blob: fcfa243618567c2787d50b2119bad58d4d137cb2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext3/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Goal-directed block allocation by Stephen Tweedie
Dave Kleikampe9ad5622006-09-27 01:49:35 -070016 * (sct@redhat.com), 1993, 1998
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
Dave Kleikampe9ad5622006-09-27 01:49:35 -070020 * (jj@sunsite.ms.mff.cuni.cz)
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 *
22 * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
23 */
24
25#include <linux/module.h>
26#include <linux/fs.h>
27#include <linux/time.h>
28#include <linux/ext3_jbd.h>
29#include <linux/jbd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/highuid.h>
31#include <linux/pagemap.h>
32#include <linux/quotaops.h>
33#include <linux/string.h>
34#include <linux/buffer_head.h>
35#include <linux/writeback.h>
36#include <linux/mpage.h>
37#include <linux/uio.h>
Jens Axboecaa38fb2006-07-23 01:41:26 +020038#include <linux/bio.h>
Josef Bacik68c9d702008-10-03 17:32:43 -040039#include <linux/fiemap.h>
Duane Griffinb5ed3112008-12-19 20:47:14 +000040#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xattr.h"
42#include "acl.h"
43
44static int ext3_writepage_trans_blocks(struct inode *inode);
45
46/*
47 * Test whether an inode is a fast symlink.
48 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080049static int ext3_inode_is_fast_symlink(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070050{
51 int ea_blocks = EXT3_I(inode)->i_file_acl ?
52 (inode->i_sb->s_blocksize >> 9) : 0;
53
Andrew Mortond6859bf2006-03-26 01:38:03 -080054 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055}
56
Andrew Mortond6859bf2006-03-26 01:38:03 -080057/*
58 * The ext3 forget function must perform a revoke if we are freeing data
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 * which has been journaled. Metadata (eg. indirect blocks) must be
Mingming Caoae6ddcc2006-09-27 01:49:27 -070060 * revoked in all cases.
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 *
62 * "bh" may be NULL: a metadata block may have been freed from memory
63 * but there may still be a record of it in the journal, and that record
64 * still needs to be revoked.
65 */
Andrew Mortond6859bf2006-03-26 01:38:03 -080066int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
Mingming Cao1c2bf372006-06-25 05:48:06 -070067 struct buffer_head *bh, ext3_fsblk_t blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
69 int err;
70
71 might_sleep();
72
73 BUFFER_TRACE(bh, "enter");
74
75 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
76 "data mode %lx\n",
77 bh, is_metadata, inode->i_mode,
78 test_opt(inode->i_sb, DATA_FLAGS));
79
80 /* Never use the revoke function if we are doing full data
81 * journaling: there is no need to, and a V1 superblock won't
82 * support it. Otherwise, only skip the revoke on un-journaled
83 * data blocks. */
84
85 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
86 (!is_metadata && !ext3_should_journal_data(inode))) {
87 if (bh) {
88 BUFFER_TRACE(bh, "call journal_forget");
89 return ext3_journal_forget(handle, bh);
90 }
91 return 0;
92 }
93
94 /*
95 * data!=journal && (is_metadata || should_journal_data(inode))
96 */
97 BUFFER_TRACE(bh, "call ext3_journal_revoke");
98 err = ext3_journal_revoke(handle, blocknr, bh);
99 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700100 ext3_abort(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 "error %d when attempting revoke", err);
102 BUFFER_TRACE(bh, "exit");
103 return err;
104}
105
106/*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800107 * Work out how many blocks we need to proceed with the next chunk of a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * truncate transaction.
109 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700110static unsigned long blocks_for_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
112 unsigned long needed;
113
114 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
115
116 /* Give ourselves just enough room to cope with inodes in which
117 * i_blocks is corrupt: we've seen disk corruptions in the past
118 * which resulted in random data in an inode which looked enough
119 * like a regular file for ext3 to try to delete it. Things
120 * will go a bit crazy if that happens, but at least we should
121 * try not to panic the whole kernel. */
122 if (needed < 2)
123 needed = 2;
124
125 /* But we need to bound the transaction so we don't overflow the
126 * journal. */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700127 if (needed > EXT3_MAX_TRANS_DATA)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 needed = EXT3_MAX_TRANS_DATA;
129
Jan Kara1f545872005-06-23 22:01:04 -0700130 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700133/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * Truncate transactions can be complex and absolutely huge. So we need to
135 * be able to restart the transaction at a conventient checkpoint to make
136 * sure we don't overflow the journal.
137 *
138 * start_transaction gets us a new handle for a truncate transaction,
139 * and extend_transaction tries to extend the existing one a bit. If
140 * extend fails, we need to propagate the failure up and restart the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700141 * transaction in the top-level truncate loop. --sct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700143static handle_t *start_transaction(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 handle_t *result;
146
147 result = ext3_journal_start(inode, blocks_for_truncate(inode));
148 if (!IS_ERR(result))
149 return result;
150
151 ext3_std_error(inode->i_sb, PTR_ERR(result));
152 return result;
153}
154
155/*
156 * Try to extend this transaction for the purposes of truncation.
157 *
158 * Returns 0 if we managed to create more room. If we can't create more
159 * room, and the transaction must be restarted we return 1.
160 */
161static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
162{
163 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
164 return 0;
165 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
166 return 0;
167 return 1;
168}
169
170/*
171 * Restart the transaction associated with *handle. This does a commit,
172 * so before we call here everything must be consistently dirtied against
173 * this transaction.
174 */
175static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
176{
177 jbd_debug(2, "restarting handle %p\n", handle);
178 return ext3_journal_restart(handle, blocks_for_truncate(inode));
179}
180
181/*
182 * Called at the last iput() if i_nlink is zero.
183 */
184void ext3_delete_inode (struct inode * inode)
185{
186 handle_t *handle;
187
Mark Fashehfef26652005-09-09 13:01:31 -0700188 truncate_inode_pages(&inode->i_data, 0);
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 if (is_bad_inode(inode))
191 goto no_delete;
192
193 handle = start_transaction(inode);
194 if (IS_ERR(handle)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800195 /*
196 * If we're going to skip the normal cleanup, we still need to
197 * make sure that the in-core orphan linked list is properly
198 * cleaned up.
199 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 ext3_orphan_del(NULL, inode);
201 goto no_delete;
202 }
203
204 if (IS_SYNC(inode))
205 handle->h_sync = 1;
206 inode->i_size = 0;
207 if (inode->i_blocks)
208 ext3_truncate(inode);
209 /*
210 * Kill off the orphan record which ext3_truncate created.
211 * AKPM: I think this can be inside the above `if'.
212 * Note that ext3_orphan_del() has to be able to cope with the
213 * deletion of a non-existent orphan - this is because we don't
214 * know if ext3_truncate() actually created an orphan record.
215 * (Well, we could do this if we need to, but heck - it works)
216 */
217 ext3_orphan_del(handle, inode);
218 EXT3_I(inode)->i_dtime = get_seconds();
219
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700220 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 * One subtle ordering requirement: if anything has gone wrong
222 * (transaction abort, IO errors, whatever), then we can still
223 * do these next steps (the fs will already have been marked as
224 * having errors), but we can't free the inode if the mark_dirty
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700225 * fails.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 */
227 if (ext3_mark_inode_dirty(handle, inode))
228 /* If that failed, just do the required in-core inode clear. */
229 clear_inode(inode);
230 else
231 ext3_free_inode(handle, inode);
232 ext3_journal_stop(handle);
233 return;
234no_delete:
235 clear_inode(inode); /* We must guarantee clearing of inode... */
236}
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238typedef struct {
239 __le32 *p;
240 __le32 key;
241 struct buffer_head *bh;
242} Indirect;
243
244static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
245{
246 p->key = *(p->p = v);
247 p->bh = bh;
248}
249
Andrew Mortond6859bf2006-03-26 01:38:03 -0800250static int verify_chain(Indirect *from, Indirect *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
252 while (from <= to && from->key == *from->p)
253 from++;
254 return (from > to);
255}
256
257/**
258 * ext3_block_to_path - parse the block number into array of offsets
259 * @inode: inode in question (we are only interested in its superblock)
260 * @i_block: block number to be parsed
261 * @offsets: array to store the offsets in
262 * @boundary: set this non-zero if the referred-to block is likely to be
263 * followed (on disk) by an indirect block.
264 *
265 * To store the locations of file's data ext3 uses a data structure common
266 * for UNIX filesystems - tree of pointers anchored in the inode, with
267 * data blocks at leaves and indirect blocks in intermediate nodes.
268 * This function translates the block number into path in that tree -
269 * return value is the path length and @offsets[n] is the offset of
270 * pointer to (n+1)th node in the nth one. If @block is out of range
271 * (negative or too large) warning is printed and zero returned.
272 *
273 * Note: function doesn't find node addresses, so no IO is needed. All
274 * we need to know is the capacity of indirect blocks (taken from the
275 * inode->i_sb).
276 */
277
278/*
279 * Portability note: the last comparison (check that we fit into triple
280 * indirect block) is spelled differently, because otherwise on an
281 * architecture with 32-bit longs and 8Kb pages we might get into trouble
282 * if our filesystem had 8Kb blocks. We might use long long, but that would
283 * kill us on x86. Oh, well, at least the sign propagation does not matter -
284 * i_block would have to be negative in the very beginning, so we would not
285 * get there at all.
286 */
287
288static int ext3_block_to_path(struct inode *inode,
289 long i_block, int offsets[4], int *boundary)
290{
291 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
292 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
293 const long direct_blocks = EXT3_NDIR_BLOCKS,
294 indirect_blocks = ptrs,
295 double_blocks = (1 << (ptrs_bits * 2));
296 int n = 0;
297 int final = 0;
298
299 if (i_block < 0) {
300 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
301 } else if (i_block < direct_blocks) {
302 offsets[n++] = i_block;
303 final = direct_blocks;
304 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
305 offsets[n++] = EXT3_IND_BLOCK;
306 offsets[n++] = i_block;
307 final = ptrs;
308 } else if ((i_block -= indirect_blocks) < double_blocks) {
309 offsets[n++] = EXT3_DIND_BLOCK;
310 offsets[n++] = i_block >> ptrs_bits;
311 offsets[n++] = i_block & (ptrs - 1);
312 final = ptrs;
313 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
314 offsets[n++] = EXT3_TIND_BLOCK;
315 offsets[n++] = i_block >> (ptrs_bits * 2);
316 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
317 offsets[n++] = i_block & (ptrs - 1);
318 final = ptrs;
319 } else {
Andrew Mortond6859bf2006-03-26 01:38:03 -0800320 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 }
322 if (boundary)
Mingming Cao89747d32006-03-26 01:37:55 -0800323 *boundary = final - 1 - (i_block & (ptrs - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 return n;
325}
326
327/**
328 * ext3_get_branch - read the chain of indirect blocks leading to data
329 * @inode: inode in question
330 * @depth: depth of the chain (1 - direct pointer, etc.)
331 * @offsets: offsets of pointers in inode/indirect blocks
332 * @chain: place to store the result
333 * @err: here we store the error value
334 *
335 * Function fills the array of triples <key, p, bh> and returns %NULL
336 * if everything went OK or the pointer to the last filled triple
337 * (incomplete one) otherwise. Upon the return chain[i].key contains
338 * the number of (i+1)-th block in the chain (as it is stored in memory,
339 * i.e. little-endian 32-bit), chain[i].p contains the address of that
340 * number (it points into struct inode for i==0 and into the bh->b_data
341 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
342 * block for i>0 and NULL for i==0. In other words, it holds the block
343 * numbers of the chain, addresses they were taken from (and where we can
344 * verify that chain did not change) and buffer_heads hosting these
345 * numbers.
346 *
347 * Function stops when it stumbles upon zero pointer (absent block)
348 * (pointer to last triple returned, *@err == 0)
349 * or when it gets an IO error reading an indirect block
350 * (ditto, *@err == -EIO)
351 * or when it notices that chain had been changed while it was reading
352 * (ditto, *@err == -EAGAIN)
353 * or when it reads all @depth-1 indirect blocks successfully and finds
354 * the whole chain, all way to the data (returns %NULL, *err == 0).
355 */
356static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
357 Indirect chain[4], int *err)
358{
359 struct super_block *sb = inode->i_sb;
360 Indirect *p = chain;
361 struct buffer_head *bh;
362
363 *err = 0;
364 /* i_data is not going away, no lock needed */
365 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
366 if (!p->key)
367 goto no_block;
368 while (--depth) {
369 bh = sb_bread(sb, le32_to_cpu(p->key));
370 if (!bh)
371 goto failure;
372 /* Reader: pointers */
373 if (!verify_chain(chain, p))
374 goto changed;
375 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
376 /* Reader: end */
377 if (!p->key)
378 goto no_block;
379 }
380 return NULL;
381
382changed:
383 brelse(bh);
384 *err = -EAGAIN;
385 goto no_block;
386failure:
387 *err = -EIO;
388no_block:
389 return p;
390}
391
392/**
393 * ext3_find_near - find a place for allocation with sufficient locality
394 * @inode: owner
395 * @ind: descriptor of indirect block.
396 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000397 * This function returns the preferred place for block allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * It is used when heuristic for sequential allocation fails.
399 * Rules are:
400 * + if there is a block to the left of our position - allocate near it.
401 * + if pointer will live in indirect block - allocate near that block.
402 * + if pointer will live in inode - allocate in the same
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700403 * cylinder group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 *
405 * In the latter case we colour the starting block by the callers PID to
406 * prevent it from clashing with concurrent allocations for a different inode
407 * in the same block group. The PID is used here so that functionally related
408 * files will be close-by on-disk.
409 *
410 * Caller must make sure that @ind is valid and will stay that way.
411 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700412static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
414 struct ext3_inode_info *ei = EXT3_I(inode);
415 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
416 __le32 *p;
Mingming Cao43d23f92006-06-25 05:48:07 -0700417 ext3_fsblk_t bg_start;
418 ext3_grpblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 /* Try to find previous block */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800421 for (p = ind->p - 1; p >= start; p--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 if (*p)
423 return le32_to_cpu(*p);
Andrew Mortond6859bf2006-03-26 01:38:03 -0800424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 /* No such thing, so let's try location of indirect block */
427 if (ind->bh)
428 return ind->bh->b_blocknr;
429
430 /*
Andrew Mortond6859bf2006-03-26 01:38:03 -0800431 * It is going to be referred to from the inode itself? OK, just put it
432 * into the same cylinder group then.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
Mingming Cao43d23f92006-06-25 05:48:07 -0700434 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 colour = (current->pid % 16) *
436 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
437 return bg_start + colour;
438}
439
440/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000441 * ext3_find_goal - find a preferred place for allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * @inode: owner
443 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000446 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800447 * returns it.
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,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800451 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
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800887 goal = ext3_find_goal(inode, iblock, 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
Jan Karabd1939d2008-02-06 01:40:21 -0800944/* Maximum number of blocks we map for direct IO at once. */
945#define DIO_MAX_BLOCKS 4096
946/*
947 * Number of credits we need for writing DIO_MAX_BLOCKS:
948 * We need sb + group descriptor + bitmap + inode -> 4
949 * For B blocks with A block pointers per block we need:
950 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
951 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
952 */
953#define DIO_CREDITS 25
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -0800955static int ext3_get_block(struct inode *inode, sector_t iblock,
956 struct buffer_head *bh_result, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800958 handle_t *handle = ext3_journal_current_handle();
Jan Karabd1939d2008-02-06 01:40:21 -0800959 int ret = 0, started = 0;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800960 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Jan Karabd1939d2008-02-06 01:40:21 -0800962 if (create && !handle) { /* Direct IO write... */
963 if (max_blocks > DIO_MAX_BLOCKS)
964 max_blocks = DIO_MAX_BLOCKS;
965 handle = ext3_journal_start(inode, DIO_CREDITS +
966 2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb));
967 if (IS_ERR(handle)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 ret = PTR_ERR(handle);
Jan Karabd1939d2008-02-06 01:40:21 -0800969 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
Jan Karabd1939d2008-02-06 01:40:21 -0800971 started = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 }
973
Jan Karabd1939d2008-02-06 01:40:21 -0800974 ret = ext3_get_blocks_handle(handle, inode, iblock,
Mingming Cao89747d32006-03-26 01:37:55 -0800975 max_blocks, bh_result, create, 0);
Jan Karabd1939d2008-02-06 01:40:21 -0800976 if (ret > 0) {
977 bh_result->b_size = (ret << inode->i_blkbits);
978 ret = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800979 }
Jan Karabd1939d2008-02-06 01:40:21 -0800980 if (started)
981 ext3_journal_stop(handle);
982out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 return ret;
984}
985
Josef Bacik68c9d702008-10-03 17:32:43 -0400986int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
987 u64 start, u64 len)
988{
989 return generic_block_fiemap(inode, fieinfo, start, len,
990 ext3_get_block);
991}
992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993/*
994 * `handle' can be NULL if create is zero
995 */
Andrew Mortond6859bf2006-03-26 01:38:03 -0800996struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
997 long block, int create, int *errp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
999 struct buffer_head dummy;
1000 int fatal = 0, err;
1001
1002 J_ASSERT(handle != NULL || create == 0);
1003
1004 dummy.b_state = 0;
1005 dummy.b_blocknr = -1000;
1006 buffer_trace_init(&dummy.b_history);
Mingming Cao89747d32006-03-26 01:37:55 -08001007 err = ext3_get_blocks_handle(handle, inode, block, 1,
1008 &dummy, create, 1);
Badari Pulavarty3665d0e2006-09-08 09:48:21 -07001009 /*
1010 * ext3_get_blocks_handle() returns number of blocks
1011 * mapped. 0 in case of a HOLE.
1012 */
1013 if (err > 0) {
1014 if (err > 1)
1015 WARN_ON(1);
Mingming Cao89747d32006-03-26 01:37:55 -08001016 err = 0;
Mingming Cao89747d32006-03-26 01:37:55 -08001017 }
1018 *errp = err;
1019 if (!err && buffer_mapped(&dummy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 struct buffer_head *bh;
1021 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001022 if (!bh) {
1023 *errp = -EIO;
1024 goto err;
1025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 if (buffer_new(&dummy)) {
1027 J_ASSERT(create != 0);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001028 J_ASSERT(handle != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Andrew Mortond6859bf2006-03-26 01:38:03 -08001030 /*
1031 * Now that we do not always journal data, we should
1032 * keep in mind whether this should always journal the
1033 * new buffer as metadata. For now, regular file
1034 * writes use ext3_get_block instead, so it's not a
1035 * problem.
1036 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 lock_buffer(bh);
1038 BUFFER_TRACE(bh, "call get_create_access");
1039 fatal = ext3_journal_get_create_access(handle, bh);
1040 if (!fatal && !buffer_uptodate(bh)) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08001041 memset(bh->b_data,0,inode->i_sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 set_buffer_uptodate(bh);
1043 }
1044 unlock_buffer(bh);
1045 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1046 err = ext3_journal_dirty_metadata(handle, bh);
1047 if (!fatal)
1048 fatal = err;
1049 } else {
1050 BUFFER_TRACE(bh, "not a new buffer");
1051 }
1052 if (fatal) {
1053 *errp = fatal;
1054 brelse(bh);
1055 bh = NULL;
1056 }
1057 return bh;
1058 }
Glauber de Oliveira Costa2973dfd2005-10-30 15:03:05 -08001059err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 return NULL;
1061}
1062
Andrew Mortond6859bf2006-03-26 01:38:03 -08001063struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 int block, int create, int *err)
1065{
1066 struct buffer_head * bh;
1067
1068 bh = ext3_getblk(handle, inode, block, create, err);
1069 if (!bh)
1070 return bh;
1071 if (buffer_uptodate(bh))
1072 return bh;
Jens Axboecaa38fb2006-07-23 01:41:26 +02001073 ll_rw_block(READ_META, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 wait_on_buffer(bh);
1075 if (buffer_uptodate(bh))
1076 return bh;
1077 put_bh(bh);
1078 *err = -EIO;
1079 return NULL;
1080}
1081
1082static int walk_page_buffers( handle_t *handle,
1083 struct buffer_head *head,
1084 unsigned from,
1085 unsigned to,
1086 int *partial,
1087 int (*fn)( handle_t *handle,
1088 struct buffer_head *bh))
1089{
1090 struct buffer_head *bh;
1091 unsigned block_start, block_end;
1092 unsigned blocksize = head->b_size;
1093 int err, ret = 0;
1094 struct buffer_head *next;
1095
1096 for ( bh = head, block_start = 0;
1097 ret == 0 && (bh != head || !block_start);
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001098 block_start = block_end, bh = next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 {
1100 next = bh->b_this_page;
1101 block_end = block_start + blocksize;
1102 if (block_end <= from || block_start >= to) {
1103 if (partial && !buffer_uptodate(bh))
1104 *partial = 1;
1105 continue;
1106 }
1107 err = (*fn)(handle, bh);
1108 if (!ret)
1109 ret = err;
1110 }
1111 return ret;
1112}
1113
1114/*
1115 * To preserve ordering, it is essential that the hole instantiation and
1116 * the data write be encapsulated in a single transaction. We cannot
1117 * close off a transaction and start a new one between the ext3_get_block()
1118 * and the commit_write(). So doing the journal_start at the start of
1119 * prepare_write() is the right place.
1120 *
1121 * Also, this function can nest inside ext3_writepage() ->
1122 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1123 * has generated enough buffer credits to do the whole page. So we won't
1124 * block on the journal in that case, which is good, because the caller may
1125 * be PF_MEMALLOC.
1126 *
1127 * By accident, ext3 can be reentered when a transaction is open via
1128 * quota file writes. If we were to commit the transaction while thus
1129 * reentered, there can be a deadlock - we would be holding a quota
1130 * lock, and the commit would never complete if another thread had a
1131 * transaction open and was blocking on the quota lock - a ranking
1132 * violation.
1133 *
1134 * So what we do is to rely on the fact that journal_stop/journal_start
1135 * will _not_ run commit under these circumstances because handle->h_ref
1136 * is elevated. We'll still have enough credits for the tiny quotafile
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001137 * write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08001139static int do_journal_get_write_access(handle_t *handle,
1140 struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
1142 if (!buffer_mapped(bh) || buffer_freed(bh))
1143 return 0;
1144 return ext3_journal_get_write_access(handle, bh);
1145}
1146
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001147static int ext3_write_begin(struct file *file, struct address_space *mapping,
1148 loff_t pos, unsigned len, unsigned flags,
1149 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150{
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001151 struct inode *inode = mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001152 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 handle_t *handle;
1154 int retries = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001155 struct page *page;
1156 pgoff_t index;
1157 unsigned from, to;
Jan Kara695f6ae2009-04-02 16:57:17 -07001158 /* Reserve one block more for addition to orphan list in case
1159 * we allocate blocks but write fails for some reason */
1160 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001161
1162 index = pos >> PAGE_CACHE_SHIFT;
1163 from = pos & (PAGE_CACHE_SIZE - 1);
1164 to = from + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166retry:
Nick Piggin54566b22009-01-04 12:00:53 -08001167 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001168 if (!page)
1169 return -ENOMEM;
1170 *pagep = page;
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 handle = ext3_journal_start(inode, needed_blocks);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001173 if (IS_ERR(handle)) {
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001174 unlock_page(page);
1175 page_cache_release(page);
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001176 ret = PTR_ERR(handle);
1177 goto out;
1178 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001179 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1180 ext3_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 if (ret)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001182 goto write_begin_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 if (ext3_should_journal_data(inode)) {
1185 ret = walk_page_buffers(handle, page_buffers(page),
1186 from, to, NULL, do_journal_get_write_access);
1187 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001188write_begin_failed:
1189 if (ret) {
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001190 /*
1191 * block_write_begin may have instantiated a few blocks
1192 * outside i_size. Trim these off again. Don't need
1193 * i_size_read because we hold i_mutex.
Jan Kara695f6ae2009-04-02 16:57:17 -07001194 *
1195 * Add inode to orphan list in case we crash before truncate
1196 * finishes.
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001197 */
1198 if (pos + len > inode->i_size)
Jan Kara695f6ae2009-04-02 16:57:17 -07001199 ext3_orphan_add(handle, inode);
1200 ext3_journal_stop(handle);
1201 unlock_page(page);
1202 page_cache_release(page);
1203 if (pos + len > inode->i_size)
Aneesh Kumar K.V5ec8b752008-10-18 20:28:00 -07001204 vmtruncate(inode, inode->i_size);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1207 goto retry;
Andrew Morton1aa9b4b2007-04-01 23:49:43 -07001208out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return ret;
1210}
1211
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001212
Andrew Mortond6859bf2006-03-26 01:38:03 -08001213int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
1215 int err = journal_dirty_data(handle, bh);
1216 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001217 ext3_journal_abort_handle(__func__, __func__,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001218 bh, handle, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 return err;
1220}
1221
Jan Kara695f6ae2009-04-02 16:57:17 -07001222/* For ordered writepage and write_end functions */
1223static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1224{
1225 /*
1226 * Write could have mapped the buffer but it didn't copy the data in
1227 * yet. So avoid filing such buffer into a transaction.
1228 */
1229 if (buffer_mapped(bh) && buffer_uptodate(bh))
1230 return ext3_journal_dirty_data(handle, bh);
1231 return 0;
1232}
1233
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001234/* For write_end() in data=journal mode */
1235static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236{
1237 if (!buffer_mapped(bh) || buffer_freed(bh))
1238 return 0;
1239 set_buffer_uptodate(bh);
1240 return ext3_journal_dirty_metadata(handle, bh);
1241}
1242
1243/*
Jan Kara695f6ae2009-04-02 16:57:17 -07001244 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1245 * for the whole page but later we failed to copy the data in. Update inode
1246 * size according to what we managed to copy. The rest is going to be
1247 * truncated in write_end function.
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001248 */
Jan Kara695f6ae2009-04-02 16:57:17 -07001249static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001250{
Jan Kara695f6ae2009-04-02 16:57:17 -07001251 /* What matters to us is i_disksize. We don't write i_size anywhere */
1252 if (pos + copied > inode->i_size)
1253 i_size_write(inode, pos + copied);
1254 if (pos + copied > EXT3_I(inode)->i_disksize) {
1255 EXT3_I(inode)->i_disksize = pos + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001256 mark_inode_dirty(inode);
1257 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001258}
1259
1260/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 * We need to pick up the new inode size which generic_commit_write gave us
1262 * `file' can be NULL - eg, when called from page_symlink().
1263 *
1264 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1265 * buffers are managed internally.
1266 */
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001267static int ext3_ordered_write_end(struct file *file,
1268 struct address_space *mapping,
1269 loff_t pos, unsigned len, unsigned copied,
1270 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271{
1272 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001273 struct inode *inode = file->f_mapping->host;
1274 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 int ret = 0, ret2;
1276
Jan Kara695f6ae2009-04-02 16:57:17 -07001277 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1278
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001279 from = pos & (PAGE_CACHE_SIZE - 1);
Jan Kara695f6ae2009-04-02 16:57:17 -07001280 to = from + copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 ret = walk_page_buffers(handle, page_buffers(page),
Jan Kara695f6ae2009-04-02 16:57:17 -07001282 from, to, NULL, journal_dirty_data_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Jan Kara695f6ae2009-04-02 16:57:17 -07001284 if (ret == 0)
1285 update_file_sizes(inode, pos, copied);
1286 /*
1287 * There may be allocated blocks outside of i_size because
1288 * we failed to copy some data. Prepare for truncate.
1289 */
1290 if (pos + len > inode->i_size)
1291 ext3_orphan_add(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 ret2 = ext3_journal_stop(handle);
1293 if (!ret)
1294 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001295 unlock_page(page);
1296 page_cache_release(page);
1297
Jan Kara695f6ae2009-04-02 16:57:17 -07001298 if (pos + len > inode->i_size)
1299 vmtruncate(inode, inode->i_size);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001300 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301}
1302
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001303static int ext3_writeback_write_end(struct file *file,
1304 struct address_space *mapping,
1305 loff_t pos, unsigned len, unsigned copied,
1306 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307{
1308 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001309 struct inode *inode = file->f_mapping->host;
Jan Kara695f6ae2009-04-02 16:57:17 -07001310 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Jan Kara695f6ae2009-04-02 16:57:17 -07001312 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1313 update_file_sizes(inode, pos, copied);
1314 /*
1315 * There may be allocated blocks outside of i_size because
1316 * we failed to copy some data. Prepare for truncate.
1317 */
1318 if (pos + len > inode->i_size)
1319 ext3_orphan_add(handle, inode);
1320 ret = ext3_journal_stop(handle);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001321 unlock_page(page);
1322 page_cache_release(page);
1323
Jan Kara695f6ae2009-04-02 16:57:17 -07001324 if (pos + len > inode->i_size)
1325 vmtruncate(inode, inode->i_size);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001326 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327}
1328
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001329static int ext3_journalled_write_end(struct file *file,
1330 struct address_space *mapping,
1331 loff_t pos, unsigned len, unsigned copied,
1332 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 handle_t *handle = ext3_journal_current_handle();
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001335 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 int ret = 0, ret2;
1337 int partial = 0;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001338 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001340 from = pos & (PAGE_CACHE_SIZE - 1);
1341 to = from + len;
1342
1343 if (copied < len) {
1344 if (!PageUptodate(page))
1345 copied = 0;
Jan Kara695f6ae2009-04-02 16:57:17 -07001346 page_zero_new_buffers(page, from + copied, to);
1347 to = from + copied;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001351 to, &partial, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 if (!partial)
1353 SetPageUptodate(page);
Jan Kara695f6ae2009-04-02 16:57:17 -07001354
1355 if (pos + copied > inode->i_size)
1356 i_size_write(inode, pos + copied);
1357 /*
1358 * There may be allocated blocks outside of i_size because
1359 * we failed to copy some data. Prepare for truncate.
1360 */
1361 if (pos + len > inode->i_size)
1362 ext3_orphan_add(handle, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1364 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1365 EXT3_I(inode)->i_disksize = inode->i_size;
1366 ret2 = ext3_mark_inode_dirty(handle, inode);
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001367 if (!ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 ret = ret2;
1369 }
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 ret2 = ext3_journal_stop(handle);
1372 if (!ret)
1373 ret = ret2;
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001374 unlock_page(page);
1375 page_cache_release(page);
1376
Jan Kara695f6ae2009-04-02 16:57:17 -07001377 if (pos + len > inode->i_size)
1378 vmtruncate(inode, inode->i_size);
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001379 return ret ? ret : copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 * bmap() is special. It gets used by applications such as lilo and by
1384 * the swapper to find the on-disk block of a specific piece of data.
1385 *
1386 * Naturally, this is dangerous if the block concerned is still in the
1387 * journal. If somebody makes a swapfile on an ext3 data-journaling
1388 * filesystem and enables swap, then they may get a nasty shock when the
1389 * data getting swapped to that swapfile suddenly gets overwritten by
1390 * the original zero's written out previously to the journal and
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001391 * awaiting writeback in the kernel's buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 *
1393 * So, if we see any bmap calls here on a modified, data-journaled file,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001394 * take extra steps to flush any blocks which might be in the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 */
1396static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1397{
1398 struct inode *inode = mapping->host;
1399 journal_t *journal;
1400 int err;
1401
1402 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001403 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 * This is a REALLY heavyweight approach, but the use of
1405 * bmap on dirty files is expected to be extremely rare:
1406 * only if we run lilo or swapon on a freshly made file
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001407 * do we expect this to happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 *
1409 * (bmap requires CAP_SYS_RAWIO so this does not
1410 * represent an unprivileged user DOS attack --- we'd be
1411 * in trouble if mortal users could trigger this path at
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001412 * will.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 *
1414 * NB. EXT3_STATE_JDATA is not set on files other than
1415 * regular files. If somebody wants to bmap a directory
1416 * or symlink and gets confused because the buffer
1417 * hasn't yet been flushed to disk, they deserve
1418 * everything they get.
1419 */
1420
1421 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1422 journal = EXT3_JOURNAL(inode);
1423 journal_lock_updates(journal);
1424 err = journal_flush(journal);
1425 journal_unlock_updates(journal);
1426
1427 if (err)
1428 return 0;
1429 }
1430
1431 return generic_block_bmap(mapping,block,ext3_get_block);
1432}
1433
1434static int bget_one(handle_t *handle, struct buffer_head *bh)
1435{
1436 get_bh(bh);
1437 return 0;
1438}
1439
1440static int bput_one(handle_t *handle, struct buffer_head *bh)
1441{
1442 put_bh(bh);
1443 return 0;
1444}
1445
Jan Kara9e80d402009-03-26 13:08:04 +01001446static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1447{
1448 return !buffer_mapped(bh);
1449}
Jan Kara695f6ae2009-04-02 16:57:17 -07001450
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451/*
1452 * Note that we always start a transaction even if we're not journalling
1453 * data. This is to preserve ordering: any hole instantiation within
1454 * __block_write_full_page -> ext3_get_block() should be journalled
1455 * along with the data so we don't crash and then get metadata which
1456 * refers to old data.
1457 *
1458 * In all journalling modes block_write_full_page() will start the I/O.
1459 *
1460 * Problem:
1461 *
1462 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1463 * ext3_writepage()
1464 *
1465 * Similar for:
1466 *
1467 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1468 *
1469 * Same applies to ext3_get_block(). We will deadlock on various things like
Arjan van de Ven97461512006-03-23 03:00:42 -08001470 * lock_journal and i_truncate_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 *
1472 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1473 * allocations fail.
1474 *
1475 * 16May01: If we're reentered then journal_current_handle() will be
1476 * non-zero. We simply *return*.
1477 *
1478 * 1 July 2001: @@@ FIXME:
1479 * In journalled data mode, a data buffer may be metadata against the
1480 * current transaction. But the same file is part of a shared mapping
1481 * and someone does a writepage() on it.
1482 *
1483 * We will move the buffer onto the async_data list, but *after* it has
1484 * been dirtied. So there's a small window where we have dirty data on
1485 * BJ_Metadata.
1486 *
1487 * Note that this only applies to the last partial page in the file. The
1488 * bit which block_write_full_page() uses prepare/commit for. (That's
1489 * broken code anyway: it's wrong for msync()).
1490 *
1491 * It's a rare case: affects the final partial page, for journalled data
1492 * where the file is subject to bith write() and writepage() in the same
1493 * transction. To fix it we'll need a custom block_write_full_page().
1494 * We'll probably need that anyway for journalling writepage() output.
1495 *
1496 * We don't honour synchronous mounts for writepage(). That would be
1497 * disastrous. Any write() or metadata operation will sync the fs for
1498 * us.
1499 *
1500 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1501 * we don't need to open a transaction here.
1502 */
1503static int ext3_ordered_writepage(struct page *page,
Andrew Mortond6859bf2006-03-26 01:38:03 -08001504 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
1506 struct inode *inode = page->mapping->host;
1507 struct buffer_head *page_bufs;
1508 handle_t *handle = NULL;
1509 int ret = 0;
1510 int err;
1511
1512 J_ASSERT(PageLocked(page));
1513
1514 /*
1515 * We give up here if we're reentered, because it might be for a
1516 * different filesystem.
1517 */
1518 if (ext3_journal_current_handle())
1519 goto out_fail;
1520
Jan Kara9e80d402009-03-26 13:08:04 +01001521 if (!page_has_buffers(page)) {
1522 create_empty_buffers(page, inode->i_sb->s_blocksize,
1523 (1 << BH_Dirty)|(1 << BH_Uptodate));
Jan Kara430db322009-04-07 18:25:01 -04001524 page_bufs = page_buffers(page);
1525 } else {
1526 page_bufs = page_buffers(page);
1527 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1528 NULL, buffer_unmapped)) {
1529 /* Provide NULL get_block() to catch bugs if buffers
1530 * weren't really mapped */
1531 return block_write_full_page(page, NULL, wbc);
1532 }
Jan Kara9e80d402009-03-26 13:08:04 +01001533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1535
1536 if (IS_ERR(handle)) {
1537 ret = PTR_ERR(handle);
1538 goto out_fail;
1539 }
1540
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 walk_page_buffers(handle, page_bufs, 0,
1542 PAGE_CACHE_SIZE, NULL, bget_one);
1543
1544 ret = block_write_full_page(page, ext3_get_block, wbc);
1545
1546 /*
1547 * The page can become unlocked at any point now, and
1548 * truncate can then come in and change things. So we
1549 * can't touch *page from now on. But *page_bufs is
1550 * safe due to elevated refcount.
1551 */
1552
1553 /*
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001554 * And attach them to the current transaction. But only if
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 * block_write_full_page() succeeded. Otherwise they are unmapped,
1556 * and generally junk.
1557 */
1558 if (ret == 0) {
1559 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1560 NULL, journal_dirty_data_fn);
1561 if (!ret)
1562 ret = err;
1563 }
1564 walk_page_buffers(handle, page_bufs, 0,
1565 PAGE_CACHE_SIZE, NULL, bput_one);
1566 err = ext3_journal_stop(handle);
1567 if (!ret)
1568 ret = err;
1569 return ret;
1570
1571out_fail:
1572 redirty_page_for_writepage(wbc, page);
1573 unlock_page(page);
1574 return ret;
1575}
1576
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577static int ext3_writeback_writepage(struct page *page,
1578 struct writeback_control *wbc)
1579{
1580 struct inode *inode = page->mapping->host;
1581 handle_t *handle = NULL;
1582 int ret = 0;
1583 int err;
1584
1585 if (ext3_journal_current_handle())
1586 goto out_fail;
1587
Jan Kara430db322009-04-07 18:25:01 -04001588 if (page_has_buffers(page)) {
1589 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1590 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1591 /* Provide NULL get_block() to catch bugs if buffers
1592 * weren't really mapped */
1593 return block_write_full_page(page, NULL, wbc);
1594 }
1595 }
1596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1598 if (IS_ERR(handle)) {
1599 ret = PTR_ERR(handle);
1600 goto out_fail;
1601 }
1602
Badari Pulavarty0e31f512006-07-30 03:04:14 -07001603 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 ret = nobh_writepage(page, ext3_get_block, wbc);
1605 else
1606 ret = block_write_full_page(page, ext3_get_block, wbc);
1607
1608 err = ext3_journal_stop(handle);
1609 if (!ret)
1610 ret = err;
1611 return ret;
1612
1613out_fail:
1614 redirty_page_for_writepage(wbc, page);
1615 unlock_page(page);
1616 return ret;
1617}
1618
1619static int ext3_journalled_writepage(struct page *page,
1620 struct writeback_control *wbc)
1621{
1622 struct inode *inode = page->mapping->host;
1623 handle_t *handle = NULL;
1624 int ret = 0;
1625 int err;
1626
1627 if (ext3_journal_current_handle())
1628 goto no_write;
1629
1630 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1631 if (IS_ERR(handle)) {
1632 ret = PTR_ERR(handle);
1633 goto no_write;
1634 }
1635
1636 if (!page_has_buffers(page) || PageChecked(page)) {
1637 /*
1638 * It's mmapped pagecache. Add buffers and journal it. There
1639 * doesn't seem much point in redirtying the page here.
1640 */
1641 ClearPageChecked(page);
1642 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1643 ext3_get_block);
Denis Lunevab4eb432005-11-13 16:07:17 -08001644 if (ret != 0) {
1645 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 goto out_unlock;
Denis Lunevab4eb432005-11-13 16:07:17 -08001647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 ret = walk_page_buffers(handle, page_buffers(page), 0,
1649 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1650
1651 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginf4fc66a2007-10-16 01:25:05 -07001652 PAGE_CACHE_SIZE, NULL, write_end_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 if (ret == 0)
1654 ret = err;
1655 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1656 unlock_page(page);
1657 } else {
1658 /*
1659 * It may be a page full of checkpoint-mode buffers. We don't
1660 * really know unless we go poke around in the buffer_heads.
1661 * But block_write_full_page will do the right thing.
1662 */
1663 ret = block_write_full_page(page, ext3_get_block, wbc);
1664 }
1665 err = ext3_journal_stop(handle);
1666 if (!ret)
1667 ret = err;
1668out:
1669 return ret;
1670
1671no_write:
1672 redirty_page_for_writepage(wbc, page);
1673out_unlock:
1674 unlock_page(page);
1675 goto out;
1676}
1677
1678static int ext3_readpage(struct file *file, struct page *page)
1679{
1680 return mpage_readpage(page, ext3_get_block);
1681}
1682
1683static int
1684ext3_readpages(struct file *file, struct address_space *mapping,
1685 struct list_head *pages, unsigned nr_pages)
1686{
1687 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1688}
1689
NeilBrown2ff28e22006-03-26 01:37:18 -08001690static void ext3_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691{
1692 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1693
1694 /*
1695 * If it's a full truncate we just forget about the pending dirtying
1696 */
1697 if (offset == 0)
1698 ClearPageChecked(page);
1699
NeilBrown2ff28e22006-03-26 01:37:18 -08001700 journal_invalidatepage(journal, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701}
1702
Al Viro27496a82005-10-21 03:20:48 -04001703static int ext3_releasepage(struct page *page, gfp_t wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
1705 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1706
1707 WARN_ON(PageChecked(page));
1708 if (!page_has_buffers(page))
1709 return 0;
1710 return journal_try_to_free_buffers(journal, page, wait);
1711}
1712
1713/*
1714 * If the O_DIRECT write will extend the file then add this inode to the
1715 * orphan list. So recovery will truncate it back to the original size
1716 * if the machine crashes during the write.
1717 *
1718 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Karabd1939d2008-02-06 01:40:21 -08001719 * crashes then stale disk data _may_ be exposed inside the file. But current
1720 * VFS code falls back into buffered path in that case so we are safe.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 */
1722static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1723 const struct iovec *iov, loff_t offset,
1724 unsigned long nr_segs)
1725{
1726 struct file *file = iocb->ki_filp;
1727 struct inode *inode = file->f_mapping->host;
1728 struct ext3_inode_info *ei = EXT3_I(inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001729 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 ssize_t ret;
1731 int orphan = 0;
1732 size_t count = iov_length(iov, nr_segs);
1733
1734 if (rw == WRITE) {
1735 loff_t final_size = offset + count;
1736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (final_size > inode->i_size) {
Jan Karabd1939d2008-02-06 01:40:21 -08001738 /* Credits for sb + inode write */
1739 handle = ext3_journal_start(inode, 2);
1740 if (IS_ERR(handle)) {
1741 ret = PTR_ERR(handle);
1742 goto out;
1743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 ret = ext3_orphan_add(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001745 if (ret) {
1746 ext3_journal_stop(handle);
1747 goto out;
1748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 orphan = 1;
1750 ei->i_disksize = inode->i_size;
Jan Karabd1939d2008-02-06 01:40:21 -08001751 ext3_journal_stop(handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 }
1753 }
1754
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001755 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 offset, nr_segs,
Badari Pulavartyf91a2ad2006-03-26 01:38:04 -08001757 ext3_get_block, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Jan Karabd1939d2008-02-06 01:40:21 -08001759 if (orphan) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 int err;
1761
Jan Karabd1939d2008-02-06 01:40:21 -08001762 /* Credits for sb + inode write */
1763 handle = ext3_journal_start(inode, 2);
1764 if (IS_ERR(handle)) {
1765 /* This is really bad luck. We've written the data
1766 * but cannot extend i_size. Bail out and pretend
1767 * the write failed... */
1768 ret = PTR_ERR(handle);
1769 goto out;
1770 }
1771 if (inode->i_nlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 ext3_orphan_del(handle, inode);
Jan Karabd1939d2008-02-06 01:40:21 -08001773 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 loff_t end = offset + ret;
1775 if (end > inode->i_size) {
1776 ei->i_disksize = end;
1777 i_size_write(inode, end);
1778 /*
1779 * We're going to return a positive `ret'
1780 * here due to non-zero-length I/O, so there's
1781 * no way of reporting error returns from
1782 * ext3_mark_inode_dirty() to userspace. So
1783 * ignore it.
1784 */
1785 ext3_mark_inode_dirty(handle, inode);
1786 }
1787 }
1788 err = ext3_journal_stop(handle);
1789 if (ret == 0)
1790 ret = err;
1791 }
1792out:
1793 return ret;
1794}
1795
1796/*
1797 * Pages can be marked dirty completely asynchronously from ext3's journalling
1798 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1799 * much here because ->set_page_dirty is called under VFS locks. The page is
1800 * not necessarily locked.
1801 *
1802 * We cannot just dirty the page and leave attached buffers clean, because the
1803 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1804 * or jbddirty because all the journalling code will explode.
1805 *
1806 * So what we do is to mark the page "pending dirty" and next time writepage
1807 * is called, propagate that into the buffers appropriately.
1808 */
1809static int ext3_journalled_set_page_dirty(struct page *page)
1810{
1811 SetPageChecked(page);
1812 return __set_page_dirty_nobuffers(page);
1813}
1814
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001815static const struct address_space_operations ext3_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001816 .readpage = ext3_readpage,
1817 .readpages = ext3_readpages,
1818 .writepage = ext3_ordered_writepage,
1819 .sync_page = block_sync_page,
1820 .write_begin = ext3_write_begin,
1821 .write_end = ext3_ordered_write_end,
1822 .bmap = ext3_bmap,
1823 .invalidatepage = ext3_invalidatepage,
1824 .releasepage = ext3_releasepage,
1825 .direct_IO = ext3_direct_IO,
1826 .migratepage = buffer_migrate_page,
1827 .is_partially_uptodate = block_is_partially_uptodate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828};
1829
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001830static const struct address_space_operations ext3_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001831 .readpage = ext3_readpage,
1832 .readpages = ext3_readpages,
1833 .writepage = ext3_writeback_writepage,
1834 .sync_page = block_sync_page,
1835 .write_begin = ext3_write_begin,
1836 .write_end = ext3_writeback_write_end,
1837 .bmap = ext3_bmap,
1838 .invalidatepage = ext3_invalidatepage,
1839 .releasepage = ext3_releasepage,
1840 .direct_IO = ext3_direct_IO,
1841 .migratepage = buffer_migrate_page,
1842 .is_partially_uptodate = block_is_partially_uptodate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843};
1844
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001845static const struct address_space_operations ext3_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001846 .readpage = ext3_readpage,
1847 .readpages = ext3_readpages,
1848 .writepage = ext3_journalled_writepage,
1849 .sync_page = block_sync_page,
1850 .write_begin = ext3_write_begin,
1851 .write_end = ext3_journalled_write_end,
1852 .set_page_dirty = ext3_journalled_set_page_dirty,
1853 .bmap = ext3_bmap,
1854 .invalidatepage = ext3_invalidatepage,
1855 .releasepage = ext3_releasepage,
1856 .is_partially_uptodate = block_is_partially_uptodate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857};
1858
1859void ext3_set_aops(struct inode *inode)
1860{
1861 if (ext3_should_order_data(inode))
1862 inode->i_mapping->a_ops = &ext3_ordered_aops;
1863 else if (ext3_should_writeback_data(inode))
1864 inode->i_mapping->a_ops = &ext3_writeback_aops;
1865 else
1866 inode->i_mapping->a_ops = &ext3_journalled_aops;
1867}
1868
1869/*
1870 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1871 * up to the end of the block which corresponds to `from'.
1872 * This required during truncate. We need to physically zero the tail end
1873 * of that block so it doesn't yield old data if the file is later grown.
1874 */
1875static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1876 struct address_space *mapping, loff_t from)
1877{
Mingming Cao43d23f92006-06-25 05:48:07 -07001878 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1880 unsigned blocksize, iblock, length, pos;
1881 struct inode *inode = mapping->host;
1882 struct buffer_head *bh;
1883 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
1885 blocksize = inode->i_sb->s_blocksize;
1886 length = blocksize - (offset & (blocksize - 1));
1887 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1888
1889 /*
1890 * For "nobh" option, we can only work if we don't need to
1891 * read-in the page - otherwise we create buffers to do the IO.
1892 */
Badari Pulavartycd6ef842006-03-11 03:27:14 -08001893 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1894 ext3_should_writeback_data(inode) && PageUptodate(page)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001895 zero_user(page, offset, length);
Badari Pulavartycd6ef842006-03-11 03:27:14 -08001896 set_page_dirty(page);
1897 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 }
1899
1900 if (!page_has_buffers(page))
1901 create_empty_buffers(page, blocksize, 0);
1902
1903 /* Find the buffer that contains "offset" */
1904 bh = page_buffers(page);
1905 pos = blocksize;
1906 while (offset >= pos) {
1907 bh = bh->b_this_page;
1908 iblock++;
1909 pos += blocksize;
1910 }
1911
1912 err = 0;
1913 if (buffer_freed(bh)) {
1914 BUFFER_TRACE(bh, "freed: skip");
1915 goto unlock;
1916 }
1917
1918 if (!buffer_mapped(bh)) {
1919 BUFFER_TRACE(bh, "unmapped");
1920 ext3_get_block(inode, iblock, bh, 0);
1921 /* unmapped? It's a hole - nothing to do */
1922 if (!buffer_mapped(bh)) {
1923 BUFFER_TRACE(bh, "still unmapped");
1924 goto unlock;
1925 }
1926 }
1927
1928 /* Ok, it's mapped. Make sure it's up-to-date */
1929 if (PageUptodate(page))
1930 set_buffer_uptodate(bh);
1931
1932 if (!buffer_uptodate(bh)) {
1933 err = -EIO;
1934 ll_rw_block(READ, 1, &bh);
1935 wait_on_buffer(bh);
1936 /* Uhhuh. Read error. Complain and punt. */
1937 if (!buffer_uptodate(bh))
1938 goto unlock;
1939 }
1940
1941 if (ext3_should_journal_data(inode)) {
1942 BUFFER_TRACE(bh, "get write access");
1943 err = ext3_journal_get_write_access(handle, bh);
1944 if (err)
1945 goto unlock;
1946 }
1947
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001948 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 BUFFER_TRACE(bh, "zeroed end of block");
1950
1951 err = 0;
1952 if (ext3_should_journal_data(inode)) {
1953 err = ext3_journal_dirty_metadata(handle, bh);
1954 } else {
1955 if (ext3_should_order_data(inode))
1956 err = ext3_journal_dirty_data(handle, bh);
1957 mark_buffer_dirty(bh);
1958 }
1959
1960unlock:
1961 unlock_page(page);
1962 page_cache_release(page);
1963 return err;
1964}
1965
1966/*
1967 * Probably it should be a library function... search for first non-zero word
1968 * or memcmp with zero_page, whatever is better for particular architecture.
1969 * Linus?
1970 */
1971static inline int all_zeroes(__le32 *p, __le32 *q)
1972{
1973 while (p < q)
1974 if (*p++)
1975 return 0;
1976 return 1;
1977}
1978
1979/**
1980 * ext3_find_shared - find the indirect blocks for partial truncation.
1981 * @inode: inode in question
1982 * @depth: depth of the affected branch
1983 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1984 * @chain: place to store the pointers to partial indirect blocks
1985 * @top: place to the (detached) top of branch
1986 *
1987 * This is a helper function used by ext3_truncate().
1988 *
1989 * When we do truncate() we may have to clean the ends of several
1990 * indirect blocks but leave the blocks themselves alive. Block is
1991 * partially truncated if some data below the new i_size is refered
1992 * from it (and it is on the path to the first completely truncated
1993 * data block, indeed). We have to free the top of that path along
1994 * with everything to the right of the path. Since no allocation
1995 * past the truncation point is possible until ext3_truncate()
1996 * finishes, we may safely do the latter, but top of branch may
1997 * require special attention - pageout below the truncation point
1998 * might try to populate it.
1999 *
2000 * We atomically detach the top of branch from the tree, store the
2001 * block number of its root in *@top, pointers to buffer_heads of
2002 * partially truncated blocks - in @chain[].bh and pointers to
2003 * their last elements that should not be removed - in
2004 * @chain[].p. Return value is the pointer to last filled element
2005 * of @chain.
2006 *
2007 * The work left to caller to do the actual freeing of subtrees:
2008 * a) free the subtree starting from *@top
2009 * b) free the subtrees whose roots are stored in
2010 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2011 * c) free the subtrees growing from the inode past the @chain[0].
2012 * (no partially truncated stuff there). */
2013
Andrew Mortond6859bf2006-03-26 01:38:03 -08002014static Indirect *ext3_find_shared(struct inode *inode, int depth,
2015 int offsets[4], Indirect chain[4], __le32 *top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016{
2017 Indirect *partial, *p;
2018 int k, err;
2019
2020 *top = 0;
2021 /* Make k index the deepest non-null offest + 1 */
2022 for (k = depth; k > 1 && !offsets[k-1]; k--)
2023 ;
2024 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2025 /* Writer: pointers */
2026 if (!partial)
2027 partial = chain + k-1;
2028 /*
2029 * If the branch acquired continuation since we've looked at it -
2030 * fine, it should all survive and (new) top doesn't belong to us.
2031 */
2032 if (!partial->key && *partial->p)
2033 /* Writer: end */
2034 goto no_top;
2035 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2036 ;
2037 /*
2038 * OK, we've found the last block that must survive. The rest of our
2039 * branch should be detached before unlocking. However, if that rest
2040 * of branch is all ours and does not grow immediately from the inode
2041 * it's easier to cheat and just decrement partial->p.
2042 */
2043 if (p == chain + k - 1 && p > chain) {
2044 p->p--;
2045 } else {
2046 *top = *p->p;
2047 /* Nope, don't do this in ext3. Must leave the tree intact */
2048#if 0
2049 *p->p = 0;
2050#endif
2051 }
2052 /* Writer: end */
2053
Andrew Mortond6859bf2006-03-26 01:38:03 -08002054 while(partial > p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 brelse(partial->bh);
2056 partial--;
2057 }
2058no_top:
2059 return partial;
2060}
2061
2062/*
2063 * Zero a number of block pointers in either an inode or an indirect block.
2064 * If we restart the transaction we must again get write access to the
2065 * indirect block for further modification.
2066 *
2067 * We release `count' blocks on disk, but (last - first) may be greater
2068 * than `count' because there can be holes in there.
2069 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002070static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
Mingming Cao43d23f92006-06-25 05:48:07 -07002071 struct buffer_head *bh, ext3_fsblk_t block_to_free,
Andrew Mortond6859bf2006-03-26 01:38:03 -08002072 unsigned long count, __le32 *first, __le32 *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073{
2074 __le32 *p;
2075 if (try_to_extend_transaction(handle, inode)) {
2076 if (bh) {
2077 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2078 ext3_journal_dirty_metadata(handle, bh);
2079 }
2080 ext3_mark_inode_dirty(handle, inode);
2081 ext3_journal_test_restart(handle, inode);
2082 if (bh) {
2083 BUFFER_TRACE(bh, "retaking write access");
2084 ext3_journal_get_write_access(handle, bh);
2085 }
2086 }
2087
2088 /*
2089 * Any buffers which are on the journal will be in memory. We find
2090 * them on the hash table so journal_revoke() will run journal_forget()
2091 * on them. We've already detached each block from the file, so
2092 * bforget() in journal_forget() should be safe.
2093 *
2094 * AKPM: turn on bforget in journal_forget()!!!
2095 */
2096 for (p = first; p < last; p++) {
2097 u32 nr = le32_to_cpu(*p);
2098 if (nr) {
2099 struct buffer_head *bh;
2100
2101 *p = 0;
2102 bh = sb_find_get_block(inode->i_sb, nr);
2103 ext3_forget(handle, 0, inode, bh, nr);
2104 }
2105 }
2106
2107 ext3_free_blocks(handle, inode, block_to_free, count);
2108}
2109
2110/**
2111 * ext3_free_data - free a list of data blocks
2112 * @handle: handle for this transaction
2113 * @inode: inode we are dealing with
2114 * @this_bh: indirect buffer_head which contains *@first and *@last
2115 * @first: array of block numbers
2116 * @last: points immediately past the end of array
2117 *
2118 * We are freeing all blocks refered from that array (numbers are stored as
2119 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2120 *
2121 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2122 * blocks are contiguous then releasing them at one time will only affect one
2123 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2124 * actually use a lot of journal space.
2125 *
2126 * @this_bh will be %NULL if @first and @last point into the inode's direct
2127 * block pointers.
2128 */
2129static void ext3_free_data(handle_t *handle, struct inode *inode,
2130 struct buffer_head *this_bh,
2131 __le32 *first, __le32 *last)
2132{
Mingming Cao43d23f92006-06-25 05:48:07 -07002133 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002134 unsigned long count = 0; /* Number of blocks in the run */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2136 corresponding to
2137 block_to_free */
Mingming Cao43d23f92006-06-25 05:48:07 -07002138 ext3_fsblk_t nr; /* Current block # */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 __le32 *p; /* Pointer into inode/ind
2140 for current block */
2141 int err;
2142
2143 if (this_bh) { /* For indirect block */
2144 BUFFER_TRACE(this_bh, "get_write_access");
2145 err = ext3_journal_get_write_access(handle, this_bh);
2146 /* Important: if we can't update the indirect pointers
2147 * to the blocks, we can't free them. */
2148 if (err)
2149 return;
2150 }
2151
2152 for (p = first; p < last; p++) {
2153 nr = le32_to_cpu(*p);
2154 if (nr) {
2155 /* accumulate blocks to free if they're contiguous */
2156 if (count == 0) {
2157 block_to_free = nr;
2158 block_to_free_p = p;
2159 count = 1;
2160 } else if (nr == block_to_free + count) {
2161 count++;
2162 } else {
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002163 ext3_clear_blocks(handle, inode, this_bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 block_to_free,
2165 count, block_to_free_p, p);
2166 block_to_free = nr;
2167 block_to_free_p = p;
2168 count = 1;
2169 }
2170 }
2171 }
2172
2173 if (count > 0)
2174 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2175 count, block_to_free_p, p);
2176
2177 if (this_bh) {
2178 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
Duane Griffin3ccc3162008-07-25 01:46:26 -07002179
2180 /*
2181 * The buffer head should have an attached journal head at this
2182 * point. However, if the data is corrupted and an indirect
2183 * block pointed to itself, it would have been detached when
2184 * the block was cleared. Check for this instead of OOPSing.
2185 */
2186 if (bh2jh(this_bh))
2187 ext3_journal_dirty_metadata(handle, this_bh);
2188 else
2189 ext3_error(inode->i_sb, "ext3_free_data",
2190 "circular indirect block detected, "
2191 "inode=%lu, block=%llu",
2192 inode->i_ino,
2193 (unsigned long long)this_bh->b_blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 }
2195}
2196
2197/**
2198 * ext3_free_branches - free an array of branches
2199 * @handle: JBD handle for this transaction
2200 * @inode: inode we are dealing with
2201 * @parent_bh: the buffer_head which contains *@first and *@last
2202 * @first: array of block numbers
2203 * @last: pointer immediately past the end of array
2204 * @depth: depth of the branches to free
2205 *
2206 * We are freeing all blocks refered from these branches (numbers are
2207 * stored as little-endian 32-bit) and updating @inode->i_blocks
2208 * appropriately.
2209 */
2210static void ext3_free_branches(handle_t *handle, struct inode *inode,
2211 struct buffer_head *parent_bh,
2212 __le32 *first, __le32 *last, int depth)
2213{
Mingming Cao43d23f92006-06-25 05:48:07 -07002214 ext3_fsblk_t nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 __le32 *p;
2216
2217 if (is_handle_aborted(handle))
2218 return;
2219
2220 if (depth--) {
2221 struct buffer_head *bh;
2222 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2223 p = last;
2224 while (--p >= first) {
2225 nr = le32_to_cpu(*p);
2226 if (!nr)
2227 continue; /* A hole */
2228
2229 /* Go read the buffer for the next level down */
2230 bh = sb_bread(inode->i_sb, nr);
2231
2232 /*
2233 * A read failure? Report error and clear slot
2234 * (should be rare).
2235 */
2236 if (!bh) {
2237 ext3_error(inode->i_sb, "ext3_free_branches",
Eric Sandeeneee194e2006-09-27 01:49:30 -07002238 "Read failure, inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 inode->i_ino, nr);
2240 continue;
2241 }
2242
2243 /* This zaps the entire block. Bottom up. */
2244 BUFFER_TRACE(bh, "free child branches");
2245 ext3_free_branches(handle, inode, bh,
2246 (__le32*)bh->b_data,
2247 (__le32*)bh->b_data + addr_per_block,
2248 depth);
2249
2250 /*
2251 * We've probably journalled the indirect block several
2252 * times during the truncate. But it's no longer
2253 * needed and we now drop it from the transaction via
2254 * journal_revoke().
2255 *
2256 * That's easy if it's exclusively part of this
2257 * transaction. But if it's part of the committing
2258 * transaction then journal_forget() will simply
2259 * brelse() it. That means that if the underlying
2260 * block is reallocated in ext3_get_block(),
2261 * unmap_underlying_metadata() will find this block
2262 * and will try to get rid of it. damn, damn.
2263 *
2264 * If this block has already been committed to the
2265 * journal, a revoke record will be written. And
2266 * revoke records must be emitted *before* clearing
2267 * this block's bit in the bitmaps.
2268 */
2269 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2270
2271 /*
2272 * Everything below this this pointer has been
2273 * released. Now let this top-of-subtree go.
2274 *
2275 * We want the freeing of this indirect block to be
2276 * atomic in the journal with the updating of the
2277 * bitmap block which owns it. So make some room in
2278 * the journal.
2279 *
2280 * We zero the parent pointer *after* freeing its
2281 * pointee in the bitmaps, so if extend_transaction()
2282 * for some reason fails to put the bitmap changes and
2283 * the release into the same transaction, recovery
2284 * will merely complain about releasing a free block,
2285 * rather than leaking blocks.
2286 */
2287 if (is_handle_aborted(handle))
2288 return;
2289 if (try_to_extend_transaction(handle, inode)) {
2290 ext3_mark_inode_dirty(handle, inode);
2291 ext3_journal_test_restart(handle, inode);
2292 }
2293
2294 ext3_free_blocks(handle, inode, nr, 1);
2295
2296 if (parent_bh) {
2297 /*
2298 * The block which we have just freed is
2299 * pointed to by an indirect block: journal it
2300 */
2301 BUFFER_TRACE(parent_bh, "get_write_access");
2302 if (!ext3_journal_get_write_access(handle,
2303 parent_bh)){
2304 *p = 0;
2305 BUFFER_TRACE(parent_bh,
2306 "call ext3_journal_dirty_metadata");
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002307 ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 parent_bh);
2309 }
2310 }
2311 }
2312 } else {
2313 /* We have reached the bottom of the tree. */
2314 BUFFER_TRACE(parent_bh, "free data blocks");
2315 ext3_free_data(handle, inode, parent_bh, first, last);
2316 }
2317}
2318
Duane Griffinae76dd92008-07-25 01:46:23 -07002319int ext3_can_truncate(struct inode *inode)
2320{
2321 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2322 return 0;
2323 if (S_ISREG(inode->i_mode))
2324 return 1;
2325 if (S_ISDIR(inode->i_mode))
2326 return 1;
2327 if (S_ISLNK(inode->i_mode))
2328 return !ext3_inode_is_fast_symlink(inode);
2329 return 0;
2330}
2331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332/*
2333 * ext3_truncate()
2334 *
2335 * We block out ext3_get_block() block instantiations across the entire
2336 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2337 * simultaneously on behalf of the same inode.
2338 *
2339 * As we work through the truncate and commmit bits of it to the journal there
2340 * is one core, guiding principle: the file's tree must always be consistent on
2341 * disk. We must be able to restart the truncate after a crash.
2342 *
2343 * The file's tree may be transiently inconsistent in memory (although it
2344 * probably isn't), but whenever we close off and commit a journal transaction,
2345 * the contents of (the filesystem + the journal) must be consistent and
2346 * restartable. It's pretty simple, really: bottom up, right to left (although
2347 * left-to-right works OK too).
2348 *
2349 * Note that at recovery time, journal replay occurs *before* the restart of
2350 * truncate against the orphan inode list.
2351 *
2352 * The committed inode has the new, desired i_size (which is the same as
2353 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2354 * that this inode's truncate did not complete and it will again call
2355 * ext3_truncate() to have another go. So there will be instantiated blocks
2356 * to the right of the truncation point in a crashed ext3 filesystem. But
2357 * that's fine - as long as they are linked from the inode, the post-crash
2358 * ext3_truncate() run will find them and release them.
2359 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08002360void ext3_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
2362 handle_t *handle;
2363 struct ext3_inode_info *ei = EXT3_I(inode);
2364 __le32 *i_data = ei->i_data;
2365 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2366 struct address_space *mapping = inode->i_mapping;
2367 int offsets[4];
2368 Indirect chain[4];
2369 Indirect *partial;
2370 __le32 nr = 0;
2371 int n;
2372 long last_block;
2373 unsigned blocksize = inode->i_sb->s_blocksize;
2374 struct page *page;
2375
Duane Griffinae76dd92008-07-25 01:46:23 -07002376 if (!ext3_can_truncate(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 return;
2378
Theodore Ts'of7ab34e2009-04-03 01:34:35 -04002379 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
2380 ei->i_state |= EXT3_STATE_FLUSH_ON_CLOSE;
2381
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 /*
2383 * We have to lock the EOF page here, because lock_page() nests
2384 * outside journal_start().
2385 */
2386 if ((inode->i_size & (blocksize - 1)) == 0) {
2387 /* Block boundary? Nothing to do */
2388 page = NULL;
2389 } else {
2390 page = grab_cache_page(mapping,
2391 inode->i_size >> PAGE_CACHE_SHIFT);
2392 if (!page)
2393 return;
2394 }
2395
2396 handle = start_transaction(inode);
2397 if (IS_ERR(handle)) {
2398 if (page) {
2399 clear_highpage(page);
2400 flush_dcache_page(page);
2401 unlock_page(page);
2402 page_cache_release(page);
2403 }
2404 return; /* AKPM: return what? */
2405 }
2406
2407 last_block = (inode->i_size + blocksize-1)
2408 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2409
2410 if (page)
2411 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2412
2413 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2414 if (n == 0)
2415 goto out_stop; /* error */
2416
2417 /*
2418 * OK. This truncate is going to happen. We add the inode to the
2419 * orphan list, so that if this truncate spans multiple transactions,
2420 * and we crash, we will resume the truncate when the filesystem
2421 * recovers. It also marks the inode dirty, to catch the new size.
2422 *
2423 * Implication: the file must always be in a sane, consistent
2424 * truncatable state while each transaction commits.
2425 */
2426 if (ext3_orphan_add(handle, inode))
2427 goto out_stop;
2428
2429 /*
2430 * The orphan list entry will now protect us from any crash which
2431 * occurs before the truncate completes, so it is now safe to propagate
2432 * the new, shorter inode size (held for now in i_size) into the
2433 * on-disk inode. We do this via i_disksize, which is the value which
2434 * ext3 *really* writes onto the disk inode.
2435 */
2436 ei->i_disksize = inode->i_size;
2437
2438 /*
2439 * From here we block out all ext3_get_block() callers who want to
2440 * modify the block allocation tree.
2441 */
Arjan van de Ven97461512006-03-23 03:00:42 -08002442 mutex_lock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
2444 if (n == 1) { /* direct blocks */
2445 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2446 i_data + EXT3_NDIR_BLOCKS);
2447 goto do_indirects;
2448 }
2449
2450 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2451 /* Kill the top of shared branch (not detached) */
2452 if (nr) {
2453 if (partial == chain) {
2454 /* Shared branch grows from the inode */
2455 ext3_free_branches(handle, inode, NULL,
2456 &nr, &nr+1, (chain+n-1) - partial);
2457 *partial->p = 0;
2458 /*
2459 * We mark the inode dirty prior to restart,
2460 * and prior to stop. No need for it here.
2461 */
2462 } else {
2463 /* Shared branch grows from an indirect block */
2464 BUFFER_TRACE(partial->bh, "get_write_access");
2465 ext3_free_branches(handle, inode, partial->bh,
2466 partial->p,
2467 partial->p+1, (chain+n-1) - partial);
2468 }
2469 }
2470 /* Clear the ends of indirect blocks on the shared branch */
2471 while (partial > chain) {
2472 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2473 (__le32*)partial->bh->b_data+addr_per_block,
2474 (chain+n-1) - partial);
2475 BUFFER_TRACE(partial->bh, "call brelse");
2476 brelse (partial->bh);
2477 partial--;
2478 }
2479do_indirects:
2480 /* Kill the remaining (whole) subtrees */
2481 switch (offsets[0]) {
Andrew Mortond6859bf2006-03-26 01:38:03 -08002482 default:
2483 nr = i_data[EXT3_IND_BLOCK];
2484 if (nr) {
2485 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2486 i_data[EXT3_IND_BLOCK] = 0;
2487 }
2488 case EXT3_IND_BLOCK:
2489 nr = i_data[EXT3_DIND_BLOCK];
2490 if (nr) {
2491 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2492 i_data[EXT3_DIND_BLOCK] = 0;
2493 }
2494 case EXT3_DIND_BLOCK:
2495 nr = i_data[EXT3_TIND_BLOCK];
2496 if (nr) {
2497 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2498 i_data[EXT3_TIND_BLOCK] = 0;
2499 }
2500 case EXT3_TIND_BLOCK:
2501 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 }
2503
2504 ext3_discard_reservation(inode);
2505
Arjan van de Ven97461512006-03-23 03:00:42 -08002506 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2508 ext3_mark_inode_dirty(handle, inode);
2509
Andrew Mortond6859bf2006-03-26 01:38:03 -08002510 /*
2511 * In a multi-transaction truncate, we only make the final transaction
2512 * synchronous
2513 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 if (IS_SYNC(inode))
2515 handle->h_sync = 1;
2516out_stop:
2517 /*
2518 * If this was a simple ftruncate(), and the file will remain alive
2519 * then we need to clear up the orphan record which we created above.
2520 * However, if this was a real unlink then we were called by
2521 * ext3_delete_inode(), and we allow that function to clean up the
2522 * orphan info for us.
2523 */
2524 if (inode->i_nlink)
2525 ext3_orphan_del(handle, inode);
2526
2527 ext3_journal_stop(handle);
2528}
2529
Mingming Cao43d23f92006-06-25 05:48:07 -07002530static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 unsigned long ino, struct ext3_iloc *iloc)
2532{
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002533 unsigned long block_group;
Mingming Cao43d23f92006-06-25 05:48:07 -07002534 unsigned long offset;
2535 ext3_fsblk_t block;
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002536 struct ext3_group_desc *gdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
Neil Brown2ccb48e2006-07-30 03:03:01 -07002538 if (!ext3_valid_inum(sb, ino)) {
2539 /*
2540 * This error is already checked for in namei.c unless we are
2541 * looking at an NFS filehandle, in which case no error
2542 * report is needed
2543 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 return 0;
2545 }
Neil Brown2ccb48e2006-07-30 03:03:01 -07002546
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002548 gdp = ext3_get_group_desc(sb, block_group, NULL);
2549 if (!gdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 /*
2552 * Figure out the offset within the block group inode table
2553 */
2554 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2555 EXT3_INODE_SIZE(sb);
Akinobu Mitae0e369a2008-04-28 02:16:08 -07002556 block = le32_to_cpu(gdp->bg_inode_table) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2558
2559 iloc->block_group = block_group;
2560 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2561 return block;
2562}
2563
2564/*
2565 * ext3_get_inode_loc returns with an extra refcount against the inode's
2566 * underlying buffer_head on success. If 'in_mem' is true, we have all
2567 * data in memory that is needed to recreate the on-disk version of this
2568 * inode.
2569 */
2570static int __ext3_get_inode_loc(struct inode *inode,
2571 struct ext3_iloc *iloc, int in_mem)
2572{
Mingming Cao43d23f92006-06-25 05:48:07 -07002573 ext3_fsblk_t block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 struct buffer_head *bh;
2575
2576 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2577 if (!block)
2578 return -EIO;
2579
2580 bh = sb_getblk(inode->i_sb, block);
2581 if (!bh) {
2582 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2583 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002584 "inode=%lu, block="E3FSBLK,
2585 inode->i_ino, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 return -EIO;
2587 }
2588 if (!buffer_uptodate(bh)) {
2589 lock_buffer(bh);
Hidehiro Kawai95450f52008-07-25 01:46:24 -07002590
2591 /*
2592 * If the buffer has the write error flag, we have failed
2593 * to write out another inode in the same block. In this
2594 * case, we don't have to read the block because we may
2595 * read the old inode data successfully.
2596 */
2597 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2598 set_buffer_uptodate(bh);
2599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 if (buffer_uptodate(bh)) {
2601 /* someone brought it uptodate while we waited */
2602 unlock_buffer(bh);
2603 goto has_buffer;
2604 }
2605
2606 /*
2607 * If we have all information of the inode in memory and this
2608 * is the only valid inode in the block, we need not read the
2609 * block.
2610 */
2611 if (in_mem) {
2612 struct buffer_head *bitmap_bh;
2613 struct ext3_group_desc *desc;
2614 int inodes_per_buffer;
2615 int inode_offset, i;
2616 int block_group;
2617 int start;
2618
2619 block_group = (inode->i_ino - 1) /
2620 EXT3_INODES_PER_GROUP(inode->i_sb);
2621 inodes_per_buffer = bh->b_size /
2622 EXT3_INODE_SIZE(inode->i_sb);
2623 inode_offset = ((inode->i_ino - 1) %
2624 EXT3_INODES_PER_GROUP(inode->i_sb));
2625 start = inode_offset & ~(inodes_per_buffer - 1);
2626
2627 /* Is the inode bitmap in cache? */
2628 desc = ext3_get_group_desc(inode->i_sb,
2629 block_group, NULL);
2630 if (!desc)
2631 goto make_io;
2632
2633 bitmap_bh = sb_getblk(inode->i_sb,
2634 le32_to_cpu(desc->bg_inode_bitmap));
2635 if (!bitmap_bh)
2636 goto make_io;
2637
2638 /*
2639 * If the inode bitmap isn't in cache then the
2640 * optimisation may end up performing two reads instead
2641 * of one, so skip it.
2642 */
2643 if (!buffer_uptodate(bitmap_bh)) {
2644 brelse(bitmap_bh);
2645 goto make_io;
2646 }
2647 for (i = start; i < start + inodes_per_buffer; i++) {
2648 if (i == inode_offset)
2649 continue;
2650 if (ext3_test_bit(i, bitmap_bh->b_data))
2651 break;
2652 }
2653 brelse(bitmap_bh);
2654 if (i == start + inodes_per_buffer) {
2655 /* all other inodes are free, so skip I/O */
2656 memset(bh->b_data, 0, bh->b_size);
2657 set_buffer_uptodate(bh);
2658 unlock_buffer(bh);
2659 goto has_buffer;
2660 }
2661 }
2662
2663make_io:
2664 /*
2665 * There are other valid inodes in the buffer, this inode
2666 * has in-inode xattrs, or we don't have this inode in memory.
2667 * Read the block from disk.
2668 */
2669 get_bh(bh);
2670 bh->b_end_io = end_buffer_read_sync;
Jens Axboecaa38fb2006-07-23 01:41:26 +02002671 submit_bh(READ_META, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 wait_on_buffer(bh);
2673 if (!buffer_uptodate(bh)) {
2674 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2675 "unable to read inode block - "
Mingming Cao43d23f92006-06-25 05:48:07 -07002676 "inode=%lu, block="E3FSBLK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 inode->i_ino, block);
2678 brelse(bh);
2679 return -EIO;
2680 }
2681 }
2682has_buffer:
2683 iloc->bh = bh;
2684 return 0;
2685}
2686
2687int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2688{
2689 /* We have all inode data except xattrs in memory here. */
2690 return __ext3_get_inode_loc(inode, iloc,
2691 !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR));
2692}
2693
2694void ext3_set_inode_flags(struct inode *inode)
2695{
2696 unsigned int flags = EXT3_I(inode)->i_flags;
2697
2698 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2699 if (flags & EXT3_SYNC_FL)
2700 inode->i_flags |= S_SYNC;
2701 if (flags & EXT3_APPEND_FL)
2702 inode->i_flags |= S_APPEND;
2703 if (flags & EXT3_IMMUTABLE_FL)
2704 inode->i_flags |= S_IMMUTABLE;
2705 if (flags & EXT3_NOATIME_FL)
2706 inode->i_flags |= S_NOATIME;
2707 if (flags & EXT3_DIRSYNC_FL)
2708 inode->i_flags |= S_DIRSYNC;
2709}
2710
Jan Kara28be5ab2007-05-08 00:30:33 -07002711/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2712void ext3_get_inode_flags(struct ext3_inode_info *ei)
2713{
2714 unsigned int flags = ei->vfs_inode.i_flags;
2715
2716 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2717 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2718 if (flags & S_SYNC)
2719 ei->i_flags |= EXT3_SYNC_FL;
2720 if (flags & S_APPEND)
2721 ei->i_flags |= EXT3_APPEND_FL;
2722 if (flags & S_IMMUTABLE)
2723 ei->i_flags |= EXT3_IMMUTABLE_FL;
2724 if (flags & S_NOATIME)
2725 ei->i_flags |= EXT3_NOATIME_FL;
2726 if (flags & S_DIRSYNC)
2727 ei->i_flags |= EXT3_DIRSYNC_FL;
2728}
2729
David Howells473043d2008-02-07 00:15:36 -08002730struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731{
2732 struct ext3_iloc iloc;
2733 struct ext3_inode *raw_inode;
David Howells473043d2008-02-07 00:15:36 -08002734 struct ext3_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 struct buffer_head *bh;
David Howells473043d2008-02-07 00:15:36 -08002736 struct inode *inode;
2737 long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 int block;
2739
David Howells473043d2008-02-07 00:15:36 -08002740 inode = iget_locked(sb, ino);
2741 if (!inode)
2742 return ERR_PTR(-ENOMEM);
2743 if (!(inode->i_state & I_NEW))
2744 return inode;
2745
2746 ei = EXT3_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747#ifdef CONFIG_EXT3_FS_POSIX_ACL
2748 ei->i_acl = EXT3_ACL_NOT_CACHED;
2749 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2750#endif
2751 ei->i_block_alloc_info = NULL;
2752
David Howells473043d2008-02-07 00:15:36 -08002753 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2754 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 goto bad_inode;
2756 bh = iloc.bh;
2757 raw_inode = ext3_raw_inode(&iloc);
2758 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2759 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2760 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2761 if(!(test_opt (inode->i_sb, NO_UID32))) {
2762 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2763 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2764 }
2765 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2766 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002767 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2768 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2769 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2771
2772 ei->i_state = 0;
2773 ei->i_dir_start_lookup = 0;
2774 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2775 /* We now have enough fields to check if the inode was active or not.
2776 * This is needed because nfsd might try to access dead inodes
2777 * the test is that same one that e2fsck uses
2778 * NeilBrown 1999oct15
2779 */
2780 if (inode->i_nlink == 0) {
2781 if (inode->i_mode == 0 ||
2782 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2783 /* this inode is deleted */
2784 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002785 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 goto bad_inode;
2787 }
2788 /* The only unlinked inodes we let through here have
2789 * valid i_mode and are being read by the orphan
2790 * recovery code: that's fine, we're about to complete
2791 * the process of deleting those. */
2792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2794 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2795#ifdef EXT3_FRAGMENTS
2796 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2797 ei->i_frag_no = raw_inode->i_frag;
2798 ei->i_frag_size = raw_inode->i_fsize;
2799#endif
2800 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2801 if (!S_ISREG(inode->i_mode)) {
2802 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2803 } else {
2804 inode->i_size |=
2805 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2806 }
2807 ei->i_disksize = inode->i_size;
2808 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2809 ei->i_block_group = iloc.block_group;
2810 /*
2811 * NOTE! The in-memory inode i_data array is in little-endian order
2812 * even on big-endian machines: we do NOT byteswap the block numbers!
2813 */
2814 for (block = 0; block < EXT3_N_BLOCKS; block++)
2815 ei->i_data[block] = raw_inode->i_block[block];
2816 INIT_LIST_HEAD(&ei->i_orphan);
2817
2818 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2819 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2820 /*
2821 * When mke2fs creates big inodes it does not zero out
2822 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2823 * so ignore those first few inodes.
2824 */
2825 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2826 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07002827 EXT3_INODE_SIZE(inode->i_sb)) {
2828 brelse (bh);
David Howells473043d2008-02-07 00:15:36 -08002829 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 goto bad_inode;
Kirill Korotaeve4a10a32007-06-23 17:16:48 -07002831 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 if (ei->i_extra_isize == 0) {
2833 /* The extra space is currently unused. Use it. */
2834 ei->i_extra_isize = sizeof(struct ext3_inode) -
2835 EXT3_GOOD_OLD_INODE_SIZE;
2836 } else {
2837 __le32 *magic = (void *)raw_inode +
2838 EXT3_GOOD_OLD_INODE_SIZE +
2839 ei->i_extra_isize;
2840 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2841 ei->i_state |= EXT3_STATE_XATTR;
2842 }
2843 } else
2844 ei->i_extra_isize = 0;
2845
2846 if (S_ISREG(inode->i_mode)) {
2847 inode->i_op = &ext3_file_inode_operations;
2848 inode->i_fop = &ext3_file_operations;
2849 ext3_set_aops(inode);
2850 } else if (S_ISDIR(inode->i_mode)) {
2851 inode->i_op = &ext3_dir_inode_operations;
2852 inode->i_fop = &ext3_dir_operations;
2853 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffinb5ed3112008-12-19 20:47:14 +00002854 if (ext3_inode_is_fast_symlink(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 inode->i_op = &ext3_fast_symlink_inode_operations;
Duane Griffinb5ed3112008-12-19 20:47:14 +00002856 nd_terminate_link(ei->i_data, inode->i_size,
2857 sizeof(ei->i_data) - 1);
2858 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 inode->i_op = &ext3_symlink_inode_operations;
2860 ext3_set_aops(inode);
2861 }
2862 } else {
2863 inode->i_op = &ext3_special_inode_operations;
2864 if (raw_inode->i_block[0])
2865 init_special_inode(inode, inode->i_mode,
2866 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002867 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 init_special_inode(inode, inode->i_mode,
2869 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2870 }
2871 brelse (iloc.bh);
2872 ext3_set_inode_flags(inode);
David Howells473043d2008-02-07 00:15:36 -08002873 unlock_new_inode(inode);
2874 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
2876bad_inode:
David Howells473043d2008-02-07 00:15:36 -08002877 iget_failed(inode);
2878 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879}
2880
2881/*
2882 * Post the struct inode info into an on-disk inode location in the
2883 * buffer-cache. This gobbles the caller's reference to the
2884 * buffer_head in the inode location struct.
2885 *
2886 * The caller must have write access to iloc->bh.
2887 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002888static int ext3_do_update_inode(handle_t *handle,
2889 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 struct ext3_iloc *iloc)
2891{
2892 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2893 struct ext3_inode_info *ei = EXT3_I(inode);
2894 struct buffer_head *bh = iloc->bh;
2895 int err = 0, rc, block;
2896
2897 /* For fields not not tracking in the in-memory inode,
2898 * initialise them to zero for new inodes. */
2899 if (ei->i_state & EXT3_STATE_NEW)
2900 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2901
Jan Kara28be5ab2007-05-08 00:30:33 -07002902 ext3_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2904 if(!(test_opt(inode->i_sb, NO_UID32))) {
2905 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2906 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2907/*
2908 * Fix up interoperability with old kernels. Otherwise, old inodes get
2909 * re-used with the upper 16 bits of the uid/gid intact
2910 */
2911 if(!ei->i_dtime) {
2912 raw_inode->i_uid_high =
2913 cpu_to_le16(high_16_bits(inode->i_uid));
2914 raw_inode->i_gid_high =
2915 cpu_to_le16(high_16_bits(inode->i_gid));
2916 } else {
2917 raw_inode->i_uid_high = 0;
2918 raw_inode->i_gid_high = 0;
2919 }
2920 } else {
2921 raw_inode->i_uid_low =
2922 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2923 raw_inode->i_gid_low =
2924 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2925 raw_inode->i_uid_high = 0;
2926 raw_inode->i_gid_high = 0;
2927 }
2928 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2929 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2930 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2931 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2932 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2933 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2934 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2935 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2936#ifdef EXT3_FRAGMENTS
2937 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2938 raw_inode->i_frag = ei->i_frag_no;
2939 raw_inode->i_fsize = ei->i_frag_size;
2940#endif
2941 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2942 if (!S_ISREG(inode->i_mode)) {
2943 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2944 } else {
2945 raw_inode->i_size_high =
2946 cpu_to_le32(ei->i_disksize >> 32);
2947 if (ei->i_disksize > 0x7fffffffULL) {
2948 struct super_block *sb = inode->i_sb;
2949 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2950 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2951 EXT3_SB(sb)->s_es->s_rev_level ==
2952 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2953 /* If this is the first large file
2954 * created, add a flag to the superblock.
2955 */
2956 err = ext3_journal_get_write_access(handle,
2957 EXT3_SB(sb)->s_sbh);
2958 if (err)
2959 goto out_brelse;
2960 ext3_update_dynamic_rev(sb);
2961 EXT3_SET_RO_COMPAT_FEATURE(sb,
2962 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2963 sb->s_dirt = 1;
2964 handle->h_sync = 1;
2965 err = ext3_journal_dirty_metadata(handle,
2966 EXT3_SB(sb)->s_sbh);
2967 }
2968 }
2969 }
2970 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2971 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2972 if (old_valid_dev(inode->i_rdev)) {
2973 raw_inode->i_block[0] =
2974 cpu_to_le32(old_encode_dev(inode->i_rdev));
2975 raw_inode->i_block[1] = 0;
2976 } else {
2977 raw_inode->i_block[0] = 0;
2978 raw_inode->i_block[1] =
2979 cpu_to_le32(new_encode_dev(inode->i_rdev));
2980 raw_inode->i_block[2] = 0;
2981 }
2982 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2983 raw_inode->i_block[block] = ei->i_data[block];
2984
Andreas Gruenbacherff87b372005-07-07 17:57:00 -07002985 if (ei->i_extra_isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2987
2988 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2989 rc = ext3_journal_dirty_metadata(handle, bh);
2990 if (!err)
2991 err = rc;
2992 ei->i_state &= ~EXT3_STATE_NEW;
2993
2994out_brelse:
2995 brelse (bh);
2996 ext3_std_error(inode->i_sb, err);
2997 return err;
2998}
2999
3000/*
3001 * ext3_write_inode()
3002 *
3003 * We are called from a few places:
3004 *
3005 * - Within generic_file_write() for O_SYNC files.
3006 * Here, there will be no transaction running. We wait for any running
3007 * trasnaction to commit.
3008 *
3009 * - Within sys_sync(), kupdate and such.
3010 * We wait on commit, if tol to.
3011 *
3012 * - Within prune_icache() (PF_MEMALLOC == true)
3013 * Here we simply return. We can't afford to block kswapd on the
3014 * journal commit.
3015 *
3016 * In all cases it is actually safe for us to return without doing anything,
3017 * because the inode has been copied into a raw inode buffer in
3018 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3019 * knfsd.
3020 *
3021 * Note that we are absolutely dependent upon all inode dirtiers doing the
3022 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3023 * which we are interested.
3024 *
3025 * It would be a bug for them to not do this. The code:
3026 *
3027 * mark_inode_dirty(inode)
3028 * stuff();
3029 * inode->i_size = expr;
3030 *
3031 * is in error because a kswapd-driven write_inode() could occur while
3032 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3033 * will no longer be on the superblock's dirty inode list.
3034 */
3035int ext3_write_inode(struct inode *inode, int wait)
3036{
3037 if (current->flags & PF_MEMALLOC)
3038 return 0;
3039
3040 if (ext3_journal_current_handle()) {
Jose R. Santos9ad163a2007-10-18 23:39:23 -07003041 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 dump_stack();
3043 return -EIO;
3044 }
3045
3046 if (!wait)
3047 return 0;
3048
3049 return ext3_force_commit(inode->i_sb);
3050}
3051
3052/*
3053 * ext3_setattr()
3054 *
3055 * Called from notify_change.
3056 *
3057 * We want to trap VFS attempts to truncate the file as soon as
3058 * possible. In particular, we want to make sure that when the VFS
3059 * shrinks i_size, we put the inode on the orphan list and modify
3060 * i_disksize immediately, so that during the subsequent flushing of
3061 * dirty pages and freeing of disk blocks, we can guarantee that any
3062 * commit will leave the blocks being flushed in an unused state on
3063 * disk. (On recovery, the inode will get truncated and the blocks will
3064 * be freed, so we have a strong guarantee that no future commit will
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003065 * leave these blocks visible to the user.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 *
3067 * Called with inode->sem down.
3068 */
3069int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3070{
3071 struct inode *inode = dentry->d_inode;
3072 int error, rc = 0;
3073 const unsigned int ia_valid = attr->ia_valid;
3074
3075 error = inode_change_ok(inode, attr);
3076 if (error)
3077 return error;
3078
3079 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3080 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3081 handle_t *handle;
3082
3083 /* (user+group)*(old+new) structure, inode write (sb,
3084 * inode block, ? - but truncate inode update has it) */
Jan Kara1f545872005-06-23 22:01:04 -07003085 handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+
3086 EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 if (IS_ERR(handle)) {
3088 error = PTR_ERR(handle);
3089 goto err_out;
3090 }
Jan Kara81a05222009-01-26 16:58:01 +01003091 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 if (error) {
3093 ext3_journal_stop(handle);
3094 return error;
3095 }
3096 /* Update corresponding info in inode so that everything is in
3097 * one transaction */
3098 if (attr->ia_valid & ATTR_UID)
3099 inode->i_uid = attr->ia_uid;
3100 if (attr->ia_valid & ATTR_GID)
3101 inode->i_gid = attr->ia_gid;
3102 error = ext3_mark_inode_dirty(handle, inode);
3103 ext3_journal_stop(handle);
3104 }
3105
3106 if (S_ISREG(inode->i_mode) &&
3107 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3108 handle_t *handle;
3109
3110 handle = ext3_journal_start(inode, 3);
3111 if (IS_ERR(handle)) {
3112 error = PTR_ERR(handle);
3113 goto err_out;
3114 }
3115
3116 error = ext3_orphan_add(handle, inode);
3117 EXT3_I(inode)->i_disksize = attr->ia_size;
3118 rc = ext3_mark_inode_dirty(handle, inode);
3119 if (!error)
3120 error = rc;
3121 ext3_journal_stop(handle);
3122 }
3123
3124 rc = inode_setattr(inode, attr);
3125
3126 /* If inode_setattr's call to ext3_truncate failed to get a
3127 * transaction handle at all, we need to clean up the in-core
3128 * orphan list manually. */
3129 if (inode->i_nlink)
3130 ext3_orphan_del(NULL, inode);
3131
3132 if (!rc && (ia_valid & ATTR_MODE))
3133 rc = ext3_acl_chmod(inode);
3134
3135err_out:
3136 ext3_std_error(inode->i_sb, error);
3137 if (!error)
3138 error = rc;
3139 return error;
3140}
3141
3142
3143/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003144 * How many blocks doth make a writepage()?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 *
3146 * With N blocks per page, it may be:
3147 * N data blocks
3148 * 2 indirect block
3149 * 2 dindirect
3150 * 1 tindirect
3151 * N+5 bitmap blocks (from the above)
3152 * N+5 group descriptor summary blocks
3153 * 1 inode block
3154 * 1 superblock.
3155 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3156 *
3157 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3158 *
3159 * With ordered or writeback data it's the same, less the N data blocks.
3160 *
3161 * If the inode's direct blocks can hold an integral number of pages then a
3162 * page cannot straddle two indirect blocks, and we can only touch one indirect
3163 * and dindirect block, and the "5" above becomes "3".
3164 *
3165 * This still overestimates under most circumstances. If we were to pass the
3166 * start and end offsets in here as well we could do block_to_path() on each
3167 * block and work out the exact number of indirects which are touched. Pah.
3168 */
3169
3170static int ext3_writepage_trans_blocks(struct inode *inode)
3171{
3172 int bpp = ext3_journal_blocks_per_page(inode);
3173 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3174 int ret;
3175
3176 if (ext3_should_journal_data(inode))
3177 ret = 3 * (bpp + indirects) + 2;
3178 else
3179 ret = 2 * (bpp + indirects) + 2;
3180
3181#ifdef CONFIG_QUOTA
Jan Kara81a05222009-01-26 16:58:01 +01003182 /* We know that structure was already allocated during vfs_dq_init so
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 * we will be updating only the data blocks + inodes */
Jan Kara1f545872005-06-23 22:01:04 -07003184 ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185#endif
3186
3187 return ret;
3188}
3189
3190/*
3191 * The caller must have previously called ext3_reserve_inode_write().
3192 * Give this, we know that the caller already has write access to iloc->bh.
3193 */
3194int ext3_mark_iloc_dirty(handle_t *handle,
3195 struct inode *inode, struct ext3_iloc *iloc)
3196{
3197 int err = 0;
3198
3199 /* the do_update_inode consumes one bh->b_count */
3200 get_bh(iloc->bh);
3201
3202 /* ext3_do_update_inode() does journal_dirty_metadata */
3203 err = ext3_do_update_inode(handle, inode, iloc);
3204 put_bh(iloc->bh);
3205 return err;
3206}
3207
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003208/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 * On success, We end up with an outstanding reference count against
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003210 * iloc->bh. This _must_ be cleaned up later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 */
3212
3213int
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003214ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 struct ext3_iloc *iloc)
3216{
3217 int err = 0;
3218 if (handle) {
3219 err = ext3_get_inode_loc(inode, iloc);
3220 if (!err) {
3221 BUFFER_TRACE(iloc->bh, "get_write_access");
3222 err = ext3_journal_get_write_access(handle, iloc->bh);
3223 if (err) {
3224 brelse(iloc->bh);
3225 iloc->bh = NULL;
3226 }
3227 }
3228 }
3229 ext3_std_error(inode->i_sb, err);
3230 return err;
3231}
3232
3233/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003234 * What we do here is to mark the in-core inode as clean with respect to inode
3235 * dirtiness (it may still be data-dirty).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 * This means that the in-core inode may be reaped by prune_icache
3237 * without having to perform any I/O. This is a very good thing,
3238 * because *any* task may call prune_icache - even ones which
3239 * have a transaction open against a different journal.
3240 *
3241 * Is this cheating? Not really. Sure, we haven't written the
3242 * inode out, but prune_icache isn't a user-visible syncing function.
3243 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3244 * we start and wait on commits.
3245 *
3246 * Is this efficient/effective? Well, we're being nice to the system
3247 * by cleaning up our inodes proactively so they can be reaped
3248 * without I/O. But we are potentially leaving up to five seconds'
3249 * worth of inodes floating about which prune_icache wants us to
3250 * write out. One way to fix that would be to get prune_icache()
3251 * to do a write_super() to free up some memory. It has the desired
3252 * effect.
3253 */
3254int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3255{
3256 struct ext3_iloc iloc;
3257 int err;
3258
3259 might_sleep();
3260 err = ext3_reserve_inode_write(handle, inode, &iloc);
3261 if (!err)
3262 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3263 return err;
3264}
3265
3266/*
Andrew Mortond6859bf2006-03-26 01:38:03 -08003267 * ext3_dirty_inode() is called from __mark_inode_dirty()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 *
3269 * We're really interested in the case where a file is being extended.
3270 * i_size has been changed by generic_commit_write() and we thus need
3271 * to include the updated inode in the current transaction.
3272 *
Jan Kara81a05222009-01-26 16:58:01 +01003273 * Also, vfs_dq_alloc_space() will always dirty the inode when blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 * are allocated to the file.
3275 *
3276 * If the inode is marked synchronous, we don't honour that here - doing
3277 * so would cause a commit on atime updates, which we don't bother doing.
3278 * We handle synchronous inodes at the highest possible level.
3279 */
3280void ext3_dirty_inode(struct inode *inode)
3281{
3282 handle_t *current_handle = ext3_journal_current_handle();
3283 handle_t *handle;
3284
3285 handle = ext3_journal_start(inode, 2);
3286 if (IS_ERR(handle))
3287 goto out;
3288 if (current_handle &&
3289 current_handle->h_transaction != handle->h_transaction) {
3290 /* This task has a transaction open against a different fs */
3291 printk(KERN_EMERG "%s: transactions do not match!\n",
Harvey Harrisone05b6b52008-04-28 02:16:15 -07003292 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 } else {
3294 jbd_debug(5, "marking dirty. outer handle=%p\n",
3295 current_handle);
3296 ext3_mark_inode_dirty(handle, inode);
3297 }
3298 ext3_journal_stop(handle);
3299out:
3300 return;
3301}
3302
Andrew Mortond6859bf2006-03-26 01:38:03 -08003303#if 0
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003304/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305 * Bind an inode's backing buffer_head into this transaction, to prevent
3306 * it from being flushed to disk early. Unlike
3307 * ext3_reserve_inode_write, this leaves behind no bh reference and
3308 * returns no iloc structure, so the caller needs to repeat the iloc
3309 * lookup to mark the inode dirty later.
3310 */
Andrew Mortond6859bf2006-03-26 01:38:03 -08003311static int ext3_pin_inode(handle_t *handle, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312{
3313 struct ext3_iloc iloc;
3314
3315 int err = 0;
3316 if (handle) {
3317 err = ext3_get_inode_loc(inode, &iloc);
3318 if (!err) {
3319 BUFFER_TRACE(iloc.bh, "get_write_access");
3320 err = journal_get_write_access(handle, iloc.bh);
3321 if (!err)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003322 err = ext3_journal_dirty_metadata(handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 iloc.bh);
3324 brelse(iloc.bh);
3325 }
3326 }
3327 ext3_std_error(inode->i_sb, err);
3328 return err;
3329}
3330#endif
3331
3332int ext3_change_inode_journal_flag(struct inode *inode, int val)
3333{
3334 journal_t *journal;
3335 handle_t *handle;
3336 int err;
3337
3338 /*
3339 * We have to be very careful here: changing a data block's
3340 * journaling status dynamically is dangerous. If we write a
3341 * data block to the journal, change the status and then delete
3342 * that block, we risk forgetting to revoke the old log record
3343 * from the journal and so a subsequent replay can corrupt data.
3344 * So, first we make sure that the journal is empty and that
3345 * nobody is changing anything.
3346 */
3347
3348 journal = EXT3_JOURNAL(inode);
Dave Hansene3a68e32007-07-15 23:41:14 -07003349 if (is_journal_aborted(journal))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 return -EROFS;
3351
3352 journal_lock_updates(journal);
3353 journal_flush(journal);
3354
3355 /*
3356 * OK, there are no updates running now, and all cached data is
3357 * synced to disk. We are now in a completely consistent state
3358 * which doesn't have anything in the journal, and we know that
3359 * no filesystem updates are running, so it is safe to modify
3360 * the inode's in-core data-journaling state flag now.
3361 */
3362
3363 if (val)
3364 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3365 else
3366 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3367 ext3_set_aops(inode);
3368
3369 journal_unlock_updates(journal);
3370
3371 /* Finally we can mark the inode as dirty. */
3372
3373 handle = ext3_journal_start(inode, 1);
3374 if (IS_ERR(handle))
3375 return PTR_ERR(handle);
3376
3377 err = ext3_mark_inode_dirty(handle, inode);
3378 handle->h_sync = 1;
3379 ext3_journal_stop(handle);
3380 ext3_std_error(inode->i_sb, err);
3381
3382 return err;
3383}