blob: eb95670a27ebafac6a9a4c593ef37e98a0f29ebf [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xattr.h"
40#include "acl.h"
41
42static int ext3_writepage_trans_blocks(struct inode *inode);
43
44/*
45 * Test whether an inode is a fast symlink.
46 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080047static int ext3_inode_is_fast_symlink(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048{
49 int ea_blocks = EXT3_I(inode)->i_file_acl ?
50 (inode->i_sb->s_blocksize >> 9) : 0;
51
Andrew Mortond6859bf2006-03-26 01:38:03 -080052 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053}
54
Andrew Mortond6859bf2006-03-26 01:38:03 -080055/*
56 * The ext3 forget function must perform a revoke if we are freeing data
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 * which has been journaled. Metadata (eg. indirect blocks) must be
Mingming Caoae6ddcc2006-09-27 01:49:27 -070058 * revoked in all cases.
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 *
60 * "bh" may be NULL: a metadata block may have been freed from memory
61 * but there may still be a record of it in the journal, and that record
62 * still needs to be revoked.
63 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080064int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -070065 struct buffer_head *bh, ext3_fsblk_t blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066{
67 int err;
68
69 might_sleep();
70
71 BUFFER_TRACE(bh, "enter");
72
73 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
74 "data mode %lx\n",
75 bh, is_metadata, inode->i_mode,
76 test_opt(inode->i_sb, DATA_FLAGS));
77
78 /* Never use the revoke function if we are doing full data
79 * journaling: there is no need to, and a V1 superblock won't
80 * support it. Otherwise, only skip the revoke on un-journaled
81 * data blocks. */
82
83 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
84 (!is_metadata && !ext3_should_journal_data(inode))) {
85 if (bh) {
86 BUFFER_TRACE(bh, "call journal_forget");
87 return ext3_journal_forget(handle, bh);
88 }
89 return 0;
90 }
91
92 /*
93 * data!=journal && (is_metadata || should_journal_data(inode))
94 */
95 BUFFER_TRACE(bh, "call ext3_journal_revoke");
96 err = ext3_journal_revoke(handle, blocknr, bh);
97 if (err)
98 ext3_abort(inode->i_sb, __FUNCTION__,
99 "error %d when attempting revoke", err);
100 BUFFER_TRACE(bh, "exit");
101 return err;
102}
103
104/*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800105 * Work out how many blocks we need to proceed with the next chunk of a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * truncate transaction.
107 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700108static unsigned long blocks_for_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
110 unsigned long needed;
111
112 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
113
114 /* Give ourselves just enough room to cope with inodes in which
115 * i_blocks is corrupt: we've seen disk corruptions in the past
116 * which resulted in random data in an inode which looked enough
117 * like a regular file for ext3 to try to delete it. Things
118 * will go a bit crazy if that happens, but at least we should
119 * try not to panic the whole kernel. */
120 if (needed < 2)
121 needed = 2;
122
123 /* But we need to bound the transaction so we don't overflow the
124 * journal. */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700125 if (needed > EXT3_MAX_TRANS_DATA)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 needed = EXT3_MAX_TRANS_DATA;
127
Jan Kara1f545872005-06-23 22:01:04 -0700128 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129}
130
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700131/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 * Truncate transactions can be complex and absolutely huge. So we need to
133 * be able to restart the transaction at a conventient checkpoint to make
134 * sure we don't overflow the journal.
135 *
136 * start_transaction gets us a new handle for a truncate transaction,
137 * and extend_transaction tries to extend the existing one a bit. If
138 * extend fails, we need to propagate the failure up and restart the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700139 * transaction in the top-level truncate loop. --sct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700141static handle_t *start_transaction(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
143 handle_t *result;
144
145 result = ext3_journal_start(inode, blocks_for_truncate(inode));
146 if (!IS_ERR(result))
147 return result;
148
149 ext3_std_error(inode->i_sb, PTR_ERR(result));
150 return result;
151}
152
153/*
154 * Try to extend this transaction for the purposes of truncation.
155 *
156 * Returns 0 if we managed to create more room. If we can't create more
157 * room, and the transaction must be restarted we return 1.
158 */
159static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
160{
161 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
162 return 0;
163 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
164 return 0;
165 return 1;
166}
167
168/*
169 * Restart the transaction associated with *handle. This does a commit,
170 * so before we call here everything must be consistently dirtied against
171 * this transaction.
172 */
173static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
174{
175 jbd_debug(2, "restarting handle %p\n", handle);
176 return ext3_journal_restart(handle, blocks_for_truncate(inode));
177}
178
179/*
180 * Called at the last iput() if i_nlink is zero.
181 */
182void ext3_delete_inode (struct inode * inode)
183{
184 handle_t *handle;
185
Mark Fashehfef26652005-09-09 13:01:31 -0700186 truncate_inode_pages(&inode->i_data, 0);
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 if (is_bad_inode(inode))
189 goto no_delete;
190
191 handle = start_transaction(inode);
192 if (IS_ERR(handle)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800193 /*
194 * If we're going to skip the normal cleanup, we still need to
195 * make sure that the in-core orphan linked list is properly
196 * cleaned up.
197 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 ext3_orphan_del(NULL, inode);
199 goto no_delete;
200 }
201
202 if (IS_SYNC(inode))
203 handle->h_sync = 1;
204 inode->i_size = 0;
205 if (inode->i_blocks)
206 ext3_truncate(inode);
207 /*
208 * Kill off the orphan record which ext3_truncate created.
209 * AKPM: I think this can be inside the above `if'.
210 * Note that ext3_orphan_del() has to be able to cope with the
211 * deletion of a non-existent orphan - this is because we don't
212 * know if ext3_truncate() actually created an orphan record.
213 * (Well, we could do this if we need to, but heck - it works)
214 */
215 ext3_orphan_del(handle, inode);
216 EXT3_I(inode)->i_dtime = get_seconds();
217
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700218 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 * One subtle ordering requirement: if anything has gone wrong
220 * (transaction abort, IO errors, whatever), then we can still
221 * do these next steps (the fs will already have been marked as
222 * having errors), but we can't free the inode if the mark_dirty
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700223 * fails.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 */
225 if (ext3_mark_inode_dirty(handle, inode))
226 /* If that failed, just do the required in-core inode clear. */
227 clear_inode(inode);
228 else
229 ext3_free_inode(handle, inode);
230 ext3_journal_stop(handle);
231 return;
232no_delete:
233 clear_inode(inode); /* We must guarantee clearing of inode... */
234}
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236typedef struct {
237 __le32 *p;
238 __le32 key;
239 struct buffer_head *bh;
240} Indirect;
241
242static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
243{
244 p->key = *(p->p = v);
245 p->bh = bh;
246}
247
Andrew Mortond6859bf2006-03-26 01:38:03 -0800248static int verify_chain(Indirect *from, Indirect *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
250 while (from <= to && from->key == *from->p)
251 from++;
252 return (from > to);
253}
254
255/**
256 * ext3_block_to_path - parse the block number into array of offsets
257 * @inode: inode in question (we are only interested in its superblock)
258 * @i_block: block number to be parsed
259 * @offsets: array to store the offsets in
260 * @boundary: set this non-zero if the referred-to block is likely to be
261 * followed (on disk) by an indirect block.
262 *
263 * To store the locations of file's data ext3 uses a data structure common
264 * for UNIX filesystems - tree of pointers anchored in the inode, with
265 * data blocks at leaves and indirect blocks in intermediate nodes.
266 * This function translates the block number into path in that tree -
267 * return value is the path length and @offsets[n] is the offset of
268 * pointer to (n+1)th node in the nth one. If @block is out of range
269 * (negative or too large) warning is printed and zero returned.
270 *
271 * Note: function doesn't find node addresses, so no IO is needed. All
272 * we need to know is the capacity of indirect blocks (taken from the
273 * inode->i_sb).
274 */
275
276/*
277 * Portability note: the last comparison (check that we fit into triple
278 * indirect block) is spelled differently, because otherwise on an
279 * architecture with 32-bit longs and 8Kb pages we might get into trouble
280 * if our filesystem had 8Kb blocks. We might use long long, but that would
281 * kill us on x86. Oh, well, at least the sign propagation does not matter -
282 * i_block would have to be negative in the very beginning, so we would not
283 * get there at all.
284 */
285
286static int ext3_block_to_path(struct inode *inode,
287 long i_block, int offsets[4], int *boundary)
288{
289 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
290 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
291 const long direct_blocks = EXT3_NDIR_BLOCKS,
292 indirect_blocks = ptrs,
293 double_blocks = (1 << (ptrs_bits * 2));
294 int n = 0;
295 int final = 0;
296
297 if (i_block < 0) {
298 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
299 } else if (i_block < direct_blocks) {
300 offsets[n++] = i_block;
301 final = direct_blocks;
302 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
303 offsets[n++] = EXT3_IND_BLOCK;
304 offsets[n++] = i_block;
305 final = ptrs;
306 } else if ((i_block -= indirect_blocks) < double_blocks) {
307 offsets[n++] = EXT3_DIND_BLOCK;
308 offsets[n++] = i_block >> ptrs_bits;
309 offsets[n++] = i_block & (ptrs - 1);
310 final = ptrs;
311 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
312 offsets[n++] = EXT3_TIND_BLOCK;
313 offsets[n++] = i_block >> (ptrs_bits * 2);
314 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
315 offsets[n++] = i_block & (ptrs - 1);
316 final = ptrs;
317 } else {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800318 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 }
320 if (boundary)
Mingming Cao89747d32006-03-26 01:37:55 -0800321 *boundary = final - 1 - (i_block & (ptrs - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 return n;
323}
324
325/**
326 * ext3_get_branch - read the chain of indirect blocks leading to data
327 * @inode: inode in question
328 * @depth: depth of the chain (1 - direct pointer, etc.)
329 * @offsets: offsets of pointers in inode/indirect blocks
330 * @chain: place to store the result
331 * @err: here we store the error value
332 *
333 * Function fills the array of triples <key, p, bh> and returns %NULL
334 * if everything went OK or the pointer to the last filled triple
335 * (incomplete one) otherwise. Upon the return chain[i].key contains
336 * the number of (i+1)-th block in the chain (as it is stored in memory,
337 * i.e. little-endian 32-bit), chain[i].p contains the address of that
338 * number (it points into struct inode for i==0 and into the bh->b_data
339 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
340 * block for i>0 and NULL for i==0. In other words, it holds the block
341 * numbers of the chain, addresses they were taken from (and where we can
342 * verify that chain did not change) and buffer_heads hosting these
343 * numbers.
344 *
345 * Function stops when it stumbles upon zero pointer (absent block)
346 * (pointer to last triple returned, *@err == 0)
347 * or when it gets an IO error reading an indirect block
348 * (ditto, *@err == -EIO)
349 * or when it notices that chain had been changed while it was reading
350 * (ditto, *@err == -EAGAIN)
351 * or when it reads all @depth-1 indirect blocks successfully and finds
352 * the whole chain, all way to the data (returns %NULL, *err == 0).
353 */
354static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
355 Indirect chain[4], int *err)
356{
357 struct super_block *sb = inode->i_sb;
358 Indirect *p = chain;
359 struct buffer_head *bh;
360
361 *err = 0;
362 /* i_data is not going away, no lock needed */
363 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
364 if (!p->key)
365 goto no_block;
366 while (--depth) {
367 bh = sb_bread(sb, le32_to_cpu(p->key));
368 if (!bh)
369 goto failure;
370 /* Reader: pointers */
371 if (!verify_chain(chain, p))
372 goto changed;
373 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
374 /* Reader: end */
375 if (!p->key)
376 goto no_block;
377 }
378 return NULL;
379
380changed:
381 brelse(bh);
382 *err = -EAGAIN;
383 goto no_block;
384failure:
385 *err = -EIO;
386no_block:
387 return p;
388}
389
390/**
391 * ext3_find_near - find a place for allocation with sufficient locality
392 * @inode: owner
393 * @ind: descriptor of indirect block.
394 *
395 * This function returns the prefered place for block allocation.
396 * It is used when heuristic for sequential allocation fails.
397 * Rules are:
398 * + if there is a block to the left of our position - allocate near it.
399 * + if pointer will live in indirect block - allocate near that block.
400 * + if pointer will live in inode - allocate in the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700401 * cylinder group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 *
403 * In the latter case we colour the starting block by the callers PID to
404 * prevent it from clashing with concurrent allocations for a different inode
405 * in the same block group. The PID is used here so that functionally related
406 * files will be close-by on-disk.
407 *
408 * Caller must make sure that @ind is valid and will stay that way.
409 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700410static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
412 struct ext3_inode_info *ei = EXT3_I(inode);
413 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
414 __le32 *p;
Mingming Cao43d23f92006-06-25 05:48:07 -0700415 ext3_fsblk_t bg_start;
416 ext3_grpblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418 /* Try to find previous block */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800419 for (p = ind->p - 1; p >= start; p--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 if (*p)
421 return le32_to_cpu(*p);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800422 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424 /* No such thing, so let's try location of indirect block */
425 if (ind->bh)
426 return ind->bh->b_blocknr;
427
428 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800429 * It is going to be referred to from the inode itself? OK, just put it
430 * into the same cylinder group then.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700432 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 colour = (current->pid % 16) *
434 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
435 return bg_start + colour;
436}
437
438/**
439 * ext3_find_goal - find a prefered place for allocation.
440 * @inode: owner
441 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 *
444 * Normally this function find the prefered place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800445 * returns it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 */
447
Mingming Cao43d23f92006-06-25 05:48:07 -0700448static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800449 Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
Andrew Mortond6859bf2006-03-26 01:38:03 -0800451 struct ext3_block_alloc_info *block_i;
452
453 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 /*
456 * try the heuristic for sequential allocation,
457 * failing that at least try to get decent locality.
458 */
459 if (block_i && (block == block_i->last_alloc_logical_block + 1)
460 && (block_i->last_alloc_physical_block != 0)) {
Mingming Caofe55c452005-05-01 08:59:20 -0700461 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463
Mingming Caofe55c452005-05-01 08:59:20 -0700464 return ext3_find_near(inode, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
Andrew Mortond6859bf2006-03-26 01:38:03 -0800466
Mingming Caob47b2472006-03-26 01:37:56 -0800467/**
468 * ext3_blks_to_allocate: Look up the block map and count the number
469 * of direct blocks need to be allocated for the given branch.
470 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700471 * @branch: chain of indirect blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800472 * @k: number of blocks need for indirect blocks
473 * @blks: number of data blocks to be mapped.
474 * @blocks_to_boundary: the offset in the indirect block
475 *
476 * return the total number of blocks to be allocate, including the
477 * direct and indirect blocks.
478 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800479static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Mingming Caob47b2472006-03-26 01:37:56 -0800480 int blocks_to_boundary)
481{
482 unsigned long count = 0;
483
484 /*
485 * Simple case, [t,d]Indirect block(s) has not allocated yet
486 * then it's clear blocks on that path have not allocated
487 */
488 if (k > 0) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800489 /* right now we don't handle cross boundary allocation */
Mingming Caob47b2472006-03-26 01:37:56 -0800490 if (blks < blocks_to_boundary + 1)
491 count += blks;
492 else
493 count += blocks_to_boundary + 1;
494 return count;
495 }
496
497 count++;
498 while (count < blks && count <= blocks_to_boundary &&
499 le32_to_cpu(*(branch[0].p + count)) == 0) {
500 count++;
501 }
502 return count;
503}
504
505/**
506 * ext3_alloc_blocks: multiple allocate blocks needed for a branch
507 * @indirect_blks: the number of blocks need to allocate for indirect
508 * blocks
509 *
510 * @new_blocks: on return it will store the new block numbers for
511 * the indirect blocks(if needed) and the first direct block,
512 * @blks: on return it will store the total number of allocated
513 * direct blocks
514 */
515static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700516 ext3_fsblk_t goal, int indirect_blks, int blks,
517 ext3_fsblk_t new_blocks[4], int *err)
Mingming Caob47b2472006-03-26 01:37:56 -0800518{
519 int target, i;
520 unsigned long count = 0;
521 int index = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700522 ext3_fsblk_t current_block = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800523 int ret = 0;
524
525 /*
526 * Here we try to allocate the requested multiple blocks at once,
527 * on a best-effort basis.
528 * To build a branch, we should allocate blocks for
529 * the indirect blocks(if not allocated yet), and at least
530 * the first direct block of this branch. That's the
531 * minimum number of blocks need to allocate(required)
532 */
533 target = blks + indirect_blks;
534
535 while (1) {
536 count = target;
537 /* allocating blocks for indirect blocks and direct blocks */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800538 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
Mingming Caob47b2472006-03-26 01:37:56 -0800539 if (*err)
540 goto failed_out;
541
542 target -= count;
543 /* allocate blocks for indirect blocks */
544 while (index < indirect_blks && count) {
545 new_blocks[index++] = current_block++;
546 count--;
547 }
548
549 if (count > 0)
550 break;
551 }
552
553 /* save the new block number for the first direct block */
554 new_blocks[index] = current_block;
555
556 /* total number of blocks allocated for direct blocks */
557 ret = count;
558 *err = 0;
559 return ret;
560failed_out:
561 for (i = 0; i <index; i++)
562 ext3_free_blocks(handle, inode, new_blocks[i], 1);
563 return ret;
564}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
566/**
567 * ext3_alloc_branch - allocate and set up a chain of blocks.
568 * @inode: owner
Mingming Caob47b2472006-03-26 01:37:56 -0800569 * @indirect_blks: number of allocated indirect blocks
570 * @blks: number of allocated direct blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 * @offsets: offsets (in the blocks) to store the pointers to next.
572 * @branch: place to store the chain in.
573 *
Mingming Caob47b2472006-03-26 01:37:56 -0800574 * This function allocates blocks, zeroes out all but the last one,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 * links them into chain and (if we are synchronous) writes them to disk.
576 * In other words, it prepares a branch that can be spliced onto the
577 * inode. It stores the information about that chain in the branch[], in
578 * the same format as ext3_get_branch() would do. We are calling it after
579 * we had read the existing part of chain and partial points to the last
580 * triple of that (one with zero ->key). Upon the exit we have the same
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -0800581 * picture as after the successful ext3_get_block(), except that in one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * place chain is disconnected - *branch->p is still zero (we did not
583 * set the last link), but branch->key contains the number that should
584 * be placed into *branch->p to fill that gap.
585 *
586 * If allocation fails we free all blocks we've allocated (and forget
587 * their buffer_heads) and return the error value the from failed
588 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
589 * as described above and return 0.
590 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700592 int indirect_blks, int *blks, ext3_fsblk_t goal,
Mingming Caob47b2472006-03-26 01:37:56 -0800593 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
595 int blocksize = inode->i_sb->s_blocksize;
Mingming Caob47b2472006-03-26 01:37:56 -0800596 int i, n = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 int err = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800598 struct buffer_head *bh;
599 int num;
Mingming Cao43d23f92006-06-25 05:48:07 -0700600 ext3_fsblk_t new_blocks[4];
601 ext3_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Mingming Caob47b2472006-03-26 01:37:56 -0800603 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
604 *blks, new_blocks, &err);
605 if (err)
606 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Mingming Caob47b2472006-03-26 01:37:56 -0800608 branch[0].key = cpu_to_le32(new_blocks[0]);
609 /*
610 * metadata blocks and data blocks are allocated.
611 */
612 for (n = 1; n <= indirect_blks; n++) {
613 /*
614 * Get buffer_head for parent block, zero it out
615 * and set the pointer to new one, then send
616 * parent to disk.
617 */
618 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
619 branch[n].bh = bh;
620 lock_buffer(bh);
621 BUFFER_TRACE(bh, "call get_create_access");
622 err = ext3_journal_get_create_access(handle, bh);
623 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 unlock_buffer(bh);
Mingming Caob47b2472006-03-26 01:37:56 -0800625 brelse(bh);
626 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Mingming Caob47b2472006-03-26 01:37:56 -0800629 memset(bh->b_data, 0, blocksize);
630 branch[n].p = (__le32 *) bh->b_data + offsets[n];
631 branch[n].key = cpu_to_le32(new_blocks[n]);
632 *branch[n].p = branch[n].key;
633 if ( n == indirect_blks) {
634 current_block = new_blocks[n];
635 /*
636 * End of chain, update the last new metablock of
637 * the chain to point to the new allocated
638 * data blocks numbers
639 */
640 for (i=1; i < num; i++)
641 *(branch[n].p + i) = cpu_to_le32(++current_block);
642 }
643 BUFFER_TRACE(bh, "marking uptodate");
644 set_buffer_uptodate(bh);
645 unlock_buffer(bh);
646
647 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
648 err = ext3_journal_dirty_metadata(handle, bh);
649 if (err)
650 goto failed;
651 }
652 *blks = num;
653 return err;
654failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 /* Allocation failed, free what we already allocated */
Mingming Caob47b2472006-03-26 01:37:56 -0800656 for (i = 1; i <= n ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 BUFFER_TRACE(branch[i].bh, "call journal_forget");
658 ext3_journal_forget(handle, branch[i].bh);
659 }
Mingming Caob47b2472006-03-26 01:37:56 -0800660 for (i = 0; i <indirect_blks; i++)
661 ext3_free_blocks(handle, inode, new_blocks[i], 1);
662
663 ext3_free_blocks(handle, inode, new_blocks[i], num);
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 return err;
666}
667
668/**
Andrew Mortond6859bf2006-03-26 01:38:03 -0800669 * ext3_splice_branch - splice the allocated branch onto inode.
670 * @inode: owner
671 * @block: (logical) number of block we are adding
672 * @chain: chain of indirect blocks (with a missing link - see
673 * ext3_alloc_branch)
674 * @where: location of missing link
675 * @num: number of indirect blocks we are adding
676 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800678 * This function fills the missing link and does all housekeeping needed in
679 * inode (->i_blocks, etc.). In case of success we end up with the full
680 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800682static int ext3_splice_branch(handle_t *handle, struct inode *inode,
683 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 int i;
686 int err = 0;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800687 struct ext3_block_alloc_info *block_i;
Mingming Cao43d23f92006-06-25 05:48:07 -0700688 ext3_fsblk_t current_block;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800689
690 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 /*
692 * If we're splicing into a [td]indirect block (as opposed to the
693 * inode) then we need to get write access to the [td]indirect block
694 * before the splice.
695 */
696 if (where->bh) {
697 BUFFER_TRACE(where->bh, "get_write_access");
698 err = ext3_journal_get_write_access(handle, where->bh);
699 if (err)
700 goto err_out;
701 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 /* That's it */
703
704 *where->p = where->key;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800705
706 /*
707 * Update the host buffer_head or inode to point to more just allocated
708 * direct blocks blocks
709 */
Mingming Caob47b2472006-03-26 01:37:56 -0800710 if (num == 0 && blks > 1) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700711 current_block = le32_to_cpu(where->key) + 1;
Mingming Caob47b2472006-03-26 01:37:56 -0800712 for (i = 1; i < blks; i++)
713 *(where->p + i ) = cpu_to_le32(current_block++);
714 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
716 /*
717 * update the most recently allocated logical & physical block
718 * in i_block_alloc_info, to assist find the proper goal block for next
719 * allocation
720 */
721 if (block_i) {
Mingming Caob47b2472006-03-26 01:37:56 -0800722 block_i->last_alloc_logical_block = block + blks - 1;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800723 block_i->last_alloc_physical_block =
Mingming Cao5dea5172006-05-03 19:55:12 -0700724 le32_to_cpu(where[num].key) + blks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726
727 /* We are done with atomic stuff, now do the rest of housekeeping */
728
729 inode->i_ctime = CURRENT_TIME_SEC;
730 ext3_mark_inode_dirty(handle, inode);
731
732 /* had we spliced it onto indirect block? */
733 if (where->bh) {
734 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800735 * If we spliced it onto an indirect block, we haven't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 * altered the inode. Note however that if it is being spliced
737 * onto an indirect block at the very end of the file (the
738 * file is growing) then we *will* alter the inode to reflect
739 * the new i_size. But that is not done here - it is done in
740 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
741 */
742 jbd_debug(5, "splicing indirect only\n");
743 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
744 err = ext3_journal_dirty_metadata(handle, where->bh);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700745 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 goto err_out;
747 } else {
748 /*
749 * OK, we spliced it into the inode itself on a direct block.
750 * Inode was dirtied above.
751 */
752 jbd_debug(5, "splicing direct\n");
753 }
754 return err;
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756err_out:
Mingming Caob47b2472006-03-26 01:37:56 -0800757 for (i = 1; i <= num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 BUFFER_TRACE(where[i].bh, "call journal_forget");
759 ext3_journal_forget(handle, where[i].bh);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800760 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 }
Mingming Caob47b2472006-03-26 01:37:56 -0800762 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return err;
765}
766
767/*
768 * Allocation strategy is simple: if we have to allocate something, we will
769 * have to go the whole way to leaf. So let's do it before attaching anything
770 * to tree, set linkage between the newborn blocks, write them if sync is
771 * required, recheck the path, free and repeat if check fails, otherwise
772 * set the last missing link (that will protect us from any truncate-generated
773 * removals - all blocks on the path are immune now) and possibly force the
774 * write on the parent block.
775 * That has a nice additional property: no special recovery from the failed
776 * allocations is needed - we simply release blocks and do not touch anything
777 * reachable from inode.
778 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800779 * `handle' can be NULL if create == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 *
781 * The BKL may not be held on entry here. Be sure to take it early.
Mingming Cao89747d32006-03-26 01:37:55 -0800782 * return > 0, # of blocks mapped or allocated.
783 * return = 0, if plain lookup failed.
784 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800786int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
787 sector_t iblock, unsigned long maxblocks,
788 struct buffer_head *bh_result,
Mingming Cao89747d32006-03-26 01:37:55 -0800789 int create, int extend_disksize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 int err = -EIO;
792 int offsets[4];
793 Indirect chain[4];
794 Indirect *partial;
Mingming Cao43d23f92006-06-25 05:48:07 -0700795 ext3_fsblk_t goal;
Mingming Caob47b2472006-03-26 01:37:56 -0800796 int indirect_blks;
Mingming Cao89747d32006-03-26 01:37:55 -0800797 int blocks_to_boundary = 0;
798 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 struct ext3_inode_info *ei = EXT3_I(inode);
Mingming Cao89747d32006-03-26 01:37:55 -0800800 int count = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700801 ext3_fsblk_t first_block = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 J_ASSERT(handle != NULL || create == 0);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800805 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
807 if (depth == 0)
808 goto out;
809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
811
812 /* Simplest case - block found, no allocation needed */
813 if (!partial) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700814 first_block = le32_to_cpu(chain[depth - 1].key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 clear_buffer_new(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800816 count++;
817 /*map more blocks*/
818 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700819 ext3_fsblk_t blk;
Mingming Cao5dea5172006-05-03 19:55:12 -0700820
Mingming Cao89747d32006-03-26 01:37:55 -0800821 if (!verify_chain(chain, partial)) {
822 /*
823 * Indirect block might be removed by
824 * truncate while we were reading it.
825 * Handling of that case: forget what we've
826 * got now. Flag the err as EAGAIN, so it
827 * will reread.
828 */
829 err = -EAGAIN;
830 count = 0;
831 break;
832 }
Mingming Cao5dea5172006-05-03 19:55:12 -0700833 blk = le32_to_cpu(*(chain[depth-1].p + count));
834
835 if (blk == first_block + count)
Mingming Cao89747d32006-03-26 01:37:55 -0800836 count++;
837 else
838 break;
839 }
840 if (err != -EAGAIN)
841 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 }
843
844 /* Next simple case - plain lookup or failed read of indirect block */
Mingming Caofe55c452005-05-01 08:59:20 -0700845 if (!create || err == -EIO)
846 goto cleanup;
847
Arjan van de Ven97461512006-03-23 03:00:42 -0800848 mutex_lock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700849
850 /*
851 * If the indirect block is missing while we are reading
852 * the chain(ext3_get_branch() returns -EAGAIN err), or
853 * if the chain has been changed after we grab the semaphore,
854 * (either because another process truncated this branch, or
855 * another get_block allocated this branch) re-grab the chain to see if
856 * the request block has been allocated or not.
857 *
858 * Since we already block the truncate/other get_block
859 * at this point, we will have the current copy of the chain when we
860 * splice the branch into the tree.
861 */
862 if (err == -EAGAIN || !verify_chain(chain, partial)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 while (partial > chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 brelse(partial->bh);
865 partial--;
866 }
Mingming Caofe55c452005-05-01 08:59:20 -0700867 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
868 if (!partial) {
Mingming Cao89747d32006-03-26 01:37:55 -0800869 count++;
Arjan van de Ven97461512006-03-23 03:00:42 -0800870 mutex_unlock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700871 if (err)
872 goto cleanup;
873 clear_buffer_new(bh_result);
874 goto got_it;
875 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 }
877
878 /*
Mingming Caofe55c452005-05-01 08:59:20 -0700879 * Okay, we need to do block allocation. Lazily initialize the block
880 * allocation info here if necessary
881 */
882 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 ext3_init_block_alloc_info(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800885 goal = ext3_find_goal(inode, iblock, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Mingming Caob47b2472006-03-26 01:37:56 -0800887 /* the number of blocks need to allocate for [d,t]indirect blocks */
888 indirect_blks = (chain + depth) - partial - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890 /*
Mingming Caob47b2472006-03-26 01:37:56 -0800891 * Next look up the indirect map to count the totoal number of
892 * direct blocks to allocate for this branch.
893 */
894 count = ext3_blks_to_allocate(partial, indirect_blks,
895 maxblocks, blocks_to_boundary);
896 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 * Block out ext3_truncate while we alter the tree
898 */
Mingming Caob47b2472006-03-26 01:37:56 -0800899 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
Mingming Caofe55c452005-05-01 08:59:20 -0700900 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Mingming Caofe55c452005-05-01 08:59:20 -0700902 /*
903 * The ext3_splice_branch call will free and forget any buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 * on the new chain if there is a failure, but that risks using
905 * up transaction credits, especially for bitmaps where the
906 * credits cannot be returned. Can we handle this somehow? We
Mingming Caofe55c452005-05-01 08:59:20 -0700907 * may need to return -EAGAIN upwards in the worst case. --sct
908 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (!err)
Mingming Caob47b2472006-03-26 01:37:56 -0800910 err = ext3_splice_branch(handle, inode, iblock,
911 partial, indirect_blks, count);
Mingming Caofe55c452005-05-01 08:59:20 -0700912 /*
Arjan van de Ven97461512006-03-23 03:00:42 -0800913 * i_disksize growing is protected by truncate_mutex. Don't forget to
Mingming Caofe55c452005-05-01 08:59:20 -0700914 * protect it if you're about to implement concurrent
915 * ext3_get_block() -bzzz
916 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
918 ei->i_disksize = inode->i_size;
Arjan van de Ven97461512006-03-23 03:00:42 -0800919 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 if (err)
921 goto cleanup;
922
923 set_buffer_new(bh_result);
Mingming Caofe55c452005-05-01 08:59:20 -0700924got_it:
925 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
Suparna Bhattacharya20acaa12006-09-16 12:15:58 -0700926 if (count > blocks_to_boundary)
Mingming Caofe55c452005-05-01 08:59:20 -0700927 set_buffer_boundary(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800928 err = count;
Mingming Caofe55c452005-05-01 08:59:20 -0700929 /* Clean up and exit */
930 partial = chain + depth - 1; /* the whole chain */
931cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 while (partial > chain) {
Mingming Caofe55c452005-05-01 08:59:20 -0700933 BUFFER_TRACE(partial->bh, "call brelse");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 brelse(partial->bh);
935 partial--;
936 }
Mingming Caofe55c452005-05-01 08:59:20 -0700937 BUFFER_TRACE(bh_result, "returned");
938out:
939 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940}
941
Jan Karabd1939d2008-02-06 01:40:21 -0800942/* Maximum number of blocks we map for direct IO at once. */
943#define DIO_MAX_BLOCKS 4096
944/*
945 * Number of credits we need for writing DIO_MAX_BLOCKS:
946 * We need sb + group descriptor + bitmap + inode -> 4
947 * For B blocks with A block pointers per block we need:
948 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
949 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
950 */
951#define DIO_CREDITS 25
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -0800953static int ext3_get_block(struct inode *inode, sector_t iblock,
954 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800956 handle_t *handle = ext3_journal_current_handle();
Jan Karabd1939d2008-02-06 01:40:21 -0800957 int ret = 0, started = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800958 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Jan Karabd1939d2008-02-06 01:40:21 -0800960 if (create && !handle) { /* Direct IO write... */
961 if (max_blocks > DIO_MAX_BLOCKS)
962 max_blocks = DIO_MAX_BLOCKS;
963 handle = ext3_journal_start(inode, DIO_CREDITS +
964 2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb));
965 if (IS_ERR(handle)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 ret = PTR_ERR(handle);
Jan Karabd1939d2008-02-06 01:40:21 -0800967 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
Jan Karabd1939d2008-02-06 01:40:21 -0800969 started = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971
Jan Karabd1939d2008-02-06 01:40:21 -0800972 ret = ext3_get_blocks_handle(handle, inode, iblock,
Mingming Cao89747d32006-03-26 01:37:55 -0800973 max_blocks, bh_result, create, 0);
Jan Karabd1939d2008-02-06 01:40:21 -0800974 if (ret > 0) {
975 bh_result->b_size = (ret << inode->i_blkbits);
976 ret = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800977 }
Jan Karabd1939d2008-02-06 01:40:21 -0800978 if (started)
979 ext3_journal_stop(handle);
980out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return ret;
982}
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984/*
985 * `handle' can be NULL if create is zero
986 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800987struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
988 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
990 struct buffer_head dummy;
991 int fatal = 0, err;
992
993 J_ASSERT(handle != NULL || create == 0);
994
995 dummy.b_state = 0;
996 dummy.b_blocknr = -1000;
997 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -0800998 err = ext3_get_blocks_handle(handle, inode, block, 1,
999 &dummy, create, 1);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001000 /*
1001 * ext3_get_blocks_handle() returns number of blocks
1002 * mapped. 0 in case of a HOLE.
1003 */
1004 if (err > 0) {
1005 if (err > 1)
1006 WARN_ON(1);
Mingming Cao89747d32006-03-26 01:37:55 -08001007 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001008 }
1009 *errp = err;
1010 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 struct buffer_head *bh;
1012 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001013 if (!bh) {
1014 *errp = -EIO;
1015 goto err;
1016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 if (buffer_new(&dummy)) {
1018 J_ASSERT(create != 0);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001019 J_ASSERT(handle != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Andrew Mortond6859bf2006-03-26 01:38:03 -08001021 /*
1022 * Now that we do not always journal data, we should
1023 * keep in mind whether this should always journal the
1024 * new buffer as metadata. For now, regular file
1025 * writes use ext3_get_block instead, so it's not a
1026 * problem.
1027 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 lock_buffer(bh);
1029 BUFFER_TRACE(bh, "call get_create_access");
1030 fatal = ext3_journal_get_create_access(handle, bh);
1031 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001032 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 set_buffer_uptodate(bh);
1034 }
1035 unlock_buffer(bh);
1036 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1037 err = ext3_journal_dirty_metadata(handle, bh);
1038 if (!fatal)
1039 fatal = err;
1040 } else {
1041 BUFFER_TRACE(bh, "not a new buffer");
1042 }
1043 if (fatal) {
1044 *errp = fatal;
1045 brelse(bh);
1046 bh = NULL;
1047 }
1048 return bh;
1049 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001050err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 return NULL;
1052}
1053
Andrew Mortond6859bf2006-03-26 01:38:03 -08001054struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 int block, int create, int *err)
1056{
1057 struct buffer_head * bh;
1058
1059 bh = ext3_getblk(handle, inode, block, create, err);
1060 if (!bh)
1061 return bh;
1062 if (buffer_uptodate(bh))
1063 return bh;
Jens Axboecaa38fb2006-07-23 01:41:26 +02001064 ll_rw_block(READ_META, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 wait_on_buffer(bh);
1066 if (buffer_uptodate(bh))
1067 return bh;
1068 put_bh(bh);
1069 *err = -EIO;
1070 return NULL;
1071}
1072
1073static int walk_page_buffers( handle_t *handle,
1074 struct buffer_head *head,
1075 unsigned from,
1076 unsigned to,
1077 int *partial,
1078 int (*fn)( handle_t *handle,
1079 struct buffer_head *bh))
1080{
1081 struct buffer_head *bh;
1082 unsigned block_start, block_end;
1083 unsigned blocksize = head->b_size;
1084 int err, ret = 0;
1085 struct buffer_head *next;
1086
1087 for ( bh = head, block_start = 0;
1088 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001089 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 {
1091 next = bh->b_this_page;
1092 block_end = block_start + blocksize;
1093 if (block_end <= from || block_start >= to) {
1094 if (partial && !buffer_uptodate(bh))
1095 *partial = 1;
1096 continue;
1097 }
1098 err = (*fn)(handle, bh);
1099 if (!ret)
1100 ret = err;
1101 }
1102 return ret;
1103}
1104
1105/*
1106 * To preserve ordering, it is essential that the hole instantiation and
1107 * the data write be encapsulated in a single transaction. We cannot
1108 * close off a transaction and start a new one between the ext3_get_block()
1109 * and the commit_write(). So doing the journal_start at the start of
1110 * prepare_write() is the right place.
1111 *
1112 * Also, this function can nest inside ext3_writepage() ->
1113 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1114 * has generated enough buffer credits to do the whole page. So we won't
1115 * block on the journal in that case, which is good, because the caller may
1116 * be PF_MEMALLOC.
1117 *
1118 * By accident, ext3 can be reentered when a transaction is open via
1119 * quota file writes. If we were to commit the transaction while thus
1120 * reentered, there can be a deadlock - we would be holding a quota
1121 * lock, and the commit would never complete if another thread had a
1122 * transaction open and was blocking on the quota lock - a ranking
1123 * violation.
1124 *
1125 * So what we do is to rely on the fact that journal_stop/journal_start
1126 * will _not_ run commit under these circumstances because handle->h_ref
1127 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001128 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001130static int do_journal_get_write_access(handle_t *handle,
1131 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
1133 if (!buffer_mapped(bh) || buffer_freed(bh))
1134 return 0;
1135 return ext3_journal_get_write_access(handle, bh);
1136}
1137
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001138static int ext3_write_begin(struct file *file, struct address_space *mapping,
1139 loff_t pos, unsigned len, unsigned flags,
1140 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001142 struct inode *inode = mapping->host;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001143 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 handle_t *handle;
1145 int retries = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001146 struct page *page;
1147 pgoff_t index;
1148 unsigned from, to;
1149
1150 index = pos >> PAGE_CACHE_SHIFT;
1151 from = pos & (PAGE_CACHE_SIZE - 1);
1152 to = from + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
1154retry:
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001155 page = __grab_cache_page(mapping, index);
1156 if (!page)
1157 return -ENOMEM;
1158 *pagep = page;
1159
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 handle = ext3_journal_start(inode, needed_blocks);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001161 if (IS_ERR(handle)) {
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001162 unlock_page(page);
1163 page_cache_release(page);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001164 ret = PTR_ERR(handle);
1165 goto out;
1166 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001167 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1168 ext3_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 if (ret)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001170 goto write_begin_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 if (ext3_should_journal_data(inode)) {
1173 ret = walk_page_buffers(handle, page_buffers(page),
1174 from, to, NULL, do_journal_get_write_access);
1175 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001176write_begin_failed:
1177 if (ret) {
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001178 ext3_journal_stop(handle);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001179 unlock_page(page);
1180 page_cache_release(page);
1181 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1183 goto retry;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001184out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return ret;
1186}
1187
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001188
Andrew Mortond6859bf2006-03-26 01:38:03 -08001189int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 int err = journal_dirty_data(handle, bh);
1192 if (err)
1193 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001194 bh, handle, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 return err;
1196}
1197
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001198/* For write_end() in data=journal mode */
1199static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
1201 if (!buffer_mapped(bh) || buffer_freed(bh))
1202 return 0;
1203 set_buffer_uptodate(bh);
1204 return ext3_journal_dirty_metadata(handle, bh);
1205}
1206
1207/*
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001208 * Generic write_end handler for ordered and writeback ext3 journal modes.
1209 * We can't use generic_write_end, because that unlocks the page and we need to
1210 * unlock the page after ext3_journal_stop, but ext3_journal_stop must run
1211 * after block_write_end.
1212 */
1213static int ext3_generic_write_end(struct file *file,
1214 struct address_space *mapping,
1215 loff_t pos, unsigned len, unsigned copied,
1216 struct page *page, void *fsdata)
1217{
1218 struct inode *inode = file->f_mapping->host;
1219
1220 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1221
1222 if (pos+copied > inode->i_size) {
1223 i_size_write(inode, pos+copied);
1224 mark_inode_dirty(inode);
1225 }
1226
1227 return copied;
1228}
1229
1230/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 * We need to pick up the new inode size which generic_commit_write gave us
1232 * `file' can be NULL - eg, when called from page_symlink().
1233 *
1234 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1235 * buffers are managed internally.
1236 */
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001237static int ext3_ordered_write_end(struct file *file,
1238 struct address_space *mapping,
1239 loff_t pos, unsigned len, unsigned copied,
1240 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
1242 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001243 struct inode *inode = file->f_mapping->host;
1244 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 int ret = 0, ret2;
1246
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001247 from = pos & (PAGE_CACHE_SIZE - 1);
1248 to = from + len;
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 ret = walk_page_buffers(handle, page_buffers(page),
1251 from, to, NULL, ext3_journal_dirty_data);
1252
1253 if (ret == 0) {
1254 /*
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001255 * generic_write_end() will run mark_inode_dirty() if i_size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 * changes. So let's piggyback the i_disksize mark_inode_dirty
1257 * into that.
1258 */
1259 loff_t new_i_size;
1260
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001261 new_i_size = pos + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 if (new_i_size > EXT3_I(inode)->i_disksize)
1263 EXT3_I(inode)->i_disksize = new_i_size;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001264 copied = ext3_generic_write_end(file, mapping, pos, len, copied,
1265 page, fsdata);
1266 if (copied < 0)
1267 ret = copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 }
1269 ret2 = ext3_journal_stop(handle);
1270 if (!ret)
1271 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001272 unlock_page(page);
1273 page_cache_release(page);
1274
1275 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276}
1277
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001278static int ext3_writeback_write_end(struct file *file,
1279 struct address_space *mapping,
1280 loff_t pos, unsigned len, unsigned copied,
1281 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
1283 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001284 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 int ret = 0, ret2;
1286 loff_t new_i_size;
1287
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001288 new_i_size = pos + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (new_i_size > EXT3_I(inode)->i_disksize)
1290 EXT3_I(inode)->i_disksize = new_i_size;
1291
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001292 copied = ext3_generic_write_end(file, mapping, pos, len, copied,
1293 page, fsdata);
1294 if (copied < 0)
1295 ret = copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 ret2 = ext3_journal_stop(handle);
1298 if (!ret)
1299 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001300 unlock_page(page);
1301 page_cache_release(page);
1302
1303 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304}
1305
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001306static int ext3_journalled_write_end(struct file *file,
1307 struct address_space *mapping,
1308 loff_t pos, unsigned len, unsigned copied,
1309 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310{
1311 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001312 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 int ret = 0, ret2;
1314 int partial = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001315 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001317 from = pos & (PAGE_CACHE_SIZE - 1);
1318 to = from + len;
1319
1320 if (copied < len) {
1321 if (!PageUptodate(page))
1322 copied = 0;
1323 page_zero_new_buffers(page, from+copied, to);
1324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001327 to, &partial, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 if (!partial)
1329 SetPageUptodate(page);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001330 if (pos+copied > inode->i_size)
1331 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1333 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1334 EXT3_I(inode)->i_disksize = inode->i_size;
1335 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001336 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 ret = ret2;
1338 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 ret2 = ext3_journal_stop(handle);
1341 if (!ret)
1342 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001343 unlock_page(page);
1344 page_cache_release(page);
1345
1346 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347}
1348
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001349/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 * bmap() is special. It gets used by applications such as lilo and by
1351 * the swapper to find the on-disk block of a specific piece of data.
1352 *
1353 * Naturally, this is dangerous if the block concerned is still in the
1354 * journal. If somebody makes a swapfile on an ext3 data-journaling
1355 * filesystem and enables swap, then they may get a nasty shock when the
1356 * data getting swapped to that swapfile suddenly gets overwritten by
1357 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001358 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 *
1360 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001361 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 */
1363static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1364{
1365 struct inode *inode = mapping->host;
1366 journal_t *journal;
1367 int err;
1368
1369 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001370 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 * This is a REALLY heavyweight approach, but the use of
1372 * bmap on dirty files is expected to be extremely rare:
1373 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001374 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 *
1376 * (bmap requires CAP_SYS_RAWIO so this does not
1377 * represent an unprivileged user DOS attack --- we'd be
1378 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001379 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 *
1381 * NB. EXT3_STATE_JDATA is not set on files other than
1382 * regular files. If somebody wants to bmap a directory
1383 * or symlink and gets confused because the buffer
1384 * hasn't yet been flushed to disk, they deserve
1385 * everything they get.
1386 */
1387
1388 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1389 journal = EXT3_JOURNAL(inode);
1390 journal_lock_updates(journal);
1391 err = journal_flush(journal);
1392 journal_unlock_updates(journal);
1393
1394 if (err)
1395 return 0;
1396 }
1397
1398 return generic_block_bmap(mapping,block,ext3_get_block);
1399}
1400
1401static int bget_one(handle_t *handle, struct buffer_head *bh)
1402{
1403 get_bh(bh);
1404 return 0;
1405}
1406
1407static int bput_one(handle_t *handle, struct buffer_head *bh)
1408{
1409 put_bh(bh);
1410 return 0;
1411}
1412
1413static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1414{
1415 if (buffer_mapped(bh))
1416 return ext3_journal_dirty_data(handle, bh);
1417 return 0;
1418}
1419
1420/*
1421 * Note that we always start a transaction even if we're not journalling
1422 * data. This is to preserve ordering: any hole instantiation within
1423 * __block_write_full_page -> ext3_get_block() should be journalled
1424 * along with the data so we don't crash and then get metadata which
1425 * refers to old data.
1426 *
1427 * In all journalling modes block_write_full_page() will start the I/O.
1428 *
1429 * Problem:
1430 *
1431 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1432 * ext3_writepage()
1433 *
1434 * Similar for:
1435 *
1436 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1437 *
1438 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001439 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 *
1441 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1442 * allocations fail.
1443 *
1444 * 16May01: If we're reentered then journal_current_handle() will be
1445 * non-zero. We simply *return*.
1446 *
1447 * 1 July 2001: @@@ FIXME:
1448 * In journalled data mode, a data buffer may be metadata against the
1449 * current transaction. But the same file is part of a shared mapping
1450 * and someone does a writepage() on it.
1451 *
1452 * We will move the buffer onto the async_data list, but *after* it has
1453 * been dirtied. So there's a small window where we have dirty data on
1454 * BJ_Metadata.
1455 *
1456 * Note that this only applies to the last partial page in the file. The
1457 * bit which block_write_full_page() uses prepare/commit for. (That's
1458 * broken code anyway: it's wrong for msync()).
1459 *
1460 * It's a rare case: affects the final partial page, for journalled data
1461 * where the file is subject to bith write() and writepage() in the same
1462 * transction. To fix it we'll need a custom block_write_full_page().
1463 * We'll probably need that anyway for journalling writepage() output.
1464 *
1465 * We don't honour synchronous mounts for writepage(). That would be
1466 * disastrous. Any write() or metadata operation will sync the fs for
1467 * us.
1468 *
1469 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1470 * we don't need to open a transaction here.
1471 */
1472static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001473 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474{
1475 struct inode *inode = page->mapping->host;
1476 struct buffer_head *page_bufs;
1477 handle_t *handle = NULL;
1478 int ret = 0;
1479 int err;
1480
1481 J_ASSERT(PageLocked(page));
1482
1483 /*
1484 * We give up here if we're reentered, because it might be for a
1485 * different filesystem.
1486 */
1487 if (ext3_journal_current_handle())
1488 goto out_fail;
1489
1490 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1491
1492 if (IS_ERR(handle)) {
1493 ret = PTR_ERR(handle);
1494 goto out_fail;
1495 }
1496
1497 if (!page_has_buffers(page)) {
1498 create_empty_buffers(page, inode->i_sb->s_blocksize,
1499 (1 << BH_Dirty)|(1 << BH_Uptodate));
1500 }
1501 page_bufs = page_buffers(page);
1502 walk_page_buffers(handle, page_bufs, 0,
1503 PAGE_CACHE_SIZE, NULL, bget_one);
1504
1505 ret = block_write_full_page(page, ext3_get_block, wbc);
1506
1507 /*
1508 * The page can become unlocked at any point now, and
1509 * truncate can then come in and change things. So we
1510 * can't touch *page from now on. But *page_bufs is
1511 * safe due to elevated refcount.
1512 */
1513
1514 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001515 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 * block_write_full_page() succeeded. Otherwise they are unmapped,
1517 * and generally junk.
1518 */
1519 if (ret == 0) {
1520 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1521 NULL, journal_dirty_data_fn);
1522 if (!ret)
1523 ret = err;
1524 }
1525 walk_page_buffers(handle, page_bufs, 0,
1526 PAGE_CACHE_SIZE, NULL, bput_one);
1527 err = ext3_journal_stop(handle);
1528 if (!ret)
1529 ret = err;
1530 return ret;
1531
1532out_fail:
1533 redirty_page_for_writepage(wbc, page);
1534 unlock_page(page);
1535 return ret;
1536}
1537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538static int ext3_writeback_writepage(struct page *page,
1539 struct writeback_control *wbc)
1540{
1541 struct inode *inode = page->mapping->host;
1542 handle_t *handle = NULL;
1543 int ret = 0;
1544 int err;
1545
1546 if (ext3_journal_current_handle())
1547 goto out_fail;
1548
1549 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1550 if (IS_ERR(handle)) {
1551 ret = PTR_ERR(handle);
1552 goto out_fail;
1553 }
1554
Badari Pulavarty0e31f512006-07-30 03:04:14 -07001555 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 ret = nobh_writepage(page, ext3_get_block, wbc);
1557 else
1558 ret = block_write_full_page(page, ext3_get_block, wbc);
1559
1560 err = ext3_journal_stop(handle);
1561 if (!ret)
1562 ret = err;
1563 return ret;
1564
1565out_fail:
1566 redirty_page_for_writepage(wbc, page);
1567 unlock_page(page);
1568 return ret;
1569}
1570
1571static int ext3_journalled_writepage(struct page *page,
1572 struct writeback_control *wbc)
1573{
1574 struct inode *inode = page->mapping->host;
1575 handle_t *handle = NULL;
1576 int ret = 0;
1577 int err;
1578
1579 if (ext3_journal_current_handle())
1580 goto no_write;
1581
1582 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1583 if (IS_ERR(handle)) {
1584 ret = PTR_ERR(handle);
1585 goto no_write;
1586 }
1587
1588 if (!page_has_buffers(page) || PageChecked(page)) {
1589 /*
1590 * It's mmapped pagecache. Add buffers and journal it. There
1591 * doesn't seem much point in redirtying the page here.
1592 */
1593 ClearPageChecked(page);
1594 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1595 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001596 if (ret != 0) {
1597 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 ret = walk_page_buffers(handle, page_buffers(page), 0,
1601 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1602
1603 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001604 PAGE_CACHE_SIZE, NULL, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 if (ret == 0)
1606 ret = err;
1607 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1608 unlock_page(page);
1609 } else {
1610 /*
1611 * It may be a page full of checkpoint-mode buffers. We don't
1612 * really know unless we go poke around in the buffer_heads.
1613 * But block_write_full_page will do the right thing.
1614 */
1615 ret = block_write_full_page(page, ext3_get_block, wbc);
1616 }
1617 err = ext3_journal_stop(handle);
1618 if (!ret)
1619 ret = err;
1620out:
1621 return ret;
1622
1623no_write:
1624 redirty_page_for_writepage(wbc, page);
1625out_unlock:
1626 unlock_page(page);
1627 goto out;
1628}
1629
1630static int ext3_readpage(struct file *file, struct page *page)
1631{
1632 return mpage_readpage(page, ext3_get_block);
1633}
1634
1635static int
1636ext3_readpages(struct file *file, struct address_space *mapping,
1637 struct list_head *pages, unsigned nr_pages)
1638{
1639 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1640}
1641
NeilBrown2ff28e22006-03-26 01:37:18 -08001642static void ext3_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
1644 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1645
1646 /*
1647 * If it's a full truncate we just forget about the pending dirtying
1648 */
1649 if (offset == 0)
1650 ClearPageChecked(page);
1651
NeilBrown2ff28e22006-03-26 01:37:18 -08001652 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653}
1654
Al Viro27496a82005-10-21 03:20:48 -04001655static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
1657 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1658
1659 WARN_ON(PageChecked(page));
1660 if (!page_has_buffers(page))
1661 return 0;
1662 return journal_try_to_free_buffers(journal, page, wait);
1663}
1664
1665/*
1666 * If the O_DIRECT write will extend the file then add this inode to the
1667 * orphan list. So recovery will truncate it back to the original size
1668 * if the machine crashes during the write.
1669 *
1670 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Karabd1939d2008-02-06 01:40:21 -08001671 * crashes then stale disk data _may_ be exposed inside the file. But current
1672 * VFS code falls back into buffered path in that case so we are safe.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 */
1674static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1675 const struct iovec *iov, loff_t offset,
1676 unsigned long nr_segs)
1677{
1678 struct file *file = iocb->ki_filp;
1679 struct inode *inode = file->f_mapping->host;
1680 struct ext3_inode_info *ei = EXT3_I(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001681 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 ssize_t ret;
1683 int orphan = 0;
1684 size_t count = iov_length(iov, nr_segs);
1685
1686 if (rw == WRITE) {
1687 loff_t final_size = offset + count;
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 if (final_size > inode->i_size) {
Jan Karabd1939d2008-02-06 01:40:21 -08001690 /* Credits for sb + inode write */
1691 handle = ext3_journal_start(inode, 2);
1692 if (IS_ERR(handle)) {
1693 ret = PTR_ERR(handle);
1694 goto out;
1695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 ret = ext3_orphan_add(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001697 if (ret) {
1698 ext3_journal_stop(handle);
1699 goto out;
1700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 orphan = 1;
1702 ei->i_disksize = inode->i_size;
Jan Karabd1939d2008-02-06 01:40:21 -08001703 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 }
1705 }
1706
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001707 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 offset, nr_segs,
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001709 ext3_get_block, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
Jan Karabd1939d2008-02-06 01:40:21 -08001711 if (orphan) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 int err;
1713
Jan Karabd1939d2008-02-06 01:40:21 -08001714 /* Credits for sb + inode write */
1715 handle = ext3_journal_start(inode, 2);
1716 if (IS_ERR(handle)) {
1717 /* This is really bad luck. We've written the data
1718 * but cannot extend i_size. Bail out and pretend
1719 * the write failed... */
1720 ret = PTR_ERR(handle);
1721 goto out;
1722 }
1723 if (inode->i_nlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 ext3_orphan_del(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001725 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 loff_t end = offset + ret;
1727 if (end > inode->i_size) {
1728 ei->i_disksize = end;
1729 i_size_write(inode, end);
1730 /*
1731 * We're going to return a positive `ret'
1732 * here due to non-zero-length I/O, so there's
1733 * no way of reporting error returns from
1734 * ext3_mark_inode_dirty() to userspace. So
1735 * ignore it.
1736 */
1737 ext3_mark_inode_dirty(handle, inode);
1738 }
1739 }
1740 err = ext3_journal_stop(handle);
1741 if (ret == 0)
1742 ret = err;
1743 }
1744out:
1745 return ret;
1746}
1747
1748/*
1749 * Pages can be marked dirty completely asynchronously from ext3's journalling
1750 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1751 * much here because ->set_page_dirty is called under VFS locks. The page is
1752 * not necessarily locked.
1753 *
1754 * We cannot just dirty the page and leave attached buffers clean, because the
1755 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1756 * or jbddirty because all the journalling code will explode.
1757 *
1758 * So what we do is to mark the page "pending dirty" and next time writepage
1759 * is called, propagate that into the buffers appropriately.
1760 */
1761static int ext3_journalled_set_page_dirty(struct page *page)
1762{
1763 SetPageChecked(page);
1764 return __set_page_dirty_nobuffers(page);
1765}
1766
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001767static const struct address_space_operations ext3_ordered_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 .readpage = ext3_readpage,
1769 .readpages = ext3_readpages,
1770 .writepage = ext3_ordered_writepage,
1771 .sync_page = block_sync_page,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001772 .write_begin = ext3_write_begin,
1773 .write_end = ext3_ordered_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 .bmap = ext3_bmap,
1775 .invalidatepage = ext3_invalidatepage,
1776 .releasepage = ext3_releasepage,
1777 .direct_IO = ext3_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001778 .migratepage = buffer_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779};
1780
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001781static const struct address_space_operations ext3_writeback_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 .readpage = ext3_readpage,
1783 .readpages = ext3_readpages,
1784 .writepage = ext3_writeback_writepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 .sync_page = block_sync_page,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001786 .write_begin = ext3_write_begin,
1787 .write_end = ext3_writeback_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 .bmap = ext3_bmap,
1789 .invalidatepage = ext3_invalidatepage,
1790 .releasepage = ext3_releasepage,
1791 .direct_IO = ext3_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001792 .migratepage = buffer_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793};
1794
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001795static const struct address_space_operations ext3_journalled_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 .readpage = ext3_readpage,
1797 .readpages = ext3_readpages,
1798 .writepage = ext3_journalled_writepage,
1799 .sync_page = block_sync_page,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001800 .write_begin = ext3_write_begin,
1801 .write_end = ext3_journalled_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 .set_page_dirty = ext3_journalled_set_page_dirty,
1803 .bmap = ext3_bmap,
1804 .invalidatepage = ext3_invalidatepage,
1805 .releasepage = ext3_releasepage,
1806};
1807
1808void ext3_set_aops(struct inode *inode)
1809{
1810 if (ext3_should_order_data(inode))
1811 inode->i_mapping->a_ops = &ext3_ordered_aops;
1812 else if (ext3_should_writeback_data(inode))
1813 inode->i_mapping->a_ops = &ext3_writeback_aops;
1814 else
1815 inode->i_mapping->a_ops = &ext3_journalled_aops;
1816}
1817
1818/*
1819 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1820 * up to the end of the block which corresponds to `from'.
1821 * This required during truncate. We need to physically zero the tail end
1822 * of that block so it doesn't yield old data if the file is later grown.
1823 */
1824static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1825 struct address_space *mapping, loff_t from)
1826{
Mingming Cao43d23f92006-06-25 05:48:07 -07001827 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1829 unsigned blocksize, iblock, length, pos;
1830 struct inode *inode = mapping->host;
1831 struct buffer_head *bh;
1832 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
1834 blocksize = inode->i_sb->s_blocksize;
1835 length = blocksize - (offset & (blocksize - 1));
1836 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1837
1838 /*
1839 * For "nobh" option, we can only work if we don't need to
1840 * read-in the page - otherwise we create buffers to do the IO.
1841 */
Badari Pulavartycd6ef842006-03-11 03:27:14 -08001842 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1843 ext3_should_writeback_data(inode) && PageUptodate(page)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001844 zero_user(page, offset, length);
Badari Pulavartycd6ef842006-03-11 03:27:14 -08001845 set_page_dirty(page);
1846 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 }
1848
1849 if (!page_has_buffers(page))
1850 create_empty_buffers(page, blocksize, 0);
1851
1852 /* Find the buffer that contains "offset" */
1853 bh = page_buffers(page);
1854 pos = blocksize;
1855 while (offset >= pos) {
1856 bh = bh->b_this_page;
1857 iblock++;
1858 pos += blocksize;
1859 }
1860
1861 err = 0;
1862 if (buffer_freed(bh)) {
1863 BUFFER_TRACE(bh, "freed: skip");
1864 goto unlock;
1865 }
1866
1867 if (!buffer_mapped(bh)) {
1868 BUFFER_TRACE(bh, "unmapped");
1869 ext3_get_block(inode, iblock, bh, 0);
1870 /* unmapped? It's a hole - nothing to do */
1871 if (!buffer_mapped(bh)) {
1872 BUFFER_TRACE(bh, "still unmapped");
1873 goto unlock;
1874 }
1875 }
1876
1877 /* Ok, it's mapped. Make sure it's up-to-date */
1878 if (PageUptodate(page))
1879 set_buffer_uptodate(bh);
1880
1881 if (!buffer_uptodate(bh)) {
1882 err = -EIO;
1883 ll_rw_block(READ, 1, &bh);
1884 wait_on_buffer(bh);
1885 /* Uhhuh. Read error. Complain and punt. */
1886 if (!buffer_uptodate(bh))
1887 goto unlock;
1888 }
1889
1890 if (ext3_should_journal_data(inode)) {
1891 BUFFER_TRACE(bh, "get write access");
1892 err = ext3_journal_get_write_access(handle, bh);
1893 if (err)
1894 goto unlock;
1895 }
1896
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001897 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 BUFFER_TRACE(bh, "zeroed end of block");
1899
1900 err = 0;
1901 if (ext3_should_journal_data(inode)) {
1902 err = ext3_journal_dirty_metadata(handle, bh);
1903 } else {
1904 if (ext3_should_order_data(inode))
1905 err = ext3_journal_dirty_data(handle, bh);
1906 mark_buffer_dirty(bh);
1907 }
1908
1909unlock:
1910 unlock_page(page);
1911 page_cache_release(page);
1912 return err;
1913}
1914
1915/*
1916 * Probably it should be a library function... search for first non-zero word
1917 * or memcmp with zero_page, whatever is better for particular architecture.
1918 * Linus?
1919 */
1920static inline int all_zeroes(__le32 *p, __le32 *q)
1921{
1922 while (p < q)
1923 if (*p++)
1924 return 0;
1925 return 1;
1926}
1927
1928/**
1929 * ext3_find_shared - find the indirect blocks for partial truncation.
1930 * @inode: inode in question
1931 * @depth: depth of the affected branch
1932 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1933 * @chain: place to store the pointers to partial indirect blocks
1934 * @top: place to the (detached) top of branch
1935 *
1936 * This is a helper function used by ext3_truncate().
1937 *
1938 * When we do truncate() we may have to clean the ends of several
1939 * indirect blocks but leave the blocks themselves alive. Block is
1940 * partially truncated if some data below the new i_size is refered
1941 * from it (and it is on the path to the first completely truncated
1942 * data block, indeed). We have to free the top of that path along
1943 * with everything to the right of the path. Since no allocation
1944 * past the truncation point is possible until ext3_truncate()
1945 * finishes, we may safely do the latter, but top of branch may
1946 * require special attention - pageout below the truncation point
1947 * might try to populate it.
1948 *
1949 * We atomically detach the top of branch from the tree, store the
1950 * block number of its root in *@top, pointers to buffer_heads of
1951 * partially truncated blocks - in @chain[].bh and pointers to
1952 * their last elements that should not be removed - in
1953 * @chain[].p. Return value is the pointer to last filled element
1954 * of @chain.
1955 *
1956 * The work left to caller to do the actual freeing of subtrees:
1957 * a) free the subtree starting from *@top
1958 * b) free the subtrees whose roots are stored in
1959 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1960 * c) free the subtrees growing from the inode past the @chain[0].
1961 * (no partially truncated stuff there). */
1962
Andrew Mortond6859bf2006-03-26 01:38:03 -08001963static Indirect *ext3_find_shared(struct inode *inode, int depth,
1964 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965{
1966 Indirect *partial, *p;
1967 int k, err;
1968
1969 *top = 0;
1970 /* Make k index the deepest non-null offest + 1 */
1971 for (k = depth; k > 1 && !offsets[k-1]; k--)
1972 ;
1973 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1974 /* Writer: pointers */
1975 if (!partial)
1976 partial = chain + k-1;
1977 /*
1978 * If the branch acquired continuation since we've looked at it -
1979 * fine, it should all survive and (new) top doesn't belong to us.
1980 */
1981 if (!partial->key && *partial->p)
1982 /* Writer: end */
1983 goto no_top;
1984 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1985 ;
1986 /*
1987 * OK, we've found the last block that must survive. The rest of our
1988 * branch should be detached before unlocking. However, if that rest
1989 * of branch is all ours and does not grow immediately from the inode
1990 * it's easier to cheat and just decrement partial->p.
1991 */
1992 if (p == chain + k - 1 && p > chain) {
1993 p->p--;
1994 } else {
1995 *top = *p->p;
1996 /* Nope, don't do this in ext3. Must leave the tree intact */
1997#if 0
1998 *p->p = 0;
1999#endif
2000 }
2001 /* Writer: end */
2002
Andrew Mortond6859bf2006-03-26 01:38:03 -08002003 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 brelse(partial->bh);
2005 partial--;
2006 }
2007no_top:
2008 return partial;
2009}
2010
2011/*
2012 * Zero a number of block pointers in either an inode or an indirect block.
2013 * If we restart the transaction we must again get write access to the
2014 * indirect block for further modification.
2015 *
2016 * We release `count' blocks on disk, but (last - first) may be greater
2017 * than `count' because there can be holes in there.
2018 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002019static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07002020 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08002021 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022{
2023 __le32 *p;
2024 if (try_to_extend_transaction(handle, inode)) {
2025 if (bh) {
2026 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2027 ext3_journal_dirty_metadata(handle, bh);
2028 }
2029 ext3_mark_inode_dirty(handle, inode);
2030 ext3_journal_test_restart(handle, inode);
2031 if (bh) {
2032 BUFFER_TRACE(bh, "retaking write access");
2033 ext3_journal_get_write_access(handle, bh);
2034 }
2035 }
2036
2037 /*
2038 * Any buffers which are on the journal will be in memory. We find
2039 * them on the hash table so journal_revoke() will run journal_forget()
2040 * on them. We've already detached each block from the file, so
2041 * bforget() in journal_forget() should be safe.
2042 *
2043 * AKPM: turn on bforget in journal_forget()!!!
2044 */
2045 for (p = first; p < last; p++) {
2046 u32 nr = le32_to_cpu(*p);
2047 if (nr) {
2048 struct buffer_head *bh;
2049
2050 *p = 0;
2051 bh = sb_find_get_block(inode->i_sb, nr);
2052 ext3_forget(handle, 0, inode, bh, nr);
2053 }
2054 }
2055
2056 ext3_free_blocks(handle, inode, block_to_free, count);
2057}
2058
2059/**
2060 * ext3_free_data - free a list of data blocks
2061 * @handle: handle for this transaction
2062 * @inode: inode we are dealing with
2063 * @this_bh: indirect buffer_head which contains *@first and *@last
2064 * @first: array of block numbers
2065 * @last: points immediately past the end of array
2066 *
2067 * We are freeing all blocks refered from that array (numbers are stored as
2068 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2069 *
2070 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2071 * blocks are contiguous then releasing them at one time will only affect one
2072 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2073 * actually use a lot of journal space.
2074 *
2075 * @this_bh will be %NULL if @first and @last point into the inode's direct
2076 * block pointers.
2077 */
2078static void ext3_free_data(handle_t *handle, struct inode *inode,
2079 struct buffer_head *this_bh,
2080 __le32 *first, __le32 *last)
2081{
Mingming Cao43d23f92006-06-25 05:48:07 -07002082 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002083 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2085 corresponding to
2086 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002087 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 __le32 *p; /* Pointer into inode/ind
2089 for current block */
2090 int err;
2091
2092 if (this_bh) { /* For indirect block */
2093 BUFFER_TRACE(this_bh, "get_write_access");
2094 err = ext3_journal_get_write_access(handle, this_bh);
2095 /* Important: if we can't update the indirect pointers
2096 * to the blocks, we can't free them. */
2097 if (err)
2098 return;
2099 }
2100
2101 for (p = first; p < last; p++) {
2102 nr = le32_to_cpu(*p);
2103 if (nr) {
2104 /* accumulate blocks to free if they're contiguous */
2105 if (count == 0) {
2106 block_to_free = nr;
2107 block_to_free_p = p;
2108 count = 1;
2109 } else if (nr == block_to_free + count) {
2110 count++;
2111 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002112 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 block_to_free,
2114 count, block_to_free_p, p);
2115 block_to_free = nr;
2116 block_to_free_p = p;
2117 count = 1;
2118 }
2119 }
2120 }
2121
2122 if (count > 0)
2123 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2124 count, block_to_free_p, p);
2125
2126 if (this_bh) {
2127 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2128 ext3_journal_dirty_metadata(handle, this_bh);
2129 }
2130}
2131
2132/**
2133 * ext3_free_branches - free an array of branches
2134 * @handle: JBD handle for this transaction
2135 * @inode: inode we are dealing with
2136 * @parent_bh: the buffer_head which contains *@first and *@last
2137 * @first: array of block numbers
2138 * @last: pointer immediately past the end of array
2139 * @depth: depth of the branches to free
2140 *
2141 * We are freeing all blocks refered from these branches (numbers are
2142 * stored as little-endian 32-bit) and updating @inode->i_blocks
2143 * appropriately.
2144 */
2145static void ext3_free_branches(handle_t *handle, struct inode *inode,
2146 struct buffer_head *parent_bh,
2147 __le32 *first, __le32 *last, int depth)
2148{
Mingming Cao43d23f92006-06-25 05:48:07 -07002149 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 __le32 *p;
2151
2152 if (is_handle_aborted(handle))
2153 return;
2154
2155 if (depth--) {
2156 struct buffer_head *bh;
2157 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2158 p = last;
2159 while (--p >= first) {
2160 nr = le32_to_cpu(*p);
2161 if (!nr)
2162 continue; /* A hole */
2163
2164 /* Go read the buffer for the next level down */
2165 bh = sb_bread(inode->i_sb, nr);
2166
2167 /*
2168 * A read failure? Report error and clear slot
2169 * (should be rare).
2170 */
2171 if (!bh) {
2172 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002173 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 inode->i_ino, nr);
2175 continue;
2176 }
2177
2178 /* This zaps the entire block. Bottom up. */
2179 BUFFER_TRACE(bh, "free child branches");
2180 ext3_free_branches(handle, inode, bh,
2181 (__le32*)bh->b_data,
2182 (__le32*)bh->b_data + addr_per_block,
2183 depth);
2184
2185 /*
2186 * We've probably journalled the indirect block several
2187 * times during the truncate. But it's no longer
2188 * needed and we now drop it from the transaction via
2189 * journal_revoke().
2190 *
2191 * That's easy if it's exclusively part of this
2192 * transaction. But if it's part of the committing
2193 * transaction then journal_forget() will simply
2194 * brelse() it. That means that if the underlying
2195 * block is reallocated in ext3_get_block(),
2196 * unmap_underlying_metadata() will find this block
2197 * and will try to get rid of it. damn, damn.
2198 *
2199 * If this block has already been committed to the
2200 * journal, a revoke record will be written. And
2201 * revoke records must be emitted *before* clearing
2202 * this block's bit in the bitmaps.
2203 */
2204 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2205
2206 /*
2207 * Everything below this this pointer has been
2208 * released. Now let this top-of-subtree go.
2209 *
2210 * We want the freeing of this indirect block to be
2211 * atomic in the journal with the updating of the
2212 * bitmap block which owns it. So make some room in
2213 * the journal.
2214 *
2215 * We zero the parent pointer *after* freeing its
2216 * pointee in the bitmaps, so if extend_transaction()
2217 * for some reason fails to put the bitmap changes and
2218 * the release into the same transaction, recovery
2219 * will merely complain about releasing a free block,
2220 * rather than leaking blocks.
2221 */
2222 if (is_handle_aborted(handle))
2223 return;
2224 if (try_to_extend_transaction(handle, inode)) {
2225 ext3_mark_inode_dirty(handle, inode);
2226 ext3_journal_test_restart(handle, inode);
2227 }
2228
2229 ext3_free_blocks(handle, inode, nr, 1);
2230
2231 if (parent_bh) {
2232 /*
2233 * The block which we have just freed is
2234 * pointed to by an indirect block: journal it
2235 */
2236 BUFFER_TRACE(parent_bh, "get_write_access");
2237 if (!ext3_journal_get_write_access(handle,
2238 parent_bh)){
2239 *p = 0;
2240 BUFFER_TRACE(parent_bh,
2241 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002242 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 parent_bh);
2244 }
2245 }
2246 }
2247 } else {
2248 /* We have reached the bottom of the tree. */
2249 BUFFER_TRACE(parent_bh, "free data blocks");
2250 ext3_free_data(handle, inode, parent_bh, first, last);
2251 }
2252}
2253
2254/*
2255 * ext3_truncate()
2256 *
2257 * We block out ext3_get_block() block instantiations across the entire
2258 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2259 * simultaneously on behalf of the same inode.
2260 *
2261 * As we work through the truncate and commmit bits of it to the journal there
2262 * is one core, guiding principle: the file's tree must always be consistent on
2263 * disk. We must be able to restart the truncate after a crash.
2264 *
2265 * The file's tree may be transiently inconsistent in memory (although it
2266 * probably isn't), but whenever we close off and commit a journal transaction,
2267 * the contents of (the filesystem + the journal) must be consistent and
2268 * restartable. It's pretty simple, really: bottom up, right to left (although
2269 * left-to-right works OK too).
2270 *
2271 * Note that at recovery time, journal replay occurs *before* the restart of
2272 * truncate against the orphan inode list.
2273 *
2274 * The committed inode has the new, desired i_size (which is the same as
2275 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2276 * that this inode's truncate did not complete and it will again call
2277 * ext3_truncate() to have another go. So there will be instantiated blocks
2278 * to the right of the truncation point in a crashed ext3 filesystem. But
2279 * that's fine - as long as they are linked from the inode, the post-crash
2280 * ext3_truncate() run will find them and release them.
2281 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002282void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283{
2284 handle_t *handle;
2285 struct ext3_inode_info *ei = EXT3_I(inode);
2286 __le32 *i_data = ei->i_data;
2287 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2288 struct address_space *mapping = inode->i_mapping;
2289 int offsets[4];
2290 Indirect chain[4];
2291 Indirect *partial;
2292 __le32 nr = 0;
2293 int n;
2294 long last_block;
2295 unsigned blocksize = inode->i_sb->s_blocksize;
2296 struct page *page;
2297
2298 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2299 S_ISLNK(inode->i_mode)))
2300 return;
2301 if (ext3_inode_is_fast_symlink(inode))
2302 return;
2303 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2304 return;
2305
2306 /*
2307 * We have to lock the EOF page here, because lock_page() nests
2308 * outside journal_start().
2309 */
2310 if ((inode->i_size & (blocksize - 1)) == 0) {
2311 /* Block boundary? Nothing to do */
2312 page = NULL;
2313 } else {
2314 page = grab_cache_page(mapping,
2315 inode->i_size >> PAGE_CACHE_SHIFT);
2316 if (!page)
2317 return;
2318 }
2319
2320 handle = start_transaction(inode);
2321 if (IS_ERR(handle)) {
2322 if (page) {
2323 clear_highpage(page);
2324 flush_dcache_page(page);
2325 unlock_page(page);
2326 page_cache_release(page);
2327 }
2328 return; /* AKPM: return what? */
2329 }
2330
2331 last_block = (inode->i_size + blocksize-1)
2332 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2333
2334 if (page)
2335 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2336
2337 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2338 if (n == 0)
2339 goto out_stop; /* error */
2340
2341 /*
2342 * OK. This truncate is going to happen. We add the inode to the
2343 * orphan list, so that if this truncate spans multiple transactions,
2344 * and we crash, we will resume the truncate when the filesystem
2345 * recovers. It also marks the inode dirty, to catch the new size.
2346 *
2347 * Implication: the file must always be in a sane, consistent
2348 * truncatable state while each transaction commits.
2349 */
2350 if (ext3_orphan_add(handle, inode))
2351 goto out_stop;
2352
2353 /*
2354 * The orphan list entry will now protect us from any crash which
2355 * occurs before the truncate completes, so it is now safe to propagate
2356 * the new, shorter inode size (held for now in i_size) into the
2357 * on-disk inode. We do this via i_disksize, which is the value which
2358 * ext3 *really* writes onto the disk inode.
2359 */
2360 ei->i_disksize = inode->i_size;
2361
2362 /*
2363 * From here we block out all ext3_get_block() callers who want to
2364 * modify the block allocation tree.
2365 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002366 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
2368 if (n == 1) { /* direct blocks */
2369 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2370 i_data + EXT3_NDIR_BLOCKS);
2371 goto do_indirects;
2372 }
2373
2374 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2375 /* Kill the top of shared branch (not detached) */
2376 if (nr) {
2377 if (partial == chain) {
2378 /* Shared branch grows from the inode */
2379 ext3_free_branches(handle, inode, NULL,
2380 &nr, &nr+1, (chain+n-1) - partial);
2381 *partial->p = 0;
2382 /*
2383 * We mark the inode dirty prior to restart,
2384 * and prior to stop. No need for it here.
2385 */
2386 } else {
2387 /* Shared branch grows from an indirect block */
2388 BUFFER_TRACE(partial->bh, "get_write_access");
2389 ext3_free_branches(handle, inode, partial->bh,
2390 partial->p,
2391 partial->p+1, (chain+n-1) - partial);
2392 }
2393 }
2394 /* Clear the ends of indirect blocks on the shared branch */
2395 while (partial > chain) {
2396 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2397 (__le32*)partial->bh->b_data+addr_per_block,
2398 (chain+n-1) - partial);
2399 BUFFER_TRACE(partial->bh, "call brelse");
2400 brelse (partial->bh);
2401 partial--;
2402 }
2403do_indirects:
2404 /* Kill the remaining (whole) subtrees */
2405 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002406 default:
2407 nr = i_data[EXT3_IND_BLOCK];
2408 if (nr) {
2409 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2410 i_data[EXT3_IND_BLOCK] = 0;
2411 }
2412 case EXT3_IND_BLOCK:
2413 nr = i_data[EXT3_DIND_BLOCK];
2414 if (nr) {
2415 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2416 i_data[EXT3_DIND_BLOCK] = 0;
2417 }
2418 case EXT3_DIND_BLOCK:
2419 nr = i_data[EXT3_TIND_BLOCK];
2420 if (nr) {
2421 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2422 i_data[EXT3_TIND_BLOCK] = 0;
2423 }
2424 case EXT3_TIND_BLOCK:
2425 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 }
2427
2428 ext3_discard_reservation(inode);
2429
Arjan van de Ven97461512006-03-23 03:00:42 -08002430 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2432 ext3_mark_inode_dirty(handle, inode);
2433
Andrew Mortond6859bf2006-03-26 01:38:03 -08002434 /*
2435 * In a multi-transaction truncate, we only make the final transaction
2436 * synchronous
2437 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (IS_SYNC(inode))
2439 handle->h_sync = 1;
2440out_stop:
2441 /*
2442 * If this was a simple ftruncate(), and the file will remain alive
2443 * then we need to clear up the orphan record which we created above.
2444 * However, if this was a real unlink then we were called by
2445 * ext3_delete_inode(), and we allow that function to clean up the
2446 * orphan info for us.
2447 */
2448 if (inode->i_nlink)
2449 ext3_orphan_del(handle, inode);
2450
2451 ext3_journal_stop(handle);
2452}
2453
Mingming Cao43d23f92006-06-25 05:48:07 -07002454static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 unsigned long ino, struct ext3_iloc *iloc)
2456{
2457 unsigned long desc, group_desc, block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002458 unsigned long offset;
2459 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 struct buffer_head *bh;
2461 struct ext3_group_desc * gdp;
2462
Neil Brown2ccb48e2006-07-30 03:03:01 -07002463 if (!ext3_valid_inum(sb, ino)) {
2464 /*
2465 * This error is already checked for in namei.c unless we are
2466 * looking at an NFS filehandle, in which case no error
2467 * report is needed
2468 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 return 0;
2470 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2473 if (block_group >= EXT3_SB(sb)->s_groups_count) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002474 ext3_error(sb,"ext3_get_inode_block","group >= groups count");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 return 0;
2476 }
2477 smp_rmb();
2478 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2479 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2480 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2481 if (!bh) {
2482 ext3_error (sb, "ext3_get_inode_block",
2483 "Descriptor not loaded");
2484 return 0;
2485 }
2486
Andrew Mortond6859bf2006-03-26 01:38:03 -08002487 gdp = (struct ext3_group_desc *)bh->b_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 /*
2489 * Figure out the offset within the block group inode table
2490 */
2491 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2492 EXT3_INODE_SIZE(sb);
2493 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2494 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2495
2496 iloc->block_group = block_group;
2497 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2498 return block;
2499}
2500
2501/*
2502 * ext3_get_inode_loc returns with an extra refcount against the inode's
2503 * underlying buffer_head on success. If 'in_mem' is true, we have all
2504 * data in memory that is needed to recreate the on-disk version of this
2505 * inode.
2506 */
2507static int __ext3_get_inode_loc(struct inode *inode,
2508 struct ext3_iloc *iloc, int in_mem)
2509{
Mingming Cao43d23f92006-06-25 05:48:07 -07002510 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 struct buffer_head *bh;
2512
2513 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2514 if (!block)
2515 return -EIO;
2516
2517 bh = sb_getblk(inode->i_sb, block);
2518 if (!bh) {
2519 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2520 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002521 "inode=%lu, block="E3FSBLK,
2522 inode->i_ino, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 return -EIO;
2524 }
2525 if (!buffer_uptodate(bh)) {
2526 lock_buffer(bh);
2527 if (buffer_uptodate(bh)) {
2528 /* someone brought it uptodate while we waited */
2529 unlock_buffer(bh);
2530 goto has_buffer;
2531 }
2532
2533 /*
2534 * If we have all information of the inode in memory and this
2535 * is the only valid inode in the block, we need not read the
2536 * block.
2537 */
2538 if (in_mem) {
2539 struct buffer_head *bitmap_bh;
2540 struct ext3_group_desc *desc;
2541 int inodes_per_buffer;
2542 int inode_offset, i;
2543 int block_group;
2544 int start;
2545
2546 block_group = (inode->i_ino - 1) /
2547 EXT3_INODES_PER_GROUP(inode->i_sb);
2548 inodes_per_buffer = bh->b_size /
2549 EXT3_INODE_SIZE(inode->i_sb);
2550 inode_offset = ((inode->i_ino - 1) %
2551 EXT3_INODES_PER_GROUP(inode->i_sb));
2552 start = inode_offset & ~(inodes_per_buffer - 1);
2553
2554 /* Is the inode bitmap in cache? */
2555 desc = ext3_get_group_desc(inode->i_sb,
2556 block_group, NULL);
2557 if (!desc)
2558 goto make_io;
2559
2560 bitmap_bh = sb_getblk(inode->i_sb,
2561 le32_to_cpu(desc->bg_inode_bitmap));
2562 if (!bitmap_bh)
2563 goto make_io;
2564
2565 /*
2566 * If the inode bitmap isn't in cache then the
2567 * optimisation may end up performing two reads instead
2568 * of one, so skip it.
2569 */
2570 if (!buffer_uptodate(bitmap_bh)) {
2571 brelse(bitmap_bh);
2572 goto make_io;
2573 }
2574 for (i = start; i < start + inodes_per_buffer; i++) {
2575 if (i == inode_offset)
2576 continue;
2577 if (ext3_test_bit(i, bitmap_bh->b_data))
2578 break;
2579 }
2580 brelse(bitmap_bh);
2581 if (i == start + inodes_per_buffer) {
2582 /* all other inodes are free, so skip I/O */
2583 memset(bh->b_data, 0, bh->b_size);
2584 set_buffer_uptodate(bh);
2585 unlock_buffer(bh);
2586 goto has_buffer;
2587 }
2588 }
2589
2590make_io:
2591 /*
2592 * There are other valid inodes in the buffer, this inode
2593 * has in-inode xattrs, or we don't have this inode in memory.
2594 * Read the block from disk.
2595 */
2596 get_bh(bh);
2597 bh->b_end_io = end_buffer_read_sync;
Jens Axboecaa38fb2006-07-23 01:41:26 +02002598 submit_bh(READ_META, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 wait_on_buffer(bh);
2600 if (!buffer_uptodate(bh)) {
2601 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2602 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002603 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 inode->i_ino, block);
2605 brelse(bh);
2606 return -EIO;
2607 }
2608 }
2609has_buffer:
2610 iloc->bh = bh;
2611 return 0;
2612}
2613
2614int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2615{
2616 /* We have all inode data except xattrs in memory here. */
2617 return __ext3_get_inode_loc(inode, iloc,
2618 !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR));
2619}
2620
2621void ext3_set_inode_flags(struct inode *inode)
2622{
2623 unsigned int flags = EXT3_I(inode)->i_flags;
2624
2625 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2626 if (flags & EXT3_SYNC_FL)
2627 inode->i_flags |= S_SYNC;
2628 if (flags & EXT3_APPEND_FL)
2629 inode->i_flags |= S_APPEND;
2630 if (flags & EXT3_IMMUTABLE_FL)
2631 inode->i_flags |= S_IMMUTABLE;
2632 if (flags & EXT3_NOATIME_FL)
2633 inode->i_flags |= S_NOATIME;
2634 if (flags & EXT3_DIRSYNC_FL)
2635 inode->i_flags |= S_DIRSYNC;
2636}
2637
Jan Kara28be5ab2007-05-08 00:30:33 -07002638/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2639void ext3_get_inode_flags(struct ext3_inode_info *ei)
2640{
2641 unsigned int flags = ei->vfs_inode.i_flags;
2642
2643 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2644 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2645 if (flags & S_SYNC)
2646 ei->i_flags |= EXT3_SYNC_FL;
2647 if (flags & S_APPEND)
2648 ei->i_flags |= EXT3_APPEND_FL;
2649 if (flags & S_IMMUTABLE)
2650 ei->i_flags |= EXT3_IMMUTABLE_FL;
2651 if (flags & S_NOATIME)
2652 ei->i_flags |= EXT3_NOATIME_FL;
2653 if (flags & S_DIRSYNC)
2654 ei->i_flags |= EXT3_DIRSYNC_FL;
2655}
2656
David Howells473043d2008-02-07 00:15:36 -08002657struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658{
2659 struct ext3_iloc iloc;
2660 struct ext3_inode *raw_inode;
David Howells473043d2008-02-07 00:15:36 -08002661 struct ext3_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 struct buffer_head *bh;
David Howells473043d2008-02-07 00:15:36 -08002663 struct inode *inode;
2664 long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 int block;
2666
David Howells473043d2008-02-07 00:15:36 -08002667 inode = iget_locked(sb, ino);
2668 if (!inode)
2669 return ERR_PTR(-ENOMEM);
2670 if (!(inode->i_state & I_NEW))
2671 return inode;
2672
2673 ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674#ifdef CONFIG_EXT3_FS_POSIX_ACL
2675 ei->i_acl = EXT3_ACL_NOT_CACHED;
2676 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2677#endif
2678 ei->i_block_alloc_info = NULL;
2679
David Howells473043d2008-02-07 00:15:36 -08002680 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2681 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 goto bad_inode;
2683 bh = iloc.bh;
2684 raw_inode = ext3_raw_inode(&iloc);
2685 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2686 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2687 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2688 if(!(test_opt (inode->i_sb, NO_UID32))) {
2689 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2690 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2691 }
2692 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2693 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002694 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2695 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2696 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2698
2699 ei->i_state = 0;
2700 ei->i_dir_start_lookup = 0;
2701 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2702 /* We now have enough fields to check if the inode was active or not.
2703 * This is needed because nfsd might try to access dead inodes
2704 * the test is that same one that e2fsck uses
2705 * NeilBrown 1999oct15
2706 */
2707 if (inode->i_nlink == 0) {
2708 if (inode->i_mode == 0 ||
2709 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2710 /* this inode is deleted */
2711 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002712 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 goto bad_inode;
2714 }
2715 /* The only unlinked inodes we let through here have
2716 * valid i_mode and are being read by the orphan
2717 * recovery code: that's fine, we're about to complete
2718 * the process of deleting those. */
2719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2721 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2722#ifdef EXT3_FRAGMENTS
2723 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2724 ei->i_frag_no = raw_inode->i_frag;
2725 ei->i_frag_size = raw_inode->i_fsize;
2726#endif
2727 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2728 if (!S_ISREG(inode->i_mode)) {
2729 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2730 } else {
2731 inode->i_size |=
2732 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2733 }
2734 ei->i_disksize = inode->i_size;
2735 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2736 ei->i_block_group = iloc.block_group;
2737 /*
2738 * NOTE! The in-memory inode i_data array is in little-endian order
2739 * even on big-endian machines: we do NOT byteswap the block numbers!
2740 */
2741 for (block = 0; block < EXT3_N_BLOCKS; block++)
2742 ei->i_data[block] = raw_inode->i_block[block];
2743 INIT_LIST_HEAD(&ei->i_orphan);
2744
2745 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2746 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2747 /*
2748 * When mke2fs creates big inodes it does not zero out
2749 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2750 * so ignore those first few inodes.
2751 */
2752 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2753 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07002754 EXT3_INODE_SIZE(inode->i_sb)) {
2755 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002756 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 goto bad_inode;
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07002758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 if (ei->i_extra_isize == 0) {
2760 /* The extra space is currently unused. Use it. */
2761 ei->i_extra_isize = sizeof(struct ext3_inode) -
2762 EXT3_GOOD_OLD_INODE_SIZE;
2763 } else {
2764 __le32 *magic = (void *)raw_inode +
2765 EXT3_GOOD_OLD_INODE_SIZE +
2766 ei->i_extra_isize;
2767 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2768 ei->i_state |= EXT3_STATE_XATTR;
2769 }
2770 } else
2771 ei->i_extra_isize = 0;
2772
2773 if (S_ISREG(inode->i_mode)) {
2774 inode->i_op = &ext3_file_inode_operations;
2775 inode->i_fop = &ext3_file_operations;
2776 ext3_set_aops(inode);
2777 } else if (S_ISDIR(inode->i_mode)) {
2778 inode->i_op = &ext3_dir_inode_operations;
2779 inode->i_fop = &ext3_dir_operations;
2780 } else if (S_ISLNK(inode->i_mode)) {
2781 if (ext3_inode_is_fast_symlink(inode))
2782 inode->i_op = &ext3_fast_symlink_inode_operations;
2783 else {
2784 inode->i_op = &ext3_symlink_inode_operations;
2785 ext3_set_aops(inode);
2786 }
2787 } else {
2788 inode->i_op = &ext3_special_inode_operations;
2789 if (raw_inode->i_block[0])
2790 init_special_inode(inode, inode->i_mode,
2791 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002792 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 init_special_inode(inode, inode->i_mode,
2794 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2795 }
2796 brelse (iloc.bh);
2797 ext3_set_inode_flags(inode);
David Howells473043d2008-02-07 00:15:36 -08002798 unlock_new_inode(inode);
2799 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800
2801bad_inode:
David Howells473043d2008-02-07 00:15:36 -08002802 iget_failed(inode);
2803 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804}
2805
2806/*
2807 * Post the struct inode info into an on-disk inode location in the
2808 * buffer-cache. This gobbles the caller's reference to the
2809 * buffer_head in the inode location struct.
2810 *
2811 * The caller must have write access to iloc->bh.
2812 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002813static int ext3_do_update_inode(handle_t *handle,
2814 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 struct ext3_iloc *iloc)
2816{
2817 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2818 struct ext3_inode_info *ei = EXT3_I(inode);
2819 struct buffer_head *bh = iloc->bh;
2820 int err = 0, rc, block;
2821
2822 /* For fields not not tracking in the in-memory inode,
2823 * initialise them to zero for new inodes. */
2824 if (ei->i_state & EXT3_STATE_NEW)
2825 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2826
Jan Kara28be5ab2007-05-08 00:30:33 -07002827 ext3_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2829 if(!(test_opt(inode->i_sb, NO_UID32))) {
2830 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2831 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2832/*
2833 * Fix up interoperability with old kernels. Otherwise, old inodes get
2834 * re-used with the upper 16 bits of the uid/gid intact
2835 */
2836 if(!ei->i_dtime) {
2837 raw_inode->i_uid_high =
2838 cpu_to_le16(high_16_bits(inode->i_uid));
2839 raw_inode->i_gid_high =
2840 cpu_to_le16(high_16_bits(inode->i_gid));
2841 } else {
2842 raw_inode->i_uid_high = 0;
2843 raw_inode->i_gid_high = 0;
2844 }
2845 } else {
2846 raw_inode->i_uid_low =
2847 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2848 raw_inode->i_gid_low =
2849 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2850 raw_inode->i_uid_high = 0;
2851 raw_inode->i_gid_high = 0;
2852 }
2853 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2854 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2855 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2856 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2857 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2858 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2859 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2860 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2861#ifdef EXT3_FRAGMENTS
2862 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2863 raw_inode->i_frag = ei->i_frag_no;
2864 raw_inode->i_fsize = ei->i_frag_size;
2865#endif
2866 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2867 if (!S_ISREG(inode->i_mode)) {
2868 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2869 } else {
2870 raw_inode->i_size_high =
2871 cpu_to_le32(ei->i_disksize >> 32);
2872 if (ei->i_disksize > 0x7fffffffULL) {
2873 struct super_block *sb = inode->i_sb;
2874 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2875 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2876 EXT3_SB(sb)->s_es->s_rev_level ==
2877 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2878 /* If this is the first large file
2879 * created, add a flag to the superblock.
2880 */
2881 err = ext3_journal_get_write_access(handle,
2882 EXT3_SB(sb)->s_sbh);
2883 if (err)
2884 goto out_brelse;
2885 ext3_update_dynamic_rev(sb);
2886 EXT3_SET_RO_COMPAT_FEATURE(sb,
2887 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2888 sb->s_dirt = 1;
2889 handle->h_sync = 1;
2890 err = ext3_journal_dirty_metadata(handle,
2891 EXT3_SB(sb)->s_sbh);
2892 }
2893 }
2894 }
2895 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2896 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2897 if (old_valid_dev(inode->i_rdev)) {
2898 raw_inode->i_block[0] =
2899 cpu_to_le32(old_encode_dev(inode->i_rdev));
2900 raw_inode->i_block[1] = 0;
2901 } else {
2902 raw_inode->i_block[0] = 0;
2903 raw_inode->i_block[1] =
2904 cpu_to_le32(new_encode_dev(inode->i_rdev));
2905 raw_inode->i_block[2] = 0;
2906 }
2907 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2908 raw_inode->i_block[block] = ei->i_data[block];
2909
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07002910 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2912
2913 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2914 rc = ext3_journal_dirty_metadata(handle, bh);
2915 if (!err)
2916 err = rc;
2917 ei->i_state &= ~EXT3_STATE_NEW;
2918
2919out_brelse:
2920 brelse (bh);
2921 ext3_std_error(inode->i_sb, err);
2922 return err;
2923}
2924
2925/*
2926 * ext3_write_inode()
2927 *
2928 * We are called from a few places:
2929 *
2930 * - Within generic_file_write() for O_SYNC files.
2931 * Here, there will be no transaction running. We wait for any running
2932 * trasnaction to commit.
2933 *
2934 * - Within sys_sync(), kupdate and such.
2935 * We wait on commit, if tol to.
2936 *
2937 * - Within prune_icache() (PF_MEMALLOC == true)
2938 * Here we simply return. We can't afford to block kswapd on the
2939 * journal commit.
2940 *
2941 * In all cases it is actually safe for us to return without doing anything,
2942 * because the inode has been copied into a raw inode buffer in
2943 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2944 * knfsd.
2945 *
2946 * Note that we are absolutely dependent upon all inode dirtiers doing the
2947 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2948 * which we are interested.
2949 *
2950 * It would be a bug for them to not do this. The code:
2951 *
2952 * mark_inode_dirty(inode)
2953 * stuff();
2954 * inode->i_size = expr;
2955 *
2956 * is in error because a kswapd-driven write_inode() could occur while
2957 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2958 * will no longer be on the superblock's dirty inode list.
2959 */
2960int ext3_write_inode(struct inode *inode, int wait)
2961{
2962 if (current->flags & PF_MEMALLOC)
2963 return 0;
2964
2965 if (ext3_journal_current_handle()) {
Jose R. Santos9ad163a2007-10-18 23:39:23 -07002966 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 dump_stack();
2968 return -EIO;
2969 }
2970
2971 if (!wait)
2972 return 0;
2973
2974 return ext3_force_commit(inode->i_sb);
2975}
2976
2977/*
2978 * ext3_setattr()
2979 *
2980 * Called from notify_change.
2981 *
2982 * We want to trap VFS attempts to truncate the file as soon as
2983 * possible. In particular, we want to make sure that when the VFS
2984 * shrinks i_size, we put the inode on the orphan list and modify
2985 * i_disksize immediately, so that during the subsequent flushing of
2986 * dirty pages and freeing of disk blocks, we can guarantee that any
2987 * commit will leave the blocks being flushed in an unused state on
2988 * disk. (On recovery, the inode will get truncated and the blocks will
2989 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002990 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 *
2992 * Called with inode->sem down.
2993 */
2994int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2995{
2996 struct inode *inode = dentry->d_inode;
2997 int error, rc = 0;
2998 const unsigned int ia_valid = attr->ia_valid;
2999
3000 error = inode_change_ok(inode, attr);
3001 if (error)
3002 return error;
3003
3004 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3005 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3006 handle_t *handle;
3007
3008 /* (user+group)*(old+new) structure, inode write (sb,
3009 * inode block, ? - but truncate inode update has it) */
Jan Kara1f545872005-06-23 22:01:04 -07003010 handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+
3011 EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 if (IS_ERR(handle)) {
3013 error = PTR_ERR(handle);
3014 goto err_out;
3015 }
3016 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3017 if (error) {
3018 ext3_journal_stop(handle);
3019 return error;
3020 }
3021 /* Update corresponding info in inode so that everything is in
3022 * one transaction */
3023 if (attr->ia_valid & ATTR_UID)
3024 inode->i_uid = attr->ia_uid;
3025 if (attr->ia_valid & ATTR_GID)
3026 inode->i_gid = attr->ia_gid;
3027 error = ext3_mark_inode_dirty(handle, inode);
3028 ext3_journal_stop(handle);
3029 }
3030
3031 if (S_ISREG(inode->i_mode) &&
3032 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3033 handle_t *handle;
3034
3035 handle = ext3_journal_start(inode, 3);
3036 if (IS_ERR(handle)) {
3037 error = PTR_ERR(handle);
3038 goto err_out;
3039 }
3040
3041 error = ext3_orphan_add(handle, inode);
3042 EXT3_I(inode)->i_disksize = attr->ia_size;
3043 rc = ext3_mark_inode_dirty(handle, inode);
3044 if (!error)
3045 error = rc;
3046 ext3_journal_stop(handle);
3047 }
3048
3049 rc = inode_setattr(inode, attr);
3050
3051 /* If inode_setattr's call to ext3_truncate failed to get a
3052 * transaction handle at all, we need to clean up the in-core
3053 * orphan list manually. */
3054 if (inode->i_nlink)
3055 ext3_orphan_del(NULL, inode);
3056
3057 if (!rc && (ia_valid & ATTR_MODE))
3058 rc = ext3_acl_chmod(inode);
3059
3060err_out:
3061 ext3_std_error(inode->i_sb, error);
3062 if (!error)
3063 error = rc;
3064 return error;
3065}
3066
3067
3068/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003069 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 *
3071 * With N blocks per page, it may be:
3072 * N data blocks
3073 * 2 indirect block
3074 * 2 dindirect
3075 * 1 tindirect
3076 * N+5 bitmap blocks (from the above)
3077 * N+5 group descriptor summary blocks
3078 * 1 inode block
3079 * 1 superblock.
3080 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3081 *
3082 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3083 *
3084 * With ordered or writeback data it's the same, less the N data blocks.
3085 *
3086 * If the inode's direct blocks can hold an integral number of pages then a
3087 * page cannot straddle two indirect blocks, and we can only touch one indirect
3088 * and dindirect block, and the "5" above becomes "3".
3089 *
3090 * This still overestimates under most circumstances. If we were to pass the
3091 * start and end offsets in here as well we could do block_to_path() on each
3092 * block and work out the exact number of indirects which are touched. Pah.
3093 */
3094
3095static int ext3_writepage_trans_blocks(struct inode *inode)
3096{
3097 int bpp = ext3_journal_blocks_per_page(inode);
3098 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3099 int ret;
3100
3101 if (ext3_should_journal_data(inode))
3102 ret = 3 * (bpp + indirects) + 2;
3103 else
3104 ret = 2 * (bpp + indirects) + 2;
3105
3106#ifdef CONFIG_QUOTA
3107 /* We know that structure was already allocated during DQUOT_INIT so
3108 * we will be updating only the data blocks + inodes */
Jan Kara1f545872005-06-23 22:01:04 -07003109 ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110#endif
3111
3112 return ret;
3113}
3114
3115/*
3116 * The caller must have previously called ext3_reserve_inode_write().
3117 * Give this, we know that the caller already has write access to iloc->bh.
3118 */
3119int ext3_mark_iloc_dirty(handle_t *handle,
3120 struct inode *inode, struct ext3_iloc *iloc)
3121{
3122 int err = 0;
3123
3124 /* the do_update_inode consumes one bh->b_count */
3125 get_bh(iloc->bh);
3126
3127 /* ext3_do_update_inode() does journal_dirty_metadata */
3128 err = ext3_do_update_inode(handle, inode, iloc);
3129 put_bh(iloc->bh);
3130 return err;
3131}
3132
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003133/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003135 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 */
3137
3138int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003139ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 struct ext3_iloc *iloc)
3141{
3142 int err = 0;
3143 if (handle) {
3144 err = ext3_get_inode_loc(inode, iloc);
3145 if (!err) {
3146 BUFFER_TRACE(iloc->bh, "get_write_access");
3147 err = ext3_journal_get_write_access(handle, iloc->bh);
3148 if (err) {
3149 brelse(iloc->bh);
3150 iloc->bh = NULL;
3151 }
3152 }
3153 }
3154 ext3_std_error(inode->i_sb, err);
3155 return err;
3156}
3157
3158/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003159 * What we do here is to mark the in-core inode as clean with respect to inode
3160 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 * This means that the in-core inode may be reaped by prune_icache
3162 * without having to perform any I/O. This is a very good thing,
3163 * because *any* task may call prune_icache - even ones which
3164 * have a transaction open against a different journal.
3165 *
3166 * Is this cheating? Not really. Sure, we haven't written the
3167 * inode out, but prune_icache isn't a user-visible syncing function.
3168 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3169 * we start and wait on commits.
3170 *
3171 * Is this efficient/effective? Well, we're being nice to the system
3172 * by cleaning up our inodes proactively so they can be reaped
3173 * without I/O. But we are potentially leaving up to five seconds'
3174 * worth of inodes floating about which prune_icache wants us to
3175 * write out. One way to fix that would be to get prune_icache()
3176 * to do a write_super() to free up some memory. It has the desired
3177 * effect.
3178 */
3179int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3180{
3181 struct ext3_iloc iloc;
3182 int err;
3183
3184 might_sleep();
3185 err = ext3_reserve_inode_write(handle, inode, &iloc);
3186 if (!err)
3187 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3188 return err;
3189}
3190
3191/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003192 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 *
3194 * We're really interested in the case where a file is being extended.
3195 * i_size has been changed by generic_commit_write() and we thus need
3196 * to include the updated inode in the current transaction.
3197 *
3198 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3199 * are allocated to the file.
3200 *
3201 * If the inode is marked synchronous, we don't honour that here - doing
3202 * so would cause a commit on atime updates, which we don't bother doing.
3203 * We handle synchronous inodes at the highest possible level.
3204 */
3205void ext3_dirty_inode(struct inode *inode)
3206{
3207 handle_t *current_handle = ext3_journal_current_handle();
3208 handle_t *handle;
3209
3210 handle = ext3_journal_start(inode, 2);
3211 if (IS_ERR(handle))
3212 goto out;
3213 if (current_handle &&
3214 current_handle->h_transaction != handle->h_transaction) {
3215 /* This task has a transaction open against a different fs */
3216 printk(KERN_EMERG "%s: transactions do not match!\n",
3217 __FUNCTION__);
3218 } else {
3219 jbd_debug(5, "marking dirty. outer handle=%p\n",
3220 current_handle);
3221 ext3_mark_inode_dirty(handle, inode);
3222 }
3223 ext3_journal_stop(handle);
3224out:
3225 return;
3226}
3227
Andrew Mortond6859bf2006-03-26 01:38:03 -08003228#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003229/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 * Bind an inode's backing buffer_head into this transaction, to prevent
3231 * it from being flushed to disk early. Unlike
3232 * ext3_reserve_inode_write, this leaves behind no bh reference and
3233 * returns no iloc structure, so the caller needs to repeat the iloc
3234 * lookup to mark the inode dirty later.
3235 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003236static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237{
3238 struct ext3_iloc iloc;
3239
3240 int err = 0;
3241 if (handle) {
3242 err = ext3_get_inode_loc(inode, &iloc);
3243 if (!err) {
3244 BUFFER_TRACE(iloc.bh, "get_write_access");
3245 err = journal_get_write_access(handle, iloc.bh);
3246 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003247 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 iloc.bh);
3249 brelse(iloc.bh);
3250 }
3251 }
3252 ext3_std_error(inode->i_sb, err);
3253 return err;
3254}
3255#endif
3256
3257int ext3_change_inode_journal_flag(struct inode *inode, int val)
3258{
3259 journal_t *journal;
3260 handle_t *handle;
3261 int err;
3262
3263 /*
3264 * We have to be very careful here: changing a data block's
3265 * journaling status dynamically is dangerous. If we write a
3266 * data block to the journal, change the status and then delete
3267 * that block, we risk forgetting to revoke the old log record
3268 * from the journal and so a subsequent replay can corrupt data.
3269 * So, first we make sure that the journal is empty and that
3270 * nobody is changing anything.
3271 */
3272
3273 journal = EXT3_JOURNAL(inode);
Dave Hansene3a68e32007-07-15 23:41:14 -07003274 if (is_journal_aborted(journal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 return -EROFS;
3276
3277 journal_lock_updates(journal);
3278 journal_flush(journal);
3279
3280 /*
3281 * OK, there are no updates running now, and all cached data is
3282 * synced to disk. We are now in a completely consistent state
3283 * which doesn't have anything in the journal, and we know that
3284 * no filesystem updates are running, so it is safe to modify
3285 * the inode's in-core data-journaling state flag now.
3286 */
3287
3288 if (val)
3289 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3290 else
3291 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3292 ext3_set_aops(inode);
3293
3294 journal_unlock_updates(journal);
3295
3296 /* Finally we can mark the inode as dirty. */
3297
3298 handle = ext3_journal_start(inode, 1);
3299 if (IS_ERR(handle))
3300 return PTR_ERR(handle);
3301
3302 err = ext3_mark_inode_dirty(handle, inode);
3303 handle->h_sync = 1;
3304 ext3_journal_stop(handle);
3305 ext3_std_error(inode->i_sb, err);
3306
3307 return err;
3308}