blob: dcf4f1dd108b21acd60c4ea3c88b0d8780b319c3 [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>
30#include <linux/smp_lock.h>
31#include <linux/highuid.h>
32#include <linux/pagemap.h>
33#include <linux/quotaops.h>
34#include <linux/string.h>
35#include <linux/buffer_head.h>
36#include <linux/writeback.h>
37#include <linux/mpage.h>
38#include <linux/uio.h>
39#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
442 * @chain: chain of indirect blocks
443 * @partial: pointer to the last triple within a chain
444 * @goal: place to store the result.
445 *
446 * Normally this function find the prefered place for block allocation,
Mingming Caofe55c452005-05-01 08:59:20 -0700447 * stores it in *@goal and returns zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 */
449
Mingming Cao43d23f92006-06-25 05:48:07 -0700450static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
Mingming Caofe55c452005-05-01 08:59:20 -0700451 Indirect chain[4], Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Andrew Mortond6859bf2006-03-26 01:38:03 -0800453 struct ext3_block_alloc_info *block_i;
454
455 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 /*
458 * try the heuristic for sequential allocation,
459 * failing that at least try to get decent locality.
460 */
461 if (block_i && (block == block_i->last_alloc_logical_block + 1)
462 && (block_i->last_alloc_physical_block != 0)) {
Mingming Caofe55c452005-05-01 08:59:20 -0700463 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
465
Mingming Caofe55c452005-05-01 08:59:20 -0700466 return ext3_find_near(inode, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
Andrew Mortond6859bf2006-03-26 01:38:03 -0800468
Mingming Caob47b2472006-03-26 01:37:56 -0800469/**
470 * ext3_blks_to_allocate: Look up the block map and count the number
471 * of direct blocks need to be allocated for the given branch.
472 *
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700473 * @branch: chain of indirect blocks
Mingming Caob47b2472006-03-26 01:37:56 -0800474 * @k: number of blocks need for indirect blocks
475 * @blks: number of data blocks to be mapped.
476 * @blocks_to_boundary: the offset in the indirect block
477 *
478 * return the total number of blocks to be allocate, including the
479 * direct and indirect blocks.
480 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800481static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Mingming Caob47b2472006-03-26 01:37:56 -0800482 int blocks_to_boundary)
483{
484 unsigned long count = 0;
485
486 /*
487 * Simple case, [t,d]Indirect block(s) has not allocated yet
488 * then it's clear blocks on that path have not allocated
489 */
490 if (k > 0) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800491 /* right now we don't handle cross boundary allocation */
Mingming Caob47b2472006-03-26 01:37:56 -0800492 if (blks < blocks_to_boundary + 1)
493 count += blks;
494 else
495 count += blocks_to_boundary + 1;
496 return count;
497 }
498
499 count++;
500 while (count < blks && count <= blocks_to_boundary &&
501 le32_to_cpu(*(branch[0].p + count)) == 0) {
502 count++;
503 }
504 return count;
505}
506
507/**
508 * ext3_alloc_blocks: multiple allocate blocks needed for a branch
509 * @indirect_blks: the number of blocks need to allocate for indirect
510 * blocks
511 *
512 * @new_blocks: on return it will store the new block numbers for
513 * the indirect blocks(if needed) and the first direct block,
514 * @blks: on return it will store the total number of allocated
515 * direct blocks
516 */
517static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700518 ext3_fsblk_t goal, int indirect_blks, int blks,
519 ext3_fsblk_t new_blocks[4], int *err)
Mingming Caob47b2472006-03-26 01:37:56 -0800520{
521 int target, i;
522 unsigned long count = 0;
523 int index = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700524 ext3_fsblk_t current_block = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800525 int ret = 0;
526
527 /*
528 * Here we try to allocate the requested multiple blocks at once,
529 * on a best-effort basis.
530 * To build a branch, we should allocate blocks for
531 * the indirect blocks(if not allocated yet), and at least
532 * the first direct block of this branch. That's the
533 * minimum number of blocks need to allocate(required)
534 */
535 target = blks + indirect_blks;
536
537 while (1) {
538 count = target;
539 /* allocating blocks for indirect blocks and direct blocks */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800540 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
Mingming Caob47b2472006-03-26 01:37:56 -0800541 if (*err)
542 goto failed_out;
543
544 target -= count;
545 /* allocate blocks for indirect blocks */
546 while (index < indirect_blks && count) {
547 new_blocks[index++] = current_block++;
548 count--;
549 }
550
551 if (count > 0)
552 break;
553 }
554
555 /* save the new block number for the first direct block */
556 new_blocks[index] = current_block;
557
558 /* total number of blocks allocated for direct blocks */
559 ret = count;
560 *err = 0;
561 return ret;
562failed_out:
563 for (i = 0; i <index; i++)
564 ext3_free_blocks(handle, inode, new_blocks[i], 1);
565 return ret;
566}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568/**
569 * ext3_alloc_branch - allocate and set up a chain of blocks.
570 * @inode: owner
Mingming Caob47b2472006-03-26 01:37:56 -0800571 * @indirect_blks: number of allocated indirect blocks
572 * @blks: number of allocated direct blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 * @offsets: offsets (in the blocks) to store the pointers to next.
574 * @branch: place to store the chain in.
575 *
Mingming Caob47b2472006-03-26 01:37:56 -0800576 * This function allocates blocks, zeroes out all but the last one,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 * links them into chain and (if we are synchronous) writes them to disk.
578 * In other words, it prepares a branch that can be spliced onto the
579 * inode. It stores the information about that chain in the branch[], in
580 * the same format as ext3_get_branch() would do. We are calling it after
581 * we had read the existing part of chain and partial points to the last
582 * triple of that (one with zero ->key). Upon the exit we have the same
Glauber de Oliveira Costa5b116872005-10-30 15:02:48 -0800583 * picture as after the successful ext3_get_block(), except that in one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 * place chain is disconnected - *branch->p is still zero (we did not
585 * set the last link), but branch->key contains the number that should
586 * be placed into *branch->p to fill that gap.
587 *
588 * If allocation fails we free all blocks we've allocated (and forget
589 * their buffer_heads) and return the error value the from failed
590 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
591 * as described above and return 0.
592 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -0700594 int indirect_blks, int *blks, ext3_fsblk_t goal,
Mingming Caob47b2472006-03-26 01:37:56 -0800595 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
597 int blocksize = inode->i_sb->s_blocksize;
Mingming Caob47b2472006-03-26 01:37:56 -0800598 int i, n = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 int err = 0;
Mingming Caob47b2472006-03-26 01:37:56 -0800600 struct buffer_head *bh;
601 int num;
Mingming Cao43d23f92006-06-25 05:48:07 -0700602 ext3_fsblk_t new_blocks[4];
603 ext3_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Mingming Caob47b2472006-03-26 01:37:56 -0800605 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
606 *blks, new_blocks, &err);
607 if (err)
608 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Mingming Caob47b2472006-03-26 01:37:56 -0800610 branch[0].key = cpu_to_le32(new_blocks[0]);
611 /*
612 * metadata blocks and data blocks are allocated.
613 */
614 for (n = 1; n <= indirect_blks; n++) {
615 /*
616 * Get buffer_head for parent block, zero it out
617 * and set the pointer to new one, then send
618 * parent to disk.
619 */
620 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
621 branch[n].bh = bh;
622 lock_buffer(bh);
623 BUFFER_TRACE(bh, "call get_create_access");
624 err = ext3_journal_get_create_access(handle, bh);
625 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 unlock_buffer(bh);
Mingming Caob47b2472006-03-26 01:37:56 -0800627 brelse(bh);
628 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Mingming Caob47b2472006-03-26 01:37:56 -0800631 memset(bh->b_data, 0, blocksize);
632 branch[n].p = (__le32 *) bh->b_data + offsets[n];
633 branch[n].key = cpu_to_le32(new_blocks[n]);
634 *branch[n].p = branch[n].key;
635 if ( n == indirect_blks) {
636 current_block = new_blocks[n];
637 /*
638 * End of chain, update the last new metablock of
639 * the chain to point to the new allocated
640 * data blocks numbers
641 */
642 for (i=1; i < num; i++)
643 *(branch[n].p + i) = cpu_to_le32(++current_block);
644 }
645 BUFFER_TRACE(bh, "marking uptodate");
646 set_buffer_uptodate(bh);
647 unlock_buffer(bh);
648
649 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
650 err = ext3_journal_dirty_metadata(handle, bh);
651 if (err)
652 goto failed;
653 }
654 *blks = num;
655 return err;
656failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 /* Allocation failed, free what we already allocated */
Mingming Caob47b2472006-03-26 01:37:56 -0800658 for (i = 1; i <= n ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 BUFFER_TRACE(branch[i].bh, "call journal_forget");
660 ext3_journal_forget(handle, branch[i].bh);
661 }
Mingming Caob47b2472006-03-26 01:37:56 -0800662 for (i = 0; i <indirect_blks; i++)
663 ext3_free_blocks(handle, inode, new_blocks[i], 1);
664
665 ext3_free_blocks(handle, inode, new_blocks[i], num);
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 return err;
668}
669
670/**
Andrew Mortond6859bf2006-03-26 01:38:03 -0800671 * ext3_splice_branch - splice the allocated branch onto inode.
672 * @inode: owner
673 * @block: (logical) number of block we are adding
674 * @chain: chain of indirect blocks (with a missing link - see
675 * ext3_alloc_branch)
676 * @where: location of missing link
677 * @num: number of indirect blocks we are adding
678 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800680 * This function fills the missing link and does all housekeeping needed in
681 * inode (->i_blocks, etc.). In case of success we end up with the full
682 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800684static int ext3_splice_branch(handle_t *handle, struct inode *inode,
685 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
687 int i;
688 int err = 0;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800689 struct ext3_block_alloc_info *block_i;
Mingming Cao43d23f92006-06-25 05:48:07 -0700690 ext3_fsblk_t current_block;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800691
692 block_i = EXT3_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 /*
694 * If we're splicing into a [td]indirect block (as opposed to the
695 * inode) then we need to get write access to the [td]indirect block
696 * before the splice.
697 */
698 if (where->bh) {
699 BUFFER_TRACE(where->bh, "get_write_access");
700 err = ext3_journal_get_write_access(handle, where->bh);
701 if (err)
702 goto err_out;
703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 /* That's it */
705
706 *where->p = where->key;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800707
708 /*
709 * Update the host buffer_head or inode to point to more just allocated
710 * direct blocks blocks
711 */
Mingming Caob47b2472006-03-26 01:37:56 -0800712 if (num == 0 && blks > 1) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700713 current_block = le32_to_cpu(where->key) + 1;
Mingming Caob47b2472006-03-26 01:37:56 -0800714 for (i = 1; i < blks; i++)
715 *(where->p + i ) = cpu_to_le32(current_block++);
716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 /*
719 * update the most recently allocated logical & physical block
720 * in i_block_alloc_info, to assist find the proper goal block for next
721 * allocation
722 */
723 if (block_i) {
Mingming Caob47b2472006-03-26 01:37:56 -0800724 block_i->last_alloc_logical_block = block + blks - 1;
Andrew Mortond6859bf2006-03-26 01:38:03 -0800725 block_i->last_alloc_physical_block =
Mingming Cao5dea5172006-05-03 19:55:12 -0700726 le32_to_cpu(where[num].key) + blks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
728
729 /* We are done with atomic stuff, now do the rest of housekeeping */
730
731 inode->i_ctime = CURRENT_TIME_SEC;
732 ext3_mark_inode_dirty(handle, inode);
733
734 /* had we spliced it onto indirect block? */
735 if (where->bh) {
736 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800737 * If we spliced it onto an indirect block, we haven't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 * altered the inode. Note however that if it is being spliced
739 * onto an indirect block at the very end of the file (the
740 * file is growing) then we *will* alter the inode to reflect
741 * the new i_size. But that is not done here - it is done in
742 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
743 */
744 jbd_debug(5, "splicing indirect only\n");
745 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
746 err = ext3_journal_dirty_metadata(handle, where->bh);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700747 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 goto err_out;
749 } else {
750 /*
751 * OK, we spliced it into the inode itself on a direct block.
752 * Inode was dirtied above.
753 */
754 jbd_debug(5, "splicing direct\n");
755 }
756 return err;
757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758err_out:
Mingming Caob47b2472006-03-26 01:37:56 -0800759 for (i = 1; i <= num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 BUFFER_TRACE(where[i].bh, "call journal_forget");
761 ext3_journal_forget(handle, where[i].bh);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800762 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 }
Mingming Caob47b2472006-03-26 01:37:56 -0800764 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 return err;
767}
768
769/*
770 * Allocation strategy is simple: if we have to allocate something, we will
771 * have to go the whole way to leaf. So let's do it before attaching anything
772 * to tree, set linkage between the newborn blocks, write them if sync is
773 * required, recheck the path, free and repeat if check fails, otherwise
774 * set the last missing link (that will protect us from any truncate-generated
775 * removals - all blocks on the path are immune now) and possibly force the
776 * write on the parent block.
777 * That has a nice additional property: no special recovery from the failed
778 * allocations is needed - we simply release blocks and do not touch anything
779 * reachable from inode.
780 *
Andrew Mortond6859bf2006-03-26 01:38:03 -0800781 * `handle' can be NULL if create == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 *
783 * The BKL may not be held on entry here. Be sure to take it early.
Mingming Cao89747d32006-03-26 01:37:55 -0800784 * return > 0, # of blocks mapped or allocated.
785 * return = 0, if plain lookup failed.
786 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800788int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
789 sector_t iblock, unsigned long maxblocks,
790 struct buffer_head *bh_result,
Mingming Cao89747d32006-03-26 01:37:55 -0800791 int create, int extend_disksize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
793 int err = -EIO;
794 int offsets[4];
795 Indirect chain[4];
796 Indirect *partial;
Mingming Cao43d23f92006-06-25 05:48:07 -0700797 ext3_fsblk_t goal;
Mingming Caob47b2472006-03-26 01:37:56 -0800798 int indirect_blks;
Mingming Cao89747d32006-03-26 01:37:55 -0800799 int blocks_to_boundary = 0;
800 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 struct ext3_inode_info *ei = EXT3_I(inode);
Mingming Cao89747d32006-03-26 01:37:55 -0800802 int count = 0;
Mingming Cao43d23f92006-06-25 05:48:07 -0700803 ext3_fsblk_t first_block = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
806 J_ASSERT(handle != NULL || create == 0);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800807 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 if (depth == 0)
810 goto out;
811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
813
814 /* Simplest case - block found, no allocation needed */
815 if (!partial) {
Mingming Cao5dea5172006-05-03 19:55:12 -0700816 first_block = le32_to_cpu(chain[depth - 1].key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 clear_buffer_new(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800818 count++;
819 /*map more blocks*/
820 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao43d23f92006-06-25 05:48:07 -0700821 ext3_fsblk_t blk;
Mingming Cao5dea5172006-05-03 19:55:12 -0700822
Mingming Cao89747d32006-03-26 01:37:55 -0800823 if (!verify_chain(chain, partial)) {
824 /*
825 * Indirect block might be removed by
826 * truncate while we were reading it.
827 * Handling of that case: forget what we've
828 * got now. Flag the err as EAGAIN, so it
829 * will reread.
830 */
831 err = -EAGAIN;
832 count = 0;
833 break;
834 }
Mingming Cao5dea5172006-05-03 19:55:12 -0700835 blk = le32_to_cpu(*(chain[depth-1].p + count));
836
837 if (blk == first_block + count)
Mingming Cao89747d32006-03-26 01:37:55 -0800838 count++;
839 else
840 break;
841 }
842 if (err != -EAGAIN)
843 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 }
845
846 /* Next simple case - plain lookup or failed read of indirect block */
Mingming Caofe55c452005-05-01 08:59:20 -0700847 if (!create || err == -EIO)
848 goto cleanup;
849
Arjan van de Ven97461512006-03-23 03:00:42 -0800850 mutex_lock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700851
852 /*
853 * If the indirect block is missing while we are reading
854 * the chain(ext3_get_branch() returns -EAGAIN err), or
855 * if the chain has been changed after we grab the semaphore,
856 * (either because another process truncated this branch, or
857 * another get_block allocated this branch) re-grab the chain to see if
858 * the request block has been allocated or not.
859 *
860 * Since we already block the truncate/other get_block
861 * at this point, we will have the current copy of the chain when we
862 * splice the branch into the tree.
863 */
864 if (err == -EAGAIN || !verify_chain(chain, partial)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 while (partial > chain) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 brelse(partial->bh);
867 partial--;
868 }
Mingming Caofe55c452005-05-01 08:59:20 -0700869 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
870 if (!partial) {
Mingming Cao89747d32006-03-26 01:37:55 -0800871 count++;
Arjan van de Ven97461512006-03-23 03:00:42 -0800872 mutex_unlock(&ei->truncate_mutex);
Mingming Caofe55c452005-05-01 08:59:20 -0700873 if (err)
874 goto cleanup;
875 clear_buffer_new(bh_result);
876 goto got_it;
877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879
880 /*
Mingming Caofe55c452005-05-01 08:59:20 -0700881 * Okay, we need to do block allocation. Lazily initialize the block
882 * allocation info here if necessary
883 */
884 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 ext3_init_block_alloc_info(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Mingming Caofe55c452005-05-01 08:59:20 -0700887 goal = ext3_find_goal(inode, iblock, chain, partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Mingming Caob47b2472006-03-26 01:37:56 -0800889 /* the number of blocks need to allocate for [d,t]indirect blocks */
890 indirect_blks = (chain + depth) - partial - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 /*
Mingming Caob47b2472006-03-26 01:37:56 -0800893 * Next look up the indirect map to count the totoal number of
894 * direct blocks to allocate for this branch.
895 */
896 count = ext3_blks_to_allocate(partial, indirect_blks,
897 maxblocks, blocks_to_boundary);
898 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 * Block out ext3_truncate while we alter the tree
900 */
Mingming Caob47b2472006-03-26 01:37:56 -0800901 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
Mingming Caofe55c452005-05-01 08:59:20 -0700902 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Mingming Caofe55c452005-05-01 08:59:20 -0700904 /*
905 * The ext3_splice_branch call will free and forget any buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 * on the new chain if there is a failure, but that risks using
907 * up transaction credits, especially for bitmaps where the
908 * credits cannot be returned. Can we handle this somehow? We
Mingming Caofe55c452005-05-01 08:59:20 -0700909 * may need to return -EAGAIN upwards in the worst case. --sct
910 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (!err)
Mingming Caob47b2472006-03-26 01:37:56 -0800912 err = ext3_splice_branch(handle, inode, iblock,
913 partial, indirect_blks, count);
Mingming Caofe55c452005-05-01 08:59:20 -0700914 /*
Arjan van de Ven97461512006-03-23 03:00:42 -0800915 * i_disksize growing is protected by truncate_mutex. Don't forget to
Mingming Caofe55c452005-05-01 08:59:20 -0700916 * protect it if you're about to implement concurrent
917 * ext3_get_block() -bzzz
918 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
920 ei->i_disksize = inode->i_size;
Arjan van de Ven97461512006-03-23 03:00:42 -0800921 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 if (err)
923 goto cleanup;
924
925 set_buffer_new(bh_result);
Mingming Caofe55c452005-05-01 08:59:20 -0700926got_it:
927 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
Suparna Bhattacharya20acaa12006-09-16 12:15:58 -0700928 if (count > blocks_to_boundary)
Mingming Caofe55c452005-05-01 08:59:20 -0700929 set_buffer_boundary(bh_result);
Mingming Cao89747d32006-03-26 01:37:55 -0800930 err = count;
Mingming Caofe55c452005-05-01 08:59:20 -0700931 /* Clean up and exit */
932 partial = chain + depth - 1; /* the whole chain */
933cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 while (partial > chain) {
Mingming Caofe55c452005-05-01 08:59:20 -0700935 BUFFER_TRACE(partial->bh, "call brelse");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 brelse(partial->bh);
937 partial--;
938 }
Mingming Caofe55c452005-05-01 08:59:20 -0700939 BUFFER_TRACE(bh_result, "returned");
940out:
941 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942}
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944#define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
945
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -0800946static int ext3_get_block(struct inode *inode, sector_t iblock,
947 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
949 handle_t *handle = journal_current_handle();
950 int ret = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800951 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Mingming Cao89747d32006-03-26 01:37:55 -0800953 if (!create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 goto get_block; /* A read */
955
Mingming Cao89747d32006-03-26 01:37:55 -0800956 if (max_blocks == 1)
957 goto get_block; /* A single block get */
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (handle->h_transaction->t_state == T_LOCKED) {
960 /*
961 * Huge direct-io writes can hold off commits for long
962 * periods of time. Let this commit run.
963 */
964 ext3_journal_stop(handle);
965 handle = ext3_journal_start(inode, DIO_CREDITS);
966 if (IS_ERR(handle))
967 ret = PTR_ERR(handle);
968 goto get_block;
969 }
970
971 if (handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
972 /*
973 * Getting low on buffer credits...
974 */
975 ret = ext3_journal_extend(handle, DIO_CREDITS);
976 if (ret > 0) {
977 /*
978 * Couldn't extend the transaction. Start a new one.
979 */
980 ret = ext3_journal_restart(handle, DIO_CREDITS);
981 }
982 }
983
984get_block:
Mingming Cao89747d32006-03-26 01:37:55 -0800985 if (ret == 0) {
986 ret = ext3_get_blocks_handle(handle, inode, iblock,
987 max_blocks, bh_result, create, 0);
988 if (ret > 0) {
989 bh_result->b_size = (ret << inode->i_blkbits);
990 ret = 0;
991 }
992 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 return ret;
994}
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996/*
997 * `handle' can be NULL if create is zero
998 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800999struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1000 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001{
1002 struct buffer_head dummy;
1003 int fatal = 0, err;
1004
1005 J_ASSERT(handle != NULL || create == 0);
1006
1007 dummy.b_state = 0;
1008 dummy.b_blocknr = -1000;
1009 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -08001010 err = ext3_get_blocks_handle(handle, inode, block, 1,
1011 &dummy, create, 1);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001012 /*
1013 * ext3_get_blocks_handle() returns number of blocks
1014 * mapped. 0 in case of a HOLE.
1015 */
1016 if (err > 0) {
1017 if (err > 1)
1018 WARN_ON(1);
Mingming Cao89747d32006-03-26 01:37:55 -08001019 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001020 }
1021 *errp = err;
1022 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 struct buffer_head *bh;
1024 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001025 if (!bh) {
1026 *errp = -EIO;
1027 goto err;
1028 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 if (buffer_new(&dummy)) {
1030 J_ASSERT(create != 0);
1031 J_ASSERT(handle != 0);
1032
Andrew Mortond6859bf2006-03-26 01:38:03 -08001033 /*
1034 * Now that we do not always journal data, we should
1035 * keep in mind whether this should always journal the
1036 * new buffer as metadata. For now, regular file
1037 * writes use ext3_get_block instead, so it's not a
1038 * problem.
1039 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 lock_buffer(bh);
1041 BUFFER_TRACE(bh, "call get_create_access");
1042 fatal = ext3_journal_get_create_access(handle, bh);
1043 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001044 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 set_buffer_uptodate(bh);
1046 }
1047 unlock_buffer(bh);
1048 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1049 err = ext3_journal_dirty_metadata(handle, bh);
1050 if (!fatal)
1051 fatal = err;
1052 } else {
1053 BUFFER_TRACE(bh, "not a new buffer");
1054 }
1055 if (fatal) {
1056 *errp = fatal;
1057 brelse(bh);
1058 bh = NULL;
1059 }
1060 return bh;
1061 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001062err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 return NULL;
1064}
1065
Andrew Mortond6859bf2006-03-26 01:38:03 -08001066struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 int block, int create, int *err)
1068{
1069 struct buffer_head * bh;
1070
1071 bh = ext3_getblk(handle, inode, block, create, err);
1072 if (!bh)
1073 return bh;
1074 if (buffer_uptodate(bh))
1075 return bh;
1076 ll_rw_block(READ, 1, &bh);
1077 wait_on_buffer(bh);
1078 if (buffer_uptodate(bh))
1079 return bh;
1080 put_bh(bh);
1081 *err = -EIO;
1082 return NULL;
1083}
1084
1085static int walk_page_buffers( handle_t *handle,
1086 struct buffer_head *head,
1087 unsigned from,
1088 unsigned to,
1089 int *partial,
1090 int (*fn)( handle_t *handle,
1091 struct buffer_head *bh))
1092{
1093 struct buffer_head *bh;
1094 unsigned block_start, block_end;
1095 unsigned blocksize = head->b_size;
1096 int err, ret = 0;
1097 struct buffer_head *next;
1098
1099 for ( bh = head, block_start = 0;
1100 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001101 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 {
1103 next = bh->b_this_page;
1104 block_end = block_start + blocksize;
1105 if (block_end <= from || block_start >= to) {
1106 if (partial && !buffer_uptodate(bh))
1107 *partial = 1;
1108 continue;
1109 }
1110 err = (*fn)(handle, bh);
1111 if (!ret)
1112 ret = err;
1113 }
1114 return ret;
1115}
1116
1117/*
1118 * To preserve ordering, it is essential that the hole instantiation and
1119 * the data write be encapsulated in a single transaction. We cannot
1120 * close off a transaction and start a new one between the ext3_get_block()
1121 * and the commit_write(). So doing the journal_start at the start of
1122 * prepare_write() is the right place.
1123 *
1124 * Also, this function can nest inside ext3_writepage() ->
1125 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1126 * has generated enough buffer credits to do the whole page. So we won't
1127 * block on the journal in that case, which is good, because the caller may
1128 * be PF_MEMALLOC.
1129 *
1130 * By accident, ext3 can be reentered when a transaction is open via
1131 * quota file writes. If we were to commit the transaction while thus
1132 * reentered, there can be a deadlock - we would be holding a quota
1133 * lock, and the commit would never complete if another thread had a
1134 * transaction open and was blocking on the quota lock - a ranking
1135 * violation.
1136 *
1137 * So what we do is to rely on the fact that journal_stop/journal_start
1138 * will _not_ run commit under these circumstances because handle->h_ref
1139 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001140 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001142static int do_journal_get_write_access(handle_t *handle,
1143 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
1145 if (!buffer_mapped(bh) || buffer_freed(bh))
1146 return 0;
1147 return ext3_journal_get_write_access(handle, bh);
1148}
1149
1150static int ext3_prepare_write(struct file *file, struct page *page,
1151 unsigned from, unsigned to)
1152{
1153 struct inode *inode = page->mapping->host;
1154 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
1155 handle_t *handle;
1156 int retries = 0;
1157
1158retry:
1159 handle = ext3_journal_start(inode, needed_blocks);
1160 if (IS_ERR(handle)) {
1161 ret = PTR_ERR(handle);
1162 goto out;
1163 }
Badari Pulavarty0e31f512006-07-30 03:04:14 -07001164 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 ret = nobh_prepare_write(page, from, to, ext3_get_block);
1166 else
1167 ret = block_prepare_write(page, from, to, ext3_get_block);
1168 if (ret)
1169 goto prepare_write_failed;
1170
1171 if (ext3_should_journal_data(inode)) {
1172 ret = walk_page_buffers(handle, page_buffers(page),
1173 from, to, NULL, do_journal_get_write_access);
1174 }
1175prepare_write_failed:
1176 if (ret)
1177 ext3_journal_stop(handle);
1178 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1179 goto retry;
1180out:
1181 return ret;
1182}
1183
Andrew Mortond6859bf2006-03-26 01:38:03 -08001184int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185{
1186 int err = journal_dirty_data(handle, bh);
1187 if (err)
1188 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1189 bh, handle,err);
1190 return err;
1191}
1192
1193/* For commit_write() in data=journal mode */
1194static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1195{
1196 if (!buffer_mapped(bh) || buffer_freed(bh))
1197 return 0;
1198 set_buffer_uptodate(bh);
1199 return ext3_journal_dirty_metadata(handle, bh);
1200}
1201
1202/*
1203 * We need to pick up the new inode size which generic_commit_write gave us
1204 * `file' can be NULL - eg, when called from page_symlink().
1205 *
1206 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1207 * buffers are managed internally.
1208 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209static int ext3_ordered_commit_write(struct file *file, struct page *page,
1210 unsigned from, unsigned to)
1211{
1212 handle_t *handle = ext3_journal_current_handle();
1213 struct inode *inode = page->mapping->host;
1214 int ret = 0, ret2;
1215
1216 ret = walk_page_buffers(handle, page_buffers(page),
1217 from, to, NULL, ext3_journal_dirty_data);
1218
1219 if (ret == 0) {
1220 /*
1221 * generic_commit_write() will run mark_inode_dirty() if i_size
1222 * changes. So let's piggyback the i_disksize mark_inode_dirty
1223 * into that.
1224 */
1225 loff_t new_i_size;
1226
1227 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1228 if (new_i_size > EXT3_I(inode)->i_disksize)
1229 EXT3_I(inode)->i_disksize = new_i_size;
1230 ret = generic_commit_write(file, page, from, to);
1231 }
1232 ret2 = ext3_journal_stop(handle);
1233 if (!ret)
1234 ret = ret2;
1235 return ret;
1236}
1237
1238static int ext3_writeback_commit_write(struct file *file, struct page *page,
1239 unsigned from, unsigned to)
1240{
1241 handle_t *handle = ext3_journal_current_handle();
1242 struct inode *inode = page->mapping->host;
1243 int ret = 0, ret2;
1244 loff_t new_i_size;
1245
1246 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1247 if (new_i_size > EXT3_I(inode)->i_disksize)
1248 EXT3_I(inode)->i_disksize = new_i_size;
1249
Badari Pulavarty0e31f512006-07-30 03:04:14 -07001250 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 ret = nobh_commit_write(file, page, from, to);
1252 else
1253 ret = generic_commit_write(file, page, from, to);
1254
1255 ret2 = ext3_journal_stop(handle);
1256 if (!ret)
1257 ret = ret2;
1258 return ret;
1259}
1260
1261static int ext3_journalled_commit_write(struct file *file,
1262 struct page *page, unsigned from, unsigned to)
1263{
1264 handle_t *handle = ext3_journal_current_handle();
1265 struct inode *inode = page->mapping->host;
1266 int ret = 0, ret2;
1267 int partial = 0;
1268 loff_t pos;
1269
1270 /*
1271 * Here we duplicate the generic_commit_write() functionality
1272 */
1273 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1274
1275 ret = walk_page_buffers(handle, page_buffers(page), from,
1276 to, &partial, commit_write_fn);
1277 if (!partial)
1278 SetPageUptodate(page);
1279 if (pos > inode->i_size)
1280 i_size_write(inode, pos);
1281 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1282 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1283 EXT3_I(inode)->i_disksize = inode->i_size;
1284 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001285 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 ret = ret2;
1287 }
1288 ret2 = ext3_journal_stop(handle);
1289 if (!ret)
1290 ret = ret2;
1291 return ret;
1292}
1293
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001294/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 * bmap() is special. It gets used by applications such as lilo and by
1296 * the swapper to find the on-disk block of a specific piece of data.
1297 *
1298 * Naturally, this is dangerous if the block concerned is still in the
1299 * journal. If somebody makes a swapfile on an ext3 data-journaling
1300 * filesystem and enables swap, then they may get a nasty shock when the
1301 * data getting swapped to that swapfile suddenly gets overwritten by
1302 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001303 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 *
1305 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001306 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 */
1308static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1309{
1310 struct inode *inode = mapping->host;
1311 journal_t *journal;
1312 int err;
1313
1314 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001315 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 * This is a REALLY heavyweight approach, but the use of
1317 * bmap on dirty files is expected to be extremely rare:
1318 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001319 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 *
1321 * (bmap requires CAP_SYS_RAWIO so this does not
1322 * represent an unprivileged user DOS attack --- we'd be
1323 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001324 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 *
1326 * NB. EXT3_STATE_JDATA is not set on files other than
1327 * regular files. If somebody wants to bmap a directory
1328 * or symlink and gets confused because the buffer
1329 * hasn't yet been flushed to disk, they deserve
1330 * everything they get.
1331 */
1332
1333 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1334 journal = EXT3_JOURNAL(inode);
1335 journal_lock_updates(journal);
1336 err = journal_flush(journal);
1337 journal_unlock_updates(journal);
1338
1339 if (err)
1340 return 0;
1341 }
1342
1343 return generic_block_bmap(mapping,block,ext3_get_block);
1344}
1345
1346static int bget_one(handle_t *handle, struct buffer_head *bh)
1347{
1348 get_bh(bh);
1349 return 0;
1350}
1351
1352static int bput_one(handle_t *handle, struct buffer_head *bh)
1353{
1354 put_bh(bh);
1355 return 0;
1356}
1357
1358static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1359{
1360 if (buffer_mapped(bh))
1361 return ext3_journal_dirty_data(handle, bh);
1362 return 0;
1363}
1364
1365/*
1366 * Note that we always start a transaction even if we're not journalling
1367 * data. This is to preserve ordering: any hole instantiation within
1368 * __block_write_full_page -> ext3_get_block() should be journalled
1369 * along with the data so we don't crash and then get metadata which
1370 * refers to old data.
1371 *
1372 * In all journalling modes block_write_full_page() will start the I/O.
1373 *
1374 * Problem:
1375 *
1376 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1377 * ext3_writepage()
1378 *
1379 * Similar for:
1380 *
1381 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1382 *
1383 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001384 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 *
1386 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1387 * allocations fail.
1388 *
1389 * 16May01: If we're reentered then journal_current_handle() will be
1390 * non-zero. We simply *return*.
1391 *
1392 * 1 July 2001: @@@ FIXME:
1393 * In journalled data mode, a data buffer may be metadata against the
1394 * current transaction. But the same file is part of a shared mapping
1395 * and someone does a writepage() on it.
1396 *
1397 * We will move the buffer onto the async_data list, but *after* it has
1398 * been dirtied. So there's a small window where we have dirty data on
1399 * BJ_Metadata.
1400 *
1401 * Note that this only applies to the last partial page in the file. The
1402 * bit which block_write_full_page() uses prepare/commit for. (That's
1403 * broken code anyway: it's wrong for msync()).
1404 *
1405 * It's a rare case: affects the final partial page, for journalled data
1406 * where the file is subject to bith write() and writepage() in the same
1407 * transction. To fix it we'll need a custom block_write_full_page().
1408 * We'll probably need that anyway for journalling writepage() output.
1409 *
1410 * We don't honour synchronous mounts for writepage(). That would be
1411 * disastrous. Any write() or metadata operation will sync the fs for
1412 * us.
1413 *
1414 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1415 * we don't need to open a transaction here.
1416 */
1417static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001418 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
1420 struct inode *inode = page->mapping->host;
1421 struct buffer_head *page_bufs;
1422 handle_t *handle = NULL;
1423 int ret = 0;
1424 int err;
1425
1426 J_ASSERT(PageLocked(page));
1427
1428 /*
1429 * We give up here if we're reentered, because it might be for a
1430 * different filesystem.
1431 */
1432 if (ext3_journal_current_handle())
1433 goto out_fail;
1434
1435 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1436
1437 if (IS_ERR(handle)) {
1438 ret = PTR_ERR(handle);
1439 goto out_fail;
1440 }
1441
1442 if (!page_has_buffers(page)) {
1443 create_empty_buffers(page, inode->i_sb->s_blocksize,
1444 (1 << BH_Dirty)|(1 << BH_Uptodate));
1445 }
1446 page_bufs = page_buffers(page);
1447 walk_page_buffers(handle, page_bufs, 0,
1448 PAGE_CACHE_SIZE, NULL, bget_one);
1449
1450 ret = block_write_full_page(page, ext3_get_block, wbc);
1451
1452 /*
1453 * The page can become unlocked at any point now, and
1454 * truncate can then come in and change things. So we
1455 * can't touch *page from now on. But *page_bufs is
1456 * safe due to elevated refcount.
1457 */
1458
1459 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001460 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 * block_write_full_page() succeeded. Otherwise they are unmapped,
1462 * and generally junk.
1463 */
1464 if (ret == 0) {
1465 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1466 NULL, journal_dirty_data_fn);
1467 if (!ret)
1468 ret = err;
1469 }
1470 walk_page_buffers(handle, page_bufs, 0,
1471 PAGE_CACHE_SIZE, NULL, bput_one);
1472 err = ext3_journal_stop(handle);
1473 if (!ret)
1474 ret = err;
1475 return ret;
1476
1477out_fail:
1478 redirty_page_for_writepage(wbc, page);
1479 unlock_page(page);
1480 return ret;
1481}
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483static int ext3_writeback_writepage(struct page *page,
1484 struct writeback_control *wbc)
1485{
1486 struct inode *inode = page->mapping->host;
1487 handle_t *handle = NULL;
1488 int ret = 0;
1489 int err;
1490
1491 if (ext3_journal_current_handle())
1492 goto out_fail;
1493
1494 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1495 if (IS_ERR(handle)) {
1496 ret = PTR_ERR(handle);
1497 goto out_fail;
1498 }
1499
Badari Pulavarty0e31f512006-07-30 03:04:14 -07001500 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 ret = nobh_writepage(page, ext3_get_block, wbc);
1502 else
1503 ret = block_write_full_page(page, ext3_get_block, wbc);
1504
1505 err = ext3_journal_stop(handle);
1506 if (!ret)
1507 ret = err;
1508 return ret;
1509
1510out_fail:
1511 redirty_page_for_writepage(wbc, page);
1512 unlock_page(page);
1513 return ret;
1514}
1515
1516static int ext3_journalled_writepage(struct page *page,
1517 struct writeback_control *wbc)
1518{
1519 struct inode *inode = page->mapping->host;
1520 handle_t *handle = NULL;
1521 int ret = 0;
1522 int err;
1523
1524 if (ext3_journal_current_handle())
1525 goto no_write;
1526
1527 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1528 if (IS_ERR(handle)) {
1529 ret = PTR_ERR(handle);
1530 goto no_write;
1531 }
1532
1533 if (!page_has_buffers(page) || PageChecked(page)) {
1534 /*
1535 * It's mmapped pagecache. Add buffers and journal it. There
1536 * doesn't seem much point in redirtying the page here.
1537 */
1538 ClearPageChecked(page);
1539 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1540 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001541 if (ret != 0) {
1542 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 ret = walk_page_buffers(handle, page_buffers(page), 0,
1546 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1547
1548 err = walk_page_buffers(handle, page_buffers(page), 0,
1549 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1550 if (ret == 0)
1551 ret = err;
1552 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1553 unlock_page(page);
1554 } else {
1555 /*
1556 * It may be a page full of checkpoint-mode buffers. We don't
1557 * really know unless we go poke around in the buffer_heads.
1558 * But block_write_full_page will do the right thing.
1559 */
1560 ret = block_write_full_page(page, ext3_get_block, wbc);
1561 }
1562 err = ext3_journal_stop(handle);
1563 if (!ret)
1564 ret = err;
1565out:
1566 return ret;
1567
1568no_write:
1569 redirty_page_for_writepage(wbc, page);
1570out_unlock:
1571 unlock_page(page);
1572 goto out;
1573}
1574
1575static int ext3_readpage(struct file *file, struct page *page)
1576{
1577 return mpage_readpage(page, ext3_get_block);
1578}
1579
1580static int
1581ext3_readpages(struct file *file, struct address_space *mapping,
1582 struct list_head *pages, unsigned nr_pages)
1583{
1584 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1585}
1586
NeilBrown2ff28e22006-03-26 01:37:18 -08001587static void ext3_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
1589 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1590
1591 /*
1592 * If it's a full truncate we just forget about the pending dirtying
1593 */
1594 if (offset == 0)
1595 ClearPageChecked(page);
1596
NeilBrown2ff28e22006-03-26 01:37:18 -08001597 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598}
1599
Al Viro27496a82005-10-21 03:20:48 -04001600static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1603
1604 WARN_ON(PageChecked(page));
1605 if (!page_has_buffers(page))
1606 return 0;
1607 return journal_try_to_free_buffers(journal, page, wait);
1608}
1609
1610/*
1611 * If the O_DIRECT write will extend the file then add this inode to the
1612 * orphan list. So recovery will truncate it back to the original size
1613 * if the machine crashes during the write.
1614 *
1615 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1616 * crashes then stale disk data _may_ be exposed inside the file.
1617 */
1618static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1619 const struct iovec *iov, loff_t offset,
1620 unsigned long nr_segs)
1621{
1622 struct file *file = iocb->ki_filp;
1623 struct inode *inode = file->f_mapping->host;
1624 struct ext3_inode_info *ei = EXT3_I(inode);
1625 handle_t *handle = NULL;
1626 ssize_t ret;
1627 int orphan = 0;
1628 size_t count = iov_length(iov, nr_segs);
1629
1630 if (rw == WRITE) {
1631 loff_t final_size = offset + count;
1632
1633 handle = ext3_journal_start(inode, DIO_CREDITS);
1634 if (IS_ERR(handle)) {
1635 ret = PTR_ERR(handle);
1636 goto out;
1637 }
1638 if (final_size > inode->i_size) {
1639 ret = ext3_orphan_add(handle, inode);
1640 if (ret)
1641 goto out_stop;
1642 orphan = 1;
1643 ei->i_disksize = inode->i_size;
1644 }
1645 }
1646
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001647 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 offset, nr_segs,
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001649 ext3_get_block, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 /*
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001652 * Reacquire the handle: ext3_get_block() can restart the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 */
1654 handle = journal_current_handle();
1655
1656out_stop:
1657 if (handle) {
1658 int err;
1659
1660 if (orphan && inode->i_nlink)
1661 ext3_orphan_del(handle, inode);
1662 if (orphan && ret > 0) {
1663 loff_t end = offset + ret;
1664 if (end > inode->i_size) {
1665 ei->i_disksize = end;
1666 i_size_write(inode, end);
1667 /*
1668 * We're going to return a positive `ret'
1669 * here due to non-zero-length I/O, so there's
1670 * no way of reporting error returns from
1671 * ext3_mark_inode_dirty() to userspace. So
1672 * ignore it.
1673 */
1674 ext3_mark_inode_dirty(handle, inode);
1675 }
1676 }
1677 err = ext3_journal_stop(handle);
1678 if (ret == 0)
1679 ret = err;
1680 }
1681out:
1682 return ret;
1683}
1684
1685/*
1686 * Pages can be marked dirty completely asynchronously from ext3's journalling
1687 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1688 * much here because ->set_page_dirty is called under VFS locks. The page is
1689 * not necessarily locked.
1690 *
1691 * We cannot just dirty the page and leave attached buffers clean, because the
1692 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1693 * or jbddirty because all the journalling code will explode.
1694 *
1695 * So what we do is to mark the page "pending dirty" and next time writepage
1696 * is called, propagate that into the buffers appropriately.
1697 */
1698static int ext3_journalled_set_page_dirty(struct page *page)
1699{
1700 SetPageChecked(page);
1701 return __set_page_dirty_nobuffers(page);
1702}
1703
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001704static const struct address_space_operations ext3_ordered_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 .readpage = ext3_readpage,
1706 .readpages = ext3_readpages,
1707 .writepage = ext3_ordered_writepage,
1708 .sync_page = block_sync_page,
1709 .prepare_write = ext3_prepare_write,
1710 .commit_write = ext3_ordered_commit_write,
1711 .bmap = ext3_bmap,
1712 .invalidatepage = ext3_invalidatepage,
1713 .releasepage = ext3_releasepage,
1714 .direct_IO = ext3_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001715 .migratepage = buffer_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716};
1717
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001718static const struct address_space_operations ext3_writeback_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 .readpage = ext3_readpage,
1720 .readpages = ext3_readpages,
1721 .writepage = ext3_writeback_writepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 .sync_page = block_sync_page,
1723 .prepare_write = ext3_prepare_write,
1724 .commit_write = ext3_writeback_commit_write,
1725 .bmap = ext3_bmap,
1726 .invalidatepage = ext3_invalidatepage,
1727 .releasepage = ext3_releasepage,
1728 .direct_IO = ext3_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001729 .migratepage = buffer_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730};
1731
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001732static const struct address_space_operations ext3_journalled_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 .readpage = ext3_readpage,
1734 .readpages = ext3_readpages,
1735 .writepage = ext3_journalled_writepage,
1736 .sync_page = block_sync_page,
1737 .prepare_write = ext3_prepare_write,
1738 .commit_write = ext3_journalled_commit_write,
1739 .set_page_dirty = ext3_journalled_set_page_dirty,
1740 .bmap = ext3_bmap,
1741 .invalidatepage = ext3_invalidatepage,
1742 .releasepage = ext3_releasepage,
1743};
1744
1745void ext3_set_aops(struct inode *inode)
1746{
1747 if (ext3_should_order_data(inode))
1748 inode->i_mapping->a_ops = &ext3_ordered_aops;
1749 else if (ext3_should_writeback_data(inode))
1750 inode->i_mapping->a_ops = &ext3_writeback_aops;
1751 else
1752 inode->i_mapping->a_ops = &ext3_journalled_aops;
1753}
1754
1755/*
1756 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1757 * up to the end of the block which corresponds to `from'.
1758 * This required during truncate. We need to physically zero the tail end
1759 * of that block so it doesn't yield old data if the file is later grown.
1760 */
1761static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1762 struct address_space *mapping, loff_t from)
1763{
Mingming Cao43d23f92006-06-25 05:48:07 -07001764 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1766 unsigned blocksize, iblock, length, pos;
1767 struct inode *inode = mapping->host;
1768 struct buffer_head *bh;
1769 int err = 0;
1770 void *kaddr;
1771
1772 blocksize = inode->i_sb->s_blocksize;
1773 length = blocksize - (offset & (blocksize - 1));
1774 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1775
1776 /*
1777 * For "nobh" option, we can only work if we don't need to
1778 * read-in the page - otherwise we create buffers to do the IO.
1779 */
Badari Pulavartycd6ef842006-03-11 03:27:14 -08001780 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1781 ext3_should_writeback_data(inode) && PageUptodate(page)) {
1782 kaddr = kmap_atomic(page, KM_USER0);
1783 memset(kaddr + offset, 0, length);
1784 flush_dcache_page(page);
1785 kunmap_atomic(kaddr, KM_USER0);
1786 set_page_dirty(page);
1787 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 }
1789
1790 if (!page_has_buffers(page))
1791 create_empty_buffers(page, blocksize, 0);
1792
1793 /* Find the buffer that contains "offset" */
1794 bh = page_buffers(page);
1795 pos = blocksize;
1796 while (offset >= pos) {
1797 bh = bh->b_this_page;
1798 iblock++;
1799 pos += blocksize;
1800 }
1801
1802 err = 0;
1803 if (buffer_freed(bh)) {
1804 BUFFER_TRACE(bh, "freed: skip");
1805 goto unlock;
1806 }
1807
1808 if (!buffer_mapped(bh)) {
1809 BUFFER_TRACE(bh, "unmapped");
1810 ext3_get_block(inode, iblock, bh, 0);
1811 /* unmapped? It's a hole - nothing to do */
1812 if (!buffer_mapped(bh)) {
1813 BUFFER_TRACE(bh, "still unmapped");
1814 goto unlock;
1815 }
1816 }
1817
1818 /* Ok, it's mapped. Make sure it's up-to-date */
1819 if (PageUptodate(page))
1820 set_buffer_uptodate(bh);
1821
1822 if (!buffer_uptodate(bh)) {
1823 err = -EIO;
1824 ll_rw_block(READ, 1, &bh);
1825 wait_on_buffer(bh);
1826 /* Uhhuh. Read error. Complain and punt. */
1827 if (!buffer_uptodate(bh))
1828 goto unlock;
1829 }
1830
1831 if (ext3_should_journal_data(inode)) {
1832 BUFFER_TRACE(bh, "get write access");
1833 err = ext3_journal_get_write_access(handle, bh);
1834 if (err)
1835 goto unlock;
1836 }
1837
1838 kaddr = kmap_atomic(page, KM_USER0);
1839 memset(kaddr + offset, 0, length);
1840 flush_dcache_page(page);
1841 kunmap_atomic(kaddr, KM_USER0);
1842
1843 BUFFER_TRACE(bh, "zeroed end of block");
1844
1845 err = 0;
1846 if (ext3_should_journal_data(inode)) {
1847 err = ext3_journal_dirty_metadata(handle, bh);
1848 } else {
1849 if (ext3_should_order_data(inode))
1850 err = ext3_journal_dirty_data(handle, bh);
1851 mark_buffer_dirty(bh);
1852 }
1853
1854unlock:
1855 unlock_page(page);
1856 page_cache_release(page);
1857 return err;
1858}
1859
1860/*
1861 * Probably it should be a library function... search for first non-zero word
1862 * or memcmp with zero_page, whatever is better for particular architecture.
1863 * Linus?
1864 */
1865static inline int all_zeroes(__le32 *p, __le32 *q)
1866{
1867 while (p < q)
1868 if (*p++)
1869 return 0;
1870 return 1;
1871}
1872
1873/**
1874 * ext3_find_shared - find the indirect blocks for partial truncation.
1875 * @inode: inode in question
1876 * @depth: depth of the affected branch
1877 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1878 * @chain: place to store the pointers to partial indirect blocks
1879 * @top: place to the (detached) top of branch
1880 *
1881 * This is a helper function used by ext3_truncate().
1882 *
1883 * When we do truncate() we may have to clean the ends of several
1884 * indirect blocks but leave the blocks themselves alive. Block is
1885 * partially truncated if some data below the new i_size is refered
1886 * from it (and it is on the path to the first completely truncated
1887 * data block, indeed). We have to free the top of that path along
1888 * with everything to the right of the path. Since no allocation
1889 * past the truncation point is possible until ext3_truncate()
1890 * finishes, we may safely do the latter, but top of branch may
1891 * require special attention - pageout below the truncation point
1892 * might try to populate it.
1893 *
1894 * We atomically detach the top of branch from the tree, store the
1895 * block number of its root in *@top, pointers to buffer_heads of
1896 * partially truncated blocks - in @chain[].bh and pointers to
1897 * their last elements that should not be removed - in
1898 * @chain[].p. Return value is the pointer to last filled element
1899 * of @chain.
1900 *
1901 * The work left to caller to do the actual freeing of subtrees:
1902 * a) free the subtree starting from *@top
1903 * b) free the subtrees whose roots are stored in
1904 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1905 * c) free the subtrees growing from the inode past the @chain[0].
1906 * (no partially truncated stuff there). */
1907
Andrew Mortond6859bf2006-03-26 01:38:03 -08001908static Indirect *ext3_find_shared(struct inode *inode, int depth,
1909 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910{
1911 Indirect *partial, *p;
1912 int k, err;
1913
1914 *top = 0;
1915 /* Make k index the deepest non-null offest + 1 */
1916 for (k = depth; k > 1 && !offsets[k-1]; k--)
1917 ;
1918 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1919 /* Writer: pointers */
1920 if (!partial)
1921 partial = chain + k-1;
1922 /*
1923 * If the branch acquired continuation since we've looked at it -
1924 * fine, it should all survive and (new) top doesn't belong to us.
1925 */
1926 if (!partial->key && *partial->p)
1927 /* Writer: end */
1928 goto no_top;
1929 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1930 ;
1931 /*
1932 * OK, we've found the last block that must survive. The rest of our
1933 * branch should be detached before unlocking. However, if that rest
1934 * of branch is all ours and does not grow immediately from the inode
1935 * it's easier to cheat and just decrement partial->p.
1936 */
1937 if (p == chain + k - 1 && p > chain) {
1938 p->p--;
1939 } else {
1940 *top = *p->p;
1941 /* Nope, don't do this in ext3. Must leave the tree intact */
1942#if 0
1943 *p->p = 0;
1944#endif
1945 }
1946 /* Writer: end */
1947
Andrew Mortond6859bf2006-03-26 01:38:03 -08001948 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 brelse(partial->bh);
1950 partial--;
1951 }
1952no_top:
1953 return partial;
1954}
1955
1956/*
1957 * Zero a number of block pointers in either an inode or an indirect block.
1958 * If we restart the transaction we must again get write access to the
1959 * indirect block for further modification.
1960 *
1961 * We release `count' blocks on disk, but (last - first) may be greater
1962 * than `count' because there can be holes in there.
1963 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001964static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07001965 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001966 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
1968 __le32 *p;
1969 if (try_to_extend_transaction(handle, inode)) {
1970 if (bh) {
1971 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1972 ext3_journal_dirty_metadata(handle, bh);
1973 }
1974 ext3_mark_inode_dirty(handle, inode);
1975 ext3_journal_test_restart(handle, inode);
1976 if (bh) {
1977 BUFFER_TRACE(bh, "retaking write access");
1978 ext3_journal_get_write_access(handle, bh);
1979 }
1980 }
1981
1982 /*
1983 * Any buffers which are on the journal will be in memory. We find
1984 * them on the hash table so journal_revoke() will run journal_forget()
1985 * on them. We've already detached each block from the file, so
1986 * bforget() in journal_forget() should be safe.
1987 *
1988 * AKPM: turn on bforget in journal_forget()!!!
1989 */
1990 for (p = first; p < last; p++) {
1991 u32 nr = le32_to_cpu(*p);
1992 if (nr) {
1993 struct buffer_head *bh;
1994
1995 *p = 0;
1996 bh = sb_find_get_block(inode->i_sb, nr);
1997 ext3_forget(handle, 0, inode, bh, nr);
1998 }
1999 }
2000
2001 ext3_free_blocks(handle, inode, block_to_free, count);
2002}
2003
2004/**
2005 * ext3_free_data - free a list of data blocks
2006 * @handle: handle for this transaction
2007 * @inode: inode we are dealing with
2008 * @this_bh: indirect buffer_head which contains *@first and *@last
2009 * @first: array of block numbers
2010 * @last: points immediately past the end of array
2011 *
2012 * We are freeing all blocks refered from that array (numbers are stored as
2013 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2014 *
2015 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2016 * blocks are contiguous then releasing them at one time will only affect one
2017 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2018 * actually use a lot of journal space.
2019 *
2020 * @this_bh will be %NULL if @first and @last point into the inode's direct
2021 * block pointers.
2022 */
2023static void ext3_free_data(handle_t *handle, struct inode *inode,
2024 struct buffer_head *this_bh,
2025 __le32 *first, __le32 *last)
2026{
Mingming Cao43d23f92006-06-25 05:48:07 -07002027 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002028 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2030 corresponding to
2031 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002032 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 __le32 *p; /* Pointer into inode/ind
2034 for current block */
2035 int err;
2036
2037 if (this_bh) { /* For indirect block */
2038 BUFFER_TRACE(this_bh, "get_write_access");
2039 err = ext3_journal_get_write_access(handle, this_bh);
2040 /* Important: if we can't update the indirect pointers
2041 * to the blocks, we can't free them. */
2042 if (err)
2043 return;
2044 }
2045
2046 for (p = first; p < last; p++) {
2047 nr = le32_to_cpu(*p);
2048 if (nr) {
2049 /* accumulate blocks to free if they're contiguous */
2050 if (count == 0) {
2051 block_to_free = nr;
2052 block_to_free_p = p;
2053 count = 1;
2054 } else if (nr == block_to_free + count) {
2055 count++;
2056 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002057 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 block_to_free,
2059 count, block_to_free_p, p);
2060 block_to_free = nr;
2061 block_to_free_p = p;
2062 count = 1;
2063 }
2064 }
2065 }
2066
2067 if (count > 0)
2068 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2069 count, block_to_free_p, p);
2070
2071 if (this_bh) {
2072 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2073 ext3_journal_dirty_metadata(handle, this_bh);
2074 }
2075}
2076
2077/**
2078 * ext3_free_branches - free an array of branches
2079 * @handle: JBD handle for this transaction
2080 * @inode: inode we are dealing with
2081 * @parent_bh: the buffer_head which contains *@first and *@last
2082 * @first: array of block numbers
2083 * @last: pointer immediately past the end of array
2084 * @depth: depth of the branches to free
2085 *
2086 * We are freeing all blocks refered from these branches (numbers are
2087 * stored as little-endian 32-bit) and updating @inode->i_blocks
2088 * appropriately.
2089 */
2090static void ext3_free_branches(handle_t *handle, struct inode *inode,
2091 struct buffer_head *parent_bh,
2092 __le32 *first, __le32 *last, int depth)
2093{
Mingming Cao43d23f92006-06-25 05:48:07 -07002094 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 __le32 *p;
2096
2097 if (is_handle_aborted(handle))
2098 return;
2099
2100 if (depth--) {
2101 struct buffer_head *bh;
2102 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2103 p = last;
2104 while (--p >= first) {
2105 nr = le32_to_cpu(*p);
2106 if (!nr)
2107 continue; /* A hole */
2108
2109 /* Go read the buffer for the next level down */
2110 bh = sb_bread(inode->i_sb, nr);
2111
2112 /*
2113 * A read failure? Report error and clear slot
2114 * (should be rare).
2115 */
2116 if (!bh) {
2117 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002118 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 inode->i_ino, nr);
2120 continue;
2121 }
2122
2123 /* This zaps the entire block. Bottom up. */
2124 BUFFER_TRACE(bh, "free child branches");
2125 ext3_free_branches(handle, inode, bh,
2126 (__le32*)bh->b_data,
2127 (__le32*)bh->b_data + addr_per_block,
2128 depth);
2129
2130 /*
2131 * We've probably journalled the indirect block several
2132 * times during the truncate. But it's no longer
2133 * needed and we now drop it from the transaction via
2134 * journal_revoke().
2135 *
2136 * That's easy if it's exclusively part of this
2137 * transaction. But if it's part of the committing
2138 * transaction then journal_forget() will simply
2139 * brelse() it. That means that if the underlying
2140 * block is reallocated in ext3_get_block(),
2141 * unmap_underlying_metadata() will find this block
2142 * and will try to get rid of it. damn, damn.
2143 *
2144 * If this block has already been committed to the
2145 * journal, a revoke record will be written. And
2146 * revoke records must be emitted *before* clearing
2147 * this block's bit in the bitmaps.
2148 */
2149 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2150
2151 /*
2152 * Everything below this this pointer has been
2153 * released. Now let this top-of-subtree go.
2154 *
2155 * We want the freeing of this indirect block to be
2156 * atomic in the journal with the updating of the
2157 * bitmap block which owns it. So make some room in
2158 * the journal.
2159 *
2160 * We zero the parent pointer *after* freeing its
2161 * pointee in the bitmaps, so if extend_transaction()
2162 * for some reason fails to put the bitmap changes and
2163 * the release into the same transaction, recovery
2164 * will merely complain about releasing a free block,
2165 * rather than leaking blocks.
2166 */
2167 if (is_handle_aborted(handle))
2168 return;
2169 if (try_to_extend_transaction(handle, inode)) {
2170 ext3_mark_inode_dirty(handle, inode);
2171 ext3_journal_test_restart(handle, inode);
2172 }
2173
2174 ext3_free_blocks(handle, inode, nr, 1);
2175
2176 if (parent_bh) {
2177 /*
2178 * The block which we have just freed is
2179 * pointed to by an indirect block: journal it
2180 */
2181 BUFFER_TRACE(parent_bh, "get_write_access");
2182 if (!ext3_journal_get_write_access(handle,
2183 parent_bh)){
2184 *p = 0;
2185 BUFFER_TRACE(parent_bh,
2186 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002187 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 parent_bh);
2189 }
2190 }
2191 }
2192 } else {
2193 /* We have reached the bottom of the tree. */
2194 BUFFER_TRACE(parent_bh, "free data blocks");
2195 ext3_free_data(handle, inode, parent_bh, first, last);
2196 }
2197}
2198
2199/*
2200 * ext3_truncate()
2201 *
2202 * We block out ext3_get_block() block instantiations across the entire
2203 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2204 * simultaneously on behalf of the same inode.
2205 *
2206 * As we work through the truncate and commmit bits of it to the journal there
2207 * is one core, guiding principle: the file's tree must always be consistent on
2208 * disk. We must be able to restart the truncate after a crash.
2209 *
2210 * The file's tree may be transiently inconsistent in memory (although it
2211 * probably isn't), but whenever we close off and commit a journal transaction,
2212 * the contents of (the filesystem + the journal) must be consistent and
2213 * restartable. It's pretty simple, really: bottom up, right to left (although
2214 * left-to-right works OK too).
2215 *
2216 * Note that at recovery time, journal replay occurs *before* the restart of
2217 * truncate against the orphan inode list.
2218 *
2219 * The committed inode has the new, desired i_size (which is the same as
2220 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2221 * that this inode's truncate did not complete and it will again call
2222 * ext3_truncate() to have another go. So there will be instantiated blocks
2223 * to the right of the truncation point in a crashed ext3 filesystem. But
2224 * that's fine - as long as they are linked from the inode, the post-crash
2225 * ext3_truncate() run will find them and release them.
2226 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002227void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228{
2229 handle_t *handle;
2230 struct ext3_inode_info *ei = EXT3_I(inode);
2231 __le32 *i_data = ei->i_data;
2232 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2233 struct address_space *mapping = inode->i_mapping;
2234 int offsets[4];
2235 Indirect chain[4];
2236 Indirect *partial;
2237 __le32 nr = 0;
2238 int n;
2239 long last_block;
2240 unsigned blocksize = inode->i_sb->s_blocksize;
2241 struct page *page;
2242
2243 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2244 S_ISLNK(inode->i_mode)))
2245 return;
2246 if (ext3_inode_is_fast_symlink(inode))
2247 return;
2248 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2249 return;
2250
2251 /*
2252 * We have to lock the EOF page here, because lock_page() nests
2253 * outside journal_start().
2254 */
2255 if ((inode->i_size & (blocksize - 1)) == 0) {
2256 /* Block boundary? Nothing to do */
2257 page = NULL;
2258 } else {
2259 page = grab_cache_page(mapping,
2260 inode->i_size >> PAGE_CACHE_SHIFT);
2261 if (!page)
2262 return;
2263 }
2264
2265 handle = start_transaction(inode);
2266 if (IS_ERR(handle)) {
2267 if (page) {
2268 clear_highpage(page);
2269 flush_dcache_page(page);
2270 unlock_page(page);
2271 page_cache_release(page);
2272 }
2273 return; /* AKPM: return what? */
2274 }
2275
2276 last_block = (inode->i_size + blocksize-1)
2277 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2278
2279 if (page)
2280 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2281
2282 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2283 if (n == 0)
2284 goto out_stop; /* error */
2285
2286 /*
2287 * OK. This truncate is going to happen. We add the inode to the
2288 * orphan list, so that if this truncate spans multiple transactions,
2289 * and we crash, we will resume the truncate when the filesystem
2290 * recovers. It also marks the inode dirty, to catch the new size.
2291 *
2292 * Implication: the file must always be in a sane, consistent
2293 * truncatable state while each transaction commits.
2294 */
2295 if (ext3_orphan_add(handle, inode))
2296 goto out_stop;
2297
2298 /*
2299 * The orphan list entry will now protect us from any crash which
2300 * occurs before the truncate completes, so it is now safe to propagate
2301 * the new, shorter inode size (held for now in i_size) into the
2302 * on-disk inode. We do this via i_disksize, which is the value which
2303 * ext3 *really* writes onto the disk inode.
2304 */
2305 ei->i_disksize = inode->i_size;
2306
2307 /*
2308 * From here we block out all ext3_get_block() callers who want to
2309 * modify the block allocation tree.
2310 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002311 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
2313 if (n == 1) { /* direct blocks */
2314 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2315 i_data + EXT3_NDIR_BLOCKS);
2316 goto do_indirects;
2317 }
2318
2319 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2320 /* Kill the top of shared branch (not detached) */
2321 if (nr) {
2322 if (partial == chain) {
2323 /* Shared branch grows from the inode */
2324 ext3_free_branches(handle, inode, NULL,
2325 &nr, &nr+1, (chain+n-1) - partial);
2326 *partial->p = 0;
2327 /*
2328 * We mark the inode dirty prior to restart,
2329 * and prior to stop. No need for it here.
2330 */
2331 } else {
2332 /* Shared branch grows from an indirect block */
2333 BUFFER_TRACE(partial->bh, "get_write_access");
2334 ext3_free_branches(handle, inode, partial->bh,
2335 partial->p,
2336 partial->p+1, (chain+n-1) - partial);
2337 }
2338 }
2339 /* Clear the ends of indirect blocks on the shared branch */
2340 while (partial > chain) {
2341 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2342 (__le32*)partial->bh->b_data+addr_per_block,
2343 (chain+n-1) - partial);
2344 BUFFER_TRACE(partial->bh, "call brelse");
2345 brelse (partial->bh);
2346 partial--;
2347 }
2348do_indirects:
2349 /* Kill the remaining (whole) subtrees */
2350 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002351 default:
2352 nr = i_data[EXT3_IND_BLOCK];
2353 if (nr) {
2354 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2355 i_data[EXT3_IND_BLOCK] = 0;
2356 }
2357 case EXT3_IND_BLOCK:
2358 nr = i_data[EXT3_DIND_BLOCK];
2359 if (nr) {
2360 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2361 i_data[EXT3_DIND_BLOCK] = 0;
2362 }
2363 case EXT3_DIND_BLOCK:
2364 nr = i_data[EXT3_TIND_BLOCK];
2365 if (nr) {
2366 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2367 i_data[EXT3_TIND_BLOCK] = 0;
2368 }
2369 case EXT3_TIND_BLOCK:
2370 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 }
2372
2373 ext3_discard_reservation(inode);
2374
Arjan van de Ven97461512006-03-23 03:00:42 -08002375 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2377 ext3_mark_inode_dirty(handle, inode);
2378
Andrew Mortond6859bf2006-03-26 01:38:03 -08002379 /*
2380 * In a multi-transaction truncate, we only make the final transaction
2381 * synchronous
2382 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 if (IS_SYNC(inode))
2384 handle->h_sync = 1;
2385out_stop:
2386 /*
2387 * If this was a simple ftruncate(), and the file will remain alive
2388 * then we need to clear up the orphan record which we created above.
2389 * However, if this was a real unlink then we were called by
2390 * ext3_delete_inode(), and we allow that function to clean up the
2391 * orphan info for us.
2392 */
2393 if (inode->i_nlink)
2394 ext3_orphan_del(handle, inode);
2395
2396 ext3_journal_stop(handle);
2397}
2398
Mingming Cao43d23f92006-06-25 05:48:07 -07002399static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 unsigned long ino, struct ext3_iloc *iloc)
2401{
2402 unsigned long desc, group_desc, block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002403 unsigned long offset;
2404 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 struct buffer_head *bh;
2406 struct ext3_group_desc * gdp;
2407
Neil Brown2ccb48e2006-07-30 03:03:01 -07002408 if (!ext3_valid_inum(sb, ino)) {
2409 /*
2410 * This error is already checked for in namei.c unless we are
2411 * looking at an NFS filehandle, in which case no error
2412 * report is needed
2413 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 return 0;
2415 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002416
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2418 if (block_group >= EXT3_SB(sb)->s_groups_count) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002419 ext3_error(sb,"ext3_get_inode_block","group >= groups count");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 return 0;
2421 }
2422 smp_rmb();
2423 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2424 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2425 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2426 if (!bh) {
2427 ext3_error (sb, "ext3_get_inode_block",
2428 "Descriptor not loaded");
2429 return 0;
2430 }
2431
Andrew Mortond6859bf2006-03-26 01:38:03 -08002432 gdp = (struct ext3_group_desc *)bh->b_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 /*
2434 * Figure out the offset within the block group inode table
2435 */
2436 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2437 EXT3_INODE_SIZE(sb);
2438 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2439 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2440
2441 iloc->block_group = block_group;
2442 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2443 return block;
2444}
2445
2446/*
2447 * ext3_get_inode_loc returns with an extra refcount against the inode's
2448 * underlying buffer_head on success. If 'in_mem' is true, we have all
2449 * data in memory that is needed to recreate the on-disk version of this
2450 * inode.
2451 */
2452static int __ext3_get_inode_loc(struct inode *inode,
2453 struct ext3_iloc *iloc, int in_mem)
2454{
Mingming Cao43d23f92006-06-25 05:48:07 -07002455 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 struct buffer_head *bh;
2457
2458 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2459 if (!block)
2460 return -EIO;
2461
2462 bh = sb_getblk(inode->i_sb, block);
2463 if (!bh) {
2464 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2465 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002466 "inode=%lu, block="E3FSBLK,
2467 inode->i_ino, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 return -EIO;
2469 }
2470 if (!buffer_uptodate(bh)) {
2471 lock_buffer(bh);
2472 if (buffer_uptodate(bh)) {
2473 /* someone brought it uptodate while we waited */
2474 unlock_buffer(bh);
2475 goto has_buffer;
2476 }
2477
2478 /*
2479 * If we have all information of the inode in memory and this
2480 * is the only valid inode in the block, we need not read the
2481 * block.
2482 */
2483 if (in_mem) {
2484 struct buffer_head *bitmap_bh;
2485 struct ext3_group_desc *desc;
2486 int inodes_per_buffer;
2487 int inode_offset, i;
2488 int block_group;
2489 int start;
2490
2491 block_group = (inode->i_ino - 1) /
2492 EXT3_INODES_PER_GROUP(inode->i_sb);
2493 inodes_per_buffer = bh->b_size /
2494 EXT3_INODE_SIZE(inode->i_sb);
2495 inode_offset = ((inode->i_ino - 1) %
2496 EXT3_INODES_PER_GROUP(inode->i_sb));
2497 start = inode_offset & ~(inodes_per_buffer - 1);
2498
2499 /* Is the inode bitmap in cache? */
2500 desc = ext3_get_group_desc(inode->i_sb,
2501 block_group, NULL);
2502 if (!desc)
2503 goto make_io;
2504
2505 bitmap_bh = sb_getblk(inode->i_sb,
2506 le32_to_cpu(desc->bg_inode_bitmap));
2507 if (!bitmap_bh)
2508 goto make_io;
2509
2510 /*
2511 * If the inode bitmap isn't in cache then the
2512 * optimisation may end up performing two reads instead
2513 * of one, so skip it.
2514 */
2515 if (!buffer_uptodate(bitmap_bh)) {
2516 brelse(bitmap_bh);
2517 goto make_io;
2518 }
2519 for (i = start; i < start + inodes_per_buffer; i++) {
2520 if (i == inode_offset)
2521 continue;
2522 if (ext3_test_bit(i, bitmap_bh->b_data))
2523 break;
2524 }
2525 brelse(bitmap_bh);
2526 if (i == start + inodes_per_buffer) {
2527 /* all other inodes are free, so skip I/O */
2528 memset(bh->b_data, 0, bh->b_size);
2529 set_buffer_uptodate(bh);
2530 unlock_buffer(bh);
2531 goto has_buffer;
2532 }
2533 }
2534
2535make_io:
2536 /*
2537 * There are other valid inodes in the buffer, this inode
2538 * has in-inode xattrs, or we don't have this inode in memory.
2539 * Read the block from disk.
2540 */
2541 get_bh(bh);
2542 bh->b_end_io = end_buffer_read_sync;
2543 submit_bh(READ, bh);
2544 wait_on_buffer(bh);
2545 if (!buffer_uptodate(bh)) {
2546 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2547 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002548 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 inode->i_ino, block);
2550 brelse(bh);
2551 return -EIO;
2552 }
2553 }
2554has_buffer:
2555 iloc->bh = bh;
2556 return 0;
2557}
2558
2559int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2560{
2561 /* We have all inode data except xattrs in memory here. */
2562 return __ext3_get_inode_loc(inode, iloc,
2563 !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR));
2564}
2565
2566void ext3_set_inode_flags(struct inode *inode)
2567{
2568 unsigned int flags = EXT3_I(inode)->i_flags;
2569
2570 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2571 if (flags & EXT3_SYNC_FL)
2572 inode->i_flags |= S_SYNC;
2573 if (flags & EXT3_APPEND_FL)
2574 inode->i_flags |= S_APPEND;
2575 if (flags & EXT3_IMMUTABLE_FL)
2576 inode->i_flags |= S_IMMUTABLE;
2577 if (flags & EXT3_NOATIME_FL)
2578 inode->i_flags |= S_NOATIME;
2579 if (flags & EXT3_DIRSYNC_FL)
2580 inode->i_flags |= S_DIRSYNC;
2581}
2582
2583void ext3_read_inode(struct inode * inode)
2584{
2585 struct ext3_iloc iloc;
2586 struct ext3_inode *raw_inode;
2587 struct ext3_inode_info *ei = EXT3_I(inode);
2588 struct buffer_head *bh;
2589 int block;
2590
2591#ifdef CONFIG_EXT3_FS_POSIX_ACL
2592 ei->i_acl = EXT3_ACL_NOT_CACHED;
2593 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2594#endif
2595 ei->i_block_alloc_info = NULL;
2596
2597 if (__ext3_get_inode_loc(inode, &iloc, 0))
2598 goto bad_inode;
2599 bh = iloc.bh;
2600 raw_inode = ext3_raw_inode(&iloc);
2601 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2602 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2603 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2604 if(!(test_opt (inode->i_sb, NO_UID32))) {
2605 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2606 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2607 }
2608 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2609 inode->i_size = le32_to_cpu(raw_inode->i_size);
2610 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2611 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2612 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2613 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2614
2615 ei->i_state = 0;
2616 ei->i_dir_start_lookup = 0;
2617 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2618 /* We now have enough fields to check if the inode was active or not.
2619 * This is needed because nfsd might try to access dead inodes
2620 * the test is that same one that e2fsck uses
2621 * NeilBrown 1999oct15
2622 */
2623 if (inode->i_nlink == 0) {
2624 if (inode->i_mode == 0 ||
2625 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2626 /* this inode is deleted */
2627 brelse (bh);
2628 goto bad_inode;
2629 }
2630 /* The only unlinked inodes we let through here have
2631 * valid i_mode and are being read by the orphan
2632 * recovery code: that's fine, we're about to complete
2633 * the process of deleting those. */
2634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2636 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2637#ifdef EXT3_FRAGMENTS
2638 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2639 ei->i_frag_no = raw_inode->i_frag;
2640 ei->i_frag_size = raw_inode->i_fsize;
2641#endif
2642 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2643 if (!S_ISREG(inode->i_mode)) {
2644 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2645 } else {
2646 inode->i_size |=
2647 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2648 }
2649 ei->i_disksize = inode->i_size;
2650 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2651 ei->i_block_group = iloc.block_group;
2652 /*
2653 * NOTE! The in-memory inode i_data array is in little-endian order
2654 * even on big-endian machines: we do NOT byteswap the block numbers!
2655 */
2656 for (block = 0; block < EXT3_N_BLOCKS; block++)
2657 ei->i_data[block] = raw_inode->i_block[block];
2658 INIT_LIST_HEAD(&ei->i_orphan);
2659
2660 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2661 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2662 /*
2663 * When mke2fs creates big inodes it does not zero out
2664 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2665 * so ignore those first few inodes.
2666 */
2667 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2668 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2669 EXT3_INODE_SIZE(inode->i_sb))
2670 goto bad_inode;
2671 if (ei->i_extra_isize == 0) {
2672 /* The extra space is currently unused. Use it. */
2673 ei->i_extra_isize = sizeof(struct ext3_inode) -
2674 EXT3_GOOD_OLD_INODE_SIZE;
2675 } else {
2676 __le32 *magic = (void *)raw_inode +
2677 EXT3_GOOD_OLD_INODE_SIZE +
2678 ei->i_extra_isize;
2679 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2680 ei->i_state |= EXT3_STATE_XATTR;
2681 }
2682 } else
2683 ei->i_extra_isize = 0;
2684
2685 if (S_ISREG(inode->i_mode)) {
2686 inode->i_op = &ext3_file_inode_operations;
2687 inode->i_fop = &ext3_file_operations;
2688 ext3_set_aops(inode);
2689 } else if (S_ISDIR(inode->i_mode)) {
2690 inode->i_op = &ext3_dir_inode_operations;
2691 inode->i_fop = &ext3_dir_operations;
2692 } else if (S_ISLNK(inode->i_mode)) {
2693 if (ext3_inode_is_fast_symlink(inode))
2694 inode->i_op = &ext3_fast_symlink_inode_operations;
2695 else {
2696 inode->i_op = &ext3_symlink_inode_operations;
2697 ext3_set_aops(inode);
2698 }
2699 } else {
2700 inode->i_op = &ext3_special_inode_operations;
2701 if (raw_inode->i_block[0])
2702 init_special_inode(inode, inode->i_mode,
2703 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002704 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 init_special_inode(inode, inode->i_mode,
2706 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2707 }
2708 brelse (iloc.bh);
2709 ext3_set_inode_flags(inode);
2710 return;
2711
2712bad_inode:
2713 make_bad_inode(inode);
2714 return;
2715}
2716
2717/*
2718 * Post the struct inode info into an on-disk inode location in the
2719 * buffer-cache. This gobbles the caller's reference to the
2720 * buffer_head in the inode location struct.
2721 *
2722 * The caller must have write access to iloc->bh.
2723 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002724static int ext3_do_update_inode(handle_t *handle,
2725 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 struct ext3_iloc *iloc)
2727{
2728 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2729 struct ext3_inode_info *ei = EXT3_I(inode);
2730 struct buffer_head *bh = iloc->bh;
2731 int err = 0, rc, block;
2732
2733 /* For fields not not tracking in the in-memory inode,
2734 * initialise them to zero for new inodes. */
2735 if (ei->i_state & EXT3_STATE_NEW)
2736 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2737
2738 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2739 if(!(test_opt(inode->i_sb, NO_UID32))) {
2740 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2741 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2742/*
2743 * Fix up interoperability with old kernels. Otherwise, old inodes get
2744 * re-used with the upper 16 bits of the uid/gid intact
2745 */
2746 if(!ei->i_dtime) {
2747 raw_inode->i_uid_high =
2748 cpu_to_le16(high_16_bits(inode->i_uid));
2749 raw_inode->i_gid_high =
2750 cpu_to_le16(high_16_bits(inode->i_gid));
2751 } else {
2752 raw_inode->i_uid_high = 0;
2753 raw_inode->i_gid_high = 0;
2754 }
2755 } else {
2756 raw_inode->i_uid_low =
2757 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2758 raw_inode->i_gid_low =
2759 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2760 raw_inode->i_uid_high = 0;
2761 raw_inode->i_gid_high = 0;
2762 }
2763 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2764 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2765 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2766 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2767 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2768 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2769 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2770 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2771#ifdef EXT3_FRAGMENTS
2772 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2773 raw_inode->i_frag = ei->i_frag_no;
2774 raw_inode->i_fsize = ei->i_frag_size;
2775#endif
2776 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2777 if (!S_ISREG(inode->i_mode)) {
2778 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2779 } else {
2780 raw_inode->i_size_high =
2781 cpu_to_le32(ei->i_disksize >> 32);
2782 if (ei->i_disksize > 0x7fffffffULL) {
2783 struct super_block *sb = inode->i_sb;
2784 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2785 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2786 EXT3_SB(sb)->s_es->s_rev_level ==
2787 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2788 /* If this is the first large file
2789 * created, add a flag to the superblock.
2790 */
2791 err = ext3_journal_get_write_access(handle,
2792 EXT3_SB(sb)->s_sbh);
2793 if (err)
2794 goto out_brelse;
2795 ext3_update_dynamic_rev(sb);
2796 EXT3_SET_RO_COMPAT_FEATURE(sb,
2797 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2798 sb->s_dirt = 1;
2799 handle->h_sync = 1;
2800 err = ext3_journal_dirty_metadata(handle,
2801 EXT3_SB(sb)->s_sbh);
2802 }
2803 }
2804 }
2805 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2806 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2807 if (old_valid_dev(inode->i_rdev)) {
2808 raw_inode->i_block[0] =
2809 cpu_to_le32(old_encode_dev(inode->i_rdev));
2810 raw_inode->i_block[1] = 0;
2811 } else {
2812 raw_inode->i_block[0] = 0;
2813 raw_inode->i_block[1] =
2814 cpu_to_le32(new_encode_dev(inode->i_rdev));
2815 raw_inode->i_block[2] = 0;
2816 }
2817 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2818 raw_inode->i_block[block] = ei->i_data[block];
2819
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07002820 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2822
2823 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2824 rc = ext3_journal_dirty_metadata(handle, bh);
2825 if (!err)
2826 err = rc;
2827 ei->i_state &= ~EXT3_STATE_NEW;
2828
2829out_brelse:
2830 brelse (bh);
2831 ext3_std_error(inode->i_sb, err);
2832 return err;
2833}
2834
2835/*
2836 * ext3_write_inode()
2837 *
2838 * We are called from a few places:
2839 *
2840 * - Within generic_file_write() for O_SYNC files.
2841 * Here, there will be no transaction running. We wait for any running
2842 * trasnaction to commit.
2843 *
2844 * - Within sys_sync(), kupdate and such.
2845 * We wait on commit, if tol to.
2846 *
2847 * - Within prune_icache() (PF_MEMALLOC == true)
2848 * Here we simply return. We can't afford to block kswapd on the
2849 * journal commit.
2850 *
2851 * In all cases it is actually safe for us to return without doing anything,
2852 * because the inode has been copied into a raw inode buffer in
2853 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2854 * knfsd.
2855 *
2856 * Note that we are absolutely dependent upon all inode dirtiers doing the
2857 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2858 * which we are interested.
2859 *
2860 * It would be a bug for them to not do this. The code:
2861 *
2862 * mark_inode_dirty(inode)
2863 * stuff();
2864 * inode->i_size = expr;
2865 *
2866 * is in error because a kswapd-driven write_inode() could occur while
2867 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2868 * will no longer be on the superblock's dirty inode list.
2869 */
2870int ext3_write_inode(struct inode *inode, int wait)
2871{
2872 if (current->flags & PF_MEMALLOC)
2873 return 0;
2874
2875 if (ext3_journal_current_handle()) {
2876 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2877 dump_stack();
2878 return -EIO;
2879 }
2880
2881 if (!wait)
2882 return 0;
2883
2884 return ext3_force_commit(inode->i_sb);
2885}
2886
2887/*
2888 * ext3_setattr()
2889 *
2890 * Called from notify_change.
2891 *
2892 * We want to trap VFS attempts to truncate the file as soon as
2893 * possible. In particular, we want to make sure that when the VFS
2894 * shrinks i_size, we put the inode on the orphan list and modify
2895 * i_disksize immediately, so that during the subsequent flushing of
2896 * dirty pages and freeing of disk blocks, we can guarantee that any
2897 * commit will leave the blocks being flushed in an unused state on
2898 * disk. (On recovery, the inode will get truncated and the blocks will
2899 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002900 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 *
2902 * Called with inode->sem down.
2903 */
2904int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2905{
2906 struct inode *inode = dentry->d_inode;
2907 int error, rc = 0;
2908 const unsigned int ia_valid = attr->ia_valid;
2909
2910 error = inode_change_ok(inode, attr);
2911 if (error)
2912 return error;
2913
2914 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2915 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2916 handle_t *handle;
2917
2918 /* (user+group)*(old+new) structure, inode write (sb,
2919 * inode block, ? - but truncate inode update has it) */
Jan Kara1f545872005-06-23 22:01:04 -07002920 handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+
2921 EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 if (IS_ERR(handle)) {
2923 error = PTR_ERR(handle);
2924 goto err_out;
2925 }
2926 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2927 if (error) {
2928 ext3_journal_stop(handle);
2929 return error;
2930 }
2931 /* Update corresponding info in inode so that everything is in
2932 * one transaction */
2933 if (attr->ia_valid & ATTR_UID)
2934 inode->i_uid = attr->ia_uid;
2935 if (attr->ia_valid & ATTR_GID)
2936 inode->i_gid = attr->ia_gid;
2937 error = ext3_mark_inode_dirty(handle, inode);
2938 ext3_journal_stop(handle);
2939 }
2940
2941 if (S_ISREG(inode->i_mode) &&
2942 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2943 handle_t *handle;
2944
2945 handle = ext3_journal_start(inode, 3);
2946 if (IS_ERR(handle)) {
2947 error = PTR_ERR(handle);
2948 goto err_out;
2949 }
2950
2951 error = ext3_orphan_add(handle, inode);
2952 EXT3_I(inode)->i_disksize = attr->ia_size;
2953 rc = ext3_mark_inode_dirty(handle, inode);
2954 if (!error)
2955 error = rc;
2956 ext3_journal_stop(handle);
2957 }
2958
2959 rc = inode_setattr(inode, attr);
2960
2961 /* If inode_setattr's call to ext3_truncate failed to get a
2962 * transaction handle at all, we need to clean up the in-core
2963 * orphan list manually. */
2964 if (inode->i_nlink)
2965 ext3_orphan_del(NULL, inode);
2966
2967 if (!rc && (ia_valid & ATTR_MODE))
2968 rc = ext3_acl_chmod(inode);
2969
2970err_out:
2971 ext3_std_error(inode->i_sb, error);
2972 if (!error)
2973 error = rc;
2974 return error;
2975}
2976
2977
2978/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08002979 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 *
2981 * With N blocks per page, it may be:
2982 * N data blocks
2983 * 2 indirect block
2984 * 2 dindirect
2985 * 1 tindirect
2986 * N+5 bitmap blocks (from the above)
2987 * N+5 group descriptor summary blocks
2988 * 1 inode block
2989 * 1 superblock.
2990 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2991 *
2992 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2993 *
2994 * With ordered or writeback data it's the same, less the N data blocks.
2995 *
2996 * If the inode's direct blocks can hold an integral number of pages then a
2997 * page cannot straddle two indirect blocks, and we can only touch one indirect
2998 * and dindirect block, and the "5" above becomes "3".
2999 *
3000 * This still overestimates under most circumstances. If we were to pass the
3001 * start and end offsets in here as well we could do block_to_path() on each
3002 * block and work out the exact number of indirects which are touched. Pah.
3003 */
3004
3005static int ext3_writepage_trans_blocks(struct inode *inode)
3006{
3007 int bpp = ext3_journal_blocks_per_page(inode);
3008 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3009 int ret;
3010
3011 if (ext3_should_journal_data(inode))
3012 ret = 3 * (bpp + indirects) + 2;
3013 else
3014 ret = 2 * (bpp + indirects) + 2;
3015
3016#ifdef CONFIG_QUOTA
3017 /* We know that structure was already allocated during DQUOT_INIT so
3018 * we will be updating only the data blocks + inodes */
Jan Kara1f545872005-06-23 22:01:04 -07003019 ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020#endif
3021
3022 return ret;
3023}
3024
3025/*
3026 * The caller must have previously called ext3_reserve_inode_write().
3027 * Give this, we know that the caller already has write access to iloc->bh.
3028 */
3029int ext3_mark_iloc_dirty(handle_t *handle,
3030 struct inode *inode, struct ext3_iloc *iloc)
3031{
3032 int err = 0;
3033
3034 /* the do_update_inode consumes one bh->b_count */
3035 get_bh(iloc->bh);
3036
3037 /* ext3_do_update_inode() does journal_dirty_metadata */
3038 err = ext3_do_update_inode(handle, inode, iloc);
3039 put_bh(iloc->bh);
3040 return err;
3041}
3042
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003043/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003045 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 */
3047
3048int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003049ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 struct ext3_iloc *iloc)
3051{
3052 int err = 0;
3053 if (handle) {
3054 err = ext3_get_inode_loc(inode, iloc);
3055 if (!err) {
3056 BUFFER_TRACE(iloc->bh, "get_write_access");
3057 err = ext3_journal_get_write_access(handle, iloc->bh);
3058 if (err) {
3059 brelse(iloc->bh);
3060 iloc->bh = NULL;
3061 }
3062 }
3063 }
3064 ext3_std_error(inode->i_sb, err);
3065 return err;
3066}
3067
3068/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003069 * What we do here is to mark the in-core inode as clean with respect to inode
3070 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 * This means that the in-core inode may be reaped by prune_icache
3072 * without having to perform any I/O. This is a very good thing,
3073 * because *any* task may call prune_icache - even ones which
3074 * have a transaction open against a different journal.
3075 *
3076 * Is this cheating? Not really. Sure, we haven't written the
3077 * inode out, but prune_icache isn't a user-visible syncing function.
3078 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3079 * we start and wait on commits.
3080 *
3081 * Is this efficient/effective? Well, we're being nice to the system
3082 * by cleaning up our inodes proactively so they can be reaped
3083 * without I/O. But we are potentially leaving up to five seconds'
3084 * worth of inodes floating about which prune_icache wants us to
3085 * write out. One way to fix that would be to get prune_icache()
3086 * to do a write_super() to free up some memory. It has the desired
3087 * effect.
3088 */
3089int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3090{
3091 struct ext3_iloc iloc;
3092 int err;
3093
3094 might_sleep();
3095 err = ext3_reserve_inode_write(handle, inode, &iloc);
3096 if (!err)
3097 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3098 return err;
3099}
3100
3101/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003102 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 *
3104 * We're really interested in the case where a file is being extended.
3105 * i_size has been changed by generic_commit_write() and we thus need
3106 * to include the updated inode in the current transaction.
3107 *
3108 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3109 * are allocated to the file.
3110 *
3111 * If the inode is marked synchronous, we don't honour that here - doing
3112 * so would cause a commit on atime updates, which we don't bother doing.
3113 * We handle synchronous inodes at the highest possible level.
3114 */
3115void ext3_dirty_inode(struct inode *inode)
3116{
3117 handle_t *current_handle = ext3_journal_current_handle();
3118 handle_t *handle;
3119
3120 handle = ext3_journal_start(inode, 2);
3121 if (IS_ERR(handle))
3122 goto out;
3123 if (current_handle &&
3124 current_handle->h_transaction != handle->h_transaction) {
3125 /* This task has a transaction open against a different fs */
3126 printk(KERN_EMERG "%s: transactions do not match!\n",
3127 __FUNCTION__);
3128 } else {
3129 jbd_debug(5, "marking dirty. outer handle=%p\n",
3130 current_handle);
3131 ext3_mark_inode_dirty(handle, inode);
3132 }
3133 ext3_journal_stop(handle);
3134out:
3135 return;
3136}
3137
Andrew Mortond6859bf2006-03-26 01:38:03 -08003138#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003139/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 * Bind an inode's backing buffer_head into this transaction, to prevent
3141 * it from being flushed to disk early. Unlike
3142 * ext3_reserve_inode_write, this leaves behind no bh reference and
3143 * returns no iloc structure, so the caller needs to repeat the iloc
3144 * lookup to mark the inode dirty later.
3145 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003146static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147{
3148 struct ext3_iloc iloc;
3149
3150 int err = 0;
3151 if (handle) {
3152 err = ext3_get_inode_loc(inode, &iloc);
3153 if (!err) {
3154 BUFFER_TRACE(iloc.bh, "get_write_access");
3155 err = journal_get_write_access(handle, iloc.bh);
3156 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003157 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 iloc.bh);
3159 brelse(iloc.bh);
3160 }
3161 }
3162 ext3_std_error(inode->i_sb, err);
3163 return err;
3164}
3165#endif
3166
3167int ext3_change_inode_journal_flag(struct inode *inode, int val)
3168{
3169 journal_t *journal;
3170 handle_t *handle;
3171 int err;
3172
3173 /*
3174 * We have to be very careful here: changing a data block's
3175 * journaling status dynamically is dangerous. If we write a
3176 * data block to the journal, change the status and then delete
3177 * that block, we risk forgetting to revoke the old log record
3178 * from the journal and so a subsequent replay can corrupt data.
3179 * So, first we make sure that the journal is empty and that
3180 * nobody is changing anything.
3181 */
3182
3183 journal = EXT3_JOURNAL(inode);
3184 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3185 return -EROFS;
3186
3187 journal_lock_updates(journal);
3188 journal_flush(journal);
3189
3190 /*
3191 * OK, there are no updates running now, and all cached data is
3192 * synced to disk. We are now in a completely consistent state
3193 * which doesn't have anything in the journal, and we know that
3194 * no filesystem updates are running, so it is safe to modify
3195 * the inode's in-core data-journaling state flag now.
3196 */
3197
3198 if (val)
3199 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3200 else
3201 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3202 ext3_set_aops(inode);
3203
3204 journal_unlock_updates(journal);
3205
3206 /* Finally we can mark the inode as dirty. */
3207
3208 handle = ext3_journal_start(inode, 1);
3209 if (IS_ERR(handle))
3210 return PTR_ERR(handle);
3211
3212 err = ext3_mark_inode_dirty(handle, inode);
3213 handle->h_sync = 1;
3214 ext3_journal_stop(handle);
3215 ext3_std_error(inode->i_sb, err);
3216
3217 return err;
3218}