1 |
1275 |
phoenix |
/*
|
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
|
16 |
|
|
* (sct@redhat.com), 1993, 1998
|
17 |
|
|
* 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
|
20 |
|
|
* (jj@sunsite.ms.mff.cuni.cz)
|
21 |
|
|
*
|
22 |
|
|
* Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
|
23 |
|
|
*/
|
24 |
|
|
|
25 |
|
|
#include <linux/fs.h>
|
26 |
|
|
#include <linux/sched.h>
|
27 |
|
|
#include <linux/ext3_jbd.h>
|
28 |
|
|
#include <linux/jbd.h>
|
29 |
|
|
#include <linux/locks.h>
|
30 |
|
|
#include <linux/smp_lock.h>
|
31 |
|
|
#include <linux/highuid.h>
|
32 |
|
|
#include <linux/quotaops.h>
|
33 |
|
|
#include <linux/module.h>
|
34 |
|
|
|
35 |
|
|
/*
|
36 |
|
|
* SEARCH_FROM_ZERO forces each block allocation to search from the start
|
37 |
|
|
* of the filesystem. This is to force rapid reallocation of recently-freed
|
38 |
|
|
* blocks. The file fragmentation is horrendous.
|
39 |
|
|
*/
|
40 |
|
|
#undef SEARCH_FROM_ZERO
|
41 |
|
|
|
42 |
|
|
/*
|
43 |
|
|
* Test whether an inode is a fast symlink.
|
44 |
|
|
*/
|
45 |
|
|
static inline int ext3_inode_is_fast_symlink(struct inode *inode)
|
46 |
|
|
{
|
47 |
|
|
int ea_blocks = EXT3_I(inode)->i_file_acl ?
|
48 |
|
|
(inode->i_sb->s_blocksize >> 9) : 0;
|
49 |
|
|
|
50 |
|
|
return (S_ISLNK(inode->i_mode) &&
|
51 |
|
|
inode->i_blocks - ea_blocks == 0);
|
52 |
|
|
}
|
53 |
|
|
|
54 |
|
|
/* The ext3 forget function must perform a revoke if we are freeing data
|
55 |
|
|
* which has been journaled. Metadata (eg. indirect blocks) must be
|
56 |
|
|
* revoked in all cases.
|
57 |
|
|
*
|
58 |
|
|
* "bh" may be NULL: a metadata block may have been freed from memory
|
59 |
|
|
* but there may still be a record of it in the journal, and that record
|
60 |
|
|
* still needs to be revoked.
|
61 |
|
|
*/
|
62 |
|
|
|
63 |
|
|
static int ext3_forget(handle_t *handle, int is_metadata,
|
64 |
|
|
struct inode *inode, struct buffer_head *bh,
|
65 |
|
|
int blocknr)
|
66 |
|
|
{
|
67 |
|
|
int err;
|
68 |
|
|
|
69 |
|
|
BUFFER_TRACE(bh, "enter");
|
70 |
|
|
|
71 |
|
|
jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
|
72 |
|
|
"data mode %lx\n",
|
73 |
|
|
bh, is_metadata, inode->i_mode,
|
74 |
|
|
test_opt(inode->i_sb, DATA_FLAGS));
|
75 |
|
|
|
76 |
|
|
/* Never use the revoke function if we are doing full data
|
77 |
|
|
* journaling: there is no need to, and a V1 superblock won't
|
78 |
|
|
* support it. Otherwise, only skip the revoke on un-journaled
|
79 |
|
|
* data blocks. */
|
80 |
|
|
|
81 |
|
|
if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
|
82 |
|
|
(!is_metadata && !ext3_should_journal_data(inode))) {
|
83 |
|
|
if (bh) {
|
84 |
|
|
BUFFER_TRACE(bh, "call journal_forget");
|
85 |
|
|
ext3_journal_forget(handle, bh);
|
86 |
|
|
}
|
87 |
|
|
return 0;
|
88 |
|
|
}
|
89 |
|
|
|
90 |
|
|
/*
|
91 |
|
|
* data!=journal && (is_metadata || should_journal_data(inode))
|
92 |
|
|
*/
|
93 |
|
|
BUFFER_TRACE(bh, "call ext3_journal_revoke");
|
94 |
|
|
err = ext3_journal_revoke(handle, blocknr, bh);
|
95 |
|
|
if (err)
|
96 |
|
|
ext3_abort(inode->i_sb, __FUNCTION__,
|
97 |
|
|
"error %d when attempting revoke", err);
|
98 |
|
|
BUFFER_TRACE(bh, "exit");
|
99 |
|
|
return err;
|
100 |
|
|
}
|
101 |
|
|
|
102 |
|
|
/*
|
103 |
|
|
* Work out how many blocks we need to progress with the next chunk of a
|
104 |
|
|
* truncate transaction.
|
105 |
|
|
*/
|
106 |
|
|
|
107 |
|
|
static unsigned long blocks_for_truncate(struct inode *inode)
|
108 |
|
|
{
|
109 |
|
|
unsigned long needed;
|
110 |
|
|
|
111 |
|
|
needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
|
112 |
|
|
|
113 |
|
|
/* Give ourselves just enough room to cope with inodes in which
|
114 |
|
|
* i_blocks is corrupt: we've seen disk corruptions in the past
|
115 |
|
|
* which resulted in random data in an inode which looked enough
|
116 |
|
|
* like a regular file for ext3 to try to delete it. Things
|
117 |
|
|
* will go a bit crazy if that happens, but at least we should
|
118 |
|
|
* try not to panic the whole kernel. */
|
119 |
|
|
if (needed < 2)
|
120 |
|
|
needed = 2;
|
121 |
|
|
|
122 |
|
|
/* But we need to bound the transaction so we don't overflow the
|
123 |
|
|
* journal. */
|
124 |
|
|
if (needed > EXT3_MAX_TRANS_DATA)
|
125 |
|
|
needed = EXT3_MAX_TRANS_DATA;
|
126 |
|
|
|
127 |
|
|
return EXT3_DATA_TRANS_BLOCKS + needed;
|
128 |
|
|
}
|
129 |
|
|
|
130 |
|
|
/*
|
131 |
|
|
* Truncate transactions can be complex and absolutely huge. So we need to
|
132 |
|
|
* be able to restart the transaction at a conventient checkpoint to make
|
133 |
|
|
* sure we don't overflow the journal.
|
134 |
|
|
*
|
135 |
|
|
* start_transaction gets us a new handle for a truncate transaction,
|
136 |
|
|
* and extend_transaction tries to extend the existing one a bit. If
|
137 |
|
|
* extend fails, we need to propagate the failure up and restart the
|
138 |
|
|
* transaction in the top-level truncate loop. --sct
|
139 |
|
|
*/
|
140 |
|
|
|
141 |
|
|
static handle_t *start_transaction(struct inode *inode)
|
142 |
|
|
{
|
143 |
|
|
handle_t *result;
|
144 |
|
|
|
145 |
|
|
result = ext3_journal_start(inode, blocks_for_truncate(inode));
|
146 |
|
|
if (!IS_ERR(result))
|
147 |
|
|
return result;
|
148 |
|
|
|
149 |
|
|
ext3_std_error(inode->i_sb, PTR_ERR(result));
|
150 |
|
|
return result;
|
151 |
|
|
}
|
152 |
|
|
|
153 |
|
|
/*
|
154 |
|
|
* Try to extend this transaction for the purposes of truncation.
|
155 |
|
|
*
|
156 |
|
|
* Returns 0 if we managed to create more room. If we can't create more
|
157 |
|
|
* room, and the transaction must be restarted we return 1.
|
158 |
|
|
*/
|
159 |
|
|
static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
|
160 |
|
|
{
|
161 |
|
|
if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
|
162 |
|
|
return 0;
|
163 |
|
|
if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
|
164 |
|
|
return 0;
|
165 |
|
|
return 1;
|
166 |
|
|
}
|
167 |
|
|
|
168 |
|
|
/*
|
169 |
|
|
* Restart the transaction associated with *handle. This does a commit,
|
170 |
|
|
* so before we call here everything must be consistently dirtied against
|
171 |
|
|
* this transaction.
|
172 |
|
|
*/
|
173 |
|
|
static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
|
174 |
|
|
{
|
175 |
|
|
jbd_debug(2, "restarting handle %p\n", handle);
|
176 |
|
|
return ext3_journal_restart(handle, blocks_for_truncate(inode));
|
177 |
|
|
}
|
178 |
|
|
|
179 |
|
|
/*
|
180 |
|
|
* Called at each iput()
|
181 |
|
|
*/
|
182 |
|
|
void ext3_put_inode (struct inode * inode)
|
183 |
|
|
{
|
184 |
|
|
ext3_discard_prealloc (inode);
|
185 |
|
|
}
|
186 |
|
|
|
187 |
|
|
/*
|
188 |
|
|
* Called at the last iput() if i_nlink is zero.
|
189 |
|
|
*/
|
190 |
|
|
void ext3_delete_inode (struct inode * inode)
|
191 |
|
|
{
|
192 |
|
|
handle_t *handle;
|
193 |
|
|
|
194 |
|
|
if (is_bad_inode(inode) ||
|
195 |
|
|
inode->i_ino == EXT3_ACL_IDX_INO ||
|
196 |
|
|
inode->i_ino == EXT3_ACL_DATA_INO)
|
197 |
|
|
goto no_delete;
|
198 |
|
|
|
199 |
|
|
lock_kernel();
|
200 |
|
|
handle = start_transaction(inode);
|
201 |
|
|
if (IS_ERR(handle)) {
|
202 |
|
|
/* If we're going to skip the normal cleanup, we still
|
203 |
|
|
* need to make sure that the in-core orphan linked list
|
204 |
|
|
* is properly cleaned up. */
|
205 |
|
|
ext3_orphan_del(NULL, inode);
|
206 |
|
|
|
207 |
|
|
ext3_std_error(inode->i_sb, PTR_ERR(handle));
|
208 |
|
|
unlock_kernel();
|
209 |
|
|
goto no_delete;
|
210 |
|
|
}
|
211 |
|
|
|
212 |
|
|
if (IS_SYNC(inode))
|
213 |
|
|
handle->h_sync = 1;
|
214 |
|
|
inode->i_size = 0;
|
215 |
|
|
if (inode->i_blocks)
|
216 |
|
|
ext3_truncate(inode);
|
217 |
|
|
/*
|
218 |
|
|
* Kill off the orphan record which ext3_truncate created.
|
219 |
|
|
* AKPM: I think this can be inside the above `if'.
|
220 |
|
|
* Note that ext3_orphan_del() has to be able to cope with the
|
221 |
|
|
* deletion of a non-existent orphan - this is because we don't
|
222 |
|
|
* know if ext3_truncate() actually created an orphan record.
|
223 |
|
|
* (Well, we could do this if we need to, but heck - it works)
|
224 |
|
|
*/
|
225 |
|
|
ext3_orphan_del(handle, inode);
|
226 |
|
|
inode->u.ext3_i.i_dtime = CURRENT_TIME;
|
227 |
|
|
|
228 |
|
|
/*
|
229 |
|
|
* One subtle ordering requirement: if anything has gone wrong
|
230 |
|
|
* (transaction abort, IO errors, whatever), then we can still
|
231 |
|
|
* do these next steps (the fs will already have been marked as
|
232 |
|
|
* having errors), but we can't free the inode if the mark_dirty
|
233 |
|
|
* fails.
|
234 |
|
|
*/
|
235 |
|
|
if (ext3_mark_inode_dirty(handle, inode))
|
236 |
|
|
/* If that failed, just do the required in-core inode clear. */
|
237 |
|
|
clear_inode(inode);
|
238 |
|
|
else
|
239 |
|
|
ext3_free_inode(handle, inode);
|
240 |
|
|
ext3_journal_stop(handle, inode);
|
241 |
|
|
unlock_kernel();
|
242 |
|
|
return;
|
243 |
|
|
no_delete:
|
244 |
|
|
clear_inode(inode); /* We must guarantee clearing of inode... */
|
245 |
|
|
}
|
246 |
|
|
|
247 |
|
|
void ext3_discard_prealloc (struct inode * inode)
|
248 |
|
|
{
|
249 |
|
|
#ifdef EXT3_PREALLOCATE
|
250 |
|
|
lock_kernel();
|
251 |
|
|
/* Writer: ->i_prealloc* */
|
252 |
|
|
if (inode->u.ext3_i.i_prealloc_count) {
|
253 |
|
|
unsigned short total = inode->u.ext3_i.i_prealloc_count;
|
254 |
|
|
unsigned long block = inode->u.ext3_i.i_prealloc_block;
|
255 |
|
|
inode->u.ext3_i.i_prealloc_count = 0;
|
256 |
|
|
inode->u.ext3_i.i_prealloc_block = 0;
|
257 |
|
|
/* Writer: end */
|
258 |
|
|
ext3_free_blocks (inode, block, total);
|
259 |
|
|
}
|
260 |
|
|
unlock_kernel();
|
261 |
|
|
#endif
|
262 |
|
|
}
|
263 |
|
|
|
264 |
|
|
static int ext3_alloc_block (handle_t *handle,
|
265 |
|
|
struct inode * inode, unsigned long goal, int *err)
|
266 |
|
|
{
|
267 |
|
|
#ifdef EXT3FS_DEBUG
|
268 |
|
|
static unsigned long alloc_hits = 0, alloc_attempts = 0;
|
269 |
|
|
#endif
|
270 |
|
|
unsigned long result;
|
271 |
|
|
|
272 |
|
|
#ifdef EXT3_PREALLOCATE
|
273 |
|
|
/* Writer: ->i_prealloc* */
|
274 |
|
|
if (inode->u.ext3_i.i_prealloc_count &&
|
275 |
|
|
(goal == inode->u.ext3_i.i_prealloc_block ||
|
276 |
|
|
goal + 1 == inode->u.ext3_i.i_prealloc_block))
|
277 |
|
|
{
|
278 |
|
|
result = inode->u.ext3_i.i_prealloc_block++;
|
279 |
|
|
inode->u.ext3_i.i_prealloc_count--;
|
280 |
|
|
/* Writer: end */
|
281 |
|
|
ext3_debug ("preallocation hit (%lu/%lu).\n",
|
282 |
|
|
++alloc_hits, ++alloc_attempts);
|
283 |
|
|
} else {
|
284 |
|
|
ext3_discard_prealloc (inode);
|
285 |
|
|
ext3_debug ("preallocation miss (%lu/%lu).\n",
|
286 |
|
|
alloc_hits, ++alloc_attempts);
|
287 |
|
|
if (S_ISREG(inode->i_mode))
|
288 |
|
|
result = ext3_new_block (inode, goal,
|
289 |
|
|
&inode->u.ext3_i.i_prealloc_count,
|
290 |
|
|
&inode->u.ext3_i.i_prealloc_block, err);
|
291 |
|
|
else
|
292 |
|
|
result = ext3_new_block (inode, goal, 0, 0, err);
|
293 |
|
|
/*
|
294 |
|
|
* AKPM: this is somewhat sticky. I'm not surprised it was
|
295 |
|
|
* disabled in 2.2's ext3. Need to integrate b_committed_data
|
296 |
|
|
* guarding with preallocation, if indeed preallocation is
|
297 |
|
|
* effective.
|
298 |
|
|
*/
|
299 |
|
|
}
|
300 |
|
|
#else
|
301 |
|
|
result = ext3_new_block (handle, inode, goal, 0, 0, err);
|
302 |
|
|
#endif
|
303 |
|
|
return result;
|
304 |
|
|
}
|
305 |
|
|
|
306 |
|
|
|
307 |
|
|
typedef struct {
|
308 |
|
|
u32 *p;
|
309 |
|
|
u32 key;
|
310 |
|
|
struct buffer_head *bh;
|
311 |
|
|
} Indirect;
|
312 |
|
|
|
313 |
|
|
static inline void add_chain(Indirect *p, struct buffer_head *bh, u32 *v)
|
314 |
|
|
{
|
315 |
|
|
p->key = *(p->p = v);
|
316 |
|
|
p->bh = bh;
|
317 |
|
|
}
|
318 |
|
|
|
319 |
|
|
static inline int verify_chain(Indirect *from, Indirect *to)
|
320 |
|
|
{
|
321 |
|
|
while (from <= to && from->key == *from->p)
|
322 |
|
|
from++;
|
323 |
|
|
return (from > to);
|
324 |
|
|
}
|
325 |
|
|
|
326 |
|
|
/**
|
327 |
|
|
* ext3_block_to_path - parse the block number into array of offsets
|
328 |
|
|
* @inode: inode in question (we are only interested in its superblock)
|
329 |
|
|
* @i_block: block number to be parsed
|
330 |
|
|
* @offsets: array to store the offsets in
|
331 |
|
|
*
|
332 |
|
|
* To store the locations of file's data ext3 uses a data structure common
|
333 |
|
|
* for UNIX filesystems - tree of pointers anchored in the inode, with
|
334 |
|
|
* data blocks at leaves and indirect blocks in intermediate nodes.
|
335 |
|
|
* This function translates the block number into path in that tree -
|
336 |
|
|
* return value is the path length and @offsets[n] is the offset of
|
337 |
|
|
* pointer to (n+1)th node in the nth one. If @block is out of range
|
338 |
|
|
* (negative or too large) warning is printed and zero returned.
|
339 |
|
|
*
|
340 |
|
|
* Note: function doesn't find node addresses, so no IO is needed. All
|
341 |
|
|
* we need to know is the capacity of indirect blocks (taken from the
|
342 |
|
|
* inode->i_sb).
|
343 |
|
|
*/
|
344 |
|
|
|
345 |
|
|
/*
|
346 |
|
|
* Portability note: the last comparison (check that we fit into triple
|
347 |
|
|
* indirect block) is spelled differently, because otherwise on an
|
348 |
|
|
* architecture with 32-bit longs and 8Kb pages we might get into trouble
|
349 |
|
|
* if our filesystem had 8Kb blocks. We might use long long, but that would
|
350 |
|
|
* kill us on x86. Oh, well, at least the sign propagation does not matter -
|
351 |
|
|
* i_block would have to be negative in the very beginning, so we would not
|
352 |
|
|
* get there at all.
|
353 |
|
|
*/
|
354 |
|
|
|
355 |
|
|
static int ext3_block_to_path(struct inode *inode, long i_block, int offsets[4])
|
356 |
|
|
{
|
357 |
|
|
int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
|
358 |
|
|
int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
|
359 |
|
|
const long direct_blocks = EXT3_NDIR_BLOCKS,
|
360 |
|
|
indirect_blocks = ptrs,
|
361 |
|
|
double_blocks = (1 << (ptrs_bits * 2));
|
362 |
|
|
int n = 0;
|
363 |
|
|
|
364 |
|
|
if (i_block < 0) {
|
365 |
|
|
ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
|
366 |
|
|
} else if (i_block < direct_blocks) {
|
367 |
|
|
offsets[n++] = i_block;
|
368 |
|
|
} else if ( (i_block -= direct_blocks) < indirect_blocks) {
|
369 |
|
|
offsets[n++] = EXT3_IND_BLOCK;
|
370 |
|
|
offsets[n++] = i_block;
|
371 |
|
|
} else if ((i_block -= indirect_blocks) < double_blocks) {
|
372 |
|
|
offsets[n++] = EXT3_DIND_BLOCK;
|
373 |
|
|
offsets[n++] = i_block >> ptrs_bits;
|
374 |
|
|
offsets[n++] = i_block & (ptrs - 1);
|
375 |
|
|
} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
|
376 |
|
|
offsets[n++] = EXT3_TIND_BLOCK;
|
377 |
|
|
offsets[n++] = i_block >> (ptrs_bits * 2);
|
378 |
|
|
offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
|
379 |
|
|
offsets[n++] = i_block & (ptrs - 1);
|
380 |
|
|
} else {
|
381 |
|
|
ext3_warning (inode->i_sb, "ext3_block_to_path", "block > big");
|
382 |
|
|
}
|
383 |
|
|
return n;
|
384 |
|
|
}
|
385 |
|
|
|
386 |
|
|
/**
|
387 |
|
|
* ext3_get_branch - read the chain of indirect blocks leading to data
|
388 |
|
|
* @inode: inode in question
|
389 |
|
|
* @depth: depth of the chain (1 - direct pointer, etc.)
|
390 |
|
|
* @offsets: offsets of pointers in inode/indirect blocks
|
391 |
|
|
* @chain: place to store the result
|
392 |
|
|
* @err: here we store the error value
|
393 |
|
|
*
|
394 |
|
|
* Function fills the array of triples <key, p, bh> and returns %NULL
|
395 |
|
|
* if everything went OK or the pointer to the last filled triple
|
396 |
|
|
* (incomplete one) otherwise. Upon the return chain[i].key contains
|
397 |
|
|
* the number of (i+1)-th block in the chain (as it is stored in memory,
|
398 |
|
|
* i.e. little-endian 32-bit), chain[i].p contains the address of that
|
399 |
|
|
* number (it points into struct inode for i==0 and into the bh->b_data
|
400 |
|
|
* for i>0) and chain[i].bh points to the buffer_head of i-th indirect
|
401 |
|
|
* block for i>0 and NULL for i==0. In other words, it holds the block
|
402 |
|
|
* numbers of the chain, addresses they were taken from (and where we can
|
403 |
|
|
* verify that chain did not change) and buffer_heads hosting these
|
404 |
|
|
* numbers.
|
405 |
|
|
*
|
406 |
|
|
* Function stops when it stumbles upon zero pointer (absent block)
|
407 |
|
|
* (pointer to last triple returned, *@err == 0)
|
408 |
|
|
* or when it gets an IO error reading an indirect block
|
409 |
|
|
* (ditto, *@err == -EIO)
|
410 |
|
|
* or when it notices that chain had been changed while it was reading
|
411 |
|
|
* (ditto, *@err == -EAGAIN)
|
412 |
|
|
* or when it reads all @depth-1 indirect blocks successfully and finds
|
413 |
|
|
* the whole chain, all way to the data (returns %NULL, *err == 0).
|
414 |
|
|
*/
|
415 |
|
|
static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
|
416 |
|
|
Indirect chain[4], int *err)
|
417 |
|
|
{
|
418 |
|
|
struct super_block *sb = inode->i_sb;
|
419 |
|
|
Indirect *p = chain;
|
420 |
|
|
struct buffer_head *bh;
|
421 |
|
|
|
422 |
|
|
*err = 0;
|
423 |
|
|
/* i_data is not going away, no lock needed */
|
424 |
|
|
add_chain (chain, NULL, inode->u.ext3_i.i_data + *offsets);
|
425 |
|
|
if (!p->key)
|
426 |
|
|
goto no_block;
|
427 |
|
|
while (--depth) {
|
428 |
|
|
bh = sb_bread(sb, le32_to_cpu(p->key));
|
429 |
|
|
if (!bh)
|
430 |
|
|
goto failure;
|
431 |
|
|
/* Reader: pointers */
|
432 |
|
|
if (!verify_chain(chain, p))
|
433 |
|
|
goto changed;
|
434 |
|
|
add_chain(++p, bh, (u32*)bh->b_data + *++offsets);
|
435 |
|
|
/* Reader: end */
|
436 |
|
|
if (!p->key)
|
437 |
|
|
goto no_block;
|
438 |
|
|
}
|
439 |
|
|
return NULL;
|
440 |
|
|
|
441 |
|
|
changed:
|
442 |
|
|
brelse(bh);
|
443 |
|
|
*err = -EAGAIN;
|
444 |
|
|
goto no_block;
|
445 |
|
|
failure:
|
446 |
|
|
*err = -EIO;
|
447 |
|
|
no_block:
|
448 |
|
|
return p;
|
449 |
|
|
}
|
450 |
|
|
|
451 |
|
|
/**
|
452 |
|
|
* ext3_find_near - find a place for allocation with sufficient locality
|
453 |
|
|
* @inode: owner
|
454 |
|
|
* @ind: descriptor of indirect block.
|
455 |
|
|
*
|
456 |
|
|
* This function returns the prefered place for block allocation.
|
457 |
|
|
* It is used when heuristic for sequential allocation fails.
|
458 |
|
|
* Rules are:
|
459 |
|
|
* + if there is a block to the left of our position - allocate near it.
|
460 |
|
|
* + if pointer will live in indirect block - allocate near that block.
|
461 |
|
|
* + if pointer will live in inode - allocate in the same
|
462 |
|
|
* cylinder group.
|
463 |
|
|
* Caller must make sure that @ind is valid and will stay that way.
|
464 |
|
|
*/
|
465 |
|
|
|
466 |
|
|
static inline unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
|
467 |
|
|
{
|
468 |
|
|
u32 *start = ind->bh ? (u32*) ind->bh->b_data : inode->u.ext3_i.i_data;
|
469 |
|
|
u32 *p;
|
470 |
|
|
|
471 |
|
|
/* Try to find previous block */
|
472 |
|
|
for (p = ind->p - 1; p >= start; p--)
|
473 |
|
|
if (*p)
|
474 |
|
|
return le32_to_cpu(*p);
|
475 |
|
|
|
476 |
|
|
/* No such thing, so let's try location of indirect block */
|
477 |
|
|
if (ind->bh)
|
478 |
|
|
return ind->bh->b_blocknr;
|
479 |
|
|
|
480 |
|
|
/*
|
481 |
|
|
* It is going to be refered from inode itself? OK, just put it into
|
482 |
|
|
* the same cylinder group then.
|
483 |
|
|
*/
|
484 |
|
|
return (inode->u.ext3_i.i_block_group *
|
485 |
|
|
EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
|
486 |
|
|
le32_to_cpu(inode->i_sb->u.ext3_sb.s_es->s_first_data_block);
|
487 |
|
|
}
|
488 |
|
|
|
489 |
|
|
/**
|
490 |
|
|
* ext3_find_goal - find a prefered place for allocation.
|
491 |
|
|
* @inode: owner
|
492 |
|
|
* @block: block we want
|
493 |
|
|
* @chain: chain of indirect blocks
|
494 |
|
|
* @partial: pointer to the last triple within a chain
|
495 |
|
|
* @goal: place to store the result.
|
496 |
|
|
*
|
497 |
|
|
* Normally this function find the prefered place for block allocation,
|
498 |
|
|
* stores it in *@goal and returns zero. If the branch had been changed
|
499 |
|
|
* under us we return -EAGAIN.
|
500 |
|
|
*/
|
501 |
|
|
|
502 |
|
|
static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
|
503 |
|
|
Indirect *partial, unsigned long *goal)
|
504 |
|
|
{
|
505 |
|
|
/* Writer: ->i_next_alloc* */
|
506 |
|
|
if (block == inode->u.ext3_i.i_next_alloc_block + 1) {
|
507 |
|
|
inode->u.ext3_i.i_next_alloc_block++;
|
508 |
|
|
inode->u.ext3_i.i_next_alloc_goal++;
|
509 |
|
|
}
|
510 |
|
|
#ifdef SEARCH_FROM_ZERO
|
511 |
|
|
inode->u.ext3_i.i_next_alloc_block = 0;
|
512 |
|
|
inode->u.ext3_i.i_next_alloc_goal = 0;
|
513 |
|
|
#endif
|
514 |
|
|
/* Writer: end */
|
515 |
|
|
/* Reader: pointers, ->i_next_alloc* */
|
516 |
|
|
if (verify_chain(chain, partial)) {
|
517 |
|
|
/*
|
518 |
|
|
* try the heuristic for sequential allocation,
|
519 |
|
|
* failing that at least try to get decent locality.
|
520 |
|
|
*/
|
521 |
|
|
if (block == inode->u.ext3_i.i_next_alloc_block)
|
522 |
|
|
*goal = inode->u.ext3_i.i_next_alloc_goal;
|
523 |
|
|
if (!*goal)
|
524 |
|
|
*goal = ext3_find_near(inode, partial);
|
525 |
|
|
#ifdef SEARCH_FROM_ZERO
|
526 |
|
|
*goal = 0;
|
527 |
|
|
#endif
|
528 |
|
|
return 0;
|
529 |
|
|
}
|
530 |
|
|
/* Reader: end */
|
531 |
|
|
return -EAGAIN;
|
532 |
|
|
}
|
533 |
|
|
|
534 |
|
|
/**
|
535 |
|
|
* ext3_alloc_branch - allocate and set up a chain of blocks.
|
536 |
|
|
* @inode: owner
|
537 |
|
|
* @num: depth of the chain (number of blocks to allocate)
|
538 |
|
|
* @offsets: offsets (in the blocks) to store the pointers to next.
|
539 |
|
|
* @branch: place to store the chain in.
|
540 |
|
|
*
|
541 |
|
|
* This function allocates @num blocks, zeroes out all but the last one,
|
542 |
|
|
* links them into chain and (if we are synchronous) writes them to disk.
|
543 |
|
|
* In other words, it prepares a branch that can be spliced onto the
|
544 |
|
|
* inode. It stores the information about that chain in the branch[], in
|
545 |
|
|
* the same format as ext3_get_branch() would do. We are calling it after
|
546 |
|
|
* we had read the existing part of chain and partial points to the last
|
547 |
|
|
* triple of that (one with zero ->key). Upon the exit we have the same
|
548 |
|
|
* picture as after the successful ext3_get_block(), excpet that in one
|
549 |
|
|
* place chain is disconnected - *branch->p is still zero (we did not
|
550 |
|
|
* set the last link), but branch->key contains the number that should
|
551 |
|
|
* be placed into *branch->p to fill that gap.
|
552 |
|
|
*
|
553 |
|
|
* If allocation fails we free all blocks we've allocated (and forget
|
554 |
|
|
* their buffer_heads) and return the error value the from failed
|
555 |
|
|
* ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
|
556 |
|
|
* as described above and return 0.
|
557 |
|
|
*/
|
558 |
|
|
|
559 |
|
|
static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
|
560 |
|
|
int num,
|
561 |
|
|
unsigned long goal,
|
562 |
|
|
int *offsets,
|
563 |
|
|
Indirect *branch)
|
564 |
|
|
{
|
565 |
|
|
int blocksize = inode->i_sb->s_blocksize;
|
566 |
|
|
int n = 0, keys = 0;
|
567 |
|
|
int err = 0;
|
568 |
|
|
int i;
|
569 |
|
|
int parent = ext3_alloc_block(handle, inode, goal, &err);
|
570 |
|
|
|
571 |
|
|
branch[0].key = cpu_to_le32(parent);
|
572 |
|
|
if (parent) {
|
573 |
|
|
for (n = 1; n < num; n++) {
|
574 |
|
|
struct buffer_head *bh;
|
575 |
|
|
/* Allocate the next block */
|
576 |
|
|
int nr = ext3_alloc_block(handle, inode, parent, &err);
|
577 |
|
|
if (!nr)
|
578 |
|
|
break;
|
579 |
|
|
branch[n].key = cpu_to_le32(nr);
|
580 |
|
|
keys = n+1;
|
581 |
|
|
|
582 |
|
|
/*
|
583 |
|
|
* Get buffer_head for parent block, zero it out
|
584 |
|
|
* and set the pointer to new one, then send
|
585 |
|
|
* parent to disk.
|
586 |
|
|
*/
|
587 |
|
|
bh = sb_getblk(inode->i_sb, parent);
|
588 |
|
|
branch[n].bh = bh;
|
589 |
|
|
lock_buffer(bh);
|
590 |
|
|
BUFFER_TRACE(bh, "call get_create_access");
|
591 |
|
|
err = ext3_journal_get_create_access(handle, bh);
|
592 |
|
|
if (err) {
|
593 |
|
|
unlock_buffer(bh);
|
594 |
|
|
brelse(bh);
|
595 |
|
|
break;
|
596 |
|
|
}
|
597 |
|
|
|
598 |
|
|
memset(bh->b_data, 0, blocksize);
|
599 |
|
|
branch[n].p = (u32*) bh->b_data + offsets[n];
|
600 |
|
|
*branch[n].p = branch[n].key;
|
601 |
|
|
BUFFER_TRACE(bh, "marking uptodate");
|
602 |
|
|
mark_buffer_uptodate(bh, 1);
|
603 |
|
|
unlock_buffer(bh);
|
604 |
|
|
|
605 |
|
|
BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
|
606 |
|
|
err = ext3_journal_dirty_metadata(handle, bh);
|
607 |
|
|
if (err)
|
608 |
|
|
break;
|
609 |
|
|
|
610 |
|
|
parent = nr;
|
611 |
|
|
}
|
612 |
|
|
}
|
613 |
|
|
if (n == num)
|
614 |
|
|
return 0;
|
615 |
|
|
|
616 |
|
|
/* Allocation failed, free what we already allocated */
|
617 |
|
|
for (i = 1; i < keys; i++) {
|
618 |
|
|
BUFFER_TRACE(branch[i].bh, "call journal_forget");
|
619 |
|
|
ext3_journal_forget(handle, branch[i].bh);
|
620 |
|
|
}
|
621 |
|
|
for (i = 0; i < keys; i++)
|
622 |
|
|
ext3_free_blocks(handle, inode, le32_to_cpu(branch[i].key), 1);
|
623 |
|
|
return err;
|
624 |
|
|
}
|
625 |
|
|
|
626 |
|
|
/**
|
627 |
|
|
* ext3_splice_branch - splice the allocated branch onto inode.
|
628 |
|
|
* @inode: owner
|
629 |
|
|
* @block: (logical) number of block we are adding
|
630 |
|
|
* @chain: chain of indirect blocks (with a missing link - see
|
631 |
|
|
* ext3_alloc_branch)
|
632 |
|
|
* @where: location of missing link
|
633 |
|
|
* @num: number of blocks we are adding
|
634 |
|
|
*
|
635 |
|
|
* This function verifies that chain (up to the missing link) had not
|
636 |
|
|
* changed, fills the missing link and does all housekeeping needed in
|
637 |
|
|
* inode (->i_blocks, etc.). In case of success we end up with the full
|
638 |
|
|
* chain to new block and return 0. Otherwise (== chain had been changed)
|
639 |
|
|
* we free the new blocks (forgetting their buffer_heads, indeed) and
|
640 |
|
|
* return -EAGAIN.
|
641 |
|
|
*/
|
642 |
|
|
|
643 |
|
|
static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
|
644 |
|
|
Indirect chain[4], Indirect *where, int num)
|
645 |
|
|
{
|
646 |
|
|
int i;
|
647 |
|
|
int err = 0;
|
648 |
|
|
|
649 |
|
|
/*
|
650 |
|
|
* If we're splicing into a [td]indirect block (as opposed to the
|
651 |
|
|
* inode) then we need to get write access to the [td]indirect block
|
652 |
|
|
* before the splice.
|
653 |
|
|
*/
|
654 |
|
|
if (where->bh) {
|
655 |
|
|
BUFFER_TRACE(where->bh, "get_write_access");
|
656 |
|
|
err = ext3_journal_get_write_access(handle, where->bh);
|
657 |
|
|
if (err)
|
658 |
|
|
goto err_out;
|
659 |
|
|
}
|
660 |
|
|
/* Verify that place we are splicing to is still there and vacant */
|
661 |
|
|
|
662 |
|
|
/* Writer: pointers, ->i_next_alloc* */
|
663 |
|
|
if (!verify_chain(chain, where-1) || *where->p)
|
664 |
|
|
/* Writer: end */
|
665 |
|
|
goto changed;
|
666 |
|
|
|
667 |
|
|
/* That's it */
|
668 |
|
|
|
669 |
|
|
*where->p = where->key;
|
670 |
|
|
inode->u.ext3_i.i_next_alloc_block = block;
|
671 |
|
|
inode->u.ext3_i.i_next_alloc_goal = le32_to_cpu(where[num-1].key);
|
672 |
|
|
#ifdef SEARCH_FROM_ZERO
|
673 |
|
|
inode->u.ext3_i.i_next_alloc_block = 0;
|
674 |
|
|
inode->u.ext3_i.i_next_alloc_goal = 0;
|
675 |
|
|
#endif
|
676 |
|
|
/* Writer: end */
|
677 |
|
|
|
678 |
|
|
/* We are done with atomic stuff, now do the rest of housekeeping */
|
679 |
|
|
|
680 |
|
|
inode->i_ctime = CURRENT_TIME;
|
681 |
|
|
ext3_mark_inode_dirty(handle, inode);
|
682 |
|
|
|
683 |
|
|
/* had we spliced it onto indirect block? */
|
684 |
|
|
if (where->bh) {
|
685 |
|
|
/*
|
686 |
|
|
* akpm: If we spliced it onto an indirect block, we haven't
|
687 |
|
|
* altered the inode. Note however that if it is being spliced
|
688 |
|
|
* onto an indirect block at the very end of the file (the
|
689 |
|
|
* file is growing) then we *will* alter the inode to reflect
|
690 |
|
|
* the new i_size. But that is not done here - it is done in
|
691 |
|
|
* generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
|
692 |
|
|
*/
|
693 |
|
|
jbd_debug(5, "splicing indirect only\n");
|
694 |
|
|
BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
|
695 |
|
|
err = ext3_journal_dirty_metadata(handle, where->bh);
|
696 |
|
|
if (err)
|
697 |
|
|
goto err_out;
|
698 |
|
|
} else {
|
699 |
|
|
/*
|
700 |
|
|
* OK, we spliced it into the inode itself on a direct block.
|
701 |
|
|
* Inode was dirtied above.
|
702 |
|
|
*/
|
703 |
|
|
jbd_debug(5, "splicing direct\n");
|
704 |
|
|
}
|
705 |
|
|
return err;
|
706 |
|
|
|
707 |
|
|
changed:
|
708 |
|
|
/*
|
709 |
|
|
* AKPM: if where[i].bh isn't part of the current updating
|
710 |
|
|
* transaction then we explode nastily. Test this code path.
|
711 |
|
|
*/
|
712 |
|
|
jbd_debug(1, "the chain changed: try again\n");
|
713 |
|
|
err = -EAGAIN;
|
714 |
|
|
|
715 |
|
|
err_out:
|
716 |
|
|
for (i = 1; i < num; i++) {
|
717 |
|
|
BUFFER_TRACE(where[i].bh, "call journal_forget");
|
718 |
|
|
ext3_journal_forget(handle, where[i].bh);
|
719 |
|
|
}
|
720 |
|
|
/* For the normal collision cleanup case, we free up the blocks.
|
721 |
|
|
* On genuine filesystem errors we don't even think about doing
|
722 |
|
|
* that. */
|
723 |
|
|
if (err == -EAGAIN)
|
724 |
|
|
for (i = 0; i < num; i++)
|
725 |
|
|
ext3_free_blocks(handle, inode,
|
726 |
|
|
le32_to_cpu(where[i].key), 1);
|
727 |
|
|
return err;
|
728 |
|
|
}
|
729 |
|
|
|
730 |
|
|
/*
|
731 |
|
|
* Allocation strategy is simple: if we have to allocate something, we will
|
732 |
|
|
* have to go the whole way to leaf. So let's do it before attaching anything
|
733 |
|
|
* to tree, set linkage between the newborn blocks, write them if sync is
|
734 |
|
|
* required, recheck the path, free and repeat if check fails, otherwise
|
735 |
|
|
* set the last missing link (that will protect us from any truncate-generated
|
736 |
|
|
* removals - all blocks on the path are immune now) and possibly force the
|
737 |
|
|
* write on the parent block.
|
738 |
|
|
* That has a nice additional property: no special recovery from the failed
|
739 |
|
|
* allocations is needed - we simply release blocks and do not touch anything
|
740 |
|
|
* reachable from inode.
|
741 |
|
|
*
|
742 |
|
|
* akpm: `handle' can be NULL if create == 0.
|
743 |
|
|
*
|
744 |
|
|
* The BKL may not be held on entry here. Be sure to take it early.
|
745 |
|
|
*/
|
746 |
|
|
|
747 |
|
|
static int ext3_get_block_handle(handle_t *handle, struct inode *inode,
|
748 |
|
|
long iblock,
|
749 |
|
|
struct buffer_head *bh_result, int create)
|
750 |
|
|
{
|
751 |
|
|
int err = -EIO;
|
752 |
|
|
int offsets[4];
|
753 |
|
|
Indirect chain[4];
|
754 |
|
|
Indirect *partial;
|
755 |
|
|
unsigned long goal;
|
756 |
|
|
int left;
|
757 |
|
|
int depth = ext3_block_to_path(inode, iblock, offsets);
|
758 |
|
|
loff_t new_size;
|
759 |
|
|
|
760 |
|
|
J_ASSERT(handle != NULL || create == 0);
|
761 |
|
|
|
762 |
|
|
if (depth == 0)
|
763 |
|
|
goto out;
|
764 |
|
|
|
765 |
|
|
lock_kernel();
|
766 |
|
|
reread:
|
767 |
|
|
partial = ext3_get_branch(inode, depth, offsets, chain, &err);
|
768 |
|
|
|
769 |
|
|
/* Simplest case - block found, no allocation needed */
|
770 |
|
|
if (!partial) {
|
771 |
|
|
bh_result->b_state &= ~(1UL << BH_New);
|
772 |
|
|
got_it:
|
773 |
|
|
bh_result->b_dev = inode->i_dev;
|
774 |
|
|
bh_result->b_blocknr = le32_to_cpu(chain[depth-1].key);
|
775 |
|
|
bh_result->b_state |= (1UL << BH_Mapped);
|
776 |
|
|
/* Clean up and exit */
|
777 |
|
|
partial = chain+depth-1; /* the whole chain */
|
778 |
|
|
goto cleanup;
|
779 |
|
|
}
|
780 |
|
|
|
781 |
|
|
/* Next simple case - plain lookup or failed read of indirect block */
|
782 |
|
|
if (!create || err == -EIO) {
|
783 |
|
|
cleanup:
|
784 |
|
|
while (partial > chain) {
|
785 |
|
|
BUFFER_TRACE(partial->bh, "call brelse");
|
786 |
|
|
brelse(partial->bh);
|
787 |
|
|
partial--;
|
788 |
|
|
}
|
789 |
|
|
BUFFER_TRACE(bh_result, "returned");
|
790 |
|
|
unlock_kernel();
|
791 |
|
|
out:
|
792 |
|
|
return err;
|
793 |
|
|
}
|
794 |
|
|
|
795 |
|
|
/*
|
796 |
|
|
* Indirect block might be removed by truncate while we were
|
797 |
|
|
* reading it. Handling of that case (forget what we've got and
|
798 |
|
|
* reread) is taken out of the main path.
|
799 |
|
|
*/
|
800 |
|
|
if (err == -EAGAIN)
|
801 |
|
|
goto changed;
|
802 |
|
|
|
803 |
|
|
if (ext3_find_goal(inode, iblock, chain, partial, &goal) < 0)
|
804 |
|
|
goto changed;
|
805 |
|
|
|
806 |
|
|
left = (chain + depth) - partial;
|
807 |
|
|
|
808 |
|
|
/*
|
809 |
|
|
* Block out ext3_truncate while we alter the tree
|
810 |
|
|
*/
|
811 |
|
|
down_read(&inode->u.ext3_i.truncate_sem);
|
812 |
|
|
err = ext3_alloc_branch(handle, inode, left, goal,
|
813 |
|
|
offsets+(partial-chain), partial);
|
814 |
|
|
|
815 |
|
|
/* The ext3_splice_branch call will free and forget any buffers
|
816 |
|
|
* on the new chain if there is a failure, but that risks using
|
817 |
|
|
* up transaction credits, especially for bitmaps where the
|
818 |
|
|
* credits cannot be returned. Can we handle this somehow? We
|
819 |
|
|
* may need to return -EAGAIN upwards in the worst case. --sct */
|
820 |
|
|
if (!err)
|
821 |
|
|
err = ext3_splice_branch(handle, inode, iblock, chain,
|
822 |
|
|
partial, left);
|
823 |
|
|
up_read(&inode->u.ext3_i.truncate_sem);
|
824 |
|
|
if (err == -EAGAIN)
|
825 |
|
|
goto changed;
|
826 |
|
|
if (err)
|
827 |
|
|
goto cleanup;
|
828 |
|
|
|
829 |
|
|
new_size = inode->i_size;
|
830 |
|
|
/*
|
831 |
|
|
* This is not racy against ext3_truncate's modification of i_disksize
|
832 |
|
|
* because VM/VFS ensures that the file cannot be extended while
|
833 |
|
|
* truncate is in progress. It is racy between multiple parallel
|
834 |
|
|
* instances of get_block, but we have the BKL.
|
835 |
|
|
*/
|
836 |
|
|
if (new_size > inode->u.ext3_i.i_disksize)
|
837 |
|
|
inode->u.ext3_i.i_disksize = new_size;
|
838 |
|
|
|
839 |
|
|
bh_result->b_state |= (1UL << BH_New);
|
840 |
|
|
goto got_it;
|
841 |
|
|
|
842 |
|
|
changed:
|
843 |
|
|
while (partial > chain) {
|
844 |
|
|
jbd_debug(1, "buffer chain changed, retrying\n");
|
845 |
|
|
BUFFER_TRACE(partial->bh, "brelsing");
|
846 |
|
|
brelse(partial->bh);
|
847 |
|
|
partial--;
|
848 |
|
|
}
|
849 |
|
|
goto reread;
|
850 |
|
|
}
|
851 |
|
|
|
852 |
|
|
/*
|
853 |
|
|
* The BKL is not held on entry here.
|
854 |
|
|
*/
|
855 |
|
|
static int ext3_get_block(struct inode *inode, long iblock,
|
856 |
|
|
struct buffer_head *bh_result, int create)
|
857 |
|
|
{
|
858 |
|
|
handle_t *handle = 0;
|
859 |
|
|
int ret;
|
860 |
|
|
|
861 |
|
|
if (create) {
|
862 |
|
|
handle = ext3_journal_current_handle();
|
863 |
|
|
J_ASSERT(handle != 0);
|
864 |
|
|
}
|
865 |
|
|
ret = ext3_get_block_handle(handle, inode, iblock, bh_result, create);
|
866 |
|
|
return ret;
|
867 |
|
|
}
|
868 |
|
|
|
869 |
|
|
/*
|
870 |
|
|
* `handle' can be NULL if create is zero
|
871 |
|
|
*/
|
872 |
|
|
struct buffer_head *ext3_getblk(handle_t *handle, struct inode * inode,
|
873 |
|
|
long block, int create, int * errp)
|
874 |
|
|
{
|
875 |
|
|
struct buffer_head dummy;
|
876 |
|
|
int fatal = 0, err;
|
877 |
|
|
|
878 |
|
|
J_ASSERT(handle != NULL || create == 0);
|
879 |
|
|
|
880 |
|
|
dummy.b_state = 0;
|
881 |
|
|
dummy.b_blocknr = -1000;
|
882 |
|
|
buffer_trace_init(&dummy.b_history);
|
883 |
|
|
*errp = ext3_get_block_handle(handle, inode, block, &dummy, create);
|
884 |
|
|
if (!*errp && buffer_mapped(&dummy)) {
|
885 |
|
|
struct buffer_head *bh;
|
886 |
|
|
bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
|
887 |
|
|
if (buffer_new(&dummy)) {
|
888 |
|
|
J_ASSERT(create != 0);
|
889 |
|
|
J_ASSERT(handle != 0);
|
890 |
|
|
|
891 |
|
|
/* Now that we do not always journal data, we
|
892 |
|
|
should keep in mind whether this should
|
893 |
|
|
always journal the new buffer as metadata.
|
894 |
|
|
For now, regular file writes use
|
895 |
|
|
ext3_get_block instead, so it's not a
|
896 |
|
|
problem. */
|
897 |
|
|
lock_kernel();
|
898 |
|
|
lock_buffer(bh);
|
899 |
|
|
BUFFER_TRACE(bh, "call get_create_access");
|
900 |
|
|
fatal = ext3_journal_get_create_access(handle, bh);
|
901 |
|
|
if (!fatal) {
|
902 |
|
|
memset(bh->b_data, 0,
|
903 |
|
|
inode->i_sb->s_blocksize);
|
904 |
|
|
mark_buffer_uptodate(bh, 1);
|
905 |
|
|
}
|
906 |
|
|
unlock_buffer(bh);
|
907 |
|
|
BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
|
908 |
|
|
err = ext3_journal_dirty_metadata(handle, bh);
|
909 |
|
|
if (!fatal) fatal = err;
|
910 |
|
|
unlock_kernel();
|
911 |
|
|
} else {
|
912 |
|
|
BUFFER_TRACE(bh, "not a new buffer");
|
913 |
|
|
}
|
914 |
|
|
if (fatal) {
|
915 |
|
|
*errp = fatal;
|
916 |
|
|
brelse(bh);
|
917 |
|
|
bh = NULL;
|
918 |
|
|
}
|
919 |
|
|
return bh;
|
920 |
|
|
}
|
921 |
|
|
return NULL;
|
922 |
|
|
}
|
923 |
|
|
|
924 |
|
|
struct buffer_head *ext3_bread(handle_t *handle, struct inode * inode,
|
925 |
|
|
int block, int create, int *err)
|
926 |
|
|
{
|
927 |
|
|
struct buffer_head * bh;
|
928 |
|
|
int prev_blocks;
|
929 |
|
|
|
930 |
|
|
prev_blocks = inode->i_blocks;
|
931 |
|
|
|
932 |
|
|
bh = ext3_getblk (handle, inode, block, create, err);
|
933 |
|
|
if (!bh)
|
934 |
|
|
return bh;
|
935 |
|
|
#ifdef EXT3_PREALLOCATE
|
936 |
|
|
/*
|
937 |
|
|
* If the inode has grown, and this is a directory, then use a few
|
938 |
|
|
* more of the preallocated blocks to keep directory fragmentation
|
939 |
|
|
* down. The preallocated blocks are guaranteed to be contiguous.
|
940 |
|
|
*/
|
941 |
|
|
if (create &&
|
942 |
|
|
S_ISDIR(inode->i_mode) &&
|
943 |
|
|
inode->i_blocks > prev_blocks &&
|
944 |
|
|
EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
|
945 |
|
|
EXT3_FEATURE_COMPAT_DIR_PREALLOC)) {
|
946 |
|
|
int i;
|
947 |
|
|
struct buffer_head *tmp_bh;
|
948 |
|
|
|
949 |
|
|
for (i = 1;
|
950 |
|
|
inode->u.ext3_i.i_prealloc_count &&
|
951 |
|
|
i < EXT3_SB(inode->i_sb)->s_es->s_prealloc_dir_blocks;
|
952 |
|
|
i++) {
|
953 |
|
|
/*
|
954 |
|
|
* ext3_getblk will zero out the contents of the
|
955 |
|
|
* directory for us
|
956 |
|
|
*/
|
957 |
|
|
tmp_bh = ext3_getblk(handle, inode,
|
958 |
|
|
block+i, create, err);
|
959 |
|
|
if (!tmp_bh) {
|
960 |
|
|
brelse (bh);
|
961 |
|
|
return 0;
|
962 |
|
|
}
|
963 |
|
|
brelse (tmp_bh);
|
964 |
|
|
}
|
965 |
|
|
}
|
966 |
|
|
#endif
|
967 |
|
|
if (buffer_uptodate(bh))
|
968 |
|
|
return bh;
|
969 |
|
|
ll_rw_block (READ, 1, &bh);
|
970 |
|
|
wait_on_buffer (bh);
|
971 |
|
|
if (buffer_uptodate(bh))
|
972 |
|
|
return bh;
|
973 |
|
|
brelse (bh);
|
974 |
|
|
*err = -EIO;
|
975 |
|
|
return NULL;
|
976 |
|
|
}
|
977 |
|
|
|
978 |
|
|
static int walk_page_buffers( handle_t *handle,
|
979 |
|
|
struct inode *inode,
|
980 |
|
|
struct buffer_head *head,
|
981 |
|
|
unsigned from,
|
982 |
|
|
unsigned to,
|
983 |
|
|
int *partial,
|
984 |
|
|
int (*fn)( handle_t *handle,
|
985 |
|
|
struct inode *inode,
|
986 |
|
|
struct buffer_head *bh))
|
987 |
|
|
{
|
988 |
|
|
struct buffer_head *bh;
|
989 |
|
|
unsigned block_start, block_end;
|
990 |
|
|
unsigned blocksize = head->b_size;
|
991 |
|
|
int err, ret = 0;
|
992 |
|
|
|
993 |
|
|
for ( bh = head, block_start = 0;
|
994 |
|
|
ret == 0 && (bh != head || !block_start);
|
995 |
|
|
block_start = block_end, bh = bh->b_this_page)
|
996 |
|
|
{
|
997 |
|
|
block_end = block_start + blocksize;
|
998 |
|
|
if (block_end <= from || block_start >= to) {
|
999 |
|
|
if (partial && !buffer_uptodate(bh))
|
1000 |
|
|
*partial = 1;
|
1001 |
|
|
continue;
|
1002 |
|
|
}
|
1003 |
|
|
err = (*fn)(handle, inode, bh);
|
1004 |
|
|
if (!ret)
|
1005 |
|
|
ret = err;
|
1006 |
|
|
}
|
1007 |
|
|
return ret;
|
1008 |
|
|
}
|
1009 |
|
|
|
1010 |
|
|
/*
|
1011 |
|
|
* To preserve ordering, it is essential that the hole instantiation and
|
1012 |
|
|
* the data write be encapsulated in a single transaction. We cannot
|
1013 |
|
|
* close off a transaction and start a new one between the ext3_get_block()
|
1014 |
|
|
* and the commit_write(). So doing the journal_start at the start of
|
1015 |
|
|
* prepare_write() is the right place.
|
1016 |
|
|
*
|
1017 |
|
|
* Also, this function can nest inside ext3_writepage() ->
|
1018 |
|
|
* block_write_full_page(). In that case, we *know* that ext3_writepage()
|
1019 |
|
|
* has generated enough buffer credits to do the whole page. So we won't
|
1020 |
|
|
* block on the journal in that case, which is good, because the caller may
|
1021 |
|
|
* be PF_MEMALLOC.
|
1022 |
|
|
*
|
1023 |
|
|
* By accident, ext3 can be reentered when a transaction is open via
|
1024 |
|
|
* quota file writes. If we were to commit the transaction while thus
|
1025 |
|
|
* reentered, there can be a deadlock - we would be holding a quota
|
1026 |
|
|
* lock, and the commit would never complete if another thread had a
|
1027 |
|
|
* transaction open and was blocking on the quota lock - a ranking
|
1028 |
|
|
* violation.
|
1029 |
|
|
*
|
1030 |
|
|
* So what we do is to rely on the fact that journal_stop/journal_start
|
1031 |
|
|
* will _not_ run commit under these circumstances because handle->h_ref
|
1032 |
|
|
* is elevated. We'll still have enough credits for the tiny quotafile
|
1033 |
|
|
* write.
|
1034 |
|
|
*/
|
1035 |
|
|
|
1036 |
|
|
static int do_journal_get_write_access(handle_t *handle, struct inode *inode,
|
1037 |
|
|
struct buffer_head *bh)
|
1038 |
|
|
{
|
1039 |
|
|
return ext3_journal_get_write_access(handle, bh);
|
1040 |
|
|
}
|
1041 |
|
|
|
1042 |
|
|
static int ext3_prepare_write(struct file *file, struct page *page,
|
1043 |
|
|
unsigned from, unsigned to)
|
1044 |
|
|
{
|
1045 |
|
|
struct inode *inode = page->mapping->host;
|
1046 |
|
|
int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
|
1047 |
|
|
handle_t *handle;
|
1048 |
|
|
|
1049 |
|
|
lock_kernel();
|
1050 |
|
|
handle = ext3_journal_start(inode, needed_blocks);
|
1051 |
|
|
if (IS_ERR(handle)) {
|
1052 |
|
|
ret = PTR_ERR(handle);
|
1053 |
|
|
goto out;
|
1054 |
|
|
}
|
1055 |
|
|
unlock_kernel();
|
1056 |
|
|
ret = block_prepare_write(page, from, to, ext3_get_block);
|
1057 |
|
|
lock_kernel();
|
1058 |
|
|
if (ret != 0)
|
1059 |
|
|
goto prepare_write_failed;
|
1060 |
|
|
|
1061 |
|
|
if (ext3_should_journal_data(inode)) {
|
1062 |
|
|
ret = walk_page_buffers(handle, inode, page->buffers,
|
1063 |
|
|
from, to, NULL, do_journal_get_write_access);
|
1064 |
|
|
if (ret) {
|
1065 |
|
|
/*
|
1066 |
|
|
* We're going to fail this prepare_write(),
|
1067 |
|
|
* so commit_write() will not be called.
|
1068 |
|
|
* We need to undo block_prepare_write()'s kmap().
|
1069 |
|
|
* AKPM: Do we need to clear PageUptodate? I don't
|
1070 |
|
|
* think so.
|
1071 |
|
|
*/
|
1072 |
|
|
kunmap(page);
|
1073 |
|
|
}
|
1074 |
|
|
}
|
1075 |
|
|
prepare_write_failed:
|
1076 |
|
|
if (ret)
|
1077 |
|
|
ext3_journal_stop(handle, inode);
|
1078 |
|
|
out:
|
1079 |
|
|
unlock_kernel();
|
1080 |
|
|
return ret;
|
1081 |
|
|
}
|
1082 |
|
|
|
1083 |
|
|
static int journal_dirty_sync_data(handle_t *handle, struct inode *inode,
|
1084 |
|
|
struct buffer_head *bh)
|
1085 |
|
|
{
|
1086 |
|
|
int ret = ext3_journal_dirty_data(handle, bh, 0);
|
1087 |
|
|
buffer_insert_inode_data_queue(bh, inode);
|
1088 |
|
|
return ret;
|
1089 |
|
|
}
|
1090 |
|
|
|
1091 |
|
|
/*
|
1092 |
|
|
* For ext3_writepage(). We also brelse() the buffer to account for
|
1093 |
|
|
* the bget() which ext3_writepage() performs.
|
1094 |
|
|
*/
|
1095 |
|
|
static int journal_dirty_async_data(handle_t *handle, struct inode *inode,
|
1096 |
|
|
struct buffer_head *bh)
|
1097 |
|
|
{
|
1098 |
|
|
int ret = ext3_journal_dirty_data(handle, bh, 1);
|
1099 |
|
|
buffer_insert_inode_data_queue(bh, inode);
|
1100 |
|
|
__brelse(bh);
|
1101 |
|
|
return ret;
|
1102 |
|
|
}
|
1103 |
|
|
|
1104 |
|
|
/* For commit_write() in data=journal mode */
|
1105 |
|
|
static int commit_write_fn(handle_t *handle, struct inode *inode,
|
1106 |
|
|
struct buffer_head *bh)
|
1107 |
|
|
{
|
1108 |
|
|
set_bit(BH_Uptodate, &bh->b_state);
|
1109 |
|
|
return ext3_journal_dirty_metadata(handle, bh);
|
1110 |
|
|
}
|
1111 |
|
|
|
1112 |
|
|
/*
|
1113 |
|
|
* We need to pick up the new inode size which generic_commit_write gave us
|
1114 |
|
|
* `file' can be NULL - eg, when called from block_symlink().
|
1115 |
|
|
*
|
1116 |
|
|
* ext3 inode->i_dirty_buffers policy: If we're journalling data we
|
1117 |
|
|
* definitely don't want them to appear on the inode at all - instead
|
1118 |
|
|
* we need to manage them at the JBD layer and we need to intercept
|
1119 |
|
|
* the relevant sync operations and translate them into journal operations.
|
1120 |
|
|
*
|
1121 |
|
|
* If we're not journalling data then we can just leave the buffers
|
1122 |
|
|
* on ->i_dirty_buffers. If someone writes them out for us then thanks.
|
1123 |
|
|
* Otherwise we'll do it in commit, if we're using ordered data.
|
1124 |
|
|
*/
|
1125 |
|
|
|
1126 |
|
|
static int ext3_commit_write(struct file *file, struct page *page,
|
1127 |
|
|
unsigned from, unsigned to)
|
1128 |
|
|
{
|
1129 |
|
|
handle_t *handle = ext3_journal_current_handle();
|
1130 |
|
|
struct inode *inode = page->mapping->host;
|
1131 |
|
|
int ret = 0, ret2;
|
1132 |
|
|
|
1133 |
|
|
lock_kernel();
|
1134 |
|
|
if (ext3_should_journal_data(inode)) {
|
1135 |
|
|
/*
|
1136 |
|
|
* Here we duplicate the generic_commit_write() functionality
|
1137 |
|
|
*/
|
1138 |
|
|
int partial = 0;
|
1139 |
|
|
loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
|
1140 |
|
|
|
1141 |
|
|
ret = walk_page_buffers(handle, inode, page->buffers,
|
1142 |
|
|
from, to, &partial, commit_write_fn);
|
1143 |
|
|
if (!partial)
|
1144 |
|
|
SetPageUptodate(page);
|
1145 |
|
|
kunmap(page);
|
1146 |
|
|
if (pos > inode->i_size)
|
1147 |
|
|
inode->i_size = pos;
|
1148 |
|
|
EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
|
1149 |
|
|
} else {
|
1150 |
|
|
if (ext3_should_order_data(inode)) {
|
1151 |
|
|
ret = walk_page_buffers(handle, inode, page->buffers,
|
1152 |
|
|
from, to, NULL, journal_dirty_sync_data);
|
1153 |
|
|
}
|
1154 |
|
|
/* Be careful here if generic_commit_write becomes a
|
1155 |
|
|
* required invocation after block_prepare_write. */
|
1156 |
|
|
if (ret == 0) {
|
1157 |
|
|
ret = generic_commit_write(file, page, from, to);
|
1158 |
|
|
} else {
|
1159 |
|
|
/*
|
1160 |
|
|
* block_prepare_write() was called, but we're not
|
1161 |
|
|
* going to call generic_commit_write(). So we
|
1162 |
|
|
* need to perform generic_commit_write()'s kunmap
|
1163 |
|
|
* by hand.
|
1164 |
|
|
*/
|
1165 |
|
|
kunmap(page);
|
1166 |
|
|
}
|
1167 |
|
|
}
|
1168 |
|
|
if (inode->i_size > inode->u.ext3_i.i_disksize) {
|
1169 |
|
|
inode->u.ext3_i.i_disksize = inode->i_size;
|
1170 |
|
|
ret2 = ext3_mark_inode_dirty(handle, inode);
|
1171 |
|
|
if (!ret)
|
1172 |
|
|
ret = ret2;
|
1173 |
|
|
}
|
1174 |
|
|
ret2 = ext3_journal_stop(handle, inode);
|
1175 |
|
|
unlock_kernel();
|
1176 |
|
|
if (!ret)
|
1177 |
|
|
ret = ret2;
|
1178 |
|
|
return ret;
|
1179 |
|
|
}
|
1180 |
|
|
|
1181 |
|
|
/*
|
1182 |
|
|
* bmap() is special. It gets used by applications such as lilo and by
|
1183 |
|
|
* the swapper to find the on-disk block of a specific piece of data.
|
1184 |
|
|
*
|
1185 |
|
|
* Naturally, this is dangerous if the block concerned is still in the
|
1186 |
|
|
* journal. If somebody makes a swapfile on an ext3 data-journaling
|
1187 |
|
|
* filesystem and enables swap, then they may get a nasty shock when the
|
1188 |
|
|
* data getting swapped to that swapfile suddenly gets overwritten by
|
1189 |
|
|
* the original zero's written out previously to the journal and
|
1190 |
|
|
* awaiting writeback in the kernel's buffer cache.
|
1191 |
|
|
*
|
1192 |
|
|
* So, if we see any bmap calls here on a modified, data-journaled file,
|
1193 |
|
|
* take extra steps to flush any blocks which might be in the cache.
|
1194 |
|
|
*/
|
1195 |
|
|
static int ext3_bmap(struct address_space *mapping, long block)
|
1196 |
|
|
{
|
1197 |
|
|
struct inode *inode = mapping->host;
|
1198 |
|
|
journal_t *journal;
|
1199 |
|
|
int err;
|
1200 |
|
|
|
1201 |
|
|
if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
|
1202 |
|
|
/*
|
1203 |
|
|
* This is a REALLY heavyweight approach, but the use of
|
1204 |
|
|
* bmap on dirty files is expected to be extremely rare:
|
1205 |
|
|
* only if we run lilo or swapon on a freshly made file
|
1206 |
|
|
* do we expect this to happen.
|
1207 |
|
|
*
|
1208 |
|
|
* (bmap requires CAP_SYS_RAWIO so this does not
|
1209 |
|
|
* represent an unprivileged user DOS attack --- we'd be
|
1210 |
|
|
* in trouble if mortal users could trigger this path at
|
1211 |
|
|
* will.)
|
1212 |
|
|
*
|
1213 |
|
|
* NB. EXT3_STATE_JDATA is not set on files other than
|
1214 |
|
|
* regular files. If somebody wants to bmap a directory
|
1215 |
|
|
* or symlink and gets confused because the buffer
|
1216 |
|
|
* hasn't yet been flushed to disk, they deserve
|
1217 |
|
|
* everything they get.
|
1218 |
|
|
*/
|
1219 |
|
|
|
1220 |
|
|
EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
|
1221 |
|
|
journal = EXT3_JOURNAL(inode);
|
1222 |
|
|
journal_lock_updates(journal);
|
1223 |
|
|
err = journal_flush(journal);
|
1224 |
|
|
journal_unlock_updates(journal);
|
1225 |
|
|
|
1226 |
|
|
if (err)
|
1227 |
|
|
return 0;
|
1228 |
|
|
}
|
1229 |
|
|
|
1230 |
|
|
return generic_block_bmap(mapping,block,ext3_get_block);
|
1231 |
|
|
}
|
1232 |
|
|
|
1233 |
|
|
static int bget_one(handle_t *handle, struct inode *inode,
|
1234 |
|
|
struct buffer_head *bh)
|
1235 |
|
|
{
|
1236 |
|
|
atomic_inc(&bh->b_count);
|
1237 |
|
|
return 0;
|
1238 |
|
|
}
|
1239 |
|
|
|
1240 |
|
|
/*
|
1241 |
|
|
* Note that we always start a transaction even if we're not journalling
|
1242 |
|
|
* data. This is to preserve ordering: any hole instantiation within
|
1243 |
|
|
* __block_write_full_page -> ext3_get_block() should be journalled
|
1244 |
|
|
* along with the data so we don't crash and then get metadata which
|
1245 |
|
|
* refers to old data.
|
1246 |
|
|
*
|
1247 |
|
|
* In all journalling modes block_write_full_page() will start the I/O.
|
1248 |
|
|
*
|
1249 |
|
|
* Problem:
|
1250 |
|
|
*
|
1251 |
|
|
* ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
|
1252 |
|
|
* ext3_writepage()
|
1253 |
|
|
*
|
1254 |
|
|
* Similar for:
|
1255 |
|
|
*
|
1256 |
|
|
* ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
|
1257 |
|
|
*
|
1258 |
|
|
* Same applies to ext3_get_block(). We will deadlock on various things like
|
1259 |
|
|
* lock_journal and i_truncate_sem.
|
1260 |
|
|
*
|
1261 |
|
|
* Setting PF_MEMALLOC here doesn't work - too many internal memory
|
1262 |
|
|
* allocations fail.
|
1263 |
|
|
*
|
1264 |
|
|
* 16May01: If we're reentered then journal_current_handle() will be
|
1265 |
|
|
* non-zero. We simply *return*.
|
1266 |
|
|
*
|
1267 |
|
|
* 1 July 2001: @@@ FIXME:
|
1268 |
|
|
* In journalled data mode, a data buffer may be metadata against the
|
1269 |
|
|
* current transaction. But the same file is part of a shared mapping
|
1270 |
|
|
* and someone does a writepage() on it.
|
1271 |
|
|
*
|
1272 |
|
|
* We will move the buffer onto the async_data list, but *after* it has
|
1273 |
|
|
* been dirtied. So there's a small window where we have dirty data on
|
1274 |
|
|
* BJ_Metadata.
|
1275 |
|
|
*
|
1276 |
|
|
* Note that this only applies to the last partial page in the file. The
|
1277 |
|
|
* bit which block_write_full_page() uses prepare/commit for. (That's
|
1278 |
|
|
* broken code anyway: it's wrong for msync()).
|
1279 |
|
|
*
|
1280 |
|
|
* It's a rare case: affects the final partial page, for journalled data
|
1281 |
|
|
* where the file is subject to bith write() and writepage() in the same
|
1282 |
|
|
* transction. To fix it we'll need a custom block_write_full_page().
|
1283 |
|
|
* We'll probably need that anyway for journalling writepage() output.
|
1284 |
|
|
*
|
1285 |
|
|
* We don't honour synchronous mounts for writepage(). That would be
|
1286 |
|
|
* disastrous. Any write() or metadata operation will sync the fs for
|
1287 |
|
|
* us.
|
1288 |
|
|
*/
|
1289 |
|
|
static int ext3_writepage(struct page *page)
|
1290 |
|
|
{
|
1291 |
|
|
struct inode *inode = page->mapping->host;
|
1292 |
|
|
struct buffer_head *page_buffers;
|
1293 |
|
|
handle_t *handle = NULL;
|
1294 |
|
|
int ret = 0, err;
|
1295 |
|
|
int needed;
|
1296 |
|
|
int order_data;
|
1297 |
|
|
|
1298 |
|
|
J_ASSERT(PageLocked(page));
|
1299 |
|
|
|
1300 |
|
|
/*
|
1301 |
|
|
* We give up here if we're reentered, because it might be
|
1302 |
|
|
* for a different filesystem. One *could* look for a
|
1303 |
|
|
* nested transaction opportunity.
|
1304 |
|
|
*/
|
1305 |
|
|
lock_kernel();
|
1306 |
|
|
if (ext3_journal_current_handle())
|
1307 |
|
|
goto out_fail;
|
1308 |
|
|
|
1309 |
|
|
needed = ext3_writepage_trans_blocks(inode);
|
1310 |
|
|
if (current->flags & PF_MEMALLOC)
|
1311 |
|
|
handle = ext3_journal_try_start(inode, needed);
|
1312 |
|
|
else
|
1313 |
|
|
handle = ext3_journal_start(inode, needed);
|
1314 |
|
|
|
1315 |
|
|
if (IS_ERR(handle)) {
|
1316 |
|
|
ret = PTR_ERR(handle);
|
1317 |
|
|
goto out_fail;
|
1318 |
|
|
}
|
1319 |
|
|
|
1320 |
|
|
order_data = ext3_should_order_data(inode) ||
|
1321 |
|
|
ext3_should_journal_data(inode);
|
1322 |
|
|
|
1323 |
|
|
unlock_kernel();
|
1324 |
|
|
|
1325 |
|
|
page_buffers = NULL; /* Purely to prevent compiler warning */
|
1326 |
|
|
|
1327 |
|
|
/* bget() all the buffers */
|
1328 |
|
|
if (order_data) {
|
1329 |
|
|
if (!page->buffers)
|
1330 |
|
|
create_empty_buffers(page,
|
1331 |
|
|
inode->i_dev, inode->i_sb->s_blocksize);
|
1332 |
|
|
page_buffers = page->buffers;
|
1333 |
|
|
walk_page_buffers(handle, inode, page_buffers, 0,
|
1334 |
|
|
PAGE_CACHE_SIZE, NULL, bget_one);
|
1335 |
|
|
}
|
1336 |
|
|
|
1337 |
|
|
ret = block_write_full_page(page, ext3_get_block);
|
1338 |
|
|
|
1339 |
|
|
/*
|
1340 |
|
|
* The page can become unlocked at any point now, and
|
1341 |
|
|
* truncate can then come in and change things. So we
|
1342 |
|
|
* can't touch *page from now on. But *page_buffers is
|
1343 |
|
|
* safe due to elevated refcount.
|
1344 |
|
|
*/
|
1345 |
|
|
|
1346 |
|
|
handle = ext3_journal_current_handle();
|
1347 |
|
|
lock_kernel();
|
1348 |
|
|
|
1349 |
|
|
/* And attach them to the current transaction */
|
1350 |
|
|
if (order_data) {
|
1351 |
|
|
err = walk_page_buffers(handle, inode, page_buffers,
|
1352 |
|
|
0, PAGE_CACHE_SIZE, NULL, journal_dirty_async_data);
|
1353 |
|
|
if (!ret)
|
1354 |
|
|
ret = err;
|
1355 |
|
|
}
|
1356 |
|
|
|
1357 |
|
|
err = ext3_journal_stop(handle, inode);
|
1358 |
|
|
if (!ret)
|
1359 |
|
|
ret = err;
|
1360 |
|
|
unlock_kernel();
|
1361 |
|
|
return ret;
|
1362 |
|
|
|
1363 |
|
|
out_fail:
|
1364 |
|
|
|
1365 |
|
|
unlock_kernel();
|
1366 |
|
|
SetPageDirty(page);
|
1367 |
|
|
UnlockPage(page);
|
1368 |
|
|
return ret;
|
1369 |
|
|
}
|
1370 |
|
|
|
1371 |
|
|
static int ext3_readpage(struct file *file, struct page *page)
|
1372 |
|
|
{
|
1373 |
|
|
return block_read_full_page(page,ext3_get_block);
|
1374 |
|
|
}
|
1375 |
|
|
|
1376 |
|
|
|
1377 |
|
|
static int ext3_flushpage(struct page *page, unsigned long offset)
|
1378 |
|
|
{
|
1379 |
|
|
journal_t *journal = EXT3_JOURNAL(page->mapping->host);
|
1380 |
|
|
return journal_flushpage(journal, page, offset);
|
1381 |
|
|
}
|
1382 |
|
|
|
1383 |
|
|
static int ext3_releasepage(struct page *page, int wait)
|
1384 |
|
|
{
|
1385 |
|
|
journal_t *journal = EXT3_JOURNAL(page->mapping->host);
|
1386 |
|
|
return journal_try_to_free_buffers(journal, page, wait);
|
1387 |
|
|
}
|
1388 |
|
|
|
1389 |
|
|
|
1390 |
|
|
struct address_space_operations ext3_aops = {
|
1391 |
|
|
readpage: ext3_readpage, /* BKL not held. Don't need */
|
1392 |
|
|
writepage: ext3_writepage, /* BKL not held. We take it */
|
1393 |
|
|
sync_page: block_sync_page,
|
1394 |
|
|
prepare_write: ext3_prepare_write, /* BKL not held. We take it */
|
1395 |
|
|
commit_write: ext3_commit_write, /* BKL not held. We take it */
|
1396 |
|
|
bmap: ext3_bmap, /* BKL held */
|
1397 |
|
|
flushpage: ext3_flushpage, /* BKL not held. Don't need */
|
1398 |
|
|
releasepage: ext3_releasepage, /* BKL not held. Don't need */
|
1399 |
|
|
};
|
1400 |
|
|
|
1401 |
|
|
/*
|
1402 |
|
|
* ext3_block_truncate_page() zeroes out a mapping from file offset `from'
|
1403 |
|
|
* up to the end of the block which corresponds to `from'.
|
1404 |
|
|
* This required during truncate. We need to physically zero the tail end
|
1405 |
|
|
* of that block so it doesn't yield old data if the file is later grown.
|
1406 |
|
|
*/
|
1407 |
|
|
static int ext3_block_truncate_page(handle_t *handle,
|
1408 |
|
|
struct address_space *mapping, loff_t from)
|
1409 |
|
|
{
|
1410 |
|
|
unsigned long index = from >> PAGE_CACHE_SHIFT;
|
1411 |
|
|
unsigned offset = from & (PAGE_CACHE_SIZE-1);
|
1412 |
|
|
unsigned blocksize, iblock, length, pos;
|
1413 |
|
|
struct inode *inode = mapping->host;
|
1414 |
|
|
struct page *page;
|
1415 |
|
|
struct buffer_head *bh;
|
1416 |
|
|
int err;
|
1417 |
|
|
|
1418 |
|
|
blocksize = inode->i_sb->s_blocksize;
|
1419 |
|
|
length = offset & (blocksize - 1);
|
1420 |
|
|
|
1421 |
|
|
/* Block boundary? Nothing to do */
|
1422 |
|
|
if (!length)
|
1423 |
|
|
return 0;
|
1424 |
|
|
|
1425 |
|
|
length = blocksize - length;
|
1426 |
|
|
iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
|
1427 |
|
|
|
1428 |
|
|
page = find_or_create_page(mapping, index, GFP_NOFS);
|
1429 |
|
|
err = -ENOMEM;
|
1430 |
|
|
if (!page)
|
1431 |
|
|
goto out;
|
1432 |
|
|
|
1433 |
|
|
if (!page->buffers)
|
1434 |
|
|
create_empty_buffers(page, inode->i_dev, blocksize);
|
1435 |
|
|
|
1436 |
|
|
/* Find the buffer that contains "offset" */
|
1437 |
|
|
bh = page->buffers;
|
1438 |
|
|
pos = blocksize;
|
1439 |
|
|
while (offset >= pos) {
|
1440 |
|
|
bh = bh->b_this_page;
|
1441 |
|
|
iblock++;
|
1442 |
|
|
pos += blocksize;
|
1443 |
|
|
}
|
1444 |
|
|
|
1445 |
|
|
err = 0;
|
1446 |
|
|
if (!buffer_mapped(bh)) {
|
1447 |
|
|
/* Hole? Nothing to do */
|
1448 |
|
|
if (buffer_uptodate(bh))
|
1449 |
|
|
goto unlock;
|
1450 |
|
|
ext3_get_block(inode, iblock, bh, 0);
|
1451 |
|
|
/* Still unmapped? Nothing to do */
|
1452 |
|
|
if (!buffer_mapped(bh))
|
1453 |
|
|
goto unlock;
|
1454 |
|
|
}
|
1455 |
|
|
|
1456 |
|
|
/* Ok, it's mapped. Make sure it's up-to-date */
|
1457 |
|
|
if (Page_Uptodate(page))
|
1458 |
|
|
set_bit(BH_Uptodate, &bh->b_state);
|
1459 |
|
|
|
1460 |
|
|
if (!buffer_uptodate(bh)) {
|
1461 |
|
|
err = -EIO;
|
1462 |
|
|
ll_rw_block(READ, 1, &bh);
|
1463 |
|
|
wait_on_buffer(bh);
|
1464 |
|
|
/* Uhhuh. Read error. Complain and punt. */
|
1465 |
|
|
if (!buffer_uptodate(bh))
|
1466 |
|
|
goto unlock;
|
1467 |
|
|
}
|
1468 |
|
|
|
1469 |
|
|
if (ext3_should_journal_data(inode)) {
|
1470 |
|
|
BUFFER_TRACE(bh, "get write access");
|
1471 |
|
|
err = ext3_journal_get_write_access(handle, bh);
|
1472 |
|
|
if (err)
|
1473 |
|
|
goto unlock;
|
1474 |
|
|
}
|
1475 |
|
|
|
1476 |
|
|
memset(kmap(page) + offset, 0, length);
|
1477 |
|
|
flush_dcache_page(page);
|
1478 |
|
|
kunmap(page);
|
1479 |
|
|
|
1480 |
|
|
BUFFER_TRACE(bh, "zeroed end of block");
|
1481 |
|
|
|
1482 |
|
|
err = 0;
|
1483 |
|
|
if (ext3_should_journal_data(inode)) {
|
1484 |
|
|
err = ext3_journal_dirty_metadata(handle, bh);
|
1485 |
|
|
} else {
|
1486 |
|
|
if (ext3_should_order_data(inode))
|
1487 |
|
|
err = ext3_journal_dirty_data(handle, bh, 0);
|
1488 |
|
|
__mark_buffer_dirty(bh);
|
1489 |
|
|
}
|
1490 |
|
|
|
1491 |
|
|
unlock:
|
1492 |
|
|
UnlockPage(page);
|
1493 |
|
|
page_cache_release(page);
|
1494 |
|
|
out:
|
1495 |
|
|
return err;
|
1496 |
|
|
}
|
1497 |
|
|
|
1498 |
|
|
/*
|
1499 |
|
|
* Probably it should be a library function... search for first non-zero word
|
1500 |
|
|
* or memcmp with zero_page, whatever is better for particular architecture.
|
1501 |
|
|
* Linus?
|
1502 |
|
|
*/
|
1503 |
|
|
static inline int all_zeroes(u32 *p, u32 *q)
|
1504 |
|
|
{
|
1505 |
|
|
while (p < q)
|
1506 |
|
|
if (*p++)
|
1507 |
|
|
return 0;
|
1508 |
|
|
return 1;
|
1509 |
|
|
}
|
1510 |
|
|
|
1511 |
|
|
/**
|
1512 |
|
|
* ext3_find_shared - find the indirect blocks for partial truncation.
|
1513 |
|
|
* @inode: inode in question
|
1514 |
|
|
* @depth: depth of the affected branch
|
1515 |
|
|
* @offsets: offsets of pointers in that branch (see ext3_block_to_path)
|
1516 |
|
|
* @chain: place to store the pointers to partial indirect blocks
|
1517 |
|
|
* @top: place to the (detached) top of branch
|
1518 |
|
|
*
|
1519 |
|
|
* This is a helper function used by ext3_truncate().
|
1520 |
|
|
*
|
1521 |
|
|
* When we do truncate() we may have to clean the ends of several
|
1522 |
|
|
* indirect blocks but leave the blocks themselves alive. Block is
|
1523 |
|
|
* partially truncated if some data below the new i_size is refered
|
1524 |
|
|
* from it (and it is on the path to the first completely truncated
|
1525 |
|
|
* data block, indeed). We have to free the top of that path along
|
1526 |
|
|
* with everything to the right of the path. Since no allocation
|
1527 |
|
|
* past the truncation point is possible until ext3_truncate()
|
1528 |
|
|
* finishes, we may safely do the latter, but top of branch may
|
1529 |
|
|
* require special attention - pageout below the truncation point
|
1530 |
|
|
* might try to populate it.
|
1531 |
|
|
*
|
1532 |
|
|
* We atomically detach the top of branch from the tree, store the
|
1533 |
|
|
* block number of its root in *@top, pointers to buffer_heads of
|
1534 |
|
|
* partially truncated blocks - in @chain[].bh and pointers to
|
1535 |
|
|
* their last elements that should not be removed - in
|
1536 |
|
|
* @chain[].p. Return value is the pointer to last filled element
|
1537 |
|
|
* of @chain.
|
1538 |
|
|
*
|
1539 |
|
|
* The work left to caller to do the actual freeing of subtrees:
|
1540 |
|
|
* a) free the subtree starting from *@top
|
1541 |
|
|
* b) free the subtrees whose roots are stored in
|
1542 |
|
|
* (@chain[i].p+1 .. end of @chain[i].bh->b_data)
|
1543 |
|
|
* c) free the subtrees growing from the inode past the @chain[0].
|
1544 |
|
|
* (no partially truncated stuff there). */
|
1545 |
|
|
|
1546 |
|
|
static Indirect *ext3_find_shared(struct inode *inode,
|
1547 |
|
|
int depth,
|
1548 |
|
|
int offsets[4],
|
1549 |
|
|
Indirect chain[4],
|
1550 |
|
|
u32 *top)
|
1551 |
|
|
{
|
1552 |
|
|
Indirect *partial, *p;
|
1553 |
|
|
int k, err;
|
1554 |
|
|
|
1555 |
|
|
*top = 0;
|
1556 |
|
|
/* Make k index the deepest non-null offest + 1 */
|
1557 |
|
|
for (k = depth; k > 1 && !offsets[k-1]; k--)
|
1558 |
|
|
;
|
1559 |
|
|
partial = ext3_get_branch(inode, k, offsets, chain, &err);
|
1560 |
|
|
/* Writer: pointers */
|
1561 |
|
|
if (!partial)
|
1562 |
|
|
partial = chain + k-1;
|
1563 |
|
|
/*
|
1564 |
|
|
* If the branch acquired continuation since we've looked at it -
|
1565 |
|
|
* fine, it should all survive and (new) top doesn't belong to us.
|
1566 |
|
|
*/
|
1567 |
|
|
if (!partial->key && *partial->p)
|
1568 |
|
|
/* Writer: end */
|
1569 |
|
|
goto no_top;
|
1570 |
|
|
for (p=partial; p>chain && all_zeroes((u32*)p->bh->b_data,p->p); p--)
|
1571 |
|
|
;
|
1572 |
|
|
/*
|
1573 |
|
|
* OK, we've found the last block that must survive. The rest of our
|
1574 |
|
|
* branch should be detached before unlocking. However, if that rest
|
1575 |
|
|
* of branch is all ours and does not grow immediately from the inode
|
1576 |
|
|
* it's easier to cheat and just decrement partial->p.
|
1577 |
|
|
*/
|
1578 |
|
|
if (p == chain + k - 1 && p > chain) {
|
1579 |
|
|
p->p--;
|
1580 |
|
|
} else {
|
1581 |
|
|
*top = *p->p;
|
1582 |
|
|
/* Nope, don't do this in ext3. Must leave the tree intact */
|
1583 |
|
|
#if 0
|
1584 |
|
|
*p->p = 0;
|
1585 |
|
|
#endif
|
1586 |
|
|
}
|
1587 |
|
|
/* Writer: end */
|
1588 |
|
|
|
1589 |
|
|
while(partial > p)
|
1590 |
|
|
{
|
1591 |
|
|
brelse(partial->bh);
|
1592 |
|
|
partial--;
|
1593 |
|
|
}
|
1594 |
|
|
no_top:
|
1595 |
|
|
return partial;
|
1596 |
|
|
}
|
1597 |
|
|
|
1598 |
|
|
/*
|
1599 |
|
|
* Zero a number of block pointers in either an inode or an indirect block.
|
1600 |
|
|
* If we restart the transaction we must again get write access to the
|
1601 |
|
|
* indirect block for further modification.
|
1602 |
|
|
*
|
1603 |
|
|
* We release `count' blocks on disk, but (last - first) may be greater
|
1604 |
|
|
* than `count' because there can be holes in there.
|
1605 |
|
|
*/
|
1606 |
|
|
static void
|
1607 |
|
|
ext3_clear_blocks(handle_t *handle, struct inode *inode, struct buffer_head *bh,
|
1608 |
|
|
unsigned long block_to_free, unsigned long count,
|
1609 |
|
|
u32 *first, u32 *last)
|
1610 |
|
|
{
|
1611 |
|
|
u32 *p;
|
1612 |
|
|
if (try_to_extend_transaction(handle, inode)) {
|
1613 |
|
|
if (bh) {
|
1614 |
|
|
BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
|
1615 |
|
|
ext3_journal_dirty_metadata(handle, bh);
|
1616 |
|
|
}
|
1617 |
|
|
ext3_mark_inode_dirty(handle, inode);
|
1618 |
|
|
ext3_journal_test_restart(handle, inode);
|
1619 |
|
|
if (bh) {
|
1620 |
|
|
BUFFER_TRACE(bh, "retaking write access");
|
1621 |
|
|
ext3_journal_get_write_access(handle, bh);
|
1622 |
|
|
}
|
1623 |
|
|
}
|
1624 |
|
|
|
1625 |
|
|
/*
|
1626 |
|
|
* Any buffers which are on the journal will be in memory. We find
|
1627 |
|
|
* them on the hash table so journal_revoke() will run journal_forget()
|
1628 |
|
|
* on them. We've already detached each block from the file, so
|
1629 |
|
|
* bforget() in journal_forget() should be safe.
|
1630 |
|
|
*
|
1631 |
|
|
* AKPM: turn on bforget in journal_forget()!!!
|
1632 |
|
|
*/
|
1633 |
|
|
for (p = first; p < last; p++) {
|
1634 |
|
|
u32 nr = le32_to_cpu(*p);
|
1635 |
|
|
if (nr) {
|
1636 |
|
|
struct buffer_head *bh;
|
1637 |
|
|
|
1638 |
|
|
*p = 0;
|
1639 |
|
|
bh = sb_get_hash_table(inode->i_sb, nr);
|
1640 |
|
|
ext3_forget(handle, 0, inode, bh, nr);
|
1641 |
|
|
}
|
1642 |
|
|
}
|
1643 |
|
|
|
1644 |
|
|
ext3_free_blocks(handle, inode, block_to_free, count);
|
1645 |
|
|
}
|
1646 |
|
|
|
1647 |
|
|
/**
|
1648 |
|
|
* ext3_free_data - free a list of data blocks
|
1649 |
|
|
* @handle: handle for this transaction
|
1650 |
|
|
* @inode: inode we are dealing with
|
1651 |
|
|
* @this_bh: indirect buffer_head which contains *@first and *@last
|
1652 |
|
|
* @first: array of block numbers
|
1653 |
|
|
* @last: points immediately past the end of array
|
1654 |
|
|
*
|
1655 |
|
|
* We are freeing all blocks refered from that array (numbers are stored as
|
1656 |
|
|
* little-endian 32-bit) and updating @inode->i_blocks appropriately.
|
1657 |
|
|
*
|
1658 |
|
|
* We accumulate contiguous runs of blocks to free. Conveniently, if these
|
1659 |
|
|
* blocks are contiguous then releasing them at one time will only affect one
|
1660 |
|
|
* or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
|
1661 |
|
|
* actually use a lot of journal space.
|
1662 |
|
|
*
|
1663 |
|
|
* @this_bh will be %NULL if @first and @last point into the inode's direct
|
1664 |
|
|
* block pointers.
|
1665 |
|
|
*/
|
1666 |
|
|
static void ext3_free_data(handle_t *handle, struct inode *inode,
|
1667 |
|
|
struct buffer_head *this_bh, u32 *first, u32 *last)
|
1668 |
|
|
{
|
1669 |
|
|
unsigned long block_to_free = 0; /* Starting block # of a run */
|
1670 |
|
|
unsigned long count = 0; /* Number of blocks in the run */
|
1671 |
|
|
u32 *block_to_free_p = NULL; /* Pointer into inode/ind
|
1672 |
|
|
corresponding to
|
1673 |
|
|
block_to_free */
|
1674 |
|
|
unsigned long nr; /* Current block # */
|
1675 |
|
|
u32 *p; /* Pointer into inode/ind
|
1676 |
|
|
for current block */
|
1677 |
|
|
int err;
|
1678 |
|
|
|
1679 |
|
|
if (this_bh) { /* For indirect block */
|
1680 |
|
|
BUFFER_TRACE(this_bh, "get_write_access");
|
1681 |
|
|
err = ext3_journal_get_write_access(handle, this_bh);
|
1682 |
|
|
/* Important: if we can't update the indirect pointers
|
1683 |
|
|
* to the blocks, we can't free them. */
|
1684 |
|
|
if (err)
|
1685 |
|
|
return;
|
1686 |
|
|
}
|
1687 |
|
|
|
1688 |
|
|
for (p = first; p < last; p++) {
|
1689 |
|
|
nr = le32_to_cpu(*p);
|
1690 |
|
|
if (nr) {
|
1691 |
|
|
/* accumulate blocks to free if they're contiguous */
|
1692 |
|
|
if (count == 0) {
|
1693 |
|
|
block_to_free = nr;
|
1694 |
|
|
block_to_free_p = p;
|
1695 |
|
|
count = 1;
|
1696 |
|
|
} else if (nr == block_to_free + count) {
|
1697 |
|
|
count++;
|
1698 |
|
|
} else {
|
1699 |
|
|
ext3_clear_blocks(handle, inode, this_bh,
|
1700 |
|
|
block_to_free,
|
1701 |
|
|
count, block_to_free_p, p);
|
1702 |
|
|
block_to_free = nr;
|
1703 |
|
|
block_to_free_p = p;
|
1704 |
|
|
count = 1;
|
1705 |
|
|
}
|
1706 |
|
|
}
|
1707 |
|
|
}
|
1708 |
|
|
|
1709 |
|
|
if (count > 0)
|
1710 |
|
|
ext3_clear_blocks(handle, inode, this_bh, block_to_free,
|
1711 |
|
|
count, block_to_free_p, p);
|
1712 |
|
|
|
1713 |
|
|
if (this_bh) {
|
1714 |
|
|
BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
|
1715 |
|
|
ext3_journal_dirty_metadata(handle, this_bh);
|
1716 |
|
|
}
|
1717 |
|
|
}
|
1718 |
|
|
|
1719 |
|
|
/**
|
1720 |
|
|
* ext3_free_branches - free an array of branches
|
1721 |
|
|
* @handle: JBD handle for this transaction
|
1722 |
|
|
* @inode: inode we are dealing with
|
1723 |
|
|
* @parent_bh: the buffer_head which contains *@first and *@last
|
1724 |
|
|
* @first: array of block numbers
|
1725 |
|
|
* @last: pointer immediately past the end of array
|
1726 |
|
|
* @depth: depth of the branches to free
|
1727 |
|
|
*
|
1728 |
|
|
* We are freeing all blocks refered from these branches (numbers are
|
1729 |
|
|
* stored as little-endian 32-bit) and updating @inode->i_blocks
|
1730 |
|
|
* appropriately.
|
1731 |
|
|
*/
|
1732 |
|
|
static void ext3_free_branches(handle_t *handle, struct inode *inode,
|
1733 |
|
|
struct buffer_head *parent_bh,
|
1734 |
|
|
u32 *first, u32 *last, int depth)
|
1735 |
|
|
{
|
1736 |
|
|
unsigned long nr;
|
1737 |
|
|
u32 *p;
|
1738 |
|
|
|
1739 |
|
|
if (is_handle_aborted(handle))
|
1740 |
|
|
return;
|
1741 |
|
|
|
1742 |
|
|
if (depth--) {
|
1743 |
|
|
struct buffer_head *bh;
|
1744 |
|
|
int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
|
1745 |
|
|
p = last;
|
1746 |
|
|
while (--p >= first) {
|
1747 |
|
|
nr = le32_to_cpu(*p);
|
1748 |
|
|
if (!nr)
|
1749 |
|
|
continue; /* A hole */
|
1750 |
|
|
|
1751 |
|
|
/* Go read the buffer for the next level down */
|
1752 |
|
|
bh = sb_bread(inode->i_sb, nr);
|
1753 |
|
|
|
1754 |
|
|
/*
|
1755 |
|
|
* A read failure? Report error and clear slot
|
1756 |
|
|
* (should be rare).
|
1757 |
|
|
*/
|
1758 |
|
|
if (!bh) {
|
1759 |
|
|
ext3_error(inode->i_sb, "ext3_free_branches",
|
1760 |
|
|
"Read failure, inode=%ld, block=%ld",
|
1761 |
|
|
inode->i_ino, nr);
|
1762 |
|
|
continue;
|
1763 |
|
|
}
|
1764 |
|
|
|
1765 |
|
|
/* This zaps the entire block. Bottom up. */
|
1766 |
|
|
BUFFER_TRACE(bh, "free child branches");
|
1767 |
|
|
ext3_free_branches(handle, inode, bh, (u32*)bh->b_data,
|
1768 |
|
|
(u32*)bh->b_data + addr_per_block,
|
1769 |
|
|
depth);
|
1770 |
|
|
|
1771 |
|
|
/*
|
1772 |
|
|
* We've probably journalled the indirect block several
|
1773 |
|
|
* times during the truncate. But it's no longer
|
1774 |
|
|
* needed and we now drop it from the transaction via
|
1775 |
|
|
* journal_revoke().
|
1776 |
|
|
*
|
1777 |
|
|
* That's easy if it's exclusively part of this
|
1778 |
|
|
* transaction. But if it's part of the committing
|
1779 |
|
|
* transaction then journal_forget() will simply
|
1780 |
|
|
* brelse() it. That means that if the underlying
|
1781 |
|
|
* block is reallocated in ext3_get_block(),
|
1782 |
|
|
* unmap_underlying_metadata() will find this block
|
1783 |
|
|
* and will try to get rid of it. damn, damn.
|
1784 |
|
|
*
|
1785 |
|
|
* If this block has already been committed to the
|
1786 |
|
|
* journal, a revoke record will be written. And
|
1787 |
|
|
* revoke records must be emitted *before* clearing
|
1788 |
|
|
* this block's bit in the bitmaps.
|
1789 |
|
|
*/
|
1790 |
|
|
ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
|
1791 |
|
|
|
1792 |
|
|
/*
|
1793 |
|
|
* Everything below this this pointer has been
|
1794 |
|
|
* released. Now let this top-of-subtree go.
|
1795 |
|
|
*
|
1796 |
|
|
* We want the freeing of this indirect block to be
|
1797 |
|
|
* atomic in the journal with the updating of the
|
1798 |
|
|
* bitmap block which owns it. So make some room in
|
1799 |
|
|
* the journal.
|
1800 |
|
|
*
|
1801 |
|
|
* We zero the parent pointer *after* freeing its
|
1802 |
|
|
* pointee in the bitmaps, so if extend_transaction()
|
1803 |
|
|
* for some reason fails to put the bitmap changes and
|
1804 |
|
|
* the release into the same transaction, recovery
|
1805 |
|
|
* will merely complain about releasing a free block,
|
1806 |
|
|
* rather than leaking blocks.
|
1807 |
|
|
*/
|
1808 |
|
|
if (is_handle_aborted(handle))
|
1809 |
|
|
return;
|
1810 |
|
|
if (try_to_extend_transaction(handle, inode)) {
|
1811 |
|
|
ext3_mark_inode_dirty(handle, inode);
|
1812 |
|
|
ext3_journal_test_restart(handle, inode);
|
1813 |
|
|
}
|
1814 |
|
|
|
1815 |
|
|
ext3_free_blocks(handle, inode, nr, 1);
|
1816 |
|
|
|
1817 |
|
|
if (parent_bh) {
|
1818 |
|
|
/*
|
1819 |
|
|
* The block which we have just freed is
|
1820 |
|
|
* pointed to by an indirect block: journal it
|
1821 |
|
|
*/
|
1822 |
|
|
BUFFER_TRACE(parent_bh, "get_write_access");
|
1823 |
|
|
if (!ext3_journal_get_write_access(handle,
|
1824 |
|
|
parent_bh)){
|
1825 |
|
|
*p = 0;
|
1826 |
|
|
BUFFER_TRACE(parent_bh,
|
1827 |
|
|
"call ext3_journal_dirty_metadata");
|
1828 |
|
|
ext3_journal_dirty_metadata(handle,
|
1829 |
|
|
parent_bh);
|
1830 |
|
|
}
|
1831 |
|
|
}
|
1832 |
|
|
}
|
1833 |
|
|
} else {
|
1834 |
|
|
/* We have reached the bottom of the tree. */
|
1835 |
|
|
BUFFER_TRACE(parent_bh, "free data blocks");
|
1836 |
|
|
ext3_free_data(handle, inode, parent_bh, first, last);
|
1837 |
|
|
}
|
1838 |
|
|
}
|
1839 |
|
|
|
1840 |
|
|
/*
|
1841 |
|
|
* ext3_truncate()
|
1842 |
|
|
*
|
1843 |
|
|
* We block out ext3_get_block() block instantiations across the entire
|
1844 |
|
|
* transaction, and VFS/VM ensures that ext3_truncate() cannot run
|
1845 |
|
|
* simultaneously on behalf of the same inode.
|
1846 |
|
|
*
|
1847 |
|
|
* As we work through the truncate and commmit bits of it to the journal there
|
1848 |
|
|
* is one core, guiding principle: the file's tree must always be consistent on
|
1849 |
|
|
* disk. We must be able to restart the truncate after a crash.
|
1850 |
|
|
*
|
1851 |
|
|
* The file's tree may be transiently inconsistent in memory (although it
|
1852 |
|
|
* probably isn't), but whenever we close off and commit a journal transaction,
|
1853 |
|
|
* the contents of (the filesystem + the journal) must be consistent and
|
1854 |
|
|
* restartable. It's pretty simple, really: bottom up, right to left (although
|
1855 |
|
|
* left-to-right works OK too).
|
1856 |
|
|
*
|
1857 |
|
|
* Note that at recovery time, journal replay occurs *before* the restart of
|
1858 |
|
|
* truncate against the orphan inode list.
|
1859 |
|
|
*
|
1860 |
|
|
* The committed inode has the new, desired i_size (which is the same as
|
1861 |
|
|
* i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
|
1862 |
|
|
* that this inode's truncate did not complete and it will again call
|
1863 |
|
|
* ext3_truncate() to have another go. So there will be instantiated blocks
|
1864 |
|
|
* to the right of the truncation point in a crashed ext3 filesystem. But
|
1865 |
|
|
* that's fine - as long as they are linked from the inode, the post-crash
|
1866 |
|
|
* ext3_truncate() run will find them and release them.
|
1867 |
|
|
*/
|
1868 |
|
|
|
1869 |
|
|
void ext3_truncate(struct inode * inode)
|
1870 |
|
|
{
|
1871 |
|
|
handle_t *handle;
|
1872 |
|
|
u32 *i_data = inode->u.ext3_i.i_data;
|
1873 |
|
|
int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
|
1874 |
|
|
int offsets[4];
|
1875 |
|
|
Indirect chain[4];
|
1876 |
|
|
Indirect *partial;
|
1877 |
|
|
int nr = 0;
|
1878 |
|
|
int n;
|
1879 |
|
|
long last_block;
|
1880 |
|
|
unsigned blocksize;
|
1881 |
|
|
|
1882 |
|
|
if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
|
1883 |
|
|
S_ISLNK(inode->i_mode)))
|
1884 |
|
|
return;
|
1885 |
|
|
if (ext3_inode_is_fast_symlink(inode))
|
1886 |
|
|
return;
|
1887 |
|
|
if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
|
1888 |
|
|
return;
|
1889 |
|
|
|
1890 |
|
|
ext3_discard_prealloc(inode);
|
1891 |
|
|
|
1892 |
|
|
handle = start_transaction(inode);
|
1893 |
|
|
if (IS_ERR(handle))
|
1894 |
|
|
return; /* AKPM: return what? */
|
1895 |
|
|
|
1896 |
|
|
blocksize = inode->i_sb->s_blocksize;
|
1897 |
|
|
last_block = (inode->i_size + blocksize-1)
|
1898 |
|
|
>> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
|
1899 |
|
|
|
1900 |
|
|
ext3_block_truncate_page(handle, inode->i_mapping, inode->i_size);
|
1901 |
|
|
|
1902 |
|
|
|
1903 |
|
|
n = ext3_block_to_path(inode, last_block, offsets);
|
1904 |
|
|
if (n == 0)
|
1905 |
|
|
goto out_stop; /* error */
|
1906 |
|
|
|
1907 |
|
|
/*
|
1908 |
|
|
* OK. This truncate is going to happen. We add the inode to the
|
1909 |
|
|
* orphan list, so that if this truncate spans multiple transactions,
|
1910 |
|
|
* and we crash, we will resume the truncate when the filesystem
|
1911 |
|
|
* recovers. It also marks the inode dirty, to catch the new size.
|
1912 |
|
|
*
|
1913 |
|
|
* Implication: the file must always be in a sane, consistent
|
1914 |
|
|
* truncatable state while each transaction commits.
|
1915 |
|
|
*/
|
1916 |
|
|
if (ext3_orphan_add(handle, inode))
|
1917 |
|
|
goto out_stop;
|
1918 |
|
|
|
1919 |
|
|
/*
|
1920 |
|
|
* The orphan list entry will now protect us from any crash which
|
1921 |
|
|
* occurs before the truncate completes, so it is now safe to propagate
|
1922 |
|
|
* the new, shorter inode size (held for now in i_size) into the
|
1923 |
|
|
* on-disk inode. We do this via i_disksize, which is the value which
|
1924 |
|
|
* ext3 *really* writes onto the disk inode.
|
1925 |
|
|
*/
|
1926 |
|
|
inode->u.ext3_i.i_disksize = inode->i_size;
|
1927 |
|
|
|
1928 |
|
|
/*
|
1929 |
|
|
* From here we block out all ext3_get_block() callers who want to
|
1930 |
|
|
* modify the block allocation tree.
|
1931 |
|
|
*/
|
1932 |
|
|
down_write(&inode->u.ext3_i.truncate_sem);
|
1933 |
|
|
|
1934 |
|
|
if (n == 1) { /* direct blocks */
|
1935 |
|
|
ext3_free_data(handle, inode, NULL, i_data+offsets[0],
|
1936 |
|
|
i_data + EXT3_NDIR_BLOCKS);
|
1937 |
|
|
goto do_indirects;
|
1938 |
|
|
}
|
1939 |
|
|
|
1940 |
|
|
partial = ext3_find_shared(inode, n, offsets, chain, &nr);
|
1941 |
|
|
/* Kill the top of shared branch (not detached) */
|
1942 |
|
|
if (nr) {
|
1943 |
|
|
if (partial == chain) {
|
1944 |
|
|
/* Shared branch grows from the inode */
|
1945 |
|
|
ext3_free_branches(handle, inode, NULL,
|
1946 |
|
|
&nr, &nr+1, (chain+n-1) - partial);
|
1947 |
|
|
*partial->p = 0;
|
1948 |
|
|
/*
|
1949 |
|
|
* We mark the inode dirty prior to restart,
|
1950 |
|
|
* and prior to stop. No need for it here.
|
1951 |
|
|
*/
|
1952 |
|
|
} else {
|
1953 |
|
|
/* Shared branch grows from an indirect block */
|
1954 |
|
|
BUFFER_TRACE(partial->bh, "get_write_access");
|
1955 |
|
|
ext3_free_branches(handle, inode, partial->bh,
|
1956 |
|
|
partial->p,
|
1957 |
|
|
partial->p+1, (chain+n-1) - partial);
|
1958 |
|
|
}
|
1959 |
|
|
}
|
1960 |
|
|
/* Clear the ends of indirect blocks on the shared branch */
|
1961 |
|
|
while (partial > chain) {
|
1962 |
|
|
ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
|
1963 |
|
|
(u32*)partial->bh->b_data + addr_per_block,
|
1964 |
|
|
(chain+n-1) - partial);
|
1965 |
|
|
BUFFER_TRACE(partial->bh, "call brelse");
|
1966 |
|
|
brelse (partial->bh);
|
1967 |
|
|
partial--;
|
1968 |
|
|
}
|
1969 |
|
|
do_indirects:
|
1970 |
|
|
/* Kill the remaining (whole) subtrees */
|
1971 |
|
|
switch (offsets[0]) {
|
1972 |
|
|
default:
|
1973 |
|
|
nr = i_data[EXT3_IND_BLOCK];
|
1974 |
|
|
if (nr) {
|
1975 |
|
|
ext3_free_branches(handle, inode, NULL,
|
1976 |
|
|
&nr, &nr+1, 1);
|
1977 |
|
|
i_data[EXT3_IND_BLOCK] = 0;
|
1978 |
|
|
}
|
1979 |
|
|
case EXT3_IND_BLOCK:
|
1980 |
|
|
nr = i_data[EXT3_DIND_BLOCK];
|
1981 |
|
|
if (nr) {
|
1982 |
|
|
ext3_free_branches(handle, inode, NULL,
|
1983 |
|
|
&nr, &nr+1, 2);
|
1984 |
|
|
i_data[EXT3_DIND_BLOCK] = 0;
|
1985 |
|
|
}
|
1986 |
|
|
case EXT3_DIND_BLOCK:
|
1987 |
|
|
nr = i_data[EXT3_TIND_BLOCK];
|
1988 |
|
|
if (nr) {
|
1989 |
|
|
ext3_free_branches(handle, inode, NULL,
|
1990 |
|
|
&nr, &nr+1, 3);
|
1991 |
|
|
i_data[EXT3_TIND_BLOCK] = 0;
|
1992 |
|
|
}
|
1993 |
|
|
case EXT3_TIND_BLOCK:
|
1994 |
|
|
;
|
1995 |
|
|
}
|
1996 |
|
|
up_write(&inode->u.ext3_i.truncate_sem);
|
1997 |
|
|
inode->i_mtime = inode->i_ctime = CURRENT_TIME;
|
1998 |
|
|
ext3_mark_inode_dirty(handle, inode);
|
1999 |
|
|
|
2000 |
|
|
/* In a multi-transaction truncate, we only make the final
|
2001 |
|
|
* transaction synchronous */
|
2002 |
|
|
if (IS_SYNC(inode))
|
2003 |
|
|
handle->h_sync = 1;
|
2004 |
|
|
out_stop:
|
2005 |
|
|
/*
|
2006 |
|
|
* If this was a simple ftruncate(), and the file will remain alive
|
2007 |
|
|
* then we need to clear up the orphan record which we created above.
|
2008 |
|
|
* However, if this was a real unlink then we were called by
|
2009 |
|
|
* ext3_delete_inode(), and we allow that function to clean up the
|
2010 |
|
|
* orphan info for us.
|
2011 |
|
|
*/
|
2012 |
|
|
if (inode->i_nlink)
|
2013 |
|
|
ext3_orphan_del(handle, inode);
|
2014 |
|
|
|
2015 |
|
|
ext3_journal_stop(handle, inode);
|
2016 |
|
|
}
|
2017 |
|
|
|
2018 |
|
|
/*
|
2019 |
|
|
* ext3_get_inode_loc returns with an extra refcount against the
|
2020 |
|
|
* inode's underlying buffer_head on success.
|
2021 |
|
|
*/
|
2022 |
|
|
|
2023 |
|
|
int ext3_get_inode_loc (struct inode *inode, struct ext3_iloc *iloc)
|
2024 |
|
|
{
|
2025 |
|
|
struct buffer_head *bh = 0;
|
2026 |
|
|
unsigned long block;
|
2027 |
|
|
unsigned long block_group;
|
2028 |
|
|
unsigned long group_desc;
|
2029 |
|
|
unsigned long desc;
|
2030 |
|
|
unsigned long offset;
|
2031 |
|
|
struct ext3_group_desc * gdp;
|
2032 |
|
|
|
2033 |
|
|
if ((inode->i_ino != EXT3_ROOT_INO &&
|
2034 |
|
|
inode->i_ino != EXT3_ACL_IDX_INO &&
|
2035 |
|
|
inode->i_ino != EXT3_ACL_DATA_INO &&
|
2036 |
|
|
inode->i_ino != EXT3_JOURNAL_INO &&
|
2037 |
|
|
inode->i_ino < EXT3_FIRST_INO(inode->i_sb)) ||
|
2038 |
|
|
inode->i_ino > le32_to_cpu(
|
2039 |
|
|
inode->i_sb->u.ext3_sb.s_es->s_inodes_count)) {
|
2040 |
|
|
ext3_error (inode->i_sb, "ext3_get_inode_loc",
|
2041 |
|
|
"bad inode number: %lu", inode->i_ino);
|
2042 |
|
|
goto bad_inode;
|
2043 |
|
|
}
|
2044 |
|
|
block_group = (inode->i_ino - 1) / EXT3_INODES_PER_GROUP(inode->i_sb);
|
2045 |
|
|
if (block_group >= inode->i_sb->u.ext3_sb.s_groups_count) {
|
2046 |
|
|
ext3_error (inode->i_sb, "ext3_get_inode_loc",
|
2047 |
|
|
"group >= groups count");
|
2048 |
|
|
goto bad_inode;
|
2049 |
|
|
}
|
2050 |
|
|
group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(inode->i_sb);
|
2051 |
|
|
desc = block_group & (EXT3_DESC_PER_BLOCK(inode->i_sb) - 1);
|
2052 |
|
|
bh = inode->i_sb->u.ext3_sb.s_group_desc[group_desc];
|
2053 |
|
|
if (!bh) {
|
2054 |
|
|
ext3_error (inode->i_sb, "ext3_get_inode_loc",
|
2055 |
|
|
"Descriptor not loaded");
|
2056 |
|
|
goto bad_inode;
|
2057 |
|
|
}
|
2058 |
|
|
|
2059 |
|
|
gdp = (struct ext3_group_desc *) bh->b_data;
|
2060 |
|
|
/*
|
2061 |
|
|
* Figure out the offset within the block group inode table
|
2062 |
|
|
*/
|
2063 |
|
|
offset = ((inode->i_ino - 1) % EXT3_INODES_PER_GROUP(inode->i_sb)) *
|
2064 |
|
|
EXT3_INODE_SIZE(inode->i_sb);
|
2065 |
|
|
block = le32_to_cpu(gdp[desc].bg_inode_table) +
|
2066 |
|
|
(offset >> EXT3_BLOCK_SIZE_BITS(inode->i_sb));
|
2067 |
|
|
if (!(bh = sb_bread(inode->i_sb, block))) {
|
2068 |
|
|
ext3_error (inode->i_sb, "ext3_get_inode_loc",
|
2069 |
|
|
"unable to read inode block - "
|
2070 |
|
|
"inode=%lu, block=%lu", inode->i_ino, block);
|
2071 |
|
|
goto bad_inode;
|
2072 |
|
|
}
|
2073 |
|
|
offset &= (EXT3_BLOCK_SIZE(inode->i_sb) - 1);
|
2074 |
|
|
|
2075 |
|
|
iloc->bh = bh;
|
2076 |
|
|
iloc->raw_inode = (struct ext3_inode *) (bh->b_data + offset);
|
2077 |
|
|
iloc->block_group = block_group;
|
2078 |
|
|
|
2079 |
|
|
return 0;
|
2080 |
|
|
|
2081 |
|
|
bad_inode:
|
2082 |
|
|
return -EIO;
|
2083 |
|
|
}
|
2084 |
|
|
|
2085 |
|
|
void ext3_set_inode_flags(struct inode *inode)
|
2086 |
|
|
{
|
2087 |
|
|
unsigned int flags = inode->u.ext3_i.i_flags;
|
2088 |
|
|
|
2089 |
|
|
inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME);
|
2090 |
|
|
if (flags & EXT3_SYNC_FL)
|
2091 |
|
|
inode->i_flags |= S_SYNC;
|
2092 |
|
|
if (flags & EXT3_APPEND_FL)
|
2093 |
|
|
inode->i_flags |= S_APPEND;
|
2094 |
|
|
if (flags & EXT3_IMMUTABLE_FL)
|
2095 |
|
|
inode->i_flags |= S_IMMUTABLE;
|
2096 |
|
|
if (flags & EXT3_NOATIME_FL)
|
2097 |
|
|
inode->i_flags |= S_NOATIME;
|
2098 |
|
|
}
|
2099 |
|
|
|
2100 |
|
|
|
2101 |
|
|
void ext3_read_inode(struct inode * inode)
|
2102 |
|
|
{
|
2103 |
|
|
struct ext3_iloc iloc;
|
2104 |
|
|
struct ext3_inode *raw_inode;
|
2105 |
|
|
struct buffer_head *bh;
|
2106 |
|
|
int block;
|
2107 |
|
|
|
2108 |
|
|
if(ext3_get_inode_loc(inode, &iloc))
|
2109 |
|
|
goto bad_inode;
|
2110 |
|
|
bh = iloc.bh;
|
2111 |
|
|
raw_inode = iloc.raw_inode;
|
2112 |
|
|
init_rwsem(&inode->u.ext3_i.truncate_sem);
|
2113 |
|
|
inode->i_mode = le16_to_cpu(raw_inode->i_mode);
|
2114 |
|
|
inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
|
2115 |
|
|
inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
|
2116 |
|
|
if(!(test_opt (inode->i_sb, NO_UID32))) {
|
2117 |
|
|
inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
|
2118 |
|
|
inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
|
2119 |
|
|
}
|
2120 |
|
|
inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
|
2121 |
|
|
inode->i_size = le32_to_cpu(raw_inode->i_size);
|
2122 |
|
|
inode->i_atime = le32_to_cpu(raw_inode->i_atime);
|
2123 |
|
|
inode->i_ctime = le32_to_cpu(raw_inode->i_ctime);
|
2124 |
|
|
inode->i_mtime = le32_to_cpu(raw_inode->i_mtime);
|
2125 |
|
|
inode->u.ext3_i.i_dtime = le32_to_cpu(raw_inode->i_dtime);
|
2126 |
|
|
/* We now have enough fields to check if the inode was active or not.
|
2127 |
|
|
* This is needed because nfsd might try to access dead inodes
|
2128 |
|
|
* the test is that same one that e2fsck uses
|
2129 |
|
|
* NeilBrown 1999oct15
|
2130 |
|
|
*/
|
2131 |
|
|
if (inode->i_nlink == 0) {
|
2132 |
|
|
if (inode->i_mode == 0 ||
|
2133 |
|
|
!(inode->i_sb->u.ext3_sb.s_mount_state & EXT3_ORPHAN_FS)) {
|
2134 |
|
|
/* this inode is deleted */
|
2135 |
|
|
brelse (bh);
|
2136 |
|
|
goto bad_inode;
|
2137 |
|
|
}
|
2138 |
|
|
/* The only unlinked inodes we let through here have
|
2139 |
|
|
* valid i_mode and are being read by the orphan
|
2140 |
|
|
* recovery code: that's fine, we're about to complete
|
2141 |
|
|
* the process of deleting those. */
|
2142 |
|
|
}
|
2143 |
|
|
inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size
|
2144 |
|
|
* (for stat), not the fs block
|
2145 |
|
|
* size */
|
2146 |
|
|
inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
|
2147 |
|
|
inode->i_version = ++event;
|
2148 |
|
|
inode->u.ext3_i.i_flags = le32_to_cpu(raw_inode->i_flags);
|
2149 |
|
|
#ifdef EXT3_FRAGMENTS
|
2150 |
|
|
inode->u.ext3_i.i_faddr = le32_to_cpu(raw_inode->i_faddr);
|
2151 |
|
|
inode->u.ext3_i.i_frag_no = raw_inode->i_frag;
|
2152 |
|
|
inode->u.ext3_i.i_frag_size = raw_inode->i_fsize;
|
2153 |
|
|
#endif
|
2154 |
|
|
inode->u.ext3_i.i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
|
2155 |
|
|
if (!S_ISREG(inode->i_mode)) {
|
2156 |
|
|
inode->u.ext3_i.i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
|
2157 |
|
|
} else {
|
2158 |
|
|
inode->i_size |=
|
2159 |
|
|
((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
|
2160 |
|
|
}
|
2161 |
|
|
inode->u.ext3_i.i_disksize = inode->i_size;
|
2162 |
|
|
inode->i_generation = le32_to_cpu(raw_inode->i_generation);
|
2163 |
|
|
#ifdef EXT3_PREALLOCATE
|
2164 |
|
|
inode->u.ext3_i.i_prealloc_count = 0;
|
2165 |
|
|
#endif
|
2166 |
|
|
inode->u.ext3_i.i_block_group = iloc.block_group;
|
2167 |
|
|
|
2168 |
|
|
/*
|
2169 |
|
|
* NOTE! The in-memory inode i_data array is in little-endian order
|
2170 |
|
|
* even on big-endian machines: we do NOT byteswap the block numbers!
|
2171 |
|
|
*/
|
2172 |
|
|
for (block = 0; block < EXT3_N_BLOCKS; block++)
|
2173 |
|
|
inode->u.ext3_i.i_data[block] = iloc.raw_inode->i_block[block];
|
2174 |
|
|
INIT_LIST_HEAD(&inode->u.ext3_i.i_orphan);
|
2175 |
|
|
|
2176 |
|
|
if (inode->i_ino == EXT3_ACL_IDX_INO ||
|
2177 |
|
|
inode->i_ino == EXT3_ACL_DATA_INO)
|
2178 |
|
|
/* Nothing to do */ ;
|
2179 |
|
|
else if (S_ISREG(inode->i_mode)) {
|
2180 |
|
|
inode->i_op = &ext3_file_inode_operations;
|
2181 |
|
|
inode->i_fop = &ext3_file_operations;
|
2182 |
|
|
inode->i_mapping->a_ops = &ext3_aops;
|
2183 |
|
|
} else if (S_ISDIR(inode->i_mode)) {
|
2184 |
|
|
inode->i_op = &ext3_dir_inode_operations;
|
2185 |
|
|
inode->i_fop = &ext3_dir_operations;
|
2186 |
|
|
} else if (S_ISLNK(inode->i_mode)) {
|
2187 |
|
|
if (ext3_inode_is_fast_symlink(inode))
|
2188 |
|
|
inode->i_op = &ext3_fast_symlink_inode_operations;
|
2189 |
|
|
else {
|
2190 |
|
|
inode->i_op = &page_symlink_inode_operations;
|
2191 |
|
|
inode->i_mapping->a_ops = &ext3_aops;
|
2192 |
|
|
}
|
2193 |
|
|
} else
|
2194 |
|
|
init_special_inode(inode, inode->i_mode,
|
2195 |
|
|
le32_to_cpu(iloc.raw_inode->i_block[0]));
|
2196 |
|
|
brelse(iloc.bh);
|
2197 |
|
|
ext3_set_inode_flags(inode);
|
2198 |
|
|
return;
|
2199 |
|
|
|
2200 |
|
|
bad_inode:
|
2201 |
|
|
make_bad_inode(inode);
|
2202 |
|
|
return;
|
2203 |
|
|
}
|
2204 |
|
|
|
2205 |
|
|
/*
|
2206 |
|
|
* Post the struct inode info into an on-disk inode location in the
|
2207 |
|
|
* buffer-cache. This gobbles the caller's reference to the
|
2208 |
|
|
* buffer_head in the inode location struct.
|
2209 |
|
|
*/
|
2210 |
|
|
|
2211 |
|
|
static int ext3_do_update_inode(handle_t *handle,
|
2212 |
|
|
struct inode *inode,
|
2213 |
|
|
struct ext3_iloc *iloc)
|
2214 |
|
|
{
|
2215 |
|
|
struct ext3_inode *raw_inode = iloc->raw_inode;
|
2216 |
|
|
struct buffer_head *bh = iloc->bh;
|
2217 |
|
|
int err = 0, rc, block;
|
2218 |
|
|
|
2219 |
|
|
if (handle) {
|
2220 |
|
|
BUFFER_TRACE(bh, "get_write_access");
|
2221 |
|
|
err = ext3_journal_get_write_access(handle, bh);
|
2222 |
|
|
if (err)
|
2223 |
|
|
goto out_brelse;
|
2224 |
|
|
}
|
2225 |
|
|
/* For fields not not tracking in the in-memory inode,
|
2226 |
|
|
* initialise them to zero for new inodes. */
|
2227 |
|
|
if (EXT3_I(inode)->i_state & EXT3_STATE_NEW)
|
2228 |
|
|
memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
|
2229 |
|
|
|
2230 |
|
|
raw_inode->i_mode = cpu_to_le16(inode->i_mode);
|
2231 |
|
|
if(!(test_opt(inode->i_sb, NO_UID32))) {
|
2232 |
|
|
raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
|
2233 |
|
|
raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
|
2234 |
|
|
/*
|
2235 |
|
|
* Fix up interoperability with old kernels. Otherwise, old inodes get
|
2236 |
|
|
* re-used with the upper 16 bits of the uid/gid intact
|
2237 |
|
|
*/
|
2238 |
|
|
if(!inode->u.ext3_i.i_dtime) {
|
2239 |
|
|
raw_inode->i_uid_high =
|
2240 |
|
|
cpu_to_le16(high_16_bits(inode->i_uid));
|
2241 |
|
|
raw_inode->i_gid_high =
|
2242 |
|
|
cpu_to_le16(high_16_bits(inode->i_gid));
|
2243 |
|
|
} else {
|
2244 |
|
|
raw_inode->i_uid_high = 0;
|
2245 |
|
|
raw_inode->i_gid_high = 0;
|
2246 |
|
|
}
|
2247 |
|
|
} else {
|
2248 |
|
|
raw_inode->i_uid_low =
|
2249 |
|
|
cpu_to_le16(fs_high2lowuid(inode->i_uid));
|
2250 |
|
|
raw_inode->i_gid_low =
|
2251 |
|
|
cpu_to_le16(fs_high2lowgid(inode->i_gid));
|
2252 |
|
|
raw_inode->i_uid_high = 0;
|
2253 |
|
|
raw_inode->i_gid_high = 0;
|
2254 |
|
|
}
|
2255 |
|
|
raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
|
2256 |
|
|
raw_inode->i_size = cpu_to_le32(inode->u.ext3_i.i_disksize);
|
2257 |
|
|
raw_inode->i_atime = cpu_to_le32(inode->i_atime);
|
2258 |
|
|
raw_inode->i_ctime = cpu_to_le32(inode->i_ctime);
|
2259 |
|
|
raw_inode->i_mtime = cpu_to_le32(inode->i_mtime);
|
2260 |
|
|
raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
|
2261 |
|
|
raw_inode->i_dtime = cpu_to_le32(inode->u.ext3_i.i_dtime);
|
2262 |
|
|
raw_inode->i_flags = cpu_to_le32(inode->u.ext3_i.i_flags);
|
2263 |
|
|
#ifdef EXT3_FRAGMENTS
|
2264 |
|
|
raw_inode->i_faddr = cpu_to_le32(inode->u.ext3_i.i_faddr);
|
2265 |
|
|
raw_inode->i_frag = inode->u.ext3_i.i_frag_no;
|
2266 |
|
|
raw_inode->i_fsize = inode->u.ext3_i.i_frag_size;
|
2267 |
|
|
#endif
|
2268 |
|
|
raw_inode->i_file_acl = cpu_to_le32(inode->u.ext3_i.i_file_acl);
|
2269 |
|
|
if (!S_ISREG(inode->i_mode)) {
|
2270 |
|
|
raw_inode->i_dir_acl = cpu_to_le32(inode->u.ext3_i.i_dir_acl);
|
2271 |
|
|
} else {
|
2272 |
|
|
raw_inode->i_size_high =
|
2273 |
|
|
cpu_to_le32(inode->u.ext3_i.i_disksize >> 32);
|
2274 |
|
|
if (inode->u.ext3_i.i_disksize > 0x7fffffffULL) {
|
2275 |
|
|
struct super_block *sb = inode->i_sb;
|
2276 |
|
|
if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
|
2277 |
|
|
EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
|
2278 |
|
|
EXT3_SB(sb)->s_es->s_rev_level ==
|
2279 |
|
|
cpu_to_le32(EXT3_GOOD_OLD_REV)) {
|
2280 |
|
|
/* If this is the first large file
|
2281 |
|
|
* created, add a flag to the superblock.
|
2282 |
|
|
*/
|
2283 |
|
|
err = ext3_journal_get_write_access(handle,
|
2284 |
|
|
sb->u.ext3_sb.s_sbh);
|
2285 |
|
|
if (err)
|
2286 |
|
|
goto out_brelse;
|
2287 |
|
|
ext3_update_dynamic_rev(sb);
|
2288 |
|
|
EXT3_SET_RO_COMPAT_FEATURE(sb,
|
2289 |
|
|
EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
|
2290 |
|
|
sb->s_dirt = 1;
|
2291 |
|
|
handle->h_sync = 1;
|
2292 |
|
|
err = ext3_journal_dirty_metadata(handle,
|
2293 |
|
|
sb->u.ext3_sb.s_sbh);
|
2294 |
|
|
}
|
2295 |
|
|
}
|
2296 |
|
|
}
|
2297 |
|
|
raw_inode->i_generation = cpu_to_le32(inode->i_generation);
|
2298 |
|
|
if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
|
2299 |
|
|
raw_inode->i_block[0] =
|
2300 |
|
|
cpu_to_le32(kdev_t_to_nr(inode->i_rdev));
|
2301 |
|
|
else for (block = 0; block < EXT3_N_BLOCKS; block++)
|
2302 |
|
|
raw_inode->i_block[block] = inode->u.ext3_i.i_data[block];
|
2303 |
|
|
|
2304 |
|
|
BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
|
2305 |
|
|
rc = ext3_journal_dirty_metadata(handle, bh);
|
2306 |
|
|
if (!err)
|
2307 |
|
|
err = rc;
|
2308 |
|
|
EXT3_I(inode)->i_state &= ~EXT3_STATE_NEW;
|
2309 |
|
|
|
2310 |
|
|
out_brelse:
|
2311 |
|
|
brelse (bh);
|
2312 |
|
|
ext3_std_error(inode->i_sb, err);
|
2313 |
|
|
return err;
|
2314 |
|
|
}
|
2315 |
|
|
|
2316 |
|
|
/*
|
2317 |
|
|
* ext3_write_inode()
|
2318 |
|
|
*
|
2319 |
|
|
* We are called from a few places:
|
2320 |
|
|
*
|
2321 |
|
|
* - Within generic_file_write() for O_SYNC files.
|
2322 |
|
|
* Here, there will be no transaction running. We wait for any running
|
2323 |
|
|
* trasnaction to commit.
|
2324 |
|
|
*
|
2325 |
|
|
* - Within sys_sync(), kupdate and such.
|
2326 |
|
|
* We wait on commit, if tol to.
|
2327 |
|
|
*
|
2328 |
|
|
* - Within prune_icache() (PF_MEMALLOC == true)
|
2329 |
|
|
* Here we simply return. We can't afford to block kswapd on the
|
2330 |
|
|
* journal commit.
|
2331 |
|
|
*
|
2332 |
|
|
* In all cases it is actually safe for us to return without doing anything,
|
2333 |
|
|
* because the inode has been copied into a raw inode buffer in
|
2334 |
|
|
* ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
|
2335 |
|
|
* knfsd.
|
2336 |
|
|
*
|
2337 |
|
|
* Note that we are absolutely dependent upon all inode dirtiers doing the
|
2338 |
|
|
* right thing: they *must* call mark_inode_dirty() after dirtying info in
|
2339 |
|
|
* which we are interested.
|
2340 |
|
|
*
|
2341 |
|
|
* It would be a bug for them to not do this. The code:
|
2342 |
|
|
*
|
2343 |
|
|
* mark_inode_dirty(inode)
|
2344 |
|
|
* stuff();
|
2345 |
|
|
* inode->i_size = expr;
|
2346 |
|
|
*
|
2347 |
|
|
* is in error because a kswapd-driven write_inode() could occur while
|
2348 |
|
|
* `stuff()' is running, and the new i_size will be lost. Plus the inode
|
2349 |
|
|
* will no longer be on the superblock's dirty inode list.
|
2350 |
|
|
*/
|
2351 |
|
|
void ext3_write_inode(struct inode *inode, int wait)
|
2352 |
|
|
{
|
2353 |
|
|
if (current->flags & PF_MEMALLOC)
|
2354 |
|
|
return;
|
2355 |
|
|
|
2356 |
|
|
if (ext3_journal_current_handle()) {
|
2357 |
|
|
jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
|
2358 |
|
|
return;
|
2359 |
|
|
}
|
2360 |
|
|
|
2361 |
|
|
if (!wait)
|
2362 |
|
|
return;
|
2363 |
|
|
|
2364 |
|
|
ext3_force_commit(inode->i_sb);
|
2365 |
|
|
}
|
2366 |
|
|
|
2367 |
|
|
/*
|
2368 |
|
|
* ext3_setattr()
|
2369 |
|
|
*
|
2370 |
|
|
* Called from notify_change.
|
2371 |
|
|
*
|
2372 |
|
|
* We want to trap VFS attempts to truncate the file as soon as
|
2373 |
|
|
* possible. In particular, we want to make sure that when the VFS
|
2374 |
|
|
* shrinks i_size, we put the inode on the orphan list and modify
|
2375 |
|
|
* i_disksize immediately, so that during the subsequent flushing of
|
2376 |
|
|
* dirty pages and freeing of disk blocks, we can guarantee that any
|
2377 |
|
|
* commit will leave the blocks being flushed in an unused state on
|
2378 |
|
|
* disk. (On recovery, the inode will get truncated and the blocks will
|
2379 |
|
|
* be freed, so we have a strong guarantee that no future commit will
|
2380 |
|
|
* leave these blocks visible to the user.)
|
2381 |
|
|
*
|
2382 |
|
|
* This is only needed for regular files. rmdir() has its own path, and
|
2383 |
|
|
* we can never truncate a direcory except on final unlink (at which
|
2384 |
|
|
* point i_nlink is zero so recovery is easy.)
|
2385 |
|
|
*
|
2386 |
|
|
* Called with the BKL.
|
2387 |
|
|
*/
|
2388 |
|
|
|
2389 |
|
|
int ext3_setattr(struct dentry *dentry, struct iattr *attr)
|
2390 |
|
|
{
|
2391 |
|
|
struct inode *inode = dentry->d_inode;
|
2392 |
|
|
int error, rc = 0;
|
2393 |
|
|
const unsigned int ia_valid = attr->ia_valid;
|
2394 |
|
|
|
2395 |
|
|
error = inode_change_ok(inode, attr);
|
2396 |
|
|
if (error)
|
2397 |
|
|
return error;
|
2398 |
|
|
|
2399 |
|
|
if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
|
2400 |
|
|
(ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
|
2401 |
|
|
error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
|
2402 |
|
|
if (error)
|
2403 |
|
|
return error;
|
2404 |
|
|
}
|
2405 |
|
|
|
2406 |
|
|
if (attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
|
2407 |
|
|
handle_t *handle;
|
2408 |
|
|
|
2409 |
|
|
handle = ext3_journal_start(inode, 3);
|
2410 |
|
|
if (IS_ERR(handle)) {
|
2411 |
|
|
error = PTR_ERR(handle);
|
2412 |
|
|
goto err_out;
|
2413 |
|
|
}
|
2414 |
|
|
|
2415 |
|
|
error = ext3_orphan_add(handle, inode);
|
2416 |
|
|
inode->u.ext3_i.i_disksize = attr->ia_size;
|
2417 |
|
|
rc = ext3_mark_inode_dirty(handle, inode);
|
2418 |
|
|
if (!error)
|
2419 |
|
|
error = rc;
|
2420 |
|
|
ext3_journal_stop(handle, inode);
|
2421 |
|
|
}
|
2422 |
|
|
|
2423 |
|
|
rc = inode_setattr(inode, attr);
|
2424 |
|
|
|
2425 |
|
|
/* If inode_setattr's call to ext3_truncate failed to get a
|
2426 |
|
|
* transaction handle at all, we need to clean up the in-core
|
2427 |
|
|
* orphan list manually. */
|
2428 |
|
|
if (inode->i_nlink)
|
2429 |
|
|
ext3_orphan_del(NULL, inode);
|
2430 |
|
|
|
2431 |
|
|
err_out:
|
2432 |
|
|
ext3_std_error(inode->i_sb, error);
|
2433 |
|
|
if (!error)
|
2434 |
|
|
error = rc;
|
2435 |
|
|
return error;
|
2436 |
|
|
}
|
2437 |
|
|
|
2438 |
|
|
|
2439 |
|
|
/*
|
2440 |
|
|
* akpm: how many blocks doth make a writepage()?
|
2441 |
|
|
*
|
2442 |
|
|
* With N blocks per page, it may be:
|
2443 |
|
|
* N data blocks
|
2444 |
|
|
* 2 indirect block
|
2445 |
|
|
* 2 dindirect
|
2446 |
|
|
* 1 tindirect
|
2447 |
|
|
* N+5 bitmap blocks (from the above)
|
2448 |
|
|
* N+5 group descriptor summary blocks
|
2449 |
|
|
* 1 inode block
|
2450 |
|
|
* 1 superblock.
|
2451 |
|
|
* 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
|
2452 |
|
|
*
|
2453 |
|
|
* 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
|
2454 |
|
|
*
|
2455 |
|
|
* With ordered or writeback data it's the same, less the N data blocks.
|
2456 |
|
|
*
|
2457 |
|
|
* If the inode's direct blocks can hold an integral number of pages then a
|
2458 |
|
|
* page cannot straddle two indirect blocks, and we can only touch one indirect
|
2459 |
|
|
* and dindirect block, and the "5" above becomes "3".
|
2460 |
|
|
*
|
2461 |
|
|
* This still overestimates under most circumstances. If we were to pass the
|
2462 |
|
|
* start and end offsets in here as well we could do block_to_path() on each
|
2463 |
|
|
* block and work out the exact number of indirects which are touched. Pah.
|
2464 |
|
|
*/
|
2465 |
|
|
|
2466 |
|
|
int ext3_writepage_trans_blocks(struct inode *inode)
|
2467 |
|
|
{
|
2468 |
|
|
int bpp = ext3_journal_blocks_per_page(inode);
|
2469 |
|
|
int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
|
2470 |
|
|
int ret;
|
2471 |
|
|
|
2472 |
|
|
if (ext3_should_journal_data(inode))
|
2473 |
|
|
ret = 3 * (bpp + indirects) + 2;
|
2474 |
|
|
else
|
2475 |
|
|
ret = 2 * (bpp + indirects) + 2;
|
2476 |
|
|
|
2477 |
|
|
#ifdef CONFIG_QUOTA
|
2478 |
|
|
ret += 2 * EXT3_SINGLEDATA_TRANS_BLOCKS;
|
2479 |
|
|
#endif
|
2480 |
|
|
|
2481 |
|
|
return ret;
|
2482 |
|
|
}
|
2483 |
|
|
|
2484 |
|
|
int
|
2485 |
|
|
ext3_mark_iloc_dirty(handle_t *handle,
|
2486 |
|
|
struct inode *inode,
|
2487 |
|
|
struct ext3_iloc *iloc)
|
2488 |
|
|
{
|
2489 |
|
|
int err = 0;
|
2490 |
|
|
|
2491 |
|
|
if (handle) {
|
2492 |
|
|
/* the do_update_inode consumes one bh->b_count */
|
2493 |
|
|
atomic_inc(&iloc->bh->b_count);
|
2494 |
|
|
err = ext3_do_update_inode(handle, inode, iloc);
|
2495 |
|
|
/* ext3_do_update_inode() does journal_dirty_metadata */
|
2496 |
|
|
brelse(iloc->bh);
|
2497 |
|
|
} else {
|
2498 |
|
|
printk(KERN_EMERG "%s: called with no handle!\n", __FUNCTION__);
|
2499 |
|
|
}
|
2500 |
|
|
return err;
|
2501 |
|
|
}
|
2502 |
|
|
|
2503 |
|
|
/*
|
2504 |
|
|
* On success, We end up with an outstanding reference count against
|
2505 |
|
|
* iloc->bh. This _must_ be cleaned up later.
|
2506 |
|
|
*/
|
2507 |
|
|
|
2508 |
|
|
int
|
2509 |
|
|
ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
|
2510 |
|
|
struct ext3_iloc *iloc)
|
2511 |
|
|
{
|
2512 |
|
|
int err = 0;
|
2513 |
|
|
if (handle) {
|
2514 |
|
|
err = ext3_get_inode_loc(inode, iloc);
|
2515 |
|
|
if (!err) {
|
2516 |
|
|
BUFFER_TRACE(iloc->bh, "get_write_access");
|
2517 |
|
|
err = ext3_journal_get_write_access(handle, iloc->bh);
|
2518 |
|
|
if (err) {
|
2519 |
|
|
brelse(iloc->bh);
|
2520 |
|
|
iloc->bh = NULL;
|
2521 |
|
|
}
|
2522 |
|
|
}
|
2523 |
|
|
}
|
2524 |
|
|
ext3_std_error(inode->i_sb, err);
|
2525 |
|
|
return err;
|
2526 |
|
|
}
|
2527 |
|
|
|
2528 |
|
|
/*
|
2529 |
|
|
* akpm: What we do here is to mark the in-core inode as clean
|
2530 |
|
|
* with respect to inode dirtiness (it may still be data-dirty).
|
2531 |
|
|
* This means that the in-core inode may be reaped by prune_icache
|
2532 |
|
|
* without having to perform any I/O. This is a very good thing,
|
2533 |
|
|
* because *any* task may call prune_icache - even ones which
|
2534 |
|
|
* have a transaction open against a different journal.
|
2535 |
|
|
*
|
2536 |
|
|
* Is this cheating? Not really. Sure, we haven't written the
|
2537 |
|
|
* inode out, but prune_icache isn't a user-visible syncing function.
|
2538 |
|
|
* Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
|
2539 |
|
|
* we start and wait on commits.
|
2540 |
|
|
*
|
2541 |
|
|
* Is this efficient/effective? Well, we're being nice to the system
|
2542 |
|
|
* by cleaning up our inodes proactively so they can be reaped
|
2543 |
|
|
* without I/O. But we are potentially leaving up to five seconds'
|
2544 |
|
|
* worth of inodes floating about which prune_icache wants us to
|
2545 |
|
|
* write out. One way to fix that would be to get prune_icache()
|
2546 |
|
|
* to do a write_super() to free up some memory. It has the desired
|
2547 |
|
|
* effect.
|
2548 |
|
|
*/
|
2549 |
|
|
int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
|
2550 |
|
|
{
|
2551 |
|
|
struct ext3_iloc iloc;
|
2552 |
|
|
int err;
|
2553 |
|
|
|
2554 |
|
|
err = ext3_reserve_inode_write(handle, inode, &iloc);
|
2555 |
|
|
if (!err)
|
2556 |
|
|
err = ext3_mark_iloc_dirty(handle, inode, &iloc);
|
2557 |
|
|
return err;
|
2558 |
|
|
}
|
2559 |
|
|
|
2560 |
|
|
/*
|
2561 |
|
|
* akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
|
2562 |
|
|
*
|
2563 |
|
|
* We're really interested in the case where a file is being extended.
|
2564 |
|
|
* i_size has been changed by generic_commit_write() and we thus need
|
2565 |
|
|
* to include the updated inode in the current transaction.
|
2566 |
|
|
*
|
2567 |
|
|
* Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
|
2568 |
|
|
* are allocated to the file.
|
2569 |
|
|
*
|
2570 |
|
|
* If the inode is marked synchronous, we don't honour that here - doing
|
2571 |
|
|
* so would cause a commit on atime updates, which we don't bother doing.
|
2572 |
|
|
* We handle synchronous inodes at the highest possible level.
|
2573 |
|
|
*/
|
2574 |
|
|
void ext3_dirty_inode(struct inode *inode)
|
2575 |
|
|
{
|
2576 |
|
|
handle_t *current_handle = ext3_journal_current_handle();
|
2577 |
|
|
handle_t *handle;
|
2578 |
|
|
|
2579 |
|
|
lock_kernel();
|
2580 |
|
|
handle = ext3_journal_start(inode, 2);
|
2581 |
|
|
if (IS_ERR(handle))
|
2582 |
|
|
goto out;
|
2583 |
|
|
if (current_handle &&
|
2584 |
|
|
current_handle->h_transaction != handle->h_transaction) {
|
2585 |
|
|
/* This task has a transaction open against a different fs */
|
2586 |
|
|
printk(KERN_EMERG "%s: transactions do not match!\n",
|
2587 |
|
|
__FUNCTION__);
|
2588 |
|
|
} else {
|
2589 |
|
|
jbd_debug(5, "marking dirty. outer handle=%p\n",
|
2590 |
|
|
current_handle);
|
2591 |
|
|
ext3_mark_inode_dirty(handle, inode);
|
2592 |
|
|
}
|
2593 |
|
|
ext3_journal_stop(handle, inode);
|
2594 |
|
|
out:
|
2595 |
|
|
unlock_kernel();
|
2596 |
|
|
}
|
2597 |
|
|
|
2598 |
|
|
#ifdef AKPM
|
2599 |
|
|
/*
|
2600 |
|
|
* Bind an inode's backing buffer_head into this transaction, to prevent
|
2601 |
|
|
* it from being flushed to disk early. Unlike
|
2602 |
|
|
* ext3_reserve_inode_write, this leaves behind no bh reference and
|
2603 |
|
|
* returns no iloc structure, so the caller needs to repeat the iloc
|
2604 |
|
|
* lookup to mark the inode dirty later.
|
2605 |
|
|
*/
|
2606 |
|
|
static inline int
|
2607 |
|
|
ext3_pin_inode(handle_t *handle, struct inode *inode)
|
2608 |
|
|
{
|
2609 |
|
|
struct ext3_iloc iloc;
|
2610 |
|
|
|
2611 |
|
|
int err = 0;
|
2612 |
|
|
if (handle) {
|
2613 |
|
|
err = ext3_get_inode_loc(inode, &iloc);
|
2614 |
|
|
if (!err) {
|
2615 |
|
|
BUFFER_TRACE(iloc.bh, "get_write_access");
|
2616 |
|
|
err = journal_get_write_access(handle, iloc.bh);
|
2617 |
|
|
if (!err)
|
2618 |
|
|
err = ext3_journal_dirty_metadata(handle,
|
2619 |
|
|
iloc.bh);
|
2620 |
|
|
brelse(iloc.bh);
|
2621 |
|
|
}
|
2622 |
|
|
}
|
2623 |
|
|
ext3_std_error(inode->i_sb, err);
|
2624 |
|
|
return err;
|
2625 |
|
|
}
|
2626 |
|
|
#endif
|
2627 |
|
|
|
2628 |
|
|
int ext3_change_inode_journal_flag(struct inode *inode, int val)
|
2629 |
|
|
{
|
2630 |
|
|
journal_t *journal;
|
2631 |
|
|
handle_t *handle;
|
2632 |
|
|
int err;
|
2633 |
|
|
|
2634 |
|
|
/*
|
2635 |
|
|
* We have to be very careful here: changing a data block's
|
2636 |
|
|
* journaling status dynamically is dangerous. If we write a
|
2637 |
|
|
* data block to the journal, change the status and then delete
|
2638 |
|
|
* that block, we risk forgetting to revoke the old log record
|
2639 |
|
|
* from the journal and so a subsequent replay can corrupt data.
|
2640 |
|
|
* So, first we make sure that the journal is empty and that
|
2641 |
|
|
* nobody is changing anything.
|
2642 |
|
|
*/
|
2643 |
|
|
|
2644 |
|
|
journal = EXT3_JOURNAL(inode);
|
2645 |
|
|
if (is_journal_aborted(journal) || IS_RDONLY(inode))
|
2646 |
|
|
return -EROFS;
|
2647 |
|
|
|
2648 |
|
|
journal_lock_updates(journal);
|
2649 |
|
|
journal_flush(journal);
|
2650 |
|
|
|
2651 |
|
|
/*
|
2652 |
|
|
* OK, there are no updates running now, and all cached data is
|
2653 |
|
|
* synced to disk. We are now in a completely consistent state
|
2654 |
|
|
* which doesn't have anything in the journal, and we know that
|
2655 |
|
|
* no filesystem updates are running, so it is safe to modify
|
2656 |
|
|
* the inode's in-core data-journaling state flag now.
|
2657 |
|
|
*/
|
2658 |
|
|
|
2659 |
|
|
if (val)
|
2660 |
|
|
inode->u.ext3_i.i_flags |= EXT3_JOURNAL_DATA_FL;
|
2661 |
|
|
else
|
2662 |
|
|
inode->u.ext3_i.i_flags &= ~EXT3_JOURNAL_DATA_FL;
|
2663 |
|
|
|
2664 |
|
|
journal_unlock_updates(journal);
|
2665 |
|
|
|
2666 |
|
|
/* Finally we can mark the inode as dirty. */
|
2667 |
|
|
|
2668 |
|
|
handle = ext3_journal_start(inode, 1);
|
2669 |
|
|
if (IS_ERR(handle))
|
2670 |
|
|
return PTR_ERR(handle);
|
2671 |
|
|
|
2672 |
|
|
err = ext3_mark_inode_dirty(handle, inode);
|
2673 |
|
|
handle->h_sync = 1;
|
2674 |
|
|
ext3_journal_stop(handle, inode);
|
2675 |
|
|
ext3_std_error(inode->i_sb, err);
|
2676 |
|
|
|
2677 |
|
|
return err;
|
2678 |
|
|
}
|
2679 |
|
|
|
2680 |
|
|
|
2681 |
|
|
/*
|
2682 |
|
|
* ext3_aops_journal_start().
|
2683 |
|
|
*
|
2684 |
|
|
* <This function died, but the comment lives on>
|
2685 |
|
|
*
|
2686 |
|
|
* We need to take the inode semaphore *outside* the
|
2687 |
|
|
* journal_start/journal_stop. Otherwise, a different task could do a
|
2688 |
|
|
* wait_for_commit() while holding ->i_sem, which deadlocks. The rule
|
2689 |
|
|
* is: transaction open/closes are considered to be a locking operation
|
2690 |
|
|
* and they nest *inside* ->i_sem.
|
2691 |
|
|
* ----------------------------------------------------------------------------
|
2692 |
|
|
* Possible problem:
|
2693 |
|
|
* ext3_file_write()
|
2694 |
|
|
* -> generic_file_write()
|
2695 |
|
|
* -> __alloc_pages()
|
2696 |
|
|
* -> page_launder()
|
2697 |
|
|
* -> ext3_writepage()
|
2698 |
|
|
*
|
2699 |
|
|
* And the writepage can be on a different fs while we have a
|
2700 |
|
|
* transaction open against this one! Bad.
|
2701 |
|
|
*
|
2702 |
|
|
* I tried making the task PF_MEMALLOC here, but that simply results in
|
2703 |
|
|
* 0-order allocation failures passed back to generic_file_write().
|
2704 |
|
|
* Instead, we rely on the reentrancy protection in ext3_writepage().
|
2705 |
|
|
* ----------------------------------------------------------------------------
|
2706 |
|
|
* When we do the journal_start() here we don't really need to reserve
|
2707 |
|
|
* any blocks - we won't need any until we hit ext3_prepare_write(),
|
2708 |
|
|
* which does all the needed journal extending. However! There is a
|
2709 |
|
|
* problem with quotas:
|
2710 |
|
|
*
|
2711 |
|
|
* Thread 1:
|
2712 |
|
|
* sys_sync
|
2713 |
|
|
* ->sync_dquots
|
2714 |
|
|
* ->commit_dquot
|
2715 |
|
|
* ->lock_dquot
|
2716 |
|
|
* ->write_dquot
|
2717 |
|
|
* ->ext3_file_write
|
2718 |
|
|
* ->journal_start
|
2719 |
|
|
* ->ext3_prepare_write
|
2720 |
|
|
* ->journal_extend
|
2721 |
|
|
* ->journal_start
|
2722 |
|
|
* Thread 2:
|
2723 |
|
|
* ext3_create (for example)
|
2724 |
|
|
* ->ext3_new_inode
|
2725 |
|
|
* ->dquot_initialize
|
2726 |
|
|
* ->lock_dquot
|
2727 |
|
|
*
|
2728 |
|
|
* Deadlock. Thread 1's journal_start blocks because thread 2 has a
|
2729 |
|
|
* transaction open. Thread 2's transaction will never close because
|
2730 |
|
|
* thread 2 is stuck waiting for the dquot lock.
|
2731 |
|
|
*
|
2732 |
|
|
* So. We must ensure that thread 1 *never* needs to extend the journal
|
2733 |
|
|
* for quota writes. We do that by reserving enough journal blocks
|
2734 |
|
|
* here, in ext3_aops_journal_start() to ensure that the forthcoming "see if we
|
2735 |
|
|
* need to extend" test in ext3_prepare_write() succeeds.
|
2736 |
|
|
*/
|