be NULL: a metadata block may have been freed from memory
65 * but there may still be a record of it in the journal, and that record
66 * still needs to be revoked.
67 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080068int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -070069 struct buffer_head *bh, ext3_fsblk_t blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
71 int err;
72
73 might_sleep();
74
Lukas Czerner785c4bc2011-05-23 18:33:01 +020075 trace_ext3_forget(inode, is_metadata, blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 BUFFER_TRACE(bh, "enter");
77
78 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
79 "data mode %lx\n",
80 bh, is_metadata, inode->i_mode,
81 test_opt(inode->i_sb, DATA_FLAGS));
82
83 /* Never use the revoke function if we are doing full data
84 * journaling: there is no need to, and a V1 superblock won't
85 * support it. Otherwise, only skip the revoke on un-journaled
86 * data blocks. */
87
88 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
89 (!is_metadata && !ext3_should_journal_data(inode))) {
90 if (bh) {
91 BUFFER_TRACE(bh, "call journal_forget");
92 return ext3_journal_forget(handle, bh);
93 }
94 return 0;
95 }
96
97 /*
98 * data!=journal && (is_metadata || should_journal_data(inode))
99 */
100 BUFFER_TRACE(bh, "call ext3_journal_revoke");
101 err = ext3_journal_revoke(handle, blocknr, bh);
102 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700103 ext3_abort(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 "error %d when attempting revoke", err);
105 BUFFER_TRACE(bh, "exit");
106 return err;
107}
108
109/*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800110 * Work out how many blocks we need to proceed with the next chunk of a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * truncate transaction.
112 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700113static unsigned long blocks_for_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
115 unsigned long needed;
116
117 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
118
119 /* Give ourselves just enough room to cope with inodes in which
120 * i_blocks is corrupt: we've seen disk corruptions in the past
121 * which resulted in random data in an inode which looked enough
122 * like a regular file for ext3 to try to delete it. Things
123 * will go a bit crazy if that happens, but at least we should
124 * try not to panic the whole kernel. */
125 if (needed < 2)
126 needed = 2;
127
128 /* But we need to bound the transaction so we don't overflow the
129 * journal. */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700130 if (needed > EXT3_MAX_TRANS_DATA)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 needed = EXT3_MAX_TRANS_DATA;
132
Jan Kara1f545872005-06-23 22:01:04 -0700133 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134}
135
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700136/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 * Truncate transactions can be complex and absolutely huge. So we need to
138 * be able to restart the transaction at a conventient checkpoint to make
139 * sure we don't overflow the journal.
140 *
141 * start_transaction gets us a new handle for a truncate transaction,
142 * and extend_transaction tries to extend the existing one a bit. If
143 * extend fails, we need to propagate the failure up and restart the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700144 * transaction in the top-level truncate loop. --sct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700146static handle_t *start_transaction(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147{
148 handle_t *result;
149
150 result = ext3_journal_start(inode, blocks_for_truncate(inode));
151 if (!IS_ERR(result))
152 return result;
153
154 ext3_std_error(inode->i_sb, PTR_ERR(result));
155 return result;
156}
157
158/*
159 * Try to extend this transaction for the purposes of truncation.
160 *
161 * Returns 0 if we managed to create more room. If we can't create more
162 * room, and the transaction must be restarted we return 1.
163 */
164static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
165{
166 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
167 return 0;
168 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
169 return 0;
170 return 1;
171}
172
173/*
174 * Restart the transaction associated with *handle. This does a commit,
175 * so before we call here everything must be consistently dirtied against
176 * this transaction.
177 */
Jan Kara00171d32009-08-11 19:06:10 +0200178static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
Jan Kara00171d32009-08-11 19:06:10 +0200180 int ret;
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara00171d32009-08-11 19:06:10 +0200183 /*
184 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
185 * At this moment, get_block can be called only for blocks inside
186 * i_size since page cache has been already dropped and writes are
187 * blocked by i_mutex. So we can safely drop the truncate_mutex.
188 */
189 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
190 ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
191 mutex_lock(&EXT3_I(inode)->truncate_mutex);
192 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
195/*
Al Viroac14a952010-06-06 07:08:19 -0400196 * Called at inode eviction from icache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 */
Al Viroac14a952010-06-06 07:08:19 -0400198void ext3_evict_inode (struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Jan Karab22570d2011-07-23 01:21:38 +0200200 struct ext3_inode_info *ei = EXT3_I(inode);
Al Viroac14a952010-06-06 07:08:19 -0400201 struct ext3_block_alloc_info *rsv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 handle_t *handle;
Al Viroac14a952010-06-06 07:08:19 -0400203 int want_delete = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200205 trace_ext3_evict_inode(inode);
Al Viroac14a952010-06-06 07:08:19 -0400206 if (!inode->i_nlink && !is_bad_inode(inode)) {
Christoph Hellwig871a2932010-03-03 09:05:07 -0500207 dquot_initialize(inode);
Al Viroac14a952010-06-06 07:08:19 -0400208 want_delete = 1;
209 }
Christoph Hellwig907f4552010-03-03 09:05:06 -0500210
Jan Karab22570d2011-07-23 01:21:38 +0200211 /*
212 * When journalling data dirty buffers are tracked only in the journal.
213 * So although mm thinks everything is clean and ready for reaping the
214 * inode might still have some pages to write in the running
215 * transaction or waiting to be checkpointed. Thus calling
216 * journal_invalidatepage() (via truncate_inode_pages()) to discard
217 * these buffers can cause data loss. Also even if we did not discard
218 * these buffers, we would have no way to find them after the inode
219 * is reaped and thus user could see stale data if he tries to read
220 * them before the transaction is checkpointed. So be careful and
221 * force everything to disk here... We use ei->i_datasync_tid to
222 * store the newest transaction containing inode's data.
223 *
224 * Note that directories do not have this problem because they don't
225 * use page cache.
Dan Carpenterbcdd0c12011-11-22 11:00:20 +0300226 *
227 * The s_journal check handles the case when ext3_get_journal() fails
228 * and puts the journal inode.
Jan Karab22570d2011-07-23 01:21:38 +0200229 */
230 if (inode->i_nlink && ext3_should_journal_data(inode) &&
Dan Carpenterbcdd0c12011-11-22 11:00:20 +0300231 EXT3_SB(inode->i_sb)->s_journal &&
Jan Karab22570d2011-07-23 01:21:38 +0200232 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
233 tid_t commit_tid = atomic_read(&ei->i_datasync_tid);
234 journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
235
236 log_start_commit(journal, commit_tid);
237 log_wait_commit(journal, commit_tid);
238 filemap_write_and_wait(&inode->i_data);
239 }
Mark Fashehfef26652005-09-09 13:01:31 -0700240 truncate_inode_pages(&inode->i_data, 0);
241
Al Viroac14a952010-06-06 07:08:19 -0400242 ext3_discard_reservation(inode);
Jan Karab22570d2011-07-23 01:21:38 +0200243 rsv = ei->i_block_alloc_info;
244 ei->i_block_alloc_info = NULL;
Al Viroac14a952010-06-06 07:08:19 -0400245 if (unlikely(rsv))
246 kfree(rsv);
247
248 if (!want_delete)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 goto no_delete;
250
251 handle = start_transaction(inode);
252 if (IS_ERR(handle)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800253 /*
254 * If we're going to skip the normal cleanup, we still need to
255 * make sure that the in-core orphan linked list is properly
256 * cleaned up.
257 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 ext3_orphan_del(NULL, inode);
259 goto no_delete;
260 }
261
262 if (IS_SYNC(inode))
263 handle->h_sync = 1;
264 inode->i_size = 0;
265 if (inode->i_blocks)
266 ext3_truncate(inode);
267 /*
Jan Kara40680f2f2011-05-24 22:24:47 +0200268 * Kill off the orphan record created when the inode lost the last
269 * link. Note that ext3_orphan_del() has to be able to cope with the
270 * deletion of a non-existent orphan - ext3_truncate() could
271 * have removed the record.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 */
273 ext3_orphan_del(handle, inode);
Jan Karab22570d2011-07-23 01:21:38 +0200274 ei->i_dtime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700276 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * One subtle ordering requirement: if anything has gone wrong
278 * (transaction abort, IO errors, whatever), then we can still
279 * do these next steps (the fs will already have been marked as
280 * having errors), but we can't free the inode if the mark_dirty
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700281 * fails.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 */
Al Viroac14a952010-06-06 07:08:19 -0400283 if (ext3_mark_inode_dirty(handle, inode)) {
284 /* If that failed, just dquot_drop() and be done with that */
285 dquot_drop(inode);
286 end_writeback(inode);
287 } else {
288 ext3_xattr_delete_inode(handle, inode);
289 dquot_free_inode(inode);
290 dquot_drop(inode);
291 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 ext3_free_inode(handle, inode);
Al Viroac14a952010-06-06 07:08:19 -0400293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 ext3_journal_stop(handle);
295 return;
296no_delete:
Al Viroac14a952010-06-06 07:08:19 -0400297 end_writeback(inode);
298 dquot_drop(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301typedef struct {
302 __le32 *p;
303 __le32 key;
304 struct buffer_head *bh;
305} Indirect;
306
307static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
308{
309 p->key = *(p->p = v);
310 p->bh = bh;
311}
312
Andrew Mortond6859bf2006-03-26 01:38:03 -0800313static int verify_chain(Indirect *from, Indirect *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
315 while (from <= to && from->key == *from->p)
316 from++;
317 return (from > to);
318}
319
320/**
321 * ext3_block_to_path - parse the block number into array of offsets
322 * @inode: inode in question (we are only interested in its superblock)
323 * @i_block: block number to be parsed
324 * @offsets: array to store the offsets in
325 * @boundary: set this non-zero if the referred-to block is likely to be
326 * followed (on disk) by an indirect block.
327 *
328 * To store the locations of file's data ext3 uses a data structure common
329 * for UNIX filesystems - tree of pointers anchored in the inode, with
330 * data blocks at leaves and indirect blocks in intermediate nodes.
331 * This function translates the block number into path in that tree -
332 * return value is the path length and @offsets[n] is the offset of
333 * pointer to (n+1)th node in the nth one. If @block is out of range
334 * (negative or too large) warning is printed and zero returned.
335 *
336 * Note: function doesn't find node addresses, so no IO is needed. All
337 * we need to know is the capacity of indirect blocks (taken from the
338 * inode->i_sb).
339 */
340
341/*
342 * Portability note: the last comparison (check that we fit into triple
343 * indirect block) is spelled differently, because otherwise on an
344 * architecture with 32-bit longs and 8Kb pages we might get into trouble
345 * if our filesystem had 8Kb blocks. We might use long long, but that would
346 * kill us on x86. Oh, well, at least the sign propagation does not matter -
347 * i_block would have to be negative in the very beginning, so we would not
348 * get there at all.
349 */
350
351static int ext3_block_to_path(struct inode *inode,
352 long i_block, int offsets[4], int *boundary)
353{
354 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
355 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
356 const long direct_blocks = EXT3_NDIR_BLOCKS,
357 indirect_blocks = ptrs,
358 double_blocks = (1 << (ptrs_bits * 2));
359 int n = 0;
360 int final = 0;
361
362 if (i_block < 0) {
363 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
364 } else if (i_block < direct_blocks) {
365 offsets[n++] = i_block;
366 final = direct_blocks;
367 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
368 offsets[n++] = EXT3_IND_BLOCK;
369 offsets[n++] = i_block;
370 final = ptrs;
371 } else if ((i_block -= indirect_blocks) < double_blocks) {
372 offsets[n++] = EXT3_DIND_BLOCK;
373 offsets[n++] = i_block >> ptrs_bits;
374 offsets[n++] = i_block & (ptrs - 1);
375 final = ptrs;
376 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
377 offsets[n++] = EXT3_TIND_BLOCK;
378 offsets[n++] = i_block >> (ptrs_bits * 2);
379 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
380 offsets[n++] = i_block & (ptrs - 1);
381 final = ptrs;
382 } else {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800383 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
385 if (boundary)
Mingming Cao89747d32006-03-26 01:37:55 -0800386 *boundary = final - 1 - (i_block & (ptrs - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return n;
388}
389
390/**
391 * ext3_get_branch - read the chain of indirect blocks leading to data
392 * @inode: inode in question
393 * @depth: depth of the chain (1 - direct pointer, etc.)
394 * @offsets: offsets of pointers in inode/indirect blocks
395 * @chain: place to store the result
396 * @err: here we store the error value
397 *
398 * Function fills the array of triples <key, p, bh> and returns %NULL
399 * if everything went OK or the pointer to the last filled triple
400 * (incomplete one) otherwise. Upon the return chain[i].key contains
401 * the number of (i+1)-th block in the chain (as it is stored in memory,
402 * i.e. little-endian 32-bit), chain[i].p contains the address of that
403 * number (it points into struct inode for i==0 and into the bh->b_data
404 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
405 * block for i>0 and NULL for i==0. In other words, it holds the block
406 * numbers of the chain, addresses they were taken from (and where we can
407 * verify that chain did not change) and buffer_heads hosting these
408 * numbers.
409 *
410 * Function stops when it stumbles upon zero pointer (absent block)
411 * (pointer to last triple returned, *@err == 0)
412 * or when it gets an IO error reading an indirect block
413 * (ditto, *@err == -EIO)
414 * or when it notices that chain had been changed while it was reading
415 * (ditto, *@err == -EAGAIN)
416 * or when it reads all @depth-1 indirect blocks successfully and finds
417 * the whole chain, all way to the data (returns %NULL, *err == 0).
418 */
419static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
420 Indirect chain[4], int *err)
421{
422 struct super_block *sb = inode->i_sb;
423 Indirect *p = chain;
424 struct buffer_head *bh;
425
426 *err = 0;
427 /* i_data is not going away, no lock needed */
428 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
429 if (!p->key)
430 goto no_block;
431 while (--depth) {
432 bh = sb_bread(sb, le32_to_cpu(p->key));
433 if (!bh)
434 goto failure;
435 /* Reader: pointers */
436 if (!verify_chain(chain, p))
437 goto changed;
438 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
439 /* Reader: end */
440 if (!p->key)
441 goto no_block;
442 }
443 return NULL;
444
445changed:
446 brelse(bh);
447 *err = -EAGAIN;
448 goto no_block;
449failure:
450 *err = -EIO;
451no_block:
452 return p;
453}
454
455/**
456 * ext3_find_near - find a place for allocation with sufficient locality
457 * @inode: owner
458 * @ind: descriptor of indirect block.
459 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000460 * This function returns the preferred place for block allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * It is used when heuristic for sequential allocation fails.
462 * Rules are:
463 * + if there is a block to the left of our position - allocate near it.
464 * + if pointer will live in indirect block - allocate near that block.
465 * + if pointer will live in inode - allocate in the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700466 * cylinder group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 *
468 * In the latter case we colour the starting block by the callers PID to
469 * prevent it from clashing with concurrent allocations for a different inode
470 * in the same block group. The PID is used here so that functionally related
471 * files will be close-by on-disk.
472 *
473 * Caller must make sure that @ind is valid and will stay that way.
474 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700475static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 struct ext3_inode_info *ei = EXT3_I(inode);
478 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
479 __le32 *p;
Mingming Cao43d23f92006-06-25 05:48:07 -0700480 ext3_fsblk_t bg_start;
481 ext3_grpblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 /* Try to find previous block */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800484 for (p = ind->p - 1; p >= start; p--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 if (*p)
486 return le32_to_cpu(*p);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 /* No such thing, so let's try location of indirect block */
490 if (ind->bh)
491 return ind->bh->b_blocknr;
492
493 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800494 * It is going to be referred to from the inode itself? OK, just put it
495 * into the same cylinder group then.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700497 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 colour = (current->pid % 16) *
499 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
500 return bg_start + colour;
501}
502
503/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000504 * ext3_find_goal - find a preferred place for allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 * @inode: owner
506 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000509 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800510 * returns it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 */
512
Mingming Cao43d23f92006-06-25 05:48:07 -0700513static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800514 Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515{
Andrew Mortond6859bf2006-03-26 01:38:03 -0800516 struct ext3_block_alloc_info *block_i;
517
518 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
520 /*
521 * try the heuristic for sequential allocation,
522 * failing that at least try to get decent locality.
523 */
524 if (block_i && (block == block_i->last_alloc_logical_block + 1)
525 && (block_i->last_alloc_physical_block != 0)) {
Mingming Caofe55c452005-05-01 08:59:20 -0700526 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 }
528
Mingming Caofe55c452005-05-01 08:59:20 -0700529 return ext3_find_near(inode, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530}
Andrew Mortond6859bf2006-03-26 01:38:03 -0800531
Mingming Caob47b2472006-03-26 01:37:56 -0800532/**
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900533 * ext3_blks_to_allocate - Look up the block map and count the number
Mingming Caob47b2472006-03-26 01:37:56 -0800534 * of direct blocks need to be allocated for the given branch.
535 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700536 * @branch: chain of indirect blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800537 * @k: number of blocks need for indirect blocks
538 * @blks: number of data blocks to be mapped.
539 * @blocks_to_boundary: the offset in the indirect block
540 *
541 * return the total number of blocks to be allocate, including the
542 * direct and indirect blocks.
543 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800544static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Mingming Caob47b2472006-03-26 01:37:56 -0800545 int blocks_to_boundary)
546{
547 unsigned long count = 0;
548
549 /*
550 * Simple case, [t,d]Indirect block(s) has not allocated yet
551 * then it's clear blocks on that path have not allocated
552 */
553 if (k > 0) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800554 /* right now we don't handle cross boundary allocation */
Mingming Caob47b2472006-03-26 01:37:56 -0800555 if (blks < blocks_to_boundary + 1)
556 count += blks;
557 else
558 count += blocks_to_boundary + 1;
559 return count;
560 }
561
562 count++;
563 while (count < blks && count <= blocks_to_boundary &&
564 le32_to_cpu(*(branch[0].p + count)) == 0) {
565 count++;
566 }
567 return count;
568}
569
570/**
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900571 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
572 * @handle: handle for this transaction
573 * @inode: owner
574 * @goal: preferred place for allocation
Mingming Caob47b2472006-03-26 01:37:56 -0800575 * @indirect_blks: the number of blocks need to allocate for indirect
576 * blocks
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900577 * @blks: number of blocks need to allocated for direct blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800578 * @new_blocks: on return it will store the new block numbers for
579 * the indirect blocks(if needed) and the first direct block,
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900580 * @err: here we store the error value
581 *
582 * return the number of direct blocks allocated
Mingming Caob47b2472006-03-26 01:37:56 -0800583 */
584static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700585 ext3_fsblk_t goal, int indirect_blks, int blks,
586 ext3_fsblk_t new_blocks[4], int *err)
Mingming Caob47b2472006-03-26 01:37:56 -0800587{
588 int target, i;
589 unsigned long count = 0;
590 int index = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700591 ext3_fsblk_t current_block = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800592 int ret = 0;
593
594 /*
595 * Here we try to allocate the requested multiple blocks at once,
596 * on a best-effort basis.
597 * To build a branch, we should allocate blocks for
598 * the indirect blocks(if not allocated yet), and at least
599 * the first direct block of this branch. That's the
600 * minimum number of blocks need to allocate(required)
601 */
602 target = blks + indirect_blks;
603
604 while (1) {
605 count = target;
606 /* allocating blocks for indirect blocks and direct blocks */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800607 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
Mingming Caob47b2472006-03-26 01:37:56 -0800608 if (*err)
609 goto failed_out;
610
611 target -= count;
612 /* allocate blocks for indirect blocks */
613 while (index < indirect_blks && count) {
614 new_blocks[index++] = current_block++;
615 count--;
616 }
617
618 if (count > 0)
619 break;
620 }
621
622 /* save the new block number for the first direct block */
623 new_blocks[index] = current_block;
624
625 /* total number of blocks allocated for direct blocks */
626 ret = count;
627 *err = 0;
628 return ret;
629failed_out:
630 for (i = 0; i <index; i++)
631 ext3_free_blocks(handle, inode, new_blocks[i], 1);
632 return ret;
633}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635/**
636 * ext3_alloc_branch - allocate and set up a chain of blocks.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900637 * @handle: handle for this transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 * @inode: owner
Mingming Caob47b2472006-03-26 01:37:56 -0800639 * @indirect_blks: number of allocated indirect blocks
640 * @blks: number of allocated direct blocks
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900641 * @goal: preferred place for allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 * @offsets: offsets (in the blocks) to store the pointers to next.
643 * @branch: place to store the chain in.
644 *
Mingming Caob47b2472006-03-26 01:37:56 -0800645 * This function allocates blocks, zeroes out all but the last one,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 * links them into chain and (if we are synchronous) writes them to disk.
647 * In other words, it prepares a branch that can be spliced onto the
648 * inode. It stores the information about that chain in the branch[], in
649 * the same format as ext3_get_branch() would do. We are calling it after
650 * we had read the existing part of chain and partial points to the last
651 * triple of that (one with zero ->key). Upon the exit we have the same
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -0800652 * picture as after the successful ext3_get_block(), except that in one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 * place chain is disconnected - *branch->p is still zero (we did not
654 * set the last link), but branch->key contains the number that should
655 * be placed into *branch->p to fill that gap.
656 *
657 * If allocation fails we free all blocks we've allocated (and forget
658 * their buffer_heads) and return the error value the from failed
659 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
660 * as described above and return 0.
661 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700663 int indirect_blks, int *blks, ext3_fsblk_t goal,
Mingming Caob47b2472006-03-26 01:37:56 -0800664 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 int blocksize = inode->i_sb->s_blocksize;
Mingming Caob47b2472006-03-26 01:37:56 -0800667 int i, n = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 int err = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800669 struct buffer_head *bh;
670 int num;
Mingming Cao43d23f92006-06-25 05:48:07 -0700671 ext3_fsblk_t new_blocks[4];
672 ext3_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Mingming Caob47b2472006-03-26 01:37:56 -0800674 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
675 *blks, new_blocks, &err);
676 if (err)
677 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Mingming Caob47b2472006-03-26 01:37:56 -0800679 branch[0].key = cpu_to_le32(new_blocks[0]);
680 /*
681 * metadata blocks and data blocks are allocated.
682 */
683 for (n = 1; n <= indirect_blks; n++) {
684 /*
685 * Get buffer_head for parent block, zero it out
686 * and set the pointer to new one, then send
687 * parent to disk.
688 */
689 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
690 branch[n].bh = bh;
691 lock_buffer(bh);
692 BUFFER_TRACE(bh, "call get_create_access");
693 err = ext3_journal_get_create_access(handle, bh);
694 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 unlock_buffer(bh);
Mingming Caob47b2472006-03-26 01:37:56 -0800696 brelse(bh);
697 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Mingming Caob47b2472006-03-26 01:37:56 -0800700 memset(bh->b_data, 0, blocksize);
701 branch[n].p = (__le32 *) bh->b_data + offsets[n];
702 branch[n].key = cpu_to_le32(new_blocks[n]);
703 *branch[n].p = branch[n].key;
704 if ( n == indirect_blks) {
705 current_block = new_blocks[n];
706 /*
707 * End of chain, update the last new metablock of
708 * the chain to point to the new allocated
709 * data blocks numbers
710 */
711 for (i=1; i < num; i++)
712 *(branch[n].p + i) = cpu_to_le32(++current_block);
713 }
714 BUFFER_TRACE(bh, "marking uptodate");
715 set_buffer_uptodate(bh);
716 unlock_buffer(bh);
717
718 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
719 err = ext3_journal_dirty_metadata(handle, bh);
720 if (err)
721 goto failed;
722 }
723 *blks = num;
724 return err;
725failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 /* Allocation failed, free what we already allocated */
Mingming Caob47b2472006-03-26 01:37:56 -0800727 for (i = 1; i <= n ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 BUFFER_TRACE(branch[i].bh, "call journal_forget");
729 ext3_journal_forget(handle, branch[i].bh);
730 }
Mingming Caob47b2472006-03-26 01:37:56 -0800731 for (i = 0; i <indirect_blks; i++)
732 ext3_free_blocks(handle, inode, new_blocks[i], 1);
733
734 ext3_free_blocks(handle, inode, new_blocks[i], num);
735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 return err;
737}
738
739/**
Andrew Mortond6859bf2006-03-26 01:38:03 -0800740 * ext3_splice_branch - splice the allocated branch onto inode.
Namhyung Kima4c18ad2010-10-19 00:34:35 +0900741 * @handle: handle for this transaction
Andrew Mortond6859bf2006-03-26 01:38:03 -0800742 * @inode: owner
743 * @block: (logical) number of block we are adding
Andrew Mortond6859bf2006-03-26 01:38:03 -0800744 * @where: location of missing link
745 * @num: number of indirect blocks we are adding
746 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800748 * This function fills the missing link and does all housekeeping needed in
749 * inode (->i_blocks, etc.). In case of success we end up with the full
750 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800752static int ext3_splice_branch(handle_t *handle, struct inode *inode,
753 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
755 int i;
756 int err = 0;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800757 struct ext3_block_alloc_info *block_i;
Mingming Cao43d23f92006-06-25 05:48:07 -0700758 ext3_fsblk_t current_block;
Jan Karafe8bc912009-10-16 19:26:15 +0200759 struct ext3_inode_info *ei = EXT3_I(inode);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800760
Jan Karafe8bc912009-10-16 19:26:15 +0200761 block_i = ei->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 /*
763 * If we're splicing into a [td]indirect block (as opposed to the
764 * inode) then we need to get write access to the [td]indirect block
765 * before the splice.
766 */
767 if (where->bh) {
768 BUFFER_TRACE(where->bh, "get_write_access");
769 err = ext3_journal_get_write_access(handle, where->bh);
770 if (err)
771 goto err_out;
772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 /* That's it */
774
775 *where->p = where->key;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800776
777 /*
778 * Update the host buffer_head or inode to point to more just allocated
779 * direct blocks blocks
780 */
Mingming Caob47b2472006-03-26 01:37:56 -0800781 if (num == 0 && blks > 1) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700782 current_block = le32_to_cpu(where->key) + 1;
Mingming Caob47b2472006-03-26 01:37:56 -0800783 for (i = 1; i < blks; i++)
784 *(where->p + i ) = cpu_to_le32(current_block++);
785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787 /*
788 * update the most recently allocated logical & physical block
789 * in i_block_alloc_info, to assist find the proper goal block for next
790 * allocation
791 */
792 if (block_i) {
Mingming Caob47b2472006-03-26 01:37:56 -0800793 block_i->last_alloc_logical_block = block + blks - 1;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800794 block_i->last_alloc_physical_block =
Mingming Cao5dea5172006-05-03 19:55:12 -0700795 le32_to_cpu(where[num].key) + blks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
797
798 /* We are done with atomic stuff, now do the rest of housekeeping */
799
800 inode->i_ctime = CURRENT_TIME_SEC;
801 ext3_mark_inode_dirty(handle, inode);
Jan Karafe8bc912009-10-16 19:26:15 +0200802 /* ext3_mark_inode_dirty already updated i_sync_tid */
803 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805 /* had we spliced it onto indirect block? */
806 if (where->bh) {
807 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800808 * If we spliced it onto an indirect block, we haven't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 * altered the inode. Note however that if it is being spliced
810 * onto an indirect block at the very end of the file (the
811 * file is growing) then we *will* alter the inode to reflect
812 * the new i_size. But that is not done here - it is done in
813 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
814 */
815 jbd_debug(5, "splicing indirect only\n");
816 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
817 err = ext3_journal_dirty_metadata(handle, where->bh);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700818 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 goto err_out;
820 } else {
821 /*
822 * OK, we spliced it into the inode itself on a direct block.
823 * Inode was dirtied above.
824 */
825 jbd_debug(5, "splicing direct\n");
826 }
827 return err;
828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829err_out:
Mingming Caob47b2472006-03-26 01:37:56 -0800830 for (i = 1; i <= num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 BUFFER_TRACE(where[i].bh, "call journal_forget");
832 ext3_journal_forget(handle, where[i].bh);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800833 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
Mingming Caob47b2472006-03-26 01:37:56 -0800835 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return err;
838}
839
840/*
841 * Allocation strategy is simple: if we have to allocate something, we will
842 * have to go the whole way to leaf. So let's do it before attaching anything
843 * to tree, set linkage between the newborn blocks, write them if sync is
844 * required, recheck the path, free and repeat if check fails, otherwise
845 * set the last missing link (that will protect us from any truncate-generated
846 * removals - all blocks on the path are immune now) and possibly force the
847 * write on the parent block.
848 * That has a nice additional property: no special recovery from the failed
849 * allocations is needed - we simply release blocks and do not touch anything
850 * reachable from inode.
851 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800852 * `handle' can be NULL if create == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 *
854 * The BKL may not be held on entry here. Be sure to take it early.
Mingming Cao89747d32006-03-26 01:37:55 -0800855 * return > 0, # of blocks mapped or allocated.
856 * return = 0, if plain lookup failed.
857 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800859int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
860 sector_t iblock, unsigned long maxblocks,
861 struct buffer_head *bh_result,
Jan Kara43237b52009-05-20 18:41:58 +0200862 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
864 int err = -EIO;
865 int offsets[4];
866 Indirect chain[4];
867 Indirect *partial;
Mingming Cao43d23f92006-06-25 05:48:07 -0700868 ext3_fsblk_t goal;
Mingming Caob47b2472006-03-26 01:37:56 -0800869 int indirect_blks;
Mingming Cao89747d32006-03-26 01:37:55 -0800870 int blocks_to_boundary = 0;
871 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 struct ext3_inode_info *ei = EXT3_I(inode);
Mingming Cao89747d32006-03-26 01:37:55 -0800873 int count = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700874 ext3_fsblk_t first_block = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Lukas Czerner785c4bc2011-05-23 18:33:01 +0200877 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 J_ASSERT(handle != NULL || create == 0);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800879 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 if (depth == 0)
882 goto out;
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
885
886 /* Simplest case - block found, no allocation needed */
887 if (!partial) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700888 first_block = le32_to_cpu(chain[depth - 1].key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 clear_buffer_new(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800890 count++;
891 /*map more blocks*/
892 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700893 ext3_fsblk_t blk;
Mingming Cao5dea5172006-05-03 19:55:12 -0700894
Jan Karae8ef7aa2009-06-17 16:26:23 -0700895 if (!verify_chain(chain, chain + depth - 1)) {
Mingming Cao89747d32006-03-26 01:37:55 -0800896 /*
897 * Indirect block might be removed by
898 * truncate while we were reading it.
899 * Handling of that case: forget what we've
900 * got now. Flag the err as EAGAIN, so it
901 * will reread.
902 */
903 err = -EAGAIN;
904 count = 0;
905 break;
906 }
Mingming Cao5dea5172006-05-03 19:55:12 -0700907 blk = le32_to_cpu(*(chain[depth-1].p + count));
908
909 if (blk == first_block + count)
Mingming Cao89747d32006-03-26 01:37:55 -0800910 count++;
911 else
912 break;
913 }
914 if (err != -EAGAIN)
915 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
917
918 /* Next simple case - plain lookup or failed read of indirect block */
Mingming Caofe55c452005-05-01 08:59:20 -0700919 if (!create || err == -EIO)
920 goto cleanup;
921
Jan Kara40680f2f2011-05-24 22:24:47 +0200922 /*
923 * Block out ext3_truncate while we alter the tree
924 */
Arjan van de Ven97461512006-03-23 03:00:42 -0800925 mutex_lock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700926
927 /*
928 * If the indirect block is missing while we are reading
929 * the chain(ext3_get_branch() returns -EAGAIN err), or
930 * if the chain has been changed after we grab the semaphore,
931 * (either because another process truncated this branch, or
932 * another get_block allocated this branch) re-grab the chain to see if
933 * the request block has been allocated or not.
934 *
935 * Since we already block the truncate/other get_block
936 * at this point, we will have the current copy of the chain when we
937 * splice the branch into the tree.
938 */
939 if (err == -EAGAIN || !verify_chain(chain, partial)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 while (partial > chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 brelse(partial->bh);
942 partial--;
943 }
Mingming Caofe55c452005-05-01 08:59:20 -0700944 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
945 if (!partial) {
Mingming Cao89747d32006-03-26 01:37:55 -0800946 count++;
Arjan van de Ven97461512006-03-23 03:00:42 -0800947 mutex_unlock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700948 if (err)
949 goto cleanup;
950 clear_buffer_new(bh_result);
951 goto got_it;
952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
954
955 /*
Mingming Caofe55c452005-05-01 08:59:20 -0700956 * Okay, we need to do block allocation. Lazily initialize the block
957 * allocation info here if necessary
958 */
959 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 ext3_init_block_alloc_info(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800962 goal = ext3_find_goal(inode, iblock, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Mingming Caob47b2472006-03-26 01:37:56 -0800964 /* the number of blocks need to allocate for [d,t]indirect blocks */
965 indirect_blks = (chain + depth) - partial - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 /*
Mingming Caob47b2472006-03-26 01:37:56 -0800968 * Next look up the indirect map to count the totoal number of
969 * direct blocks to allocate for this branch.
970 */
971 count = ext3_blks_to_allocate(partial, indirect_blks,
972 maxblocks, blocks_to_boundary);
Mingming Caob47b2472006-03-26 01:37:56 -0800973 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
Mingming Caofe55c452005-05-01 08:59:20 -0700974 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Mingming Caofe55c452005-05-01 08:59:20 -0700976 /*
977 * The ext3_splice_branch call will free and forget any buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 * on the new chain if there is a failure, but that risks using
979 * up transaction credits, especially for bitmaps where the
980 * credits cannot be returned. Can we handle this somehow? We
Mingming Caofe55c452005-05-01 08:59:20 -0700981 * may need to return -EAGAIN upwards in the worst case. --sct
982 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 if (!err)
Mingming Caob47b2472006-03-26 01:37:56 -0800984 err = ext3_splice_branch(handle, inode, iblock,
985 partial, indirect_blks, count);
Arjan van de Ven97461512006-03-23 03:00:42 -0800986 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 if (err)
988 goto cleanup;
989
990 set_buffer_new(bh_result);
Mingming Caofe55c452005-05-01 08:59:20 -0700991got_it:
992 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
Suparna Bhattacharya20acaa12006-09-16 12:15:58 -0700993 if (count > blocks_to_boundary)
Mingming Caofe55c452005-05-01 08:59:20 -0700994 set_buffer_boundary(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800995 err = count;
Mingming Caofe55c452005-05-01 08:59:20 -0700996 /* Clean up and exit */
997 partial = chain + depth - 1; /* the whole chain */
998cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 while (partial > chain) {
Mingming Caofe55c452005-05-01 08:59:20 -07001000 BUFFER_TRACE(partial->bh, "call brelse");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 brelse(partial->bh);
1002 partial--;
1003 }
Mingming Caofe55c452005-05-01 08:59:20 -07001004 BUFFER_TRACE(bh_result, "returned");
1005out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001006 trace_ext3_get_blocks_exit(inode, iblock,
1007 depth ? le32_to_cpu(chain[depth-1].key) : 0,
1008 count, err);
Mingming Caofe55c452005-05-01 08:59:20 -07001009 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010}
1011
Jan Karabd1939d2008-02-06 01:40:21 -08001012/* Maximum number of blocks we map for direct IO at once. */
1013#define DIO_MAX_BLOCKS 4096
1014/*
1015 * Number of credits we need for writing DIO_MAX_BLOCKS:
1016 * We need sb + group descriptor + bitmap + inode -> 4
1017 * For B blocks with A block pointers per block we need:
1018 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1019 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1020 */
1021#define DIO_CREDITS 25
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001023static int ext3_get_block(struct inode *inode, sector_t iblock,
1024 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001026 handle_t *handle = ext3_journal_current_handle();
Jan Karabd1939d2008-02-06 01:40:21 -08001027 int ret = 0, started = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -08001028 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Jan Karabd1939d2008-02-06 01:40:21 -08001030 if (create && !handle) { /* Direct IO write... */
1031 if (max_blocks > DIO_MAX_BLOCKS)
1032 max_blocks = DIO_MAX_BLOCKS;
1033 handle = ext3_journal_start(inode, DIO_CREDITS +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001034 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
Jan Karabd1939d2008-02-06 01:40:21 -08001035 if (IS_ERR(handle)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 ret = PTR_ERR(handle);
Jan Karabd1939d2008-02-06 01:40:21 -08001037 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
Jan Karabd1939d2008-02-06 01:40:21 -08001039 started = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
1041
Jan Karabd1939d2008-02-06 01:40:21 -08001042 ret = ext3_get_blocks_handle(handle, inode, iblock,
Jan Kara43237b52009-05-20 18:41:58 +02001043 max_blocks, bh_result, create);
Jan Karabd1939d2008-02-06 01:40:21 -08001044 if (ret > 0) {
1045 bh_result->b_size = (ret << inode->i_blkbits);
1046 ret = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001047 }
Jan Karabd1939d2008-02-06 01:40:21 -08001048 if (started)
1049 ext3_journal_stop(handle);
1050out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 return ret;
1052}
1053
Josef Bacik68c9d702008-10-03 17:32:43 -04001054int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1055 u64 start, u64 len)
1056{
1057 return generic_block_fiemap(inode, fieinfo, start, len,
1058 ext3_get_block);
1059}
1060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061/*
1062 * `handle' can be NULL if create is zero
1063 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001064struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1065 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
1067 struct buffer_head dummy;
1068 int fatal = 0, err;
1069
1070 J_ASSERT(handle != NULL || create == 0);
1071
1072 dummy.b_state = 0;
1073 dummy.b_blocknr = -1000;
1074 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -08001075 err = ext3_get_blocks_handle(handle, inode, block, 1,
Jan Kara43237b52009-05-20 18:41:58 +02001076 &dummy, create);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001077 /*
1078 * ext3_get_blocks_handle() returns number of blocks
1079 * mapped. 0 in case of a HOLE.
1080 */
1081 if (err > 0) {
1082 if (err > 1)
1083 WARN_ON(1);
Mingming Cao89747d32006-03-26 01:37:55 -08001084 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001085 }
1086 *errp = err;
1087 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 struct buffer_head *bh;
1089 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001090 if (!bh) {
1091 *errp = -EIO;
1092 goto err;
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (buffer_new(&dummy)) {
1095 J_ASSERT(create != 0);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001096 J_ASSERT(handle != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Andrew Mortond6859bf2006-03-26 01:38:03 -08001098 /*
1099 * Now that we do not always journal data, we should
1100 * keep in mind whether this should always journal the
1101 * new buffer as metadata. For now, regular file
1102 * writes use ext3_get_block instead, so it's not a
1103 * problem.
1104 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 lock_buffer(bh);
1106 BUFFER_TRACE(bh, "call get_create_access");
1107 fatal = ext3_journal_get_create_access(handle, bh);
1108 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001109 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 set_buffer_uptodate(bh);
1111 }
1112 unlock_buffer(bh);
1113 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1114 err = ext3_journal_dirty_metadata(handle, bh);
1115 if (!fatal)
1116 fatal = err;
1117 } else {
1118 BUFFER_TRACE(bh, "not a new buffer");
1119 }
1120 if (fatal) {
1121 *errp = fatal;
1122 brelse(bh);
1123 bh = NULL;
1124 }
1125 return bh;
1126 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001127err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 return NULL;
1129}
1130
Andrew Mortond6859bf2006-03-26 01:38:03 -08001131struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 int block, int create, int *err)
1133{
1134 struct buffer_head * bh;
1135
1136 bh = ext3_getblk(handle, inode, block, create, err);
1137 if (!bh)
1138 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001139 if (bh_uptodate_or_lock(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 return bh;
Zheng Liud03e1292011-12-05 15:55:11 +08001141 get_bh(bh);
1142 bh->b_end_io = end_buffer_read_sync;
1143 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 wait_on_buffer(bh);
1145 if (buffer_uptodate(bh))
1146 return bh;
1147 put_bh(bh);
1148 *err = -EIO;
1149 return NULL;
1150}
1151
1152static int walk_page_buffers( handle_t *handle,
1153 struct buffer_head *head,
1154 unsigned from,
1155 unsigned to,
1156 int *partial,
1157 int (*fn)( handle_t *handle,
1158 struct buffer_head *bh))
1159{
1160 struct buffer_head *bh;
1161 unsigned block_start, block_end;
1162 unsigned blocksize = head->b_size;
1163 int err, ret = 0;
1164 struct buffer_head *next;
1165
1166 for ( bh = head, block_start = 0;
1167 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001168 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 {
1170 next = bh->b_this_page;
1171 block_end = block_start + blocksize;
1172 if (block_end <= from || block_start >= to) {
1173 if (partial && !buffer_uptodate(bh))
1174 *partial = 1;
1175 continue;
1176 }
1177 err = (*fn)(handle, bh);
1178 if (!ret)
1179 ret = err;
1180 }
1181 return ret;
1182}
1183
1184/*
1185 * To preserve ordering, it is essential that the hole instantiation and
1186 * the data write be encapsulated in a single transaction. We cannot
1187 * close off a transaction and start a new one between the ext3_get_block()
1188 * and the commit_write(). So doing the journal_start at the start of
1189 * prepare_write() is the right place.
1190 *
1191 * Also, this function can nest inside ext3_writepage() ->
1192 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1193 * has generated enough buffer credits to do the whole page. So we won't
1194 * block on the journal in that case, which is good, because the caller may
1195 * be PF_MEMALLOC.
1196 *
1197 * By accident, ext3 can be reentered when a transaction is open via
1198 * quota file writes. If we were to commit the transaction while thus
1199 * reentered, there can be a deadlock - we would be holding a quota
1200 * lock, and the commit would never complete if another thread had a
1201 * transaction open and was blocking on the quota lock - a ranking
1202 * violation.
1203 *
1204 * So what we do is to rely on the fact that journal_stop/journal_start
1205 * will _not_ run commit under these circumstances because handle->h_ref
1206 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001207 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001209static int do_journal_get_write_access(handle_t *handle,
1210 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
Jan Kara5f11e6a2010-08-05 12:38:26 +02001212 int dirty = buffer_dirty(bh);
1213 int ret;
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (!buffer_mapped(bh) || buffer_freed(bh))
1216 return 0;
Jan Kara5f11e6a2010-08-05 12:38:26 +02001217 /*
1218 * __block_prepare_write() could have dirtied some buffers. Clean
1219 * the dirty bit as jbd2_journal_get_write_access() could complain
1220 * otherwise about fs integrity issues. Setting of the dirty bit
1221 * by __block_prepare_write() isn't a real problem here as we clear
1222 * the bit before releasing a page lock and thus writeback cannot
1223 * ever write the buffer.
1224 */
1225 if (dirty)
1226 clear_buffer_dirty(bh);
1227 ret = ext3_journal_get_write_access(handle, bh);
1228 if (!ret && dirty)
1229 ret = ext3_journal_dirty_metadata(handle, bh);
1230 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231}
1232
Jan Kara68eb3db2009-12-01 16:53:06 +01001233/*
1234 * Truncate blocks that were not used by write. We have to truncate the
1235 * pagecache as well so that corresponding buffers get properly unmapped.
1236 */
1237static void ext3_truncate_failed_write(struct inode *inode)
1238{
1239 truncate_inode_pages(inode->i_mapping, inode->i_size);
1240 ext3_truncate(inode);
1241}
1242
Jan Karaee3e77f2011-06-03 21:58:11 +02001243/*
1244 * Truncate blocks that were not used by direct IO write. We have to zero out
1245 * the last file block as well because direct IO might have written to it.
1246 */
1247static void ext3_truncate_failed_direct_write(struct inode *inode)
1248{
1249 ext3_block_truncate_page(inode, inode->i_size);
1250 ext3_truncate(inode);
1251}
1252
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001253static int ext3_write_begin(struct file *file, struct address_space *mapping,
1254 loff_t pos, unsigned len, unsigned flags,
1255 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256{
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001257 struct inode *inode = mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001258 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 handle_t *handle;
1260 int retries = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001261 struct page *page;
1262 pgoff_t index;
1263 unsigned from, to;
Jan Kara695f6ae2009-04-02 16:57:17 -07001264 /* Reserve one block more for addition to orphan list in case
1265 * we allocate blocks but write fails for some reason */
1266 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001267
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001268 trace_ext3_write_begin(inode, pos, len, flags);
1269
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001270 index = pos >> PAGE_CACHE_SHIFT;
1271 from = pos & (PAGE_CACHE_SIZE - 1);
1272 to = from + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
1274retry:
Nick Piggin54566b22009-01-04 12:00:53 -08001275 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001276 if (!page)
1277 return -ENOMEM;
1278 *pagep = page;
1279
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 handle = ext3_journal_start(inode, needed_blocks);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001281 if (IS_ERR(handle)) {
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001282 unlock_page(page);
1283 page_cache_release(page);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001284 ret = PTR_ERR(handle);
1285 goto out;
1286 }
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001287 ret = __block_write_begin(page, pos, len, ext3_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 if (ret)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001289 goto write_begin_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 if (ext3_should_journal_data(inode)) {
1292 ret = walk_page_buffers(handle, page_buffers(page),
1293 from, to, NULL, do_journal_get_write_access);
1294 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001295write_begin_failed:
1296 if (ret) {
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001297 /*
1298 * block_write_begin may have instantiated a few blocks
1299 * outside i_size. Trim these off again. Don't need
1300 * i_size_read because we hold i_mutex.
Jan Kara695f6ae2009-04-02 16:57:17 -07001301 *
1302 * Add inode to orphan list in case we crash before truncate
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001303 * finishes. Do this only if ext3_can_truncate() agrees so
1304 * that orphan processing code is happy.
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001305 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001306 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001307 ext3_orphan_add(handle, inode);
1308 ext3_journal_stop(handle);
1309 unlock_page(page);
1310 page_cache_release(page);
1311 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001312 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1315 goto retry;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001316out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 return ret;
1318}
1319
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001320
Andrew Mortond6859bf2006-03-26 01:38:03 -08001321int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
1323 int err = journal_dirty_data(handle, bh);
1324 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001325 ext3_journal_abort_handle(__func__, __func__,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001326 bh, handle, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 return err;
1328}
1329
Jan Kara695f6ae2009-04-02 16:57:17 -07001330/* For ordered writepage and write_end functions */
1331static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1332{
1333 /*
1334 * Write could have mapped the buffer but it didn't copy the data in
1335 * yet. So avoid filing such buffer into a transaction.
1336 */
1337 if (buffer_mapped(bh) && buffer_uptodate(bh))
1338 return ext3_journal_dirty_data(handle, bh);
1339 return 0;
1340}
1341
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001342/* For write_end() in data=journal mode */
1343static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
1345 if (!buffer_mapped(bh) || buffer_freed(bh))
1346 return 0;
1347 set_buffer_uptodate(bh);
1348 return ext3_journal_dirty_metadata(handle, bh);
1349}
1350
1351/*
Jan Kara695f6ae2009-04-02 16:57:17 -07001352 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1353 * for the whole page but later we failed to copy the data in. Update inode
1354 * size according to what we managed to copy. The rest is going to be
1355 * truncated in write_end function.
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001356 */
Jan Kara695f6ae2009-04-02 16:57:17 -07001357static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001358{
Jan Kara695f6ae2009-04-02 16:57:17 -07001359 /* What matters to us is i_disksize. We don't write i_size anywhere */
1360 if (pos + copied > inode->i_size)
1361 i_size_write(inode, pos + copied);
1362 if (pos + copied > EXT3_I(inode)->i_disksize) {
1363 EXT3_I(inode)->i_disksize = pos + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001364 mark_inode_dirty(inode);
1365 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001366}
1367
1368/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * We need to pick up the new inode size which generic_commit_write gave us
1370 * `file' can be NULL - eg, when called from page_symlink().
1371 *
1372 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1373 * buffers are managed internally.
1374 */
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001375static int ext3_ordered_write_end(struct file *file,
1376 struct address_space *mapping,
1377 loff_t pos, unsigned len, unsigned copied,
1378 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
1380 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001381 struct inode *inode = file->f_mapping->host;
1382 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 int ret = 0, ret2;
1384
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001385 trace_ext3_ordered_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001386 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1387
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001388 from = pos & (PAGE_CACHE_SIZE - 1);
Jan Kara695f6ae2009-04-02 16:57:17 -07001389 to = from + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 ret = walk_page_buffers(handle, page_buffers(page),
Jan Kara695f6ae2009-04-02 16:57:17 -07001391 from, to, NULL, journal_dirty_data_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Jan Kara695f6ae2009-04-02 16:57:17 -07001393 if (ret == 0)
1394 update_file_sizes(inode, pos, copied);
1395 /*
1396 * There may be allocated blocks outside of i_size because
1397 * we failed to copy some data. Prepare for truncate.
1398 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001399 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001400 ext3_orphan_add(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 ret2 = ext3_journal_stop(handle);
1402 if (!ret)
1403 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001404 unlock_page(page);
1405 page_cache_release(page);
1406
Jan Kara695f6ae2009-04-02 16:57:17 -07001407 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001408 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001409 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410}
1411
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001412static int ext3_writeback_write_end(struct file *file,
1413 struct address_space *mapping,
1414 loff_t pos, unsigned len, unsigned copied,
1415 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416{
1417 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001418 struct inode *inode = file->f_mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001419 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001421 trace_ext3_writeback_write_end(inode, pos, len, copied);
Jan Kara695f6ae2009-04-02 16:57:17 -07001422 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1423 update_file_sizes(inode, pos, copied);
1424 /*
1425 * There may be allocated blocks outside of i_size because
1426 * we failed to copy some data. Prepare for truncate.
1427 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001428 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001429 ext3_orphan_add(handle, inode);
1430 ret = ext3_journal_stop(handle);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001431 unlock_page(page);
1432 page_cache_release(page);
1433
Jan Kara695f6ae2009-04-02 16:57:17 -07001434 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001435 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001436 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437}
1438
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001439static int ext3_journalled_write_end(struct file *file,
1440 struct address_space *mapping,
1441 loff_t pos, unsigned len, unsigned copied,
1442 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443{
1444 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001445 struct inode *inode = mapping->host;
Jan Karab22570d2011-07-23 01:21:38 +02001446 struct ext3_inode_info *ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 int ret = 0, ret2;
1448 int partial = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001449 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001451 trace_ext3_journalled_write_end(inode, pos, len, copied);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001452 from = pos & (PAGE_CACHE_SIZE - 1);
1453 to = from + len;
1454
1455 if (copied < len) {
1456 if (!PageUptodate(page))
1457 copied = 0;
Jan Kara695f6ae2009-04-02 16:57:17 -07001458 page_zero_new_buffers(page, from + copied, to);
1459 to = from + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
1462 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001463 to, &partial, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (!partial)
1465 SetPageUptodate(page);
Jan Kara695f6ae2009-04-02 16:57:17 -07001466
1467 if (pos + copied > inode->i_size)
1468 i_size_write(inode, pos + copied);
1469 /*
1470 * There may be allocated blocks outside of i_size because
1471 * we failed to copy some data. Prepare for truncate.
1472 */
Jan Kara9eaaa2d2009-07-13 20:26:52 +02001473 if (pos + len > inode->i_size && ext3_can_truncate(inode))
Jan Kara695f6ae2009-04-02 16:57:17 -07001474 ext3_orphan_add(handle, inode);
Jan Kara9df93932010-01-06 21:58:48 +01001475 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001476 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1477 if (inode->i_size > ei->i_disksize) {
1478 ei->i_disksize = inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001480 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 ret = ret2;
1482 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 ret2 = ext3_journal_stop(handle);
1485 if (!ret)
1486 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001487 unlock_page(page);
1488 page_cache_release(page);
1489
Jan Kara695f6ae2009-04-02 16:57:17 -07001490 if (pos + len > inode->i_size)
Jan Kara68eb3db2009-12-01 16:53:06 +01001491 ext3_truncate_failed_write(inode);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001492 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001495/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 * bmap() is special. It gets used by applications such as lilo and by
1497 * the swapper to find the on-disk block of a specific piece of data.
1498 *
1499 * Naturally, this is dangerous if the block concerned is still in the
1500 * journal. If somebody makes a swapfile on an ext3 data-journaling
1501 * filesystem and enables swap, then they may get a nasty shock when the
1502 * data getting swapped to that swapfile suddenly gets overwritten by
1503 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001504 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 *
1506 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001507 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 */
1509static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1510{
1511 struct inode *inode = mapping->host;
1512 journal_t *journal;
1513 int err;
1514
Jan Kara9df93932010-01-06 21:58:48 +01001515 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001516 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 * This is a REALLY heavyweight approach, but the use of
1518 * bmap on dirty files is expected to be extremely rare:
1519 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001520 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
1522 * (bmap requires CAP_SYS_RAWIO so this does not
1523 * represent an unprivileged user DOS attack --- we'd be
1524 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001525 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 *
1527 * NB. EXT3_STATE_JDATA is not set on files other than
1528 * regular files. If somebody wants to bmap a directory
1529 * or symlink and gets confused because the buffer
1530 * hasn't yet been flushed to disk, they deserve
1531 * everything they get.
1532 */
1533
Jan Kara9df93932010-01-06 21:58:48 +01001534 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 journal = EXT3_JOURNAL(inode);
1536 journal_lock_updates(journal);
1537 err = journal_flush(journal);
1538 journal_unlock_updates(journal);
1539
1540 if (err)
1541 return 0;
1542 }
1543
1544 return generic_block_bmap(mapping,block,ext3_get_block);
1545}
1546
1547static int bget_one(handle_t *handle, struct buffer_head *bh)
1548{
1549 get_bh(bh);
1550 return 0;
1551}
1552
1553static int bput_one(handle_t *handle, struct buffer_head *bh)
1554{
1555 put_bh(bh);
1556 return 0;
1557}
1558
Jan Kara9e80d402009-03-26 13:08:04 +01001559static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1560{
1561 return !buffer_mapped(bh);
1562}
Jan Kara695f6ae2009-04-02 16:57:17 -07001563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564/*
1565 * Note that we always start a transaction even if we're not journalling
1566 * data. This is to preserve ordering: any hole instantiation within
1567 * __block_write_full_page -> ext3_get_block() should be journalled
1568 * along with the data so we don't crash and then get metadata which
1569 * refers to old data.
1570 *
1571 * In all journalling modes block_write_full_page() will start the I/O.
1572 *
1573 * Problem:
1574 *
1575 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1576 * ext3_writepage()
1577 *
1578 * Similar for:
1579 *
1580 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1581 *
1582 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001583 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 *
1585 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1586 * allocations fail.
1587 *
1588 * 16May01: If we're reentered then journal_current_handle() will be
1589 * non-zero. We simply *return*.
1590 *
1591 * 1 July 2001: @@@ FIXME:
1592 * In journalled data mode, a data buffer may be metadata against the
1593 * current transaction. But the same file is part of a shared mapping
1594 * and someone does a writepage() on it.
1595 *
1596 * We will move the buffer onto the async_data list, but *after* it has
1597 * been dirtied. So there's a small window where we have dirty data on
1598 * BJ_Metadata.
1599 *
1600 * Note that this only applies to the last partial page in the file. The
1601 * bit which block_write_full_page() uses prepare/commit for. (That's
1602 * broken code anyway: it's wrong for msync()).
1603 *
1604 * It's a rare case: affects the final partial page, for journalled data
1605 * where the file is subject to bith write() and writepage() in the same
1606 * transction. To fix it we'll need a custom block_write_full_page().
1607 * We'll probably need that anyway for journalling writepage() output.
1608 *
1609 * We don't honour synchronous mounts for writepage(). That would be
1610 * disastrous. Any write() or metadata operation will sync the fs for
1611 * us.
1612 *
1613 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1614 * we don't need to open a transaction here.
1615 */
1616static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001617 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618{
1619 struct inode *inode = page->mapping->host;
1620 struct buffer_head *page_bufs;
1621 handle_t *handle = NULL;
1622 int ret = 0;
1623 int err;
1624
1625 J_ASSERT(PageLocked(page));
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001626 WARN_ON_ONCE(IS_RDONLY(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628 /*
1629 * We give up here if we're reentered, because it might be for a
1630 * different filesystem.
1631 */
1632 if (ext3_journal_current_handle())
1633 goto out_fail;
1634
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001635 trace_ext3_ordered_writepage(page);
Jan Kara9e80d402009-03-26 13:08:04 +01001636 if (!page_has_buffers(page)) {
1637 create_empty_buffers(page, inode->i_sb->s_blocksize,
1638 (1 << BH_Dirty)|(1 << BH_Uptodate));
Jan Kara430db322009-04-07 18:25:01 -04001639 page_bufs = page_buffers(page);
1640 } else {
1641 page_bufs = page_buffers(page);
1642 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1643 NULL, buffer_unmapped)) {
1644 /* Provide NULL get_block() to catch bugs if buffers
1645 * weren't really mapped */
1646 return block_write_full_page(page, NULL, wbc);
1647 }
Jan Kara9e80d402009-03-26 13:08:04 +01001648 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1650
1651 if (IS_ERR(handle)) {
1652 ret = PTR_ERR(handle);
1653 goto out_fail;
1654 }
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 walk_page_buffers(handle, page_bufs, 0,
1657 PAGE_CACHE_SIZE, NULL, bget_one);
1658
1659 ret = block_write_full_page(page, ext3_get_block, wbc);
1660
1661 /*
1662 * The page can become unlocked at any point now, and
1663 * truncate can then come in and change things. So we
1664 * can't touch *page from now on. But *page_bufs is
1665 * safe due to elevated refcount.
1666 */
1667
1668 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001669 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 * block_write_full_page() succeeded. Otherwise they are unmapped,
1671 * and generally junk.
1672 */
1673 if (ret == 0) {
1674 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1675 NULL, journal_dirty_data_fn);
1676 if (!ret)
1677 ret = err;
1678 }
1679 walk_page_buffers(handle, page_bufs, 0,
1680 PAGE_CACHE_SIZE, NULL, bput_one);
1681 err = ext3_journal_stop(handle);
1682 if (!ret)
1683 ret = err;
1684 return ret;
1685
1686out_fail:
1687 redirty_page_for_writepage(wbc, page);
1688 unlock_page(page);
1689 return ret;
1690}
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692static int ext3_writeback_writepage(struct page *page,
1693 struct writeback_control *wbc)
1694{
1695 struct inode *inode = page->mapping->host;
1696 handle_t *handle = NULL;
1697 int ret = 0;
1698 int err;
1699
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001700 J_ASSERT(PageLocked(page));
1701 WARN_ON_ONCE(IS_RDONLY(inode));
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (ext3_journal_current_handle())
1704 goto out_fail;
1705
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001706 trace_ext3_writeback_writepage(page);
Jan Kara430db322009-04-07 18:25:01 -04001707 if (page_has_buffers(page)) {
1708 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1709 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1710 /* Provide NULL get_block() to catch bugs if buffers
1711 * weren't really mapped */
1712 return block_write_full_page(page, NULL, wbc);
1713 }
1714 }
1715
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1717 if (IS_ERR(handle)) {
1718 ret = PTR_ERR(handle);
1719 goto out_fail;
1720 }
1721
Christoph Hellwig4c4d3902010-06-07 10:20:39 +02001722 ret = block_write_full_page(page, ext3_get_block, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724 err = ext3_journal_stop(handle);
1725 if (!ret)
1726 ret = err;
1727 return ret;
1728
1729out_fail:
1730 redirty_page_for_writepage(wbc, page);
1731 unlock_page(page);
1732 return ret;
1733}
1734
1735static int ext3_journalled_writepage(struct page *page,
1736 struct writeback_control *wbc)
1737{
1738 struct inode *inode = page->mapping->host;
1739 handle_t *handle = NULL;
1740 int ret = 0;
1741 int err;
1742
Dmitry Monakhov49792c82010-03-02 15:51:02 +03001743 J_ASSERT(PageLocked(page));
1744 WARN_ON_ONCE(IS_RDONLY(inode));
1745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (ext3_journal_current_handle())
1747 goto no_write;
1748
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001749 trace_ext3_journalled_writepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1751 if (IS_ERR(handle)) {
1752 ret = PTR_ERR(handle);
1753 goto no_write;
1754 }
1755
1756 if (!page_has_buffers(page) || PageChecked(page)) {
1757 /*
1758 * It's mmapped pagecache. Add buffers and journal it. There
1759 * doesn't seem much point in redirtying the page here.
1760 */
1761 ClearPageChecked(page);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001762 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1763 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001764 if (ret != 0) {
1765 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 ret = walk_page_buffers(handle, page_buffers(page), 0,
1769 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1770
1771 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001772 PAGE_CACHE_SIZE, NULL, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 if (ret == 0)
1774 ret = err;
Jan Kara9df93932010-01-06 21:58:48 +01001775 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
Jan Karab22570d2011-07-23 01:21:38 +02001776 atomic_set(&EXT3_I(inode)->i_datasync_tid,
1777 handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 unlock_page(page);
1779 } else {
1780 /*
1781 * It may be a page full of checkpoint-mode buffers. We don't
1782 * really know unless we go poke around in the buffer_heads.
1783 * But block_write_full_page will do the right thing.
1784 */
1785 ret = block_write_full_page(page, ext3_get_block, wbc);
1786 }
1787 err = ext3_journal_stop(handle);
1788 if (!ret)
1789 ret = err;
1790out:
1791 return ret;
1792
1793no_write:
1794 redirty_page_for_writepage(wbc, page);
1795out_unlock:
1796 unlock_page(page);
1797 goto out;
1798}
1799
1800static int ext3_readpage(struct file *file, struct page *page)
1801{
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001802 trace_ext3_readpage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return mpage_readpage(page, ext3_get_block);
1804}
1805
1806static int
1807ext3_readpages(struct file *file, struct address_space *mapping,
1808 struct list_head *pages, unsigned nr_pages)
1809{
1810 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1811}
1812
NeilBrown2ff28e22006-03-26 01:37:18 -08001813static void ext3_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814{
1815 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1816
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001817 trace_ext3_invalidatepage(page, offset);
1818
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 /*
1820 * If it's a full truncate we just forget about the pending dirtying
1821 */
1822 if (offset == 0)
1823 ClearPageChecked(page);
1824
NeilBrown2ff28e22006-03-26 01:37:18 -08001825 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826}
1827
Al Viro27496a82005-10-21 03:20:48 -04001828static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829{
1830 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1831
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001832 trace_ext3_releasepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 WARN_ON(PageChecked(page));
1834 if (!page_has_buffers(page))
1835 return 0;
1836 return journal_try_to_free_buffers(journal, page, wait);
1837}
1838
1839/*
1840 * If the O_DIRECT write will extend the file then add this inode to the
1841 * orphan list. So recovery will truncate it back to the original size
1842 * if the machine crashes during the write.
1843 *
1844 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Karabd1939d2008-02-06 01:40:21 -08001845 * crashes then stale disk data _may_ be exposed inside the file. But current
1846 * VFS code falls back into buffered path in that case so we are safe.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 */
1848static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1849 const struct iovec *iov, loff_t offset,
1850 unsigned long nr_segs)
1851{
1852 struct file *file = iocb->ki_filp;
1853 struct inode *inode = file->f_mapping->host;
1854 struct ext3_inode_info *ei = EXT3_I(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001855 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 ssize_t ret;
1857 int orphan = 0;
1858 size_t count = iov_length(iov, nr_segs);
Eric Sandeenea0174a2009-10-12 21:34:27 -05001859 int retries = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001861 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 if (rw == WRITE) {
1864 loff_t final_size = offset + count;
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 if (final_size > inode->i_size) {
Jan Karabd1939d2008-02-06 01:40:21 -08001867 /* Credits for sb + inode write */
1868 handle = ext3_journal_start(inode, 2);
1869 if (IS_ERR(handle)) {
1870 ret = PTR_ERR(handle);
1871 goto out;
1872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 ret = ext3_orphan_add(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001874 if (ret) {
1875 ext3_journal_stop(handle);
1876 goto out;
1877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 orphan = 1;
1879 ei->i_disksize = inode->i_size;
Jan Karabd1939d2008-02-06 01:40:21 -08001880 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
1882 }
1883
Eric Sandeenea0174a2009-10-12 21:34:27 -05001884retry:
Christoph Hellwigaacfc19c2011-06-24 14:29:47 -04001885 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1886 ext3_get_block);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001887 /*
1888 * In case of error extending write may have instantiated a few
1889 * blocks outside i_size. Trim these off again.
1890 */
1891 if (unlikely((rw & WRITE) && ret < 0)) {
1892 loff_t isize = i_size_read(inode);
1893 loff_t end = offset + iov_length(iov, nr_segs);
1894
1895 if (end > isize)
Jan Karaee3e77f2011-06-03 21:58:11 +02001896 ext3_truncate_failed_direct_write(inode);
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001897 }
Eric Sandeenea0174a2009-10-12 21:34:27 -05001898 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1899 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Jan Karabd1939d2008-02-06 01:40:21 -08001901 if (orphan) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 int err;
1903
Jan Karabd1939d2008-02-06 01:40:21 -08001904 /* Credits for sb + inode write */
1905 handle = ext3_journal_start(inode, 2);
1906 if (IS_ERR(handle)) {
1907 /* This is really bad luck. We've written the data
Jan Kara7eb49692010-03-01 14:02:37 +01001908 * but cannot extend i_size. Truncate allocated blocks
1909 * and pretend the write failed... */
Jan Karaee3e77f2011-06-03 21:58:11 +02001910 ext3_truncate_failed_direct_write(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001911 ret = PTR_ERR(handle);
1912 goto out;
1913 }
1914 if (inode->i_nlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 ext3_orphan_del(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001916 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 loff_t end = offset + ret;
1918 if (end > inode->i_size) {
1919 ei->i_disksize = end;
1920 i_size_write(inode, end);
1921 /*
1922 * We're going to return a positive `ret'
1923 * here due to non-zero-length I/O, so there's
1924 * no way of reporting error returns from
1925 * ext3_mark_inode_dirty() to userspace. So
1926 * ignore it.
1927 */
1928 ext3_mark_inode_dirty(handle, inode);
1929 }
1930 }
1931 err = ext3_journal_stop(handle);
1932 if (ret == 0)
1933 ret = err;
1934 }
1935out:
Lukas Czerner785c4bc2011-05-23 18:33:01 +02001936 trace_ext3_direct_IO_exit(inode, offset,
1937 iov_length(iov, nr_segs), rw, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 return ret;
1939}
1940
1941/*
1942 * Pages can be marked dirty completely asynchronously from ext3's journalling
1943 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1944 * much here because ->set_page_dirty is called under VFS locks. The page is
1945 * not necessarily locked.
1946 *
1947 * We cannot just dirty the page and leave attached buffers clean, because the
1948 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1949 * or jbddirty because all the journalling code will explode.
1950 *
1951 * So what we do is to mark the page "pending dirty" and next time writepage
1952 * is called, propagate that into the buffers appropriately.
1953 */
1954static int ext3_journalled_set_page_dirty(struct page *page)
1955{
1956 SetPageChecked(page);
1957 return __set_page_dirty_nobuffers(page);
1958}
1959
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001960static const struct address_space_operations ext3_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001961 .readpage = ext3_readpage,
1962 .readpages = ext3_readpages,
1963 .writepage = ext3_ordered_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001964 .write_begin = ext3_write_begin,
1965 .write_end = ext3_ordered_write_end,
1966 .bmap = ext3_bmap,
1967 .invalidatepage = ext3_invalidatepage,
1968 .releasepage = ext3_releasepage,
1969 .direct_IO = ext3_direct_IO,
1970 .migratepage = buffer_migrate_page,
1971 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001972 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973};
1974
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001975static const struct address_space_operations ext3_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001976 .readpage = ext3_readpage,
1977 .readpages = ext3_readpages,
1978 .writepage = ext3_writeback_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001979 .write_begin = ext3_write_begin,
1980 .write_end = ext3_writeback_write_end,
1981 .bmap = ext3_bmap,
1982 .invalidatepage = ext3_invalidatepage,
1983 .releasepage = ext3_releasepage,
1984 .direct_IO = ext3_direct_IO,
1985 .migratepage = buffer_migrate_page,
1986 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001987 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988};
1989
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001990static const struct address_space_operations ext3_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001991 .readpage = ext3_readpage,
1992 .readpages = ext3_readpages,
1993 .writepage = ext3_journalled_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001994 .write_begin = ext3_write_begin,
1995 .write_end = ext3_journalled_write_end,
1996 .set_page_dirty = ext3_journalled_set_page_dirty,
1997 .bmap = ext3_bmap,
1998 .invalidatepage = ext3_invalidatepage,
1999 .releasepage = ext3_releasepage,
2000 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002001 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002};
2003
2004void ext3_set_aops(struct inode *inode)
2005{
2006 if (ext3_should_order_data(inode))
2007 inode->i_mapping->a_ops = &ext3_ordered_aops;
2008 else if (ext3_should_writeback_data(inode))
2009 inode->i_mapping->a_ops = &ext3_writeback_aops;
2010 else
2011 inode->i_mapping->a_ops = &ext3_journalled_aops;
2012}
2013
2014/*
2015 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2016 * up to the end of the block which corresponds to `from'.
2017 * This required during truncate. We need to physically zero the tail end
2018 * of that block so it doesn't yield old data if the file is later grown.
2019 */
Jan Karaee3e77f2011-06-03 21:58:11 +02002020static int ext3_block_truncate_page(struct inode *inode, loff_t from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021{
Mingming Cao43d23f92006-06-25 05:48:07 -07002022 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Jan Karaee3e77f2011-06-03 21:58:11 +02002023 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 unsigned blocksize, iblock, length, pos;
Jan Karaee3e77f2011-06-03 21:58:11 +02002025 struct page *page;
2026 handle_t *handle = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 struct buffer_head *bh;
2028 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Jan Karaee3e77f2011-06-03 21:58:11 +02002030 /* Truncated on block boundary - nothing to do */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 blocksize = inode->i_sb->s_blocksize;
Jan Karaee3e77f2011-06-03 21:58:11 +02002032 if ((from & (blocksize - 1)) == 0)
2033 return 0;
2034
2035 page = grab_cache_page(inode->i_mapping, index);
2036 if (!page)
2037 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 length = blocksize - (offset & (blocksize - 1));
2039 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 if (!page_has_buffers(page))
2042 create_empty_buffers(page, blocksize, 0);
2043
2044 /* Find the buffer that contains "offset" */
2045 bh = page_buffers(page);
2046 pos = blocksize;
2047 while (offset >= pos) {
2048 bh = bh->b_this_page;
2049 iblock++;
2050 pos += blocksize;
2051 }
2052
2053 err = 0;
2054 if (buffer_freed(bh)) {
2055 BUFFER_TRACE(bh, "freed: skip");
2056 goto unlock;
2057 }
2058
2059 if (!buffer_mapped(bh)) {
2060 BUFFER_TRACE(bh, "unmapped");
2061 ext3_get_block(inode, iblock, bh, 0);
2062 /* unmapped? It's a hole - nothing to do */
2063 if (!buffer_mapped(bh)) {
2064 BUFFER_TRACE(bh, "still unmapped");
2065 goto unlock;
2066 }
2067 }
2068
2069 /* Ok, it's mapped. Make sure it's up-to-date */
2070 if (PageUptodate(page))
2071 set_buffer_uptodate(bh);
2072
Zheng Liud03e1292011-12-05 15:55:11 +08002073 if (!bh_uptodate_or_lock(bh)) {
2074 err = bh_submit_read(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 /* Uhhuh. Read error. Complain and punt. */
Zheng Liud03e1292011-12-05 15:55:11 +08002076 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 goto unlock;
2078 }
2079
Jan Karaee3e77f2011-06-03 21:58:11 +02002080 /* data=writeback mode doesn't need transaction to zero-out data */
2081 if (!ext3_should_writeback_data(inode)) {
2082 /* We journal at most one block */
2083 handle = ext3_journal_start(inode, 1);
2084 if (IS_ERR(handle)) {
2085 clear_highpage(page);
2086 flush_dcache_page(page);
2087 err = PTR_ERR(handle);
2088 goto unlock;
2089 }
2090 }
2091
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 if (ext3_should_journal_data(inode)) {
2093 BUFFER_TRACE(bh, "get write access");
2094 err = ext3_journal_get_write_access(handle, bh);
2095 if (err)
Jan Karaee3e77f2011-06-03 21:58:11 +02002096 goto stop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 }
2098
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002099 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 BUFFER_TRACE(bh, "zeroed end of block");
2101
2102 err = 0;
2103 if (ext3_should_journal_data(inode)) {
2104 err = ext3_journal_dirty_metadata(handle, bh);
2105 } else {
2106 if (ext3_should_order_data(inode))
2107 err = ext3_journal_dirty_data(handle, bh);
2108 mark_buffer_dirty(bh);
2109 }
Jan Karaee3e77f2011-06-03 21:58:11 +02002110stop:
2111 if (handle)
2112 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114unlock:
2115 unlock_page(page);
2116 page_cache_release(page);
2117 return err;
2118}
2119
2120/*
2121 * Probably it should be a library function... search for first non-zero word
2122 * or memcmp with zero_page, whatever is better for particular architecture.
2123 * Linus?
2124 */
2125static inline int all_zeroes(__le32 *p, __le32 *q)
2126{
2127 while (p < q)
2128 if (*p++)
2129 return 0;
2130 return 1;
2131}
2132
2133/**
2134 * ext3_find_shared - find the indirect blocks for partial truncation.
2135 * @inode: inode in question
2136 * @depth: depth of the affected branch
2137 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2138 * @chain: place to store the pointers to partial indirect blocks
2139 * @top: place to the (detached) top of branch
2140 *
2141 * This is a helper function used by ext3_truncate().
2142 *
2143 * When we do truncate() we may have to clean the ends of several
2144 * indirect blocks but leave the blocks themselves alive. Block is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002145 * partially truncated if some data below the new i_size is referred
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 * from it (and it is on the path to the first completely truncated
2147 * data block, indeed). We have to free the top of that path along
2148 * with everything to the right of the path. Since no allocation
2149 * past the truncation point is possible until ext3_truncate()
2150 * finishes, we may safely do the latter, but top of branch may
2151 * require special attention - pageout below the truncation point
2152 * might try to populate it.
2153 *
2154 * We atomically detach the top of branch from the tree, store the
2155 * block number of its root in *@top, pointers to buffer_heads of
2156 * partially truncated blocks - in @chain[].bh and pointers to
2157 * their last elements that should not be removed - in
2158 * @chain[].p. Return value is the pointer to last filled element
2159 * of @chain.
2160 *
2161 * The work left to caller to do the actual freeing of subtrees:
2162 * a) free the subtree starting from *@top
2163 * b) free the subtrees whose roots are stored in
2164 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2165 * c) free the subtrees growing from the inode past the @chain[0].
2166 * (no partially truncated stuff there). */
2167
Andrew Mortond6859bf2006-03-26 01:38:03 -08002168static Indirect *ext3_find_shared(struct inode *inode, int depth,
2169 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170{
2171 Indirect *partial, *p;
2172 int k, err;
2173
2174 *top = 0;
Uwe Kleine-Königbf48aab2009-10-28 20:11:03 +01002175 /* Make k index the deepest non-null offset + 1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 for (k = depth; k > 1 && !offsets[k-1]; k--)
2177 ;
2178 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2179 /* Writer: pointers */
2180 if (!partial)
2181 partial = chain + k-1;
2182 /*
2183 * If the branch acquired continuation since we've looked at it -
2184 * fine, it should all survive and (new) top doesn't belong to us.
2185 */
2186 if (!partial->key && *partial->p)
2187 /* Writer: end */
2188 goto no_top;
2189 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2190 ;
2191 /*
2192 * OK, we've found the last block that must survive. The rest of our
2193 * branch should be detached before unlocking. However, if that rest
2194 * of branch is all ours and does not grow immediately from the inode
2195 * it's easier to cheat and just decrement partial->p.
2196 */
2197 if (p == chain + k - 1 && p > chain) {
2198 p->p--;
2199 } else {
2200 *top = *p->p;
2201 /* Nope, don't do this in ext3. Must leave the tree intact */
2202#if 0
2203 *p->p = 0;
2204#endif
2205 }
2206 /* Writer: end */
2207
Andrew Mortond6859bf2006-03-26 01:38:03 -08002208 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 brelse(partial->bh);
2210 partial--;
2211 }
2212no_top:
2213 return partial;
2214}
2215
2216/*
2217 * Zero a number of block pointers in either an inode or an indirect block.
2218 * If we restart the transaction we must again get write access to the
2219 * indirect block for further modification.
2220 *
2221 * We release `count' blocks on disk, but (last - first) may be greater
2222 * than `count' because there can be holes in there.
2223 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002224static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07002225 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08002226 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227{
2228 __le32 *p;
2229 if (try_to_extend_transaction(handle, inode)) {
2230 if (bh) {
2231 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Namhyung Kim156e7432010-11-25 01:53:13 +09002232 if (ext3_journal_dirty_metadata(handle, bh))
2233 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 }
2235 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002236 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 if (bh) {
2238 BUFFER_TRACE(bh, "retaking write access");
Namhyung Kim156e7432010-11-25 01:53:13 +09002239 if (ext3_journal_get_write_access(handle, bh))
2240 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 }
2242 }
2243
2244 /*
2245 * Any buffers which are on the journal will be in memory. We find
2246 * them on the hash table so journal_revoke() will run journal_forget()
2247 * on them. We've already detached each block from the file, so
2248 * bforget() in journal_forget() should be safe.
2249 *
2250 * AKPM: turn on bforget in journal_forget()!!!
2251 */
2252 for (p = first; p < last; p++) {
2253 u32 nr = le32_to_cpu(*p);
2254 if (nr) {
2255 struct buffer_head *bh;
2256
2257 *p = 0;
2258 bh = sb_find_get_block(inode->i_sb, nr);
2259 ext3_forget(handle, 0, inode, bh, nr);
2260 }
2261 }
2262
2263 ext3_free_blocks(handle, inode, block_to_free, count);
2264}
2265
2266/**
2267 * ext3_free_data - free a list of data blocks
2268 * @handle: handle for this transaction
2269 * @inode: inode we are dealing with
2270 * @this_bh: indirect buffer_head which contains *@first and *@last
2271 * @first: array of block numbers
2272 * @last: points immediately past the end of array
2273 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002274 * We are freeing all blocks referred from that array (numbers are stored as
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2276 *
2277 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2278 * blocks are contiguous then releasing them at one time will only affect one
2279 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2280 * actually use a lot of journal space.
2281 *
2282 * @this_bh will be %NULL if @first and @last point into the inode's direct
2283 * block pointers.
2284 */
2285static void ext3_free_data(handle_t *handle, struct inode *inode,
2286 struct buffer_head *this_bh,
2287 __le32 *first, __le32 *last)
2288{
Mingming Cao43d23f92006-06-25 05:48:07 -07002289 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002290 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2292 corresponding to
2293 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002294 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 __le32 *p; /* Pointer into inode/ind
2296 for current block */
2297 int err;
2298
2299 if (this_bh) { /* For indirect block */
2300 BUFFER_TRACE(this_bh, "get_write_access");
2301 err = ext3_journal_get_write_access(handle, this_bh);
2302 /* Important: if we can't update the indirect pointers
2303 * to the blocks, we can't free them. */
2304 if (err)
2305 return;
2306 }
2307
2308 for (p = first; p < last; p++) {
2309 nr = le32_to_cpu(*p);
2310 if (nr) {
2311 /* accumulate blocks to free if they're contiguous */
2312 if (count == 0) {
2313 block_to_free = nr;
2314 block_to_free_p = p;
2315 count = 1;
2316 } else if (nr == block_to_free + count) {
2317 count++;
2318 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002319 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 block_to_free,
2321 count, block_to_free_p, p);
2322 block_to_free = nr;
2323 block_to_free_p = p;
2324 count = 1;
2325 }
2326 }
2327 }
2328
2329 if (count > 0)
2330 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2331 count, block_to_free_p, p);
2332
2333 if (this_bh) {
2334 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
Duane Griffin3ccc3162008-07-25 01:46:26 -07002335
2336 /*
2337 * The buffer head should have an attached journal head at this
2338 * point. However, if the data is corrupted and an indirect
2339 * block pointed to itself, it would have been detached when
2340 * the block was cleared. Check for this instead of OOPSing.
2341 */
2342 if (bh2jh(this_bh))
2343 ext3_journal_dirty_metadata(handle, this_bh);
2344 else
2345 ext3_error(inode->i_sb, "ext3_free_data",
2346 "circular indirect block detected, "
2347 "inode=%lu, block=%llu",
2348 inode->i_ino,
2349 (unsigned long long)this_bh->b_blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 }
2351}
2352
2353/**
2354 * ext3_free_branches - free an array of branches
2355 * @handle: JBD handle for this transaction
2356 * @inode: inode we are dealing with
2357 * @parent_bh: the buffer_head which contains *@first and *@last
2358 * @first: array of block numbers
2359 * @last: pointer immediately past the end of array
2360 * @depth: depth of the branches to free
2361 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002362 * We are freeing all blocks referred from these branches (numbers are
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 * stored as little-endian 32-bit) and updating @inode->i_blocks
2364 * appropriately.
2365 */
2366static void ext3_free_branches(handle_t *handle, struct inode *inode,
2367 struct buffer_head *parent_bh,
2368 __le32 *first, __le32 *last, int depth)
2369{
Mingming Cao43d23f92006-06-25 05:48:07 -07002370 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 __le32 *p;
2372
2373 if (is_handle_aborted(handle))
2374 return;
2375
2376 if (depth--) {
2377 struct buffer_head *bh;
2378 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2379 p = last;
2380 while (--p >= first) {
2381 nr = le32_to_cpu(*p);
2382 if (!nr)
2383 continue; /* A hole */
2384
2385 /* Go read the buffer for the next level down */
2386 bh = sb_bread(inode->i_sb, nr);
2387
2388 /*
2389 * A read failure? Report error and clear slot
2390 * (should be rare).
2391 */
2392 if (!bh) {
2393 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002394 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 inode->i_ino, nr);
2396 continue;
2397 }
2398
2399 /* This zaps the entire block. Bottom up. */
2400 BUFFER_TRACE(bh, "free child branches");
2401 ext3_free_branches(handle, inode, bh,
2402 (__le32*)bh->b_data,
2403 (__le32*)bh->b_data + addr_per_block,
2404 depth);
2405
2406 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 * Everything below this this pointer has been
2408 * released. Now let this top-of-subtree go.
2409 *
2410 * We want the freeing of this indirect block to be
2411 * atomic in the journal with the updating of the
2412 * bitmap block which owns it. So make some room in
2413 * the journal.
2414 *
2415 * We zero the parent pointer *after* freeing its
2416 * pointee in the bitmaps, so if extend_transaction()
2417 * for some reason fails to put the bitmap changes and
2418 * the release into the same transaction, recovery
2419 * will merely complain about releasing a free block,
2420 * rather than leaking blocks.
2421 */
2422 if (is_handle_aborted(handle))
2423 return;
2424 if (try_to_extend_transaction(handle, inode)) {
2425 ext3_mark_inode_dirty(handle, inode);
Jan Kara00171d32009-08-11 19:06:10 +02002426 truncate_restart_transaction(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 }
2428
Jan Karaf25f6242010-07-12 21:04:31 +02002429 /*
2430 * We've probably journalled the indirect block several
2431 * times during the truncate. But it's no longer
2432 * needed and we now drop it from the transaction via
2433 * journal_revoke().
2434 *
2435 * That's easy if it's exclusively part of this
2436 * transaction. But if it's part of the committing
2437 * transaction then journal_forget() will simply
2438 * brelse() it. That means that if the underlying
2439 * block is reallocated in ext3_get_block(),
2440 * unmap_underlying_metadata() will find this block
2441 * and will try to get rid of it. damn, damn. Thus
2442 * we don't allow a block to be reallocated until
2443 * a transaction freeing it has fully committed.
2444 *
2445 * We also have to make sure journal replay after a
2446 * crash does not overwrite non-journaled data blocks
2447 * with old metadata when the block got reallocated for
2448 * data. Thus we have to store a revoke record for a
2449 * block in the same transaction in which we free the
2450 * block.
2451 */
2452 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2453
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 ext3_free_blocks(handle, inode, nr, 1);
2455
2456 if (parent_bh) {
2457 /*
2458 * The block which we have just freed is
2459 * pointed to by an indirect block: journal it
2460 */
2461 BUFFER_TRACE(parent_bh, "get_write_access");
2462 if (!ext3_journal_get_write_access(handle,
2463 parent_bh)){
2464 *p = 0;
2465 BUFFER_TRACE(parent_bh,
2466 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002467 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 parent_bh);
2469 }
2470 }
2471 }
2472 } else {
2473 /* We have reached the bottom of the tree. */
2474 BUFFER_TRACE(parent_bh, "free data blocks");
2475 ext3_free_data(handle, inode, parent_bh, first, last);
2476 }
2477}
2478
Duane Griffinae76dd92008-07-25 01:46:23 -07002479int ext3_can_truncate(struct inode *inode)
2480{
Duane Griffinae76dd92008-07-25 01:46:23 -07002481 if (S_ISREG(inode->i_mode))
2482 return 1;
2483 if (S_ISDIR(inode->i_mode))
2484 return 1;
2485 if (S_ISLNK(inode->i_mode))
2486 return !ext3_inode_is_fast_symlink(inode);
2487 return 0;
2488}
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490/*
2491 * ext3_truncate()
2492 *
2493 * We block out ext3_get_block() block instantiations across the entire
2494 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2495 * simultaneously on behalf of the same inode.
2496 *
2497 * As we work through the truncate and commmit bits of it to the journal there
2498 * is one core, guiding principle: the file's tree must always be consistent on
2499 * disk. We must be able to restart the truncate after a crash.
2500 *
2501 * The file's tree may be transiently inconsistent in memory (although it
2502 * probably isn't), but whenever we close off and commit a journal transaction,
2503 * the contents of (the filesystem + the journal) must be consistent and
2504 * restartable. It's pretty simple, really: bottom up, right to left (although
2505 * left-to-right works OK too).
2506 *
2507 * Note that at recovery time, journal replay occurs *before* the restart of
2508 * truncate against the orphan inode list.
2509 *
2510 * The committed inode has the new, desired i_size (which is the same as
2511 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2512 * that this inode's truncate did not complete and it will again call
2513 * ext3_truncate() to have another go. So there will be instantiated blocks
2514 * to the right of the truncation point in a crashed ext3 filesystem. But
2515 * that's fine - as long as they are linked from the inode, the post-crash
2516 * ext3_truncate() run will find them and release them.
2517 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002518void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519{
2520 handle_t *handle;
2521 struct ext3_inode_info *ei = EXT3_I(inode);
2522 __le32 *i_data = ei->i_data;
2523 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 int offsets[4];
2525 Indirect chain[4];
2526 Indirect *partial;
2527 __le32 nr = 0;
2528 int n;
2529 long last_block;
2530 unsigned blocksize = inode->i_sb->s_blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002532 trace_ext3_truncate_enter(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
Duane Griffinae76dd92008-07-25 01:46:23 -07002534 if (!ext3_can_truncate(inode))
Jan Karaef436182009-06-17 16:26:24 -07002535 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002537 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
Jan Kara9df93932010-01-06 21:58:48 +01002538 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 handle = start_transaction(inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02002541 if (IS_ERR(handle))
Jan Karaef436182009-06-17 16:26:24 -07002542 goto out_notrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
2544 last_block = (inode->i_size + blocksize-1)
2545 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2547 if (n == 0)
2548 goto out_stop; /* error */
2549
2550 /*
2551 * OK. This truncate is going to happen. We add the inode to the
2552 * orphan list, so that if this truncate spans multiple transactions,
2553 * and we crash, we will resume the truncate when the filesystem
2554 * recovers. It also marks the inode dirty, to catch the new size.
2555 *
2556 * Implication: the file must always be in a sane, consistent
2557 * truncatable state while each transaction commits.
2558 */
2559 if (ext3_orphan_add(handle, inode))
2560 goto out_stop;
2561
2562 /*
2563 * The orphan list entry will now protect us from any crash which
2564 * occurs before the truncate completes, so it is now safe to propagate
2565 * the new, shorter inode size (held for now in i_size) into the
2566 * on-disk inode. We do this via i_disksize, which is the value which
2567 * ext3 *really* writes onto the disk inode.
2568 */
2569 ei->i_disksize = inode->i_size;
2570
2571 /*
2572 * From here we block out all ext3_get_block() callers who want to
2573 * modify the block allocation tree.
2574 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002575 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576
2577 if (n == 1) { /* direct blocks */
2578 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2579 i_data + EXT3_NDIR_BLOCKS);
2580 goto do_indirects;
2581 }
2582
2583 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2584 /* Kill the top of shared branch (not detached) */
2585 if (nr) {
2586 if (partial == chain) {
2587 /* Shared branch grows from the inode */
2588 ext3_free_branches(handle, inode, NULL,
2589 &nr, &nr+1, (chain+n-1) - partial);
2590 *partial->p = 0;
2591 /*
2592 * We mark the inode dirty prior to restart,
2593 * and prior to stop. No need for it here.
2594 */
2595 } else {
2596 /* Shared branch grows from an indirect block */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 ext3_free_branches(handle, inode, partial->bh,
2598 partial->p,
2599 partial->p+1, (chain+n-1) - partial);
2600 }
2601 }
2602 /* Clear the ends of indirect blocks on the shared branch */
2603 while (partial > chain) {
2604 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2605 (__le32*)partial->bh->b_data+addr_per_block,
2606 (chain+n-1) - partial);
2607 BUFFER_TRACE(partial->bh, "call brelse");
2608 brelse (partial->bh);
2609 partial--;
2610 }
2611do_indirects:
2612 /* Kill the remaining (whole) subtrees */
2613 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002614 default:
2615 nr = i_data[EXT3_IND_BLOCK];
2616 if (nr) {
2617 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2618 i_data[EXT3_IND_BLOCK] = 0;
2619 }
2620 case EXT3_IND_BLOCK:
2621 nr = i_data[EXT3_DIND_BLOCK];
2622 if (nr) {
2623 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2624 i_data[EXT3_DIND_BLOCK] = 0;
2625 }
2626 case EXT3_DIND_BLOCK:
2627 nr = i_data[EXT3_TIND_BLOCK];
2628 if (nr) {
2629 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2630 i_data[EXT3_TIND_BLOCK] = 0;
2631 }
2632 case EXT3_TIND_BLOCK:
2633 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 }
2635
2636 ext3_discard_reservation(inode);
2637
Arjan van de Ven97461512006-03-23 03:00:42 -08002638 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2640 ext3_mark_inode_dirty(handle, inode);
2641
Andrew Mortond6859bf2006-03-26 01:38:03 -08002642 /*
2643 * In a multi-transaction truncate, we only make the final transaction
2644 * synchronous
2645 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 if (IS_SYNC(inode))
2647 handle->h_sync = 1;
2648out_stop:
2649 /*
2650 * If this was a simple ftruncate(), and the file will remain alive
2651 * then we need to clear up the orphan record which we created above.
2652 * However, if this was a real unlink then we were called by
Al Viroac14a952010-06-06 07:08:19 -04002653 * ext3_evict_inode(), and we allow that function to clean up the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 * orphan info for us.
2655 */
2656 if (inode->i_nlink)
2657 ext3_orphan_del(handle, inode);
2658
2659 ext3_journal_stop(handle);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002660 trace_ext3_truncate_exit(inode);
Jan Karaef436182009-06-17 16:26:24 -07002661 return;
2662out_notrans:
2663 /*
2664 * Delete the inode from orphan list so that it doesn't stay there
2665 * forever and trigger assertion on umount.
2666 */
2667 if (inode->i_nlink)
2668 ext3_orphan_del(NULL, inode);
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002669 trace_ext3_truncate_exit(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670}
2671
Mingming Cao43d23f92006-06-25 05:48:07 -07002672static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 unsigned long ino, struct ext3_iloc *iloc)
2674{
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002675 unsigned long block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002676 unsigned long offset;
2677 ext3_fsblk_t block;
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002678 struct ext3_group_desc *gdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679
Neil Brown2ccb48e2006-07-30 03:03:01 -07002680 if (!ext3_valid_inum(sb, ino)) {
2681 /*
2682 * This error is already checked for in namei.c unless we are
2683 * looking at an NFS filehandle, in which case no error
2684 * report is needed
2685 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 return 0;
2687 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002688
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002690 gdp = ext3_get_group_desc(sb, block_group, NULL);
2691 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 /*
2694 * Figure out the offset within the block group inode table
2695 */
2696 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2697 EXT3_INODE_SIZE(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002698 block = le32_to_cpu(gdp->bg_inode_table) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2700
2701 iloc->block_group = block_group;
2702 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2703 return block;
2704}
2705
2706/*
2707 * ext3_get_inode_loc returns with an extra refcount against the inode's
2708 * underlying buffer_head on success. If 'in_mem' is true, we have all
2709 * data in memory that is needed to recreate the on-disk version of this
2710 * inode.
2711 */
2712static int __ext3_get_inode_loc(struct inode *inode,
2713 struct ext3_iloc *iloc, int in_mem)
2714{
Mingming Cao43d23f92006-06-25 05:48:07 -07002715 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 struct buffer_head *bh;
2717
2718 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2719 if (!block)
2720 return -EIO;
2721
2722 bh = sb_getblk(inode->i_sb, block);
2723 if (!bh) {
2724 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2725 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002726 "inode=%lu, block="E3FSBLK,
2727 inode->i_ino, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 return -EIO;
2729 }
2730 if (!buffer_uptodate(bh)) {
2731 lock_buffer(bh);
Hidehiro Kawai95450f52008-07-25 01:46:24 -07002732
2733 /*
2734 * If the buffer has the write error flag, we have failed
2735 * to write out another inode in the same block. In this
2736 * case, we don't have to read the block because we may
2737 * read the old inode data successfully.
2738 */
2739 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2740 set_buffer_uptodate(bh);
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 if (buffer_uptodate(bh)) {
2743 /* someone brought it uptodate while we waited */
2744 unlock_buffer(bh);
2745 goto has_buffer;
2746 }
2747
2748 /*
2749 * If we have all information of the inode in memory and this
2750 * is the only valid inode in the block, we need not read the
2751 * block.
2752 */
2753 if (in_mem) {
2754 struct buffer_head *bitmap_bh;
2755 struct ext3_group_desc *desc;
2756 int inodes_per_buffer;
2757 int inode_offset, i;
2758 int block_group;
2759 int start;
2760
2761 block_group = (inode->i_ino - 1) /
2762 EXT3_INODES_PER_GROUP(inode->i_sb);
2763 inodes_per_buffer = bh->b_size /
2764 EXT3_INODE_SIZE(inode->i_sb);
2765 inode_offset = ((inode->i_ino - 1) %
2766 EXT3_INODES_PER_GROUP(inode->i_sb));
2767 start = inode_offset & ~(inodes_per_buffer - 1);
2768
2769 /* Is the inode bitmap in cache? */
2770 desc = ext3_get_group_desc(inode->i_sb,
2771 block_group, NULL);
2772 if (!desc)
2773 goto make_io;
2774
2775 bitmap_bh = sb_getblk(inode->i_sb,
2776 le32_to_cpu(desc->bg_inode_bitmap));
2777 if (!bitmap_bh)
2778 goto make_io;
2779
2780 /*
2781 * If the inode bitmap isn't in cache then the
2782 * optimisation may end up performing two reads instead
2783 * of one, so skip it.
2784 */
2785 if (!buffer_uptodate(bitmap_bh)) {
2786 brelse(bitmap_bh);
2787 goto make_io;
2788 }
2789 for (i = start; i < start + inodes_per_buffer; i++) {
2790 if (i == inode_offset)
2791 continue;
2792 if (ext3_test_bit(i, bitmap_bh->b_data))
2793 break;
2794 }
2795 brelse(bitmap_bh);
2796 if (i == start + inodes_per_buffer) {
2797 /* all other inodes are free, so skip I/O */
2798 memset(bh->b_data, 0, bh->b_size);
2799 set_buffer_uptodate(bh);
2800 unlock_buffer(bh);
2801 goto has_buffer;
2802 }
2803 }
2804
2805make_io:
2806 /*
2807 * There are other valid inodes in the buffer, this inode
2808 * has in-inode xattrs, or we don't have this inode in memory.
2809 * Read the block from disk.
2810 */
Lukas Czerner785c4bc2011-05-23 18:33:01 +02002811 trace_ext3_load_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 get_bh(bh);
2813 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02002814 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 wait_on_buffer(bh);
2816 if (!buffer_uptodate(bh)) {
2817 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2818 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002819 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 inode->i_ino, block);
2821 brelse(bh);
2822 return -EIO;
2823 }
2824 }
2825has_buffer:
2826 iloc->bh = bh;
2827 return 0;
2828}
2829
2830int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2831{
2832 /* We have all inode data except xattrs in memory here. */
2833 return __ext3_get_inode_loc(inode, iloc,
Jan Kara9df93932010-01-06 21:58:48 +01002834 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835}
2836
2837void ext3_set_inode_flags(struct inode *inode)
2838{
2839 unsigned int flags = EXT3_I(inode)->i_flags;
2840
2841 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2842 if (flags & EXT3_SYNC_FL)
2843 inode->i_flags |= S_SYNC;
2844 if (flags & EXT3_APPEND_FL)
2845 inode->i_flags |= S_APPEND;
2846 if (flags & EXT3_IMMUTABLE_FL)
2847 inode->i_flags |= S_IMMUTABLE;
2848 if (flags & EXT3_NOATIME_FL)
2849 inode->i_flags |= S_NOATIME;
2850 if (flags & EXT3_DIRSYNC_FL)
2851 inode->i_flags |= S_DIRSYNC;
2852}
2853
Jan Kara28be5ab2007-05-08 00:30:33 -07002854/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2855void ext3_get_inode_flags(struct ext3_inode_info *ei)
2856{
2857 unsigned int flags = ei->vfs_inode.i_flags;
2858
2859 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2860 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2861 if (flags & S_SYNC)
2862 ei->i_flags |= EXT3_SYNC_FL;
2863 if (flags & S_APPEND)
2864 ei->i_flags |= EXT3_APPEND_FL;
2865 if (flags & S_IMMUTABLE)
2866 ei->i_flags |= EXT3_IMMUTABLE_FL;
2867 if (flags & S_NOATIME)
2868 ei->i_flags |= EXT3_NOATIME_FL;
2869 if (flags & S_DIRSYNC)
2870 ei->i_flags |= EXT3_DIRSYNC_FL;
2871}
2872
David Howells473043d2008-02-07 00:15:36 -08002873struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874{
2875 struct ext3_iloc iloc;
2876 struct ext3_inode *raw_inode;
David Howells473043d2008-02-07 00:15:36 -08002877 struct ext3_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 struct buffer_head *bh;
David Howells473043d2008-02-07 00:15:36 -08002879 struct inode *inode;
Jan Karafe8bc912009-10-16 19:26:15 +02002880 journal_t *journal = EXT3_SB(sb)->s_journal;
2881 transaction_t *transaction;
David Howells473043d2008-02-07 00:15:36 -08002882 long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 int block;
2884
David Howells473043d2008-02-07 00:15:36 -08002885 inode = iget_locked(sb, ino);
2886 if (!inode)
2887 return ERR_PTR(-ENOMEM);
2888 if (!(inode->i_state & I_NEW))
2889 return inode;
2890
2891 ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 ei->i_block_alloc_info = NULL;
2893
David Howells473043d2008-02-07 00:15:36 -08002894 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2895 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 goto bad_inode;
2897 bh = iloc.bh;
2898 raw_inode = ext3_raw_inode(&iloc);
2899 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2900 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2901 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2902 if(!(test_opt (inode->i_sb, NO_UID32))) {
2903 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2904 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2905 }
Miklos Szeredibfe86842011-10-28 14:13:29 +02002906 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002908 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2909 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2910 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2912
Linus Torvaldsde329822010-03-29 14:30:19 -07002913 ei->i_state_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 ei->i_dir_start_lookup = 0;
2915 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2916 /* We now have enough fields to check if the inode was active or not.
2917 * This is needed because nfsd might try to access dead inodes
2918 * the test is that same one that e2fsck uses
2919 * NeilBrown 1999oct15
2920 */
2921 if (inode->i_nlink == 0) {
2922 if (inode->i_mode == 0 ||
2923 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2924 /* this inode is deleted */
2925 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002926 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 goto bad_inode;
2928 }
2929 /* The only unlinked inodes we let through here have
2930 * valid i_mode and are being read by the orphan
2931 * recovery code: that's fine, we're about to complete
2932 * the process of deleting those. */
2933 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2935 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2936#ifdef EXT3_FRAGMENTS
2937 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2938 ei->i_frag_no = raw_inode->i_frag;
2939 ei->i_frag_size = raw_inode->i_fsize;
2940#endif
2941 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2942 if (!S_ISREG(inode->i_mode)) {
2943 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2944 } else {
2945 inode->i_size |=
2946 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2947 }
2948 ei->i_disksize = inode->i_size;
2949 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2950 ei->i_block_group = iloc.block_group;
2951 /*
2952 * NOTE! The in-memory inode i_data array is in little-endian order
2953 * even on big-endian machines: we do NOT byteswap the block numbers!
2954 */
2955 for (block = 0; block < EXT3_N_BLOCKS; block++)
2956 ei->i_data[block] = raw_inode->i_block[block];
2957 INIT_LIST_HEAD(&ei->i_orphan);
2958
Jan Karafe8bc912009-10-16 19:26:15 +02002959 /*
2960 * Set transaction id's of transactions that have to be committed
2961 * to finish f[data]sync. We set them to currently running transaction
2962 * as we cannot be sure that the inode or some of its metadata isn't
2963 * part of the transaction - the inode could have been reclaimed and
2964 * now it is reread from disk.
2965 */
2966 if (journal) {
2967 tid_t tid;
2968
2969 spin_lock(&journal->j_state_lock);
2970 if (journal->j_running_transaction)
2971 transaction = journal->j_running_transaction;
2972 else
2973 transaction = journal->j_committing_transaction;
2974 if (transaction)
2975 tid = transaction->t_tid;
2976 else
2977 tid = journal->j_commit_sequence;
2978 spin_unlock(&journal->j_state_lock);
2979 atomic_set(&ei->i_sync_tid, tid);
2980 atomic_set(&ei->i_datasync_tid, tid);
2981 }
2982
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2984 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2985 /*
2986 * When mke2fs creates big inodes it does not zero out
2987 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2988 * so ignore those first few inodes.
2989 */
2990 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2991 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07002992 EXT3_INODE_SIZE(inode->i_sb)) {
2993 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002994 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 goto bad_inode;
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07002996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 if (ei->i_extra_isize == 0) {
2998 /* The extra space is currently unused. Use it. */
2999 ei->i_extra_isize = sizeof(struct ext3_inode) -
3000 EXT3_GOOD_OLD_INODE_SIZE;
3001 } else {
3002 __le32 *magic = (void *)raw_inode +
3003 EXT3_GOOD_OLD_INODE_SIZE +
3004 ei->i_extra_isize;
3005 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
Jan Kara9df93932010-01-06 21:58:48 +01003006 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 }
3008 } else
3009 ei->i_extra_isize = 0;
3010
3011 if (S_ISREG(inode->i_mode)) {
3012 inode->i_op = &ext3_file_inode_operations;
3013 inode->i_fop = &ext3_file_operations;
3014 ext3_set_aops(inode);
3015 } else if (S_ISDIR(inode->i_mode)) {
3016 inode->i_op = &ext3_dir_inode_operations;
3017 inode->i_fop = &ext3_dir_operations;
3018 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffinb5ed3112008-12-19 20:47:14 +00003019 if (ext3_inode_is_fast_symlink(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 inode->i_op = &ext3_fast_symlink_inode_operations;
Duane Griffinb5ed3112008-12-19 20:47:14 +00003021 nd_terminate_link(ei->i_data, inode->i_size,
3022 sizeof(ei->i_data) - 1);
3023 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 inode->i_op = &ext3_symlink_inode_operations;
3025 ext3_set_aops(inode);
3026 }
3027 } else {
3028 inode->i_op = &ext3_special_inode_operations;
3029 if (raw_inode->i_block[0])
3030 init_special_inode(inode, inode->i_mode,
3031 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003032 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 init_special_inode(inode, inode->i_mode,
3034 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3035 }
3036 brelse (iloc.bh);
3037 ext3_set_inode_flags(inode);
David Howells473043d2008-02-07 00:15:36 -08003038 unlock_new_inode(inode);
3039 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
3041bad_inode:
David Howells473043d2008-02-07 00:15:36 -08003042 iget_failed(inode);
3043 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044}
3045
3046/*
3047 * Post the struct inode info into an on-disk inode location in the
3048 * buffer-cache. This gobbles the caller's reference to the
3049 * buffer_head in the inode location struct.
3050 *
3051 * The caller must have write access to iloc->bh.
3052 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003053static int ext3_do_update_inode(handle_t *handle,
3054 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 struct ext3_iloc *iloc)
3056{
3057 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3058 struct ext3_inode_info *ei = EXT3_I(inode);
3059 struct buffer_head *bh = iloc->bh;
3060 int err = 0, rc, block;
3061
Chris Mason4f003fd2009-09-08 00:22:14 +02003062again:
3063 /* we can't allow multiple procs in here at once, its a bit racey */
3064 lock_buffer(bh);
3065
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 /* For fields not not tracking in the in-memory inode,
3067 * initialise them to zero for new inodes. */
Jan Kara9df93932010-01-06 21:58:48 +01003068 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3070
Jan Kara28be5ab2007-05-08 00:30:33 -07003071 ext3_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3073 if(!(test_opt(inode->i_sb, NO_UID32))) {
3074 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3075 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3076/*
3077 * Fix up interoperability with old kernels. Otherwise, old inodes get
3078 * re-used with the upper 16 bits of the uid/gid intact
3079 */
3080 if(!ei->i_dtime) {
3081 raw_inode->i_uid_high =
3082 cpu_to_le16(high_16_bits(inode->i_uid));
3083 raw_inode->i_gid_high =
3084 cpu_to_le16(high_16_bits(inode->i_gid));
3085 } else {
3086 raw_inode->i_uid_high = 0;
3087 raw_inode->i_gid_high = 0;
3088 }
3089 } else {
3090 raw_inode->i_uid_low =
3091 cpu_to_le16(fs_high2lowuid(inode->i_uid));
3092 raw_inode->i_gid_low =
3093 cpu_to_le16(fs_high2lowgid(inode->i_gid));
3094 raw_inode->i_uid_high = 0;
3095 raw_inode->i_gid_high = 0;
3096 }
3097 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3098 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
3099 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3100 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3101 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3102 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3103 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3104 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3105#ifdef EXT3_FRAGMENTS
3106 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3107 raw_inode->i_frag = ei->i_frag_no;
3108 raw_inode->i_fsize = ei->i_frag_size;
3109#endif
3110 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3111 if (!S_ISREG(inode->i_mode)) {
3112 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3113 } else {
3114 raw_inode->i_size_high =
3115 cpu_to_le32(ei->i_disksize >> 32);
3116 if (ei->i_disksize > 0x7fffffffULL) {
3117 struct super_block *sb = inode->i_sb;
3118 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3119 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3120 EXT3_SB(sb)->s_es->s_rev_level ==
3121 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3122 /* If this is the first large file
3123 * created, add a flag to the superblock.
3124 */
Chris Mason4f003fd2009-09-08 00:22:14 +02003125 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 err = ext3_journal_get_write_access(handle,
3127 EXT3_SB(sb)->s_sbh);
3128 if (err)
3129 goto out_brelse;
Chris Mason4f003fd2009-09-08 00:22:14 +02003130
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 ext3_update_dynamic_rev(sb);
3132 EXT3_SET_RO_COMPAT_FEATURE(sb,
3133 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 handle->h_sync = 1;
3135 err = ext3_journal_dirty_metadata(handle,
3136 EXT3_SB(sb)->s_sbh);
Chris Mason4f003fd2009-09-08 00:22:14 +02003137 /* get our lock and start over */
3138 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 }
3140 }
3141 }
3142 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3143 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3144 if (old_valid_dev(inode->i_rdev)) {
3145 raw_inode->i_block[0] =
3146 cpu_to_le32(old_encode_dev(inode->i_rdev));
3147 raw_inode->i_block[1] = 0;
3148 } else {
3149 raw_inode->i_block[0] = 0;
3150 raw_inode->i_block[1] =
3151 cpu_to_le32(new_encode_dev(inode->i_rdev));
3152 raw_inode->i_block[2] = 0;
3153 }
3154 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3155 raw_inode->i_block[block] = ei->i_data[block];
3156
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07003157 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3159
3160 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
Chris Mason4f003fd2009-09-08 00:22:14 +02003161 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 rc = ext3_journal_dirty_metadata(handle, bh);
3163 if (!err)
3164 err = rc;
Jan Kara9df93932010-01-06 21:58:48 +01003165 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166
Jan Karafe8bc912009-10-16 19:26:15 +02003167 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168out_brelse:
3169 brelse (bh);
3170 ext3_std_error(inode->i_sb, err);
3171 return err;
3172}
3173
3174/*
3175 * ext3_write_inode()
3176 *
3177 * We are called from a few places:
3178 *
3179 * - Within generic_file_write() for O_SYNC files.
3180 * Here, there will be no transaction running. We wait for any running
3181 * trasnaction to commit.
3182 *
3183 * - Within sys_sync(), kupdate and such.
3184 * We wait on commit, if tol to.
3185 *
3186 * - Within prune_icache() (PF_MEMALLOC == true)
3187 * Here we simply return. We can't afford to block kswapd on the
3188 * journal commit.
3189 *
3190 * In all cases it is actually safe for us to return without doing anything,
3191 * because the inode has been copied into a raw inode buffer in
3192 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3193 * knfsd.
3194 *
3195 * Note that we are absolutely dependent upon all inode dirtiers doing the
3196 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3197 * which we are interested.
3198 *
3199 * It would be a bug for them to not do this. The code:
3200 *
3201 * mark_inode_dirty(inode)
3202 * stuff();
3203 * inode->i_size = expr;
3204 *
3205 * is in error because a kswapd-driven write_inode() could occur while
3206 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3207 * will no longer be on the superblock's dirty inode list.
3208 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003209int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210{
3211 if (current->flags & PF_MEMALLOC)
3212 return 0;
3213
3214 if (ext3_journal_current_handle()) {
Jose R. Santos9ad163a2007-10-18 23:39:23 -07003215 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 dump_stack();
3217 return -EIO;
3218 }
3219
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003220 if (wbc->sync_mode != WB_SYNC_ALL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221 return 0;
3222
3223 return ext3_force_commit(inode->i_sb);
3224}
3225
3226/*
3227 * ext3_setattr()
3228 *
3229 * Called from notify_change.
3230 *
3231 * We want to trap VFS attempts to truncate the file as soon as
3232 * possible. In particular, we want to make sure that when the VFS
3233 * shrinks i_size, we put the inode on the orphan list and modify
3234 * i_disksize immediately, so that during the subsequent flushing of
3235 * dirty pages and freeing of disk blocks, we can guarantee that any
3236 * commit will leave the blocks being flushed in an unused state on
3237 * disk. (On recovery, the inode will get truncated and the blocks will
3238 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003239 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 *
3241 * Called with inode->sem down.
3242 */
3243int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3244{
3245 struct inode *inode = dentry->d_inode;
3246 int error, rc = 0;
3247 const unsigned int ia_valid = attr->ia_valid;
3248
3249 error = inode_change_ok(inode, attr);
3250 if (error)
3251 return error;
3252
Dmitry Monakhov12755622010-04-08 22:04:20 +04003253 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05003254 dquot_initialize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3256 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3257 handle_t *handle;
3258
3259 /* (user+group)*(old+new) structure, inode write (sb,
3260 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003261 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3262 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 if (IS_ERR(handle)) {
3264 error = PTR_ERR(handle);
3265 goto err_out;
3266 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05003267 error = dquot_transfer(inode, attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 if (error) {
3269 ext3_journal_stop(handle);
3270 return error;
3271 }
3272 /* Update corresponding info in inode so that everything is in
3273 * one transaction */
3274 if (attr->ia_valid & ATTR_UID)
3275 inode->i_uid = attr->ia_uid;
3276 if (attr->ia_valid & ATTR_GID)
3277 inode->i_gid = attr->ia_gid;
3278 error = ext3_mark_inode_dirty(handle, inode);
3279 ext3_journal_stop(handle);
3280 }
3281
Christoph Hellwig562c72a2011-06-24 14:29:45 -04003282 if (attr->ia_valid & ATTR_SIZE)
3283 inode_dio_wait(inode);
3284
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 if (S_ISREG(inode->i_mode) &&
3286 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3287 handle_t *handle;
3288
3289 handle = ext3_journal_start(inode, 3);
3290 if (IS_ERR(handle)) {
3291 error = PTR_ERR(handle);
3292 goto err_out;
3293 }
3294
3295 error = ext3_orphan_add(handle, inode);
Jan Karaee3e77f2011-06-03 21:58:11 +02003296 if (error) {
3297 ext3_journal_stop(handle);
3298 goto err_out;
3299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 EXT3_I(inode)->i_disksize = attr->ia_size;
Jan Karaee3e77f2011-06-03 21:58:11 +02003301 error = ext3_mark_inode_dirty(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 ext3_journal_stop(handle);
Jan Karaee3e77f2011-06-03 21:58:11 +02003303 if (error) {
3304 /* Some hard fs error must have happened. Bail out. */
3305 ext3_orphan_del(NULL, inode);
3306 goto err_out;
3307 }
3308 rc = ext3_block_truncate_page(inode, attr->ia_size);
3309 if (rc) {
3310 /* Cleanup orphan list and exit */
3311 handle = ext3_journal_start(inode, 3);
3312 if (IS_ERR(handle)) {
3313 ext3_orphan_del(NULL, inode);
3314 goto err_out;
3315 }
3316 ext3_orphan_del(handle, inode);
3317 ext3_journal_stop(handle);
3318 goto err_out;
3319 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 }
3321
Christoph Hellwig10257742010-06-04 11:30:02 +02003322 if ((attr->ia_valid & ATTR_SIZE) &&
3323 attr->ia_size != i_size_read(inode)) {
Jan Kara40680f2f2011-05-24 22:24:47 +02003324 truncate_setsize(inode, attr->ia_size);
3325 ext3_truncate(inode);
Christoph Hellwig10257742010-06-04 11:30:02 +02003326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327
Christoph Hellwig10257742010-06-04 11:30:02 +02003328 setattr_copy(inode, attr);
3329 mark_inode_dirty(inode);
3330
3331 if (ia_valid & ATTR_MODE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 rc = ext3_acl_chmod(inode);
3333
3334err_out:
3335 ext3_std_error(inode->i_sb, error);
3336 if (!error)
3337 error = rc;
3338 return error;
3339}
3340
3341
3342/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003343 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344 *
3345 * With N blocks per page, it may be:
3346 * N data blocks
3347 * 2 indirect block
3348 * 2 dindirect
3349 * 1 tindirect
3350 * N+5 bitmap blocks (from the above)
3351 * N+5 group descriptor summary blocks
3352 * 1 inode block
3353 * 1 superblock.
3354 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3355 *
3356 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3357 *
3358 * With ordered or writeback data it's the same, less the N data blocks.
3359 *
3360 * If the inode's direct blocks can hold an integral number of pages then a
3361 * page cannot straddle two indirect blocks, and we can only touch one indirect
3362 * and dindirect block, and the "5" above becomes "3".
3363 *
3364 * This still overestimates under most circumstances. If we were to pass the
3365 * start and end offsets in here as well we could do block_to_path() on each
3366 * block and work out the exact number of indirects which are touched. Pah.
3367 */
3368
3369static int ext3_writepage_trans_blocks(struct inode *inode)
3370{
3371 int bpp = ext3_journal_blocks_per_page(inode);
3372 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3373 int ret;
3374
3375 if (ext3_should_journal_data(inode))
3376 ret = 3 * (bpp + indirects) + 2;
3377 else
Yongqiang Yang523334b2011-03-24 08:48:39 +08003378 ret = 2 * (bpp + indirects) + indirects + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
3380#ifdef CONFIG_QUOTA
Christoph Hellwig871a2932010-03-03 09:05:07 -05003381 /* We know that structure was already allocated during dquot_initialize so
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 * we will be updating only the data blocks + inodes */
Dmitry Monakhovc4590012009-12-09 03:05:30 +03003383 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384#endif
3385
3386 return ret;
3387}
3388
3389/*
3390 * The caller must have previously called ext3_reserve_inode_write().
3391 * Give this, we know that the caller already has write access to iloc->bh.
3392 */
3393int ext3_mark_iloc_dirty(handle_t *handle,
3394 struct inode *inode, struct ext3_iloc *iloc)
3395{
3396 int err = 0;
3397
3398 /* the do_update_inode consumes one bh->b_count */
3399 get_bh(iloc->bh);
3400
3401 /* ext3_do_update_inode() does journal_dirty_metadata */
3402 err = ext3_do_update_inode(handle, inode, iloc);
3403 put_bh(iloc->bh);
3404 return err;
3405}
3406
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003407/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003409 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410 */
3411
3412int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003413ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 struct ext3_iloc *iloc)
3415{
3416 int err = 0;
3417 if (handle) {
3418 err = ext3_get_inode_loc(inode, iloc);
3419 if (!err) {
3420 BUFFER_TRACE(iloc->bh, "get_write_access");
3421 err = ext3_journal_get_write_access(handle, iloc->bh);
3422 if (err) {
3423 brelse(iloc->bh);
3424 iloc->bh = NULL;
3425 }
3426 }
3427 }
3428 ext3_std_error(inode->i_sb, err);
3429 return err;
3430}
3431
3432/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003433 * What we do here is to mark the in-core inode as clean with respect to inode
3434 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 * This means that the in-core inode may be reaped by prune_icache
3436 * without having to perform any I/O. This is a very good thing,
3437 * because *any* task may call prune_icache - even ones which
3438 * have a transaction open against a different journal.
3439 *
3440 * Is this cheating? Not really. Sure, we haven't written the
3441 * inode out, but prune_icache isn't a user-visible syncing function.
3442 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3443 * we start and wait on commits.
3444 *
3445 * Is this efficient/effective? Well, we're being nice to the system
3446 * by cleaning up our inodes proactively so they can be reaped
3447 * without I/O. But we are potentially leaving up to five seconds'
3448 * worth of inodes floating about which prune_icache wants us to
3449 * write out. One way to fix that would be to get prune_icache()
3450 * to do a write_super() to free up some memory. It has the desired
3451 * effect.
3452 */
3453int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3454{
3455 struct ext3_iloc iloc;
3456 int err;
3457
3458 might_sleep();
Lukas Czerner785c4bc2011-05-23 18:33:01 +02003459 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 err = ext3_reserve_inode_write(handle, inode, &iloc);
3461 if (!err)
3462 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3463 return err;
3464}
3465
3466/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003467 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468 *
3469 * We're really interested in the case where a file is being extended.
3470 * i_size has been changed by generic_commit_write() and we thus need
3471 * to include the updated inode in the current transaction.
3472 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05003473 * Also, dquot_alloc_space() will always dirty the inode when blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 * are allocated to the file.
3475 *
3476 * If the inode is marked synchronous, we don't honour that here - doing
3477 * so would cause a commit on atime updates, which we don't bother doing.
3478 * We handle synchronous inodes at the highest possible level.
3479 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04003480void ext3_dirty_inode(struct inode *inode, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481{
3482 handle_t *current_handle = ext3_journal_current_handle();
3483 handle_t *handle;
3484
3485 handle = ext3_journal_start(inode, 2);
3486 if (IS_ERR(handle))
3487 goto out;
3488 if (current_handle &&
3489 current_handle->h_transaction != handle->h_transaction) {
3490 /* This task has a transaction open against a different fs */
3491 printk(KERN_EMERG "%s: transactions do not match!\n",
Harvey Harrisone05b6b52008-04-28 02:16:15 -07003492 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 } else {
3494 jbd_debug(5, "marking dirty. outer handle=%p\n",
3495 current_handle);
3496 ext3_mark_inode_dirty(handle, inode);
3497 }
3498 ext3_journal_stop(handle);
3499out:
3500 return;
3501}
3502
Andrew Mortond6859bf2006-03-26 01:38:03 -08003503#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003504/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 * Bind an inode's backing buffer_head into this transaction, to prevent
3506 * it from being flushed to disk early. Unlike
3507 * ext3_reserve_inode_write, this leaves behind no bh reference and
3508 * returns no iloc structure, so the caller needs to repeat the iloc
3509 * lookup to mark the inode dirty later.
3510 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003511static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512{
3513 struct ext3_iloc iloc;
3514
3515 int err = 0;
3516 if (handle) {
3517 err = ext3_get_inode_loc(inode, &iloc);
3518 if (!err) {
3519 BUFFER_TRACE(iloc.bh, "get_write_access");
3520 err = journal_get_write_access(handle, iloc.bh);
3521 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003522 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 iloc.bh);
3524 brelse(iloc.bh);
3525 }
3526 }
3527 ext3_std_error(inode->i_sb, err);
3528 return err;
3529}
3530#endif
3531
3532int ext3_change_inode_journal_flag(struct inode *inode, int val)
3533{
3534 journal_t *journal;
3535 handle_t *handle;
3536 int err;
3537
3538 /*
3539 * We have to be very careful here: changing a data block's
3540 * journaling status dynamically is dangerous. If we write a
3541 * data block to the journal, change the status and then delete
3542 * that block, we risk forgetting to revoke the old log record
3543 * from the journal and so a subsequent replay can corrupt data.
3544 * So, first we make sure that the journal is empty and that
3545 * nobody is changing anything.
3546 */
3547
3548 journal = EXT3_JOURNAL(inode);
Dave Hansene3a68e32007-07-15 23:41:14 -07003549 if (is_journal_aborted(journal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 return -EROFS;
3551
3552 journal_lock_updates(journal);
3553 journal_flush(journal);
3554
3555 /*
3556 * OK, there are no updates running now, and all cached data is
3557 * synced to disk. We are now in a completely consistent state
3558 * which doesn't have anything in the journal, and we know that
3559 * no filesystem updates are running, so it is safe to modify
3560 * the inode's in-core data-journaling state flag now.
3561 */
3562
3563 if (val)
3564 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3565 else
3566 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3567 ext3_set_aops(inode);
3568
3569 journal_unlock_updates(journal);
3570
3571 /* Finally we can mark the inode as dirty. */
3572
3573 handle = ext3_journal_start(inode, 1);
3574 if (IS_ERR(handle))
3575 return PTR_ERR(handle);
3576
3577 err = ext3_mark_inode_dirty(handle, inode);
3578 handle->h_sync = 1;
3579 ext3_journal_stop(handle);
3580 ext3_std_error(inode->i_sb, err);
3581
3582 return err;
3583}