blob: 49035c5a2c43ac024557f728e5526c49f7b5fa4d [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/inode.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
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 *
Mingming Cao617ba132006-10-11 01:20:53 -070022 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
Dave Kleikampac27a0e2006-10-11 01:20:50 -070023 */
24
25#include <linux/module.h>
26#include <linux/fs.h>
27#include <linux/time.h>
Mingming Caodab291a2006-10-11 01:21:01 -070028#include <linux/ext4_jbd2.h>
29#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070030#include <linux/highuid.h>
31#include <linux/pagemap.h>
32#include <linux/quotaops.h>
33#include <linux/string.h>
34#include <linux/buffer_head.h>
35#include <linux/writeback.h>
36#include <linux/mpage.h>
37#include <linux/uio.h>
38#include <linux/bio.h>
39#include "xattr.h"
40#include "acl.h"
41
Dave Kleikampac27a0e2006-10-11 01:20:50 -070042/*
43 * Test whether an inode is a fast symlink.
44 */
Mingming Cao617ba132006-10-11 01:20:53 -070045static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070046{
Mingming Cao617ba132006-10-11 01:20:53 -070047 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -070048 (inode->i_sb->s_blocksize >> 9) : 0;
49
50 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
51}
52
53/*
Mingming Cao617ba132006-10-11 01:20:53 -070054 * The ext4 forget function must perform a revoke if we are freeing data
Dave Kleikampac27a0e2006-10-11 01:20:50 -070055 * 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 */
Mingming Cao617ba132006-10-11 01:20:53 -070062int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
63 struct buffer_head *bh, ext4_fsblk_t blocknr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070064{
65 int err;
66
67 might_sleep();
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
Mingming Cao617ba132006-10-11 01:20:53 -070081 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
82 (!is_metadata && !ext4_should_journal_data(inode))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070083 if (bh) {
Mingming Caodab291a2006-10-11 01:21:01 -070084 BUFFER_TRACE(bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -070085 return ext4_journal_forget(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070086 }
87 return 0;
88 }
89
90 /*
91 * data!=journal && (is_metadata || should_journal_data(inode))
92 */
Mingming Cao617ba132006-10-11 01:20:53 -070093 BUFFER_TRACE(bh, "call ext4_journal_revoke");
94 err = ext4_journal_revoke(handle, blocknr, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -070096 ext4_abort(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070097 "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 proceed with the next chunk of a
104 * truncate transaction.
105 */
106static unsigned long blocks_for_truncate(struct inode *inode)
107{
108 unsigned long needed;
109
110 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
111
112 /* Give ourselves just enough room to cope with inodes in which
113 * i_blocks is corrupt: we've seen disk corruptions in the past
114 * which resulted in random data in an inode which looked enough
Mingming Cao617ba132006-10-11 01:20:53 -0700115 * like a regular file for ext4 to try to delete it. Things
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116 * will go a bit crazy if that happens, but at least we should
117 * try not to panic the whole kernel. */
118 if (needed < 2)
119 needed = 2;
120
121 /* But we need to bound the transaction so we don't overflow the
122 * journal. */
Mingming Cao617ba132006-10-11 01:20:53 -0700123 if (needed > EXT4_MAX_TRANS_DATA)
124 needed = EXT4_MAX_TRANS_DATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700125
Mingming Cao617ba132006-10-11 01:20:53 -0700126 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700127}
128
129/*
130 * Truncate transactions can be complex and absolutely huge. So we need to
131 * be able to restart the transaction at a conventient checkpoint to make
132 * sure we don't overflow the journal.
133 *
134 * start_transaction gets us a new handle for a truncate transaction,
135 * and extend_transaction tries to extend the existing one a bit. If
136 * extend fails, we need to propagate the failure up and restart the
137 * transaction in the top-level truncate loop. --sct
138 */
139static handle_t *start_transaction(struct inode *inode)
140{
141 handle_t *result;
142
Mingming Cao617ba132006-10-11 01:20:53 -0700143 result = ext4_journal_start(inode, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700144 if (!IS_ERR(result))
145 return result;
146
Mingming Cao617ba132006-10-11 01:20:53 -0700147 ext4_std_error(inode->i_sb, PTR_ERR(result));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700148 return result;
149}
150
151/*
152 * Try to extend this transaction for the purposes of truncation.
153 *
154 * Returns 0 if we managed to create more room. If we can't create more
155 * room, and the transaction must be restarted we return 1.
156 */
157static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
158{
Mingming Cao617ba132006-10-11 01:20:53 -0700159 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700160 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700161 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700162 return 0;
163 return 1;
164}
165
166/*
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
169 * this transaction.
170 */
Mingming Cao617ba132006-10-11 01:20:53 -0700171static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700172{
173 jbd_debug(2, "restarting handle %p\n", handle);
Mingming Cao617ba132006-10-11 01:20:53 -0700174 return ext4_journal_restart(handle, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700175}
176
177/*
178 * Called at the last iput() if i_nlink is zero.
179 */
Mingming Cao617ba132006-10-11 01:20:53 -0700180void ext4_delete_inode (struct inode * inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700181{
182 handle_t *handle;
183
184 truncate_inode_pages(&inode->i_data, 0);
185
186 if (is_bad_inode(inode))
187 goto no_delete;
188
189 handle = start_transaction(inode);
190 if (IS_ERR(handle)) {
191 /*
192 * If we're going to skip the normal cleanup, we still need to
193 * make sure that the in-core orphan linked list is properly
194 * cleaned up.
195 */
Mingming Cao617ba132006-10-11 01:20:53 -0700196 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700197 goto no_delete;
198 }
199
200 if (IS_SYNC(inode))
201 handle->h_sync = 1;
202 inode->i_size = 0;
203 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700204 ext4_truncate(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700205 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700206 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700207 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700208 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700209 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700210 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700211 * (Well, we could do this if we need to, but heck - it works)
212 */
Mingming Cao617ba132006-10-11 01:20:53 -0700213 ext4_orphan_del(handle, inode);
214 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700215
216 /*
217 * One subtle ordering requirement: if anything has gone wrong
218 * (transaction abort, IO errors, whatever), then we can still
219 * do these next steps (the fs will already have been marked as
220 * having errors), but we can't free the inode if the mark_dirty
221 * fails.
222 */
Mingming Cao617ba132006-10-11 01:20:53 -0700223 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700224 /* If that failed, just do the required in-core inode clear. */
225 clear_inode(inode);
226 else
Mingming Cao617ba132006-10-11 01:20:53 -0700227 ext4_free_inode(handle, inode);
228 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229 return;
230no_delete:
231 clear_inode(inode); /* We must guarantee clearing of inode... */
232}
233
234typedef struct {
235 __le32 *p;
236 __le32 key;
237 struct buffer_head *bh;
238} Indirect;
239
240static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
241{
242 p->key = *(p->p = v);
243 p->bh = bh;
244}
245
246static int verify_chain(Indirect *from, Indirect *to)
247{
248 while (from <= to && from->key == *from->p)
249 from++;
250 return (from > to);
251}
252
253/**
Mingming Cao617ba132006-10-11 01:20:53 -0700254 * ext4_block_to_path - parse the block number into array of offsets
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700255 * @inode: inode in question (we are only interested in its superblock)
256 * @i_block: block number to be parsed
257 * @offsets: array to store the offsets in
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400258 * @boundary: set this non-zero if the referred-to block is likely to be
259 * followed (on disk) by an indirect block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260 *
Mingming Cao617ba132006-10-11 01:20:53 -0700261 * To store the locations of file's data ext4 uses a data structure common
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262 * for UNIX filesystems - tree of pointers anchored in the inode, with
263 * data blocks at leaves and indirect blocks in intermediate nodes.
264 * This function translates the block number into path in that tree -
265 * return value is the path length and @offsets[n] is the offset of
266 * pointer to (n+1)th node in the nth one. If @block is out of range
267 * (negative or too large) warning is printed and zero returned.
268 *
269 * Note: function doesn't find node addresses, so no IO is needed. All
270 * we need to know is the capacity of indirect blocks (taken from the
271 * inode->i_sb).
272 */
273
274/*
275 * Portability note: the last comparison (check that we fit into triple
276 * indirect block) is spelled differently, because otherwise on an
277 * architecture with 32-bit longs and 8Kb pages we might get into trouble
278 * if our filesystem had 8Kb blocks. We might use long long, but that would
279 * kill us on x86. Oh, well, at least the sign propagation does not matter -
280 * i_block would have to be negative in the very beginning, so we would not
281 * get there at all.
282 */
283
Mingming Cao617ba132006-10-11 01:20:53 -0700284static int ext4_block_to_path(struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700285 long i_block, int offsets[4], int *boundary)
286{
Mingming Cao617ba132006-10-11 01:20:53 -0700287 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
288 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
289 const long direct_blocks = EXT4_NDIR_BLOCKS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700290 indirect_blocks = ptrs,
291 double_blocks = (1 << (ptrs_bits * 2));
292 int n = 0;
293 int final = 0;
294
295 if (i_block < 0) {
Mingming Cao617ba132006-10-11 01:20:53 -0700296 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700297 } else if (i_block < direct_blocks) {
298 offsets[n++] = i_block;
299 final = direct_blocks;
300 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700301 offsets[n++] = EXT4_IND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700302 offsets[n++] = i_block;
303 final = ptrs;
304 } else if ((i_block -= indirect_blocks) < double_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700305 offsets[n++] = EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700306 offsets[n++] = i_block >> ptrs_bits;
307 offsets[n++] = i_block & (ptrs - 1);
308 final = ptrs;
309 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
Mingming Cao617ba132006-10-11 01:20:53 -0700310 offsets[n++] = EXT4_TIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700311 offsets[n++] = i_block >> (ptrs_bits * 2);
312 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
313 offsets[n++] = i_block & (ptrs - 1);
314 final = ptrs;
315 } else {
Mingming Cao617ba132006-10-11 01:20:53 -0700316 ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700317 }
318 if (boundary)
319 *boundary = final - 1 - (i_block & (ptrs - 1));
320 return n;
321}
322
323/**
Mingming Cao617ba132006-10-11 01:20:53 -0700324 * ext4_get_branch - read the chain of indirect blocks leading to data
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700325 * @inode: inode in question
326 * @depth: depth of the chain (1 - direct pointer, etc.)
327 * @offsets: offsets of pointers in inode/indirect blocks
328 * @chain: place to store the result
329 * @err: here we store the error value
330 *
331 * Function fills the array of triples <key, p, bh> and returns %NULL
332 * if everything went OK or the pointer to the last filled triple
333 * (incomplete one) otherwise. Upon the return chain[i].key contains
334 * the number of (i+1)-th block in the chain (as it is stored in memory,
335 * i.e. little-endian 32-bit), chain[i].p contains the address of that
336 * number (it points into struct inode for i==0 and into the bh->b_data
337 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
338 * block for i>0 and NULL for i==0. In other words, it holds the block
339 * numbers of the chain, addresses they were taken from (and where we can
340 * verify that chain did not change) and buffer_heads hosting these
341 * numbers.
342 *
343 * Function stops when it stumbles upon zero pointer (absent block)
344 * (pointer to last triple returned, *@err == 0)
345 * or when it gets an IO error reading an indirect block
346 * (ditto, *@err == -EIO)
347 * or when it notices that chain had been changed while it was reading
348 * (ditto, *@err == -EAGAIN)
349 * or when it reads all @depth-1 indirect blocks successfully and finds
350 * the whole chain, all way to the data (returns %NULL, *err == 0).
351 */
Mingming Cao617ba132006-10-11 01:20:53 -0700352static Indirect *ext4_get_branch(struct inode *inode, int depth, int *offsets,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700353 Indirect chain[4], int *err)
354{
355 struct super_block *sb = inode->i_sb;
356 Indirect *p = chain;
357 struct buffer_head *bh;
358
359 *err = 0;
360 /* i_data is not going away, no lock needed */
Mingming Cao617ba132006-10-11 01:20:53 -0700361 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700362 if (!p->key)
363 goto no_block;
364 while (--depth) {
365 bh = sb_bread(sb, le32_to_cpu(p->key));
366 if (!bh)
367 goto failure;
368 /* Reader: pointers */
369 if (!verify_chain(chain, p))
370 goto changed;
371 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
372 /* Reader: end */
373 if (!p->key)
374 goto no_block;
375 }
376 return NULL;
377
378changed:
379 brelse(bh);
380 *err = -EAGAIN;
381 goto no_block;
382failure:
383 *err = -EIO;
384no_block:
385 return p;
386}
387
388/**
Mingming Cao617ba132006-10-11 01:20:53 -0700389 * ext4_find_near - find a place for allocation with sufficient locality
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700390 * @inode: owner
391 * @ind: descriptor of indirect block.
392 *
393 * This function returns the prefered place for block allocation.
394 * It is used when heuristic for sequential allocation fails.
395 * Rules are:
396 * + if there is a block to the left of our position - allocate near it.
397 * + if pointer will live in indirect block - allocate near that block.
398 * + if pointer will live in inode - allocate in the same
399 * cylinder group.
400 *
401 * In the latter case we colour the starting block by the callers PID to
402 * prevent it from clashing with concurrent allocations for a different inode
403 * in the same block group. The PID is used here so that functionally related
404 * files will be close-by on-disk.
405 *
406 * Caller must make sure that @ind is valid and will stay that way.
407 */
Mingming Cao617ba132006-10-11 01:20:53 -0700408static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700409{
Mingming Cao617ba132006-10-11 01:20:53 -0700410 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700411 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
412 __le32 *p;
Mingming Cao617ba132006-10-11 01:20:53 -0700413 ext4_fsblk_t bg_start;
414 ext4_grpblk_t colour;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415
416 /* Try to find previous block */
417 for (p = ind->p - 1; p >= start; p--) {
418 if (*p)
419 return le32_to_cpu(*p);
420 }
421
422 /* No such thing, so let's try location of indirect block */
423 if (ind->bh)
424 return ind->bh->b_blocknr;
425
426 /*
427 * It is going to be referred to from the inode itself? OK, just put it
428 * into the same cylinder group then.
429 */
Mingming Cao617ba132006-10-11 01:20:53 -0700430 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700431 colour = (current->pid % 16) *
Mingming Cao617ba132006-10-11 01:20:53 -0700432 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700433 return bg_start + colour;
434}
435
436/**
Mingming Cao617ba132006-10-11 01:20:53 -0700437 * ext4_find_goal - find a prefered place for allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700438 * @inode: owner
439 * @block: block we want
440 * @chain: chain of indirect blocks
441 * @partial: pointer to the last triple within a chain
442 * @goal: place to store the result.
443 *
444 * Normally this function find the prefered place for block allocation,
445 * stores it in *@goal and returns zero.
446 */
447
Mingming Cao617ba132006-10-11 01:20:53 -0700448static ext4_fsblk_t ext4_find_goal(struct inode *inode, long block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700449 Indirect chain[4], Indirect *partial)
450{
Mingming Cao617ba132006-10-11 01:20:53 -0700451 struct ext4_block_alloc_info *block_i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700452
Mingming Cao617ba132006-10-11 01:20:53 -0700453 block_i = EXT4_I(inode)->i_block_alloc_info;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700454
455 /*
456 * try the heuristic for sequential allocation,
457 * failing that at least try to get decent locality.
458 */
459 if (block_i && (block == block_i->last_alloc_logical_block + 1)
460 && (block_i->last_alloc_physical_block != 0)) {
461 return block_i->last_alloc_physical_block + 1;
462 }
463
Mingming Cao617ba132006-10-11 01:20:53 -0700464 return ext4_find_near(inode, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700465}
466
467/**
Mingming Cao617ba132006-10-11 01:20:53 -0700468 * ext4_blks_to_allocate: Look up the block map and count the number
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700469 * of direct blocks need to be allocated for the given branch.
470 *
471 * @branch: chain of indirect blocks
472 * @k: number of blocks need for indirect blocks
473 * @blks: number of data blocks to be mapped.
474 * @blocks_to_boundary: the offset in the indirect block
475 *
476 * return the total number of blocks to be allocate, including the
477 * direct and indirect blocks.
478 */
Mingming Cao617ba132006-10-11 01:20:53 -0700479static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700480 int blocks_to_boundary)
481{
482 unsigned long count = 0;
483
484 /*
485 * Simple case, [t,d]Indirect block(s) has not allocated yet
486 * then it's clear blocks on that path have not allocated
487 */
488 if (k > 0) {
489 /* right now we don't handle cross boundary allocation */
490 if (blks < blocks_to_boundary + 1)
491 count += blks;
492 else
493 count += blocks_to_boundary + 1;
494 return count;
495 }
496
497 count++;
498 while (count < blks && count <= blocks_to_boundary &&
499 le32_to_cpu(*(branch[0].p + count)) == 0) {
500 count++;
501 }
502 return count;
503}
504
505/**
Mingming Cao617ba132006-10-11 01:20:53 -0700506 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700507 * @indirect_blks: the number of blocks need to allocate for indirect
508 * blocks
509 *
510 * @new_blocks: on return it will store the new block numbers for
511 * the indirect blocks(if needed) and the first direct block,
512 * @blks: on return it will store the total number of allocated
513 * direct blocks
514 */
Mingming Cao617ba132006-10-11 01:20:53 -0700515static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
516 ext4_fsblk_t goal, int indirect_blks, int blks,
517 ext4_fsblk_t new_blocks[4], int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700518{
519 int target, i;
520 unsigned long count = 0;
521 int index = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700522 ext4_fsblk_t current_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700523 int ret = 0;
524
525 /*
526 * Here we try to allocate the requested multiple blocks at once,
527 * on a best-effort basis.
528 * To build a branch, we should allocate blocks for
529 * the indirect blocks(if not allocated yet), and at least
530 * the first direct block of this branch. That's the
531 * minimum number of blocks need to allocate(required)
532 */
533 target = blks + indirect_blks;
534
535 while (1) {
536 count = target;
537 /* allocating blocks for indirect blocks and direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -0700538 current_block = ext4_new_blocks(handle,inode,goal,&count,err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700539 if (*err)
540 goto failed_out;
541
542 target -= count;
543 /* allocate blocks for indirect blocks */
544 while (index < indirect_blks && count) {
545 new_blocks[index++] = current_block++;
546 count--;
547 }
548
549 if (count > 0)
550 break;
551 }
552
553 /* save the new block number for the first direct block */
554 new_blocks[index] = current_block;
555
556 /* total number of blocks allocated for direct blocks */
557 ret = count;
558 *err = 0;
559 return ret;
560failed_out:
561 for (i = 0; i <index; i++)
Mingming Cao617ba132006-10-11 01:20:53 -0700562 ext4_free_blocks(handle, inode, new_blocks[i], 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700563 return ret;
564}
565
566/**
Mingming Cao617ba132006-10-11 01:20:53 -0700567 * ext4_alloc_branch - allocate and set up a chain of blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700568 * @inode: owner
569 * @indirect_blks: number of allocated indirect blocks
570 * @blks: number of allocated direct blocks
571 * @offsets: offsets (in the blocks) to store the pointers to next.
572 * @branch: place to store the chain in.
573 *
574 * This function allocates blocks, zeroes out all but the last one,
575 * links them into chain and (if we are synchronous) writes them to disk.
576 * In other words, it prepares a branch that can be spliced onto the
577 * inode. It stores the information about that chain in the branch[], in
Mingming Cao617ba132006-10-11 01:20:53 -0700578 * the same format as ext4_get_branch() would do. We are calling it after
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700579 * we had read the existing part of chain and partial points to the last
580 * triple of that (one with zero ->key). Upon the exit we have the same
Mingming Cao617ba132006-10-11 01:20:53 -0700581 * picture as after the successful ext4_get_block(), except that in one
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700582 * place chain is disconnected - *branch->p is still zero (we did not
583 * set the last link), but branch->key contains the number that should
584 * be placed into *branch->p to fill that gap.
585 *
586 * If allocation fails we free all blocks we've allocated (and forget
587 * their buffer_heads) and return the error value the from failed
Mingming Cao617ba132006-10-11 01:20:53 -0700588 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700589 * as described above and return 0.
590 */
Mingming Cao617ba132006-10-11 01:20:53 -0700591static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
592 int indirect_blks, int *blks, ext4_fsblk_t goal,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700593 int *offsets, Indirect *branch)
594{
595 int blocksize = inode->i_sb->s_blocksize;
596 int i, n = 0;
597 int err = 0;
598 struct buffer_head *bh;
599 int num;
Mingming Cao617ba132006-10-11 01:20:53 -0700600 ext4_fsblk_t new_blocks[4];
601 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700602
Mingming Cao617ba132006-10-11 01:20:53 -0700603 num = ext4_alloc_blocks(handle, inode, goal, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700604 *blks, new_blocks, &err);
605 if (err)
606 return err;
607
608 branch[0].key = cpu_to_le32(new_blocks[0]);
609 /*
610 * metadata blocks and data blocks are allocated.
611 */
612 for (n = 1; n <= indirect_blks; n++) {
613 /*
614 * Get buffer_head for parent block, zero it out
615 * and set the pointer to new one, then send
616 * parent to disk.
617 */
618 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
619 branch[n].bh = bh;
620 lock_buffer(bh);
621 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700622 err = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700623 if (err) {
624 unlock_buffer(bh);
625 brelse(bh);
626 goto failed;
627 }
628
629 memset(bh->b_data, 0, blocksize);
630 branch[n].p = (__le32 *) bh->b_data + offsets[n];
631 branch[n].key = cpu_to_le32(new_blocks[n]);
632 *branch[n].p = branch[n].key;
633 if ( n == indirect_blks) {
634 current_block = new_blocks[n];
635 /*
636 * End of chain, update the last new metablock of
637 * the chain to point to the new allocated
638 * data blocks numbers
639 */
640 for (i=1; i < num; i++)
641 *(branch[n].p + i) = cpu_to_le32(++current_block);
642 }
643 BUFFER_TRACE(bh, "marking uptodate");
644 set_buffer_uptodate(bh);
645 unlock_buffer(bh);
646
Mingming Cao617ba132006-10-11 01:20:53 -0700647 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
648 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700649 if (err)
650 goto failed;
651 }
652 *blks = num;
653 return err;
654failed:
655 /* Allocation failed, free what we already allocated */
656 for (i = 1; i <= n ; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700657 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700658 ext4_journal_forget(handle, branch[i].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700659 }
660 for (i = 0; i <indirect_blks; i++)
Mingming Cao617ba132006-10-11 01:20:53 -0700661 ext4_free_blocks(handle, inode, new_blocks[i], 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700662
Mingming Cao617ba132006-10-11 01:20:53 -0700663 ext4_free_blocks(handle, inode, new_blocks[i], num);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664
665 return err;
666}
667
668/**
Mingming Cao617ba132006-10-11 01:20:53 -0700669 * ext4_splice_branch - splice the allocated branch onto inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700670 * @inode: owner
671 * @block: (logical) number of block we are adding
672 * @chain: chain of indirect blocks (with a missing link - see
Mingming Cao617ba132006-10-11 01:20:53 -0700673 * ext4_alloc_branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700674 * @where: location of missing link
675 * @num: number of indirect blocks we are adding
676 * @blks: number of direct blocks we are adding
677 *
678 * This function fills the missing link and does all housekeeping needed in
679 * inode (->i_blocks, etc.). In case of success we end up with the full
680 * chain to new block and return 0.
681 */
Mingming Cao617ba132006-10-11 01:20:53 -0700682static int ext4_splice_branch(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700683 long block, Indirect *where, int num, int blks)
684{
685 int i;
686 int err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700687 struct ext4_block_alloc_info *block_i;
688 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700689
Mingming Cao617ba132006-10-11 01:20:53 -0700690 block_i = EXT4_I(inode)->i_block_alloc_info;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700691 /*
692 * If we're splicing into a [td]indirect block (as opposed to the
693 * inode) then we need to get write access to the [td]indirect block
694 * before the splice.
695 */
696 if (where->bh) {
697 BUFFER_TRACE(where->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700698 err = ext4_journal_get_write_access(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700699 if (err)
700 goto err_out;
701 }
702 /* That's it */
703
704 *where->p = where->key;
705
706 /*
707 * Update the host buffer_head or inode to point to more just allocated
708 * direct blocks blocks
709 */
710 if (num == 0 && blks > 1) {
711 current_block = le32_to_cpu(where->key) + 1;
712 for (i = 1; i < blks; i++)
713 *(where->p + i ) = cpu_to_le32(current_block++);
714 }
715
716 /*
717 * update the most recently allocated logical & physical block
718 * in i_block_alloc_info, to assist find the proper goal block for next
719 * allocation
720 */
721 if (block_i) {
722 block_i->last_alloc_logical_block = block + blks - 1;
723 block_i->last_alloc_physical_block =
724 le32_to_cpu(where[num].key) + blks - 1;
725 }
726
727 /* We are done with atomic stuff, now do the rest of housekeeping */
728
729 inode->i_ctime = CURRENT_TIME_SEC;
Mingming Cao617ba132006-10-11 01:20:53 -0700730 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700731
732 /* had we spliced it onto indirect block? */
733 if (where->bh) {
734 /*
735 * If we spliced it onto an indirect block, we haven't
736 * altered the inode. Note however that if it is being spliced
737 * onto an indirect block at the very end of the file (the
738 * file is growing) then we *will* alter the inode to reflect
739 * the new i_size. But that is not done here - it is done in
Mingming Cao617ba132006-10-11 01:20:53 -0700740 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700741 */
742 jbd_debug(5, "splicing indirect only\n");
Mingming Cao617ba132006-10-11 01:20:53 -0700743 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
744 err = ext4_journal_dirty_metadata(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700745 if (err)
746 goto err_out;
747 } else {
748 /*
749 * OK, we spliced it into the inode itself on a direct block.
750 * Inode was dirtied above.
751 */
752 jbd_debug(5, "splicing direct\n");
753 }
754 return err;
755
756err_out:
757 for (i = 1; i <= num; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700758 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700759 ext4_journal_forget(handle, where[i].bh);
760 ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700761 }
Mingming Cao617ba132006-10-11 01:20:53 -0700762 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700763
764 return err;
765}
766
767/*
768 * Allocation strategy is simple: if we have to allocate something, we will
769 * have to go the whole way to leaf. So let's do it before attaching anything
770 * to tree, set linkage between the newborn blocks, write them if sync is
771 * required, recheck the path, free and repeat if check fails, otherwise
772 * set the last missing link (that will protect us from any truncate-generated
773 * removals - all blocks on the path are immune now) and possibly force the
774 * write on the parent block.
775 * That has a nice additional property: no special recovery from the failed
776 * allocations is needed - we simply release blocks and do not touch anything
777 * reachable from inode.
778 *
779 * `handle' can be NULL if create == 0.
780 *
781 * The BKL may not be held on entry here. Be sure to take it early.
782 * return > 0, # of blocks mapped or allocated.
783 * return = 0, if plain lookup failed.
784 * return < 0, error case.
785 */
Mingming Cao617ba132006-10-11 01:20:53 -0700786int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700787 sector_t iblock, unsigned long maxblocks,
788 struct buffer_head *bh_result,
789 int create, int extend_disksize)
790{
791 int err = -EIO;
792 int offsets[4];
793 Indirect chain[4];
794 Indirect *partial;
Mingming Cao617ba132006-10-11 01:20:53 -0700795 ext4_fsblk_t goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700796 int indirect_blks;
797 int blocks_to_boundary = 0;
798 int depth;
Mingming Cao617ba132006-10-11 01:20:53 -0700799 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700800 int count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700801 ext4_fsblk_t first_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700802
803
Alex Tomasa86c6182006-10-11 01:21:03 -0700804 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700805 J_ASSERT(handle != NULL || create == 0);
Mingming Cao617ba132006-10-11 01:20:53 -0700806 depth = ext4_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700807
808 if (depth == 0)
809 goto out;
810
Mingming Cao617ba132006-10-11 01:20:53 -0700811 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700812
813 /* Simplest case - block found, no allocation needed */
814 if (!partial) {
815 first_block = le32_to_cpu(chain[depth - 1].key);
816 clear_buffer_new(bh_result);
817 count++;
818 /*map more blocks*/
819 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao617ba132006-10-11 01:20:53 -0700820 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700821
822 if (!verify_chain(chain, partial)) {
823 /*
824 * Indirect block might be removed by
825 * truncate while we were reading it.
826 * Handling of that case: forget what we've
827 * got now. Flag the err as EAGAIN, so it
828 * will reread.
829 */
830 err = -EAGAIN;
831 count = 0;
832 break;
833 }
834 blk = le32_to_cpu(*(chain[depth-1].p + count));
835
836 if (blk == first_block + count)
837 count++;
838 else
839 break;
840 }
841 if (err != -EAGAIN)
842 goto got_it;
843 }
844
845 /* Next simple case - plain lookup or failed read of indirect block */
846 if (!create || err == -EIO)
847 goto cleanup;
848
849 mutex_lock(&ei->truncate_mutex);
850
851 /*
852 * If the indirect block is missing while we are reading
Mingming Cao617ba132006-10-11 01:20:53 -0700853 * the chain(ext4_get_branch() returns -EAGAIN err), or
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700854 * if the chain has been changed after we grab the semaphore,
855 * (either because another process truncated this branch, or
856 * another get_block allocated this branch) re-grab the chain to see if
857 * the request block has been allocated or not.
858 *
859 * Since we already block the truncate/other get_block
860 * at this point, we will have the current copy of the chain when we
861 * splice the branch into the tree.
862 */
863 if (err == -EAGAIN || !verify_chain(chain, partial)) {
864 while (partial > chain) {
865 brelse(partial->bh);
866 partial--;
867 }
Mingming Cao617ba132006-10-11 01:20:53 -0700868 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 if (!partial) {
870 count++;
871 mutex_unlock(&ei->truncate_mutex);
872 if (err)
873 goto cleanup;
874 clear_buffer_new(bh_result);
875 goto got_it;
876 }
877 }
878
879 /*
880 * Okay, we need to do block allocation. Lazily initialize the block
881 * allocation info here if necessary
882 */
883 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Mingming Cao617ba132006-10-11 01:20:53 -0700884 ext4_init_block_alloc_info(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700885
Mingming Cao617ba132006-10-11 01:20:53 -0700886 goal = ext4_find_goal(inode, iblock, chain, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700887
888 /* the number of blocks need to allocate for [d,t]indirect blocks */
889 indirect_blks = (chain + depth) - partial - 1;
890
891 /*
892 * Next look up the indirect map to count the totoal number of
893 * direct blocks to allocate for this branch.
894 */
Mingming Cao617ba132006-10-11 01:20:53 -0700895 count = ext4_blks_to_allocate(partial, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896 maxblocks, blocks_to_boundary);
897 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700898 * Block out ext4_truncate while we alter the tree
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700899 */
Mingming Cao617ba132006-10-11 01:20:53 -0700900 err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700901 offsets + (partial - chain), partial);
902
903 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700904 * The ext4_splice_branch call will free and forget any buffers
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700905 * on the new chain if there is a failure, but that risks using
906 * up transaction credits, especially for bitmaps where the
907 * credits cannot be returned. Can we handle this somehow? We
908 * may need to return -EAGAIN upwards in the worst case. --sct
909 */
910 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -0700911 err = ext4_splice_branch(handle, inode, iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700912 partial, indirect_blks, count);
913 /*
914 * i_disksize growing is protected by truncate_mutex. Don't forget to
915 * protect it if you're about to implement concurrent
Mingming Cao617ba132006-10-11 01:20:53 -0700916 * ext4_get_block() -bzzz
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700917 */
918 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
919 ei->i_disksize = inode->i_size;
920 mutex_unlock(&ei->truncate_mutex);
921 if (err)
922 goto cleanup;
923
924 set_buffer_new(bh_result);
925got_it:
926 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
927 if (count > blocks_to_boundary)
928 set_buffer_boundary(bh_result);
929 err = count;
930 /* Clean up and exit */
931 partial = chain + depth - 1; /* the whole chain */
932cleanup:
933 while (partial > chain) {
934 BUFFER_TRACE(partial->bh, "call brelse");
935 brelse(partial->bh);
936 partial--;
937 }
938 BUFFER_TRACE(bh_result, "returned");
939out:
940 return err;
941}
942
Mingming Cao617ba132006-10-11 01:20:53 -0700943#define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700944
Mingming Cao617ba132006-10-11 01:20:53 -0700945static int ext4_get_block(struct inode *inode, sector_t iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700946 struct buffer_head *bh_result, int create)
947{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800948 handle_t *handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700949 int ret = 0;
950 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
951
952 if (!create)
953 goto get_block; /* A read */
954
955 if (max_blocks == 1)
956 goto get_block; /* A single block get */
957
958 if (handle->h_transaction->t_state == T_LOCKED) {
959 /*
960 * Huge direct-io writes can hold off commits for long
961 * periods of time. Let this commit run.
962 */
Mingming Cao617ba132006-10-11 01:20:53 -0700963 ext4_journal_stop(handle);
964 handle = ext4_journal_start(inode, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700965 if (IS_ERR(handle))
966 ret = PTR_ERR(handle);
967 goto get_block;
968 }
969
Mingming Cao617ba132006-10-11 01:20:53 -0700970 if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700971 /*
972 * Getting low on buffer credits...
973 */
Mingming Cao617ba132006-10-11 01:20:53 -0700974 ret = ext4_journal_extend(handle, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700975 if (ret > 0) {
976 /*
977 * Couldn't extend the transaction. Start a new one.
978 */
Mingming Cao617ba132006-10-11 01:20:53 -0700979 ret = ext4_journal_restart(handle, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980 }
981 }
982
983get_block:
984 if (ret == 0) {
Alex Tomasa86c6182006-10-11 01:21:03 -0700985 ret = ext4_get_blocks_wrap(handle, inode, iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700986 max_blocks, bh_result, create, 0);
987 if (ret > 0) {
988 bh_result->b_size = (ret << inode->i_blkbits);
989 ret = 0;
990 }
991 }
992 return ret;
993}
994
995/*
996 * `handle' can be NULL if create is zero
997 */
Mingming Cao617ba132006-10-11 01:20:53 -0700998struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700999 long block, int create, int *errp)
1000{
1001 struct buffer_head dummy;
1002 int fatal = 0, err;
1003
1004 J_ASSERT(handle != NULL || create == 0);
1005
1006 dummy.b_state = 0;
1007 dummy.b_blocknr = -1000;
1008 buffer_trace_init(&dummy.b_history);
Alex Tomasa86c6182006-10-11 01:21:03 -07001009 err = ext4_get_blocks_wrap(handle, inode, block, 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001010 &dummy, create, 1);
1011 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001012 * ext4_get_blocks_handle() returns number of blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013 * mapped. 0 in case of a HOLE.
1014 */
1015 if (err > 0) {
1016 if (err > 1)
1017 WARN_ON(1);
1018 err = 0;
1019 }
1020 *errp = err;
1021 if (!err && buffer_mapped(&dummy)) {
1022 struct buffer_head *bh;
1023 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1024 if (!bh) {
1025 *errp = -EIO;
1026 goto err;
1027 }
1028 if (buffer_new(&dummy)) {
1029 J_ASSERT(create != 0);
1030 J_ASSERT(handle != 0);
1031
1032 /*
1033 * Now that we do not always journal data, we should
1034 * keep in mind whether this should always journal the
1035 * new buffer as metadata. For now, regular file
Mingming Cao617ba132006-10-11 01:20:53 -07001036 * writes use ext4_get_block instead, so it's not a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001037 * problem.
1038 */
1039 lock_buffer(bh);
1040 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001041 fatal = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001042 if (!fatal && !buffer_uptodate(bh)) {
1043 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1044 set_buffer_uptodate(bh);
1045 }
1046 unlock_buffer(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001047 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1048 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001049 if (!fatal)
1050 fatal = err;
1051 } else {
1052 BUFFER_TRACE(bh, "not a new buffer");
1053 }
1054 if (fatal) {
1055 *errp = fatal;
1056 brelse(bh);
1057 bh = NULL;
1058 }
1059 return bh;
1060 }
1061err:
1062 return NULL;
1063}
1064
Mingming Cao617ba132006-10-11 01:20:53 -07001065struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066 int block, int create, int *err)
1067{
1068 struct buffer_head * bh;
1069
Mingming Cao617ba132006-10-11 01:20:53 -07001070 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001071 if (!bh)
1072 return bh;
1073 if (buffer_uptodate(bh))
1074 return bh;
1075 ll_rw_block(READ_META, 1, &bh);
1076 wait_on_buffer(bh);
1077 if (buffer_uptodate(bh))
1078 return bh;
1079 put_bh(bh);
1080 *err = -EIO;
1081 return NULL;
1082}
1083
1084static int walk_page_buffers( handle_t *handle,
1085 struct buffer_head *head,
1086 unsigned from,
1087 unsigned to,
1088 int *partial,
1089 int (*fn)( handle_t *handle,
1090 struct buffer_head *bh))
1091{
1092 struct buffer_head *bh;
1093 unsigned block_start, block_end;
1094 unsigned blocksize = head->b_size;
1095 int err, ret = 0;
1096 struct buffer_head *next;
1097
1098 for ( bh = head, block_start = 0;
1099 ret == 0 && (bh != head || !block_start);
1100 block_start = block_end, bh = next)
1101 {
1102 next = bh->b_this_page;
1103 block_end = block_start + blocksize;
1104 if (block_end <= from || block_start >= to) {
1105 if (partial && !buffer_uptodate(bh))
1106 *partial = 1;
1107 continue;
1108 }
1109 err = (*fn)(handle, bh);
1110 if (!ret)
1111 ret = err;
1112 }
1113 return ret;
1114}
1115
1116/*
1117 * To preserve ordering, it is essential that the hole instantiation and
1118 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -07001119 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -07001120 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001121 * prepare_write() is the right place.
1122 *
Mingming Cao617ba132006-10-11 01:20:53 -07001123 * Also, this function can nest inside ext4_writepage() ->
1124 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001125 * has generated enough buffer credits to do the whole page. So we won't
1126 * block on the journal in that case, which is good, because the caller may
1127 * be PF_MEMALLOC.
1128 *
Mingming Cao617ba132006-10-11 01:20:53 -07001129 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001130 * quota file writes. If we were to commit the transaction while thus
1131 * reentered, there can be a deadlock - we would be holding a quota
1132 * lock, and the commit would never complete if another thread had a
1133 * transaction open and was blocking on the quota lock - a ranking
1134 * violation.
1135 *
Mingming Caodab291a2006-10-11 01:21:01 -07001136 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137 * will _not_ run commit under these circumstances because handle->h_ref
1138 * is elevated. We'll still have enough credits for the tiny quotafile
1139 * write.
1140 */
1141static int do_journal_get_write_access(handle_t *handle,
1142 struct buffer_head *bh)
1143{
1144 if (!buffer_mapped(bh) || buffer_freed(bh))
1145 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001146 return ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001147}
1148
Mingming Cao617ba132006-10-11 01:20:53 -07001149static int ext4_prepare_write(struct file *file, struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001150 unsigned from, unsigned to)
1151{
1152 struct inode *inode = page->mapping->host;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001153 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001154 handle_t *handle;
1155 int retries = 0;
1156
1157retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001158 handle = ext4_journal_start(inode, needed_blocks);
Andrew Morton7479d2b2007-04-01 23:49:44 -07001159 if (IS_ERR(handle)) {
1160 ret = PTR_ERR(handle);
1161 goto out;
1162 }
Mingming Cao617ba132006-10-11 01:20:53 -07001163 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1164 ret = nobh_prepare_write(page, from, to, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001165 else
Mingming Cao617ba132006-10-11 01:20:53 -07001166 ret = block_prepare_write(page, from, to, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001167 if (ret)
Andrew Morton7479d2b2007-04-01 23:49:44 -07001168 goto prepare_write_failed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001169
Mingming Cao617ba132006-10-11 01:20:53 -07001170 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001171 ret = walk_page_buffers(handle, page_buffers(page),
1172 from, to, NULL, do_journal_get_write_access);
1173 }
Andrew Morton7479d2b2007-04-01 23:49:44 -07001174prepare_write_failed:
1175 if (ret)
1176 ext4_journal_stop(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07001177 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001178 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001179out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001180 return ret;
1181}
1182
Mingming Cao617ba132006-10-11 01:20:53 -07001183int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001184{
Mingming Caodab291a2006-10-11 01:21:01 -07001185 int err = jbd2_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001186 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001187 ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001188 bh, handle,err);
1189 return err;
1190}
1191
1192/* For commit_write() in data=journal mode */
1193static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1194{
1195 if (!buffer_mapped(bh) || buffer_freed(bh))
1196 return 0;
1197 set_buffer_uptodate(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001198 return ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001199}
1200
1201/*
1202 * We need to pick up the new inode size which generic_commit_write gave us
1203 * `file' can be NULL - eg, when called from page_symlink().
1204 *
Mingming Cao617ba132006-10-11 01:20:53 -07001205 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001206 * buffers are managed internally.
1207 */
Mingming Cao617ba132006-10-11 01:20:53 -07001208static int ext4_ordered_commit_write(struct file *file, struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001209 unsigned from, unsigned to)
1210{
Mingming Cao617ba132006-10-11 01:20:53 -07001211 handle_t *handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001212 struct inode *inode = page->mapping->host;
1213 int ret = 0, ret2;
1214
1215 ret = walk_page_buffers(handle, page_buffers(page),
Mingming Cao617ba132006-10-11 01:20:53 -07001216 from, to, NULL, ext4_journal_dirty_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001217
1218 if (ret == 0) {
1219 /*
1220 * generic_commit_write() will run mark_inode_dirty() if i_size
1221 * changes. So let's piggyback the i_disksize mark_inode_dirty
1222 * into that.
1223 */
1224 loff_t new_i_size;
1225
1226 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
Mingming Cao617ba132006-10-11 01:20:53 -07001227 if (new_i_size > EXT4_I(inode)->i_disksize)
1228 EXT4_I(inode)->i_disksize = new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001229 ret = generic_commit_write(file, page, from, to);
1230 }
Mingming Cao617ba132006-10-11 01:20:53 -07001231 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001232 if (!ret)
1233 ret = ret2;
1234 return ret;
1235}
1236
Mingming Cao617ba132006-10-11 01:20:53 -07001237static int ext4_writeback_commit_write(struct file *file, struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001238 unsigned from, unsigned to)
1239{
Mingming Cao617ba132006-10-11 01:20:53 -07001240 handle_t *handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001241 struct inode *inode = page->mapping->host;
1242 int ret = 0, ret2;
1243 loff_t new_i_size;
1244
1245 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
Mingming Cao617ba132006-10-11 01:20:53 -07001246 if (new_i_size > EXT4_I(inode)->i_disksize)
1247 EXT4_I(inode)->i_disksize = new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001248
Mingming Cao617ba132006-10-11 01:20:53 -07001249 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001250 ret = nobh_commit_write(file, page, from, to);
1251 else
1252 ret = generic_commit_write(file, page, from, to);
1253
Mingming Cao617ba132006-10-11 01:20:53 -07001254 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001255 if (!ret)
1256 ret = ret2;
1257 return ret;
1258}
1259
Mingming Cao617ba132006-10-11 01:20:53 -07001260static int ext4_journalled_commit_write(struct file *file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001261 struct page *page, unsigned from, unsigned to)
1262{
Mingming Cao617ba132006-10-11 01:20:53 -07001263 handle_t *handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001264 struct inode *inode = page->mapping->host;
1265 int ret = 0, ret2;
1266 int partial = 0;
1267 loff_t pos;
1268
1269 /*
1270 * Here we duplicate the generic_commit_write() functionality
1271 */
1272 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1273
1274 ret = walk_page_buffers(handle, page_buffers(page), from,
1275 to, &partial, commit_write_fn);
1276 if (!partial)
1277 SetPageUptodate(page);
1278 if (pos > inode->i_size)
1279 i_size_write(inode, pos);
Mingming Cao617ba132006-10-11 01:20:53 -07001280 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1281 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1282 EXT4_I(inode)->i_disksize = inode->i_size;
1283 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001284 if (!ret)
1285 ret = ret2;
1286 }
Mingming Cao617ba132006-10-11 01:20:53 -07001287 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001288 if (!ret)
1289 ret = ret2;
1290 return ret;
1291}
1292
1293/*
1294 * bmap() is special. It gets used by applications such as lilo and by
1295 * the swapper to find the on-disk block of a specific piece of data.
1296 *
1297 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07001298 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001299 * filesystem and enables swap, then they may get a nasty shock when the
1300 * data getting swapped to that swapfile suddenly gets overwritten by
1301 * the original zero's written out previously to the journal and
1302 * awaiting writeback in the kernel's buffer cache.
1303 *
1304 * So, if we see any bmap calls here on a modified, data-journaled file,
1305 * take extra steps to flush any blocks which might be in the cache.
1306 */
Mingming Cao617ba132006-10-11 01:20:53 -07001307static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001308{
1309 struct inode *inode = mapping->host;
1310 journal_t *journal;
1311 int err;
1312
Mingming Cao617ba132006-10-11 01:20:53 -07001313 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001314 /*
1315 * This is a REALLY heavyweight approach, but the use of
1316 * bmap on dirty files is expected to be extremely rare:
1317 * only if we run lilo or swapon on a freshly made file
1318 * do we expect this to happen.
1319 *
1320 * (bmap requires CAP_SYS_RAWIO so this does not
1321 * represent an unprivileged user DOS attack --- we'd be
1322 * in trouble if mortal users could trigger this path at
1323 * will.)
1324 *
Mingming Cao617ba132006-10-11 01:20:53 -07001325 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001326 * regular files. If somebody wants to bmap a directory
1327 * or symlink and gets confused because the buffer
1328 * hasn't yet been flushed to disk, they deserve
1329 * everything they get.
1330 */
1331
Mingming Cao617ba132006-10-11 01:20:53 -07001332 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1333 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07001334 jbd2_journal_lock_updates(journal);
1335 err = jbd2_journal_flush(journal);
1336 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001337
1338 if (err)
1339 return 0;
1340 }
1341
Mingming Cao617ba132006-10-11 01:20:53 -07001342 return generic_block_bmap(mapping,block,ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001343}
1344
1345static int bget_one(handle_t *handle, struct buffer_head *bh)
1346{
1347 get_bh(bh);
1348 return 0;
1349}
1350
1351static int bput_one(handle_t *handle, struct buffer_head *bh)
1352{
1353 put_bh(bh);
1354 return 0;
1355}
1356
Mingming Caodab291a2006-10-11 01:21:01 -07001357static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001358{
1359 if (buffer_mapped(bh))
Mingming Cao617ba132006-10-11 01:20:53 -07001360 return ext4_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001361 return 0;
1362}
1363
1364/*
1365 * Note that we always start a transaction even if we're not journalling
1366 * data. This is to preserve ordering: any hole instantiation within
Mingming Cao617ba132006-10-11 01:20:53 -07001367 * __block_write_full_page -> ext4_get_block() should be journalled
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001368 * along with the data so we don't crash and then get metadata which
1369 * refers to old data.
1370 *
1371 * In all journalling modes block_write_full_page() will start the I/O.
1372 *
1373 * Problem:
1374 *
Mingming Cao617ba132006-10-11 01:20:53 -07001375 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1376 * ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001377 *
1378 * Similar for:
1379 *
Mingming Cao617ba132006-10-11 01:20:53 -07001380 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001381 *
Mingming Cao617ba132006-10-11 01:20:53 -07001382 * Same applies to ext4_get_block(). We will deadlock on various things like
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001383 * lock_journal and i_truncate_mutex.
1384 *
1385 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1386 * allocations fail.
1387 *
1388 * 16May01: If we're reentered then journal_current_handle() will be
1389 * non-zero. We simply *return*.
1390 *
1391 * 1 July 2001: @@@ FIXME:
1392 * In journalled data mode, a data buffer may be metadata against the
1393 * current transaction. But the same file is part of a shared mapping
1394 * and someone does a writepage() on it.
1395 *
1396 * We will move the buffer onto the async_data list, but *after* it has
1397 * been dirtied. So there's a small window where we have dirty data on
1398 * BJ_Metadata.
1399 *
1400 * Note that this only applies to the last partial page in the file. The
1401 * bit which block_write_full_page() uses prepare/commit for. (That's
1402 * broken code anyway: it's wrong for msync()).
1403 *
1404 * It's a rare case: affects the final partial page, for journalled data
1405 * where the file is subject to bith write() and writepage() in the same
1406 * transction. To fix it we'll need a custom block_write_full_page().
1407 * We'll probably need that anyway for journalling writepage() output.
1408 *
1409 * We don't honour synchronous mounts for writepage(). That would be
1410 * disastrous. Any write() or metadata operation will sync the fs for
1411 * us.
1412 *
1413 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1414 * we don't need to open a transaction here.
1415 */
Mingming Cao617ba132006-10-11 01:20:53 -07001416static int ext4_ordered_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001417 struct writeback_control *wbc)
1418{
1419 struct inode *inode = page->mapping->host;
1420 struct buffer_head *page_bufs;
1421 handle_t *handle = NULL;
1422 int ret = 0;
1423 int err;
1424
1425 J_ASSERT(PageLocked(page));
1426
1427 /*
1428 * We give up here if we're reentered, because it might be for a
1429 * different filesystem.
1430 */
Mingming Cao617ba132006-10-11 01:20:53 -07001431 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001432 goto out_fail;
1433
Mingming Cao617ba132006-10-11 01:20:53 -07001434 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001435
1436 if (IS_ERR(handle)) {
1437 ret = PTR_ERR(handle);
1438 goto out_fail;
1439 }
1440
1441 if (!page_has_buffers(page)) {
1442 create_empty_buffers(page, inode->i_sb->s_blocksize,
1443 (1 << BH_Dirty)|(1 << BH_Uptodate));
1444 }
1445 page_bufs = page_buffers(page);
1446 walk_page_buffers(handle, page_bufs, 0,
1447 PAGE_CACHE_SIZE, NULL, bget_one);
1448
Mingming Cao617ba132006-10-11 01:20:53 -07001449 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001450
1451 /*
1452 * The page can become unlocked at any point now, and
1453 * truncate can then come in and change things. So we
1454 * can't touch *page from now on. But *page_bufs is
1455 * safe due to elevated refcount.
1456 */
1457
1458 /*
1459 * And attach them to the current transaction. But only if
1460 * block_write_full_page() succeeded. Otherwise they are unmapped,
1461 * and generally junk.
1462 */
1463 if (ret == 0) {
1464 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
Mingming Caodab291a2006-10-11 01:21:01 -07001465 NULL, jbd2_journal_dirty_data_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001466 if (!ret)
1467 ret = err;
1468 }
1469 walk_page_buffers(handle, page_bufs, 0,
1470 PAGE_CACHE_SIZE, NULL, bput_one);
Mingming Cao617ba132006-10-11 01:20:53 -07001471 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472 if (!ret)
1473 ret = err;
1474 return ret;
1475
1476out_fail:
1477 redirty_page_for_writepage(wbc, page);
1478 unlock_page(page);
1479 return ret;
1480}
1481
Mingming Cao617ba132006-10-11 01:20:53 -07001482static int ext4_writeback_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001483 struct writeback_control *wbc)
1484{
1485 struct inode *inode = page->mapping->host;
1486 handle_t *handle = NULL;
1487 int ret = 0;
1488 int err;
1489
Mingming Cao617ba132006-10-11 01:20:53 -07001490 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001491 goto out_fail;
1492
Mingming Cao617ba132006-10-11 01:20:53 -07001493 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001494 if (IS_ERR(handle)) {
1495 ret = PTR_ERR(handle);
1496 goto out_fail;
1497 }
1498
Mingming Cao617ba132006-10-11 01:20:53 -07001499 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1500 ret = nobh_writepage(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501 else
Mingming Cao617ba132006-10-11 01:20:53 -07001502 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001503
Mingming Cao617ba132006-10-11 01:20:53 -07001504 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001505 if (!ret)
1506 ret = err;
1507 return ret;
1508
1509out_fail:
1510 redirty_page_for_writepage(wbc, page);
1511 unlock_page(page);
1512 return ret;
1513}
1514
Mingming Cao617ba132006-10-11 01:20:53 -07001515static int ext4_journalled_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001516 struct writeback_control *wbc)
1517{
1518 struct inode *inode = page->mapping->host;
1519 handle_t *handle = NULL;
1520 int ret = 0;
1521 int err;
1522
Mingming Cao617ba132006-10-11 01:20:53 -07001523 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001524 goto no_write;
1525
Mingming Cao617ba132006-10-11 01:20:53 -07001526 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001527 if (IS_ERR(handle)) {
1528 ret = PTR_ERR(handle);
1529 goto no_write;
1530 }
1531
1532 if (!page_has_buffers(page) || PageChecked(page)) {
1533 /*
1534 * It's mmapped pagecache. Add buffers and journal it. There
1535 * doesn't seem much point in redirtying the page here.
1536 */
1537 ClearPageChecked(page);
1538 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
Mingming Cao617ba132006-10-11 01:20:53 -07001539 ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001540 if (ret != 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001541 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001542 goto out_unlock;
1543 }
1544 ret = walk_page_buffers(handle, page_buffers(page), 0,
1545 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1546
1547 err = walk_page_buffers(handle, page_buffers(page), 0,
1548 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1549 if (ret == 0)
1550 ret = err;
Mingming Cao617ba132006-10-11 01:20:53 -07001551 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001552 unlock_page(page);
1553 } else {
1554 /*
1555 * It may be a page full of checkpoint-mode buffers. We don't
1556 * really know unless we go poke around in the buffer_heads.
1557 * But block_write_full_page will do the right thing.
1558 */
Mingming Cao617ba132006-10-11 01:20:53 -07001559 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 }
Mingming Cao617ba132006-10-11 01:20:53 -07001561 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001562 if (!ret)
1563 ret = err;
1564out:
1565 return ret;
1566
1567no_write:
1568 redirty_page_for_writepage(wbc, page);
1569out_unlock:
1570 unlock_page(page);
1571 goto out;
1572}
1573
Mingming Cao617ba132006-10-11 01:20:53 -07001574static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001575{
Mingming Cao617ba132006-10-11 01:20:53 -07001576 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001577}
1578
1579static int
Mingming Cao617ba132006-10-11 01:20:53 -07001580ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001581 struct list_head *pages, unsigned nr_pages)
1582{
Mingming Cao617ba132006-10-11 01:20:53 -07001583 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001584}
1585
Mingming Cao617ba132006-10-11 01:20:53 -07001586static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001587{
Mingming Cao617ba132006-10-11 01:20:53 -07001588 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001589
1590 /*
1591 * If it's a full truncate we just forget about the pending dirtying
1592 */
1593 if (offset == 0)
1594 ClearPageChecked(page);
1595
Mingming Caodab291a2006-10-11 01:21:01 -07001596 jbd2_journal_invalidatepage(journal, page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001597}
1598
Mingming Cao617ba132006-10-11 01:20:53 -07001599static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001600{
Mingming Cao617ba132006-10-11 01:20:53 -07001601 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001602
1603 WARN_ON(PageChecked(page));
1604 if (!page_has_buffers(page))
1605 return 0;
Mingming Caodab291a2006-10-11 01:21:01 -07001606 return jbd2_journal_try_to_free_buffers(journal, page, wait);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001607}
1608
1609/*
1610 * If the O_DIRECT write will extend the file then add this inode to the
1611 * orphan list. So recovery will truncate it back to the original size
1612 * if the machine crashes during the write.
1613 *
1614 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1615 * crashes then stale disk data _may_ be exposed inside the file.
1616 */
Mingming Cao617ba132006-10-11 01:20:53 -07001617static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001618 const struct iovec *iov, loff_t offset,
1619 unsigned long nr_segs)
1620{
1621 struct file *file = iocb->ki_filp;
1622 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07001623 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001624 handle_t *handle = NULL;
1625 ssize_t ret;
1626 int orphan = 0;
1627 size_t count = iov_length(iov, nr_segs);
1628
1629 if (rw == WRITE) {
1630 loff_t final_size = offset + count;
1631
Mingming Cao617ba132006-10-11 01:20:53 -07001632 handle = ext4_journal_start(inode, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001633 if (IS_ERR(handle)) {
1634 ret = PTR_ERR(handle);
1635 goto out;
1636 }
1637 if (final_size > inode->i_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001638 ret = ext4_orphan_add(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639 if (ret)
1640 goto out_stop;
1641 orphan = 1;
1642 ei->i_disksize = inode->i_size;
1643 }
1644 }
1645
1646 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1647 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07001648 ext4_get_block, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001649
1650 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001651 * Reacquire the handle: ext4_get_block() can restart the transaction
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001652 */
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001653 handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001654
1655out_stop:
1656 if (handle) {
1657 int err;
1658
1659 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07001660 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001661 if (orphan && ret > 0) {
1662 loff_t end = offset + ret;
1663 if (end > inode->i_size) {
1664 ei->i_disksize = end;
1665 i_size_write(inode, end);
1666 /*
1667 * We're going to return a positive `ret'
1668 * here due to non-zero-length I/O, so there's
1669 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07001670 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001671 * ignore it.
1672 */
Mingming Cao617ba132006-10-11 01:20:53 -07001673 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001674 }
1675 }
Mingming Cao617ba132006-10-11 01:20:53 -07001676 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001677 if (ret == 0)
1678 ret = err;
1679 }
1680out:
1681 return ret;
1682}
1683
1684/*
Mingming Cao617ba132006-10-11 01:20:53 -07001685 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001686 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1687 * much here because ->set_page_dirty is called under VFS locks. The page is
1688 * not necessarily locked.
1689 *
1690 * We cannot just dirty the page and leave attached buffers clean, because the
1691 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1692 * or jbddirty because all the journalling code will explode.
1693 *
1694 * So what we do is to mark the page "pending dirty" and next time writepage
1695 * is called, propagate that into the buffers appropriately.
1696 */
Mingming Cao617ba132006-10-11 01:20:53 -07001697static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001698{
1699 SetPageChecked(page);
1700 return __set_page_dirty_nobuffers(page);
1701}
1702
Mingming Cao617ba132006-10-11 01:20:53 -07001703static const struct address_space_operations ext4_ordered_aops = {
1704 .readpage = ext4_readpage,
1705 .readpages = ext4_readpages,
1706 .writepage = ext4_ordered_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001707 .sync_page = block_sync_page,
Mingming Cao617ba132006-10-11 01:20:53 -07001708 .prepare_write = ext4_prepare_write,
1709 .commit_write = ext4_ordered_commit_write,
1710 .bmap = ext4_bmap,
1711 .invalidatepage = ext4_invalidatepage,
1712 .releasepage = ext4_releasepage,
1713 .direct_IO = ext4_direct_IO,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001714 .migratepage = buffer_migrate_page,
1715};
1716
Mingming Cao617ba132006-10-11 01:20:53 -07001717static const struct address_space_operations ext4_writeback_aops = {
1718 .readpage = ext4_readpage,
1719 .readpages = ext4_readpages,
1720 .writepage = ext4_writeback_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001721 .sync_page = block_sync_page,
Mingming Cao617ba132006-10-11 01:20:53 -07001722 .prepare_write = ext4_prepare_write,
1723 .commit_write = ext4_writeback_commit_write,
1724 .bmap = ext4_bmap,
1725 .invalidatepage = ext4_invalidatepage,
1726 .releasepage = ext4_releasepage,
1727 .direct_IO = ext4_direct_IO,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001728 .migratepage = buffer_migrate_page,
1729};
1730
Mingming Cao617ba132006-10-11 01:20:53 -07001731static const struct address_space_operations ext4_journalled_aops = {
1732 .readpage = ext4_readpage,
1733 .readpages = ext4_readpages,
1734 .writepage = ext4_journalled_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001735 .sync_page = block_sync_page,
Mingming Cao617ba132006-10-11 01:20:53 -07001736 .prepare_write = ext4_prepare_write,
1737 .commit_write = ext4_journalled_commit_write,
1738 .set_page_dirty = ext4_journalled_set_page_dirty,
1739 .bmap = ext4_bmap,
1740 .invalidatepage = ext4_invalidatepage,
1741 .releasepage = ext4_releasepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001742};
1743
Mingming Cao617ba132006-10-11 01:20:53 -07001744void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001745{
Mingming Cao617ba132006-10-11 01:20:53 -07001746 if (ext4_should_order_data(inode))
1747 inode->i_mapping->a_ops = &ext4_ordered_aops;
1748 else if (ext4_should_writeback_data(inode))
1749 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001750 else
Mingming Cao617ba132006-10-11 01:20:53 -07001751 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752}
1753
1754/*
Mingming Cao617ba132006-10-11 01:20:53 -07001755 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001756 * up to the end of the block which corresponds to `from'.
1757 * This required during truncate. We need to physically zero the tail end
1758 * of that block so it doesn't yield old data if the file is later grown.
1759 */
Alex Tomasa86c6182006-10-11 01:21:03 -07001760int ext4_block_truncate_page(handle_t *handle, struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001761 struct address_space *mapping, loff_t from)
1762{
Mingming Cao617ba132006-10-11 01:20:53 -07001763 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1765 unsigned blocksize, iblock, length, pos;
1766 struct inode *inode = mapping->host;
1767 struct buffer_head *bh;
1768 int err = 0;
1769 void *kaddr;
1770
1771 blocksize = inode->i_sb->s_blocksize;
1772 length = blocksize - (offset & (blocksize - 1));
1773 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1774
1775 /*
1776 * For "nobh" option, we can only work if we don't need to
1777 * read-in the page - otherwise we create buffers to do the IO.
1778 */
1779 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
Mingming Cao617ba132006-10-11 01:20:53 -07001780 ext4_should_writeback_data(inode) && PageUptodate(page)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001781 kaddr = kmap_atomic(page, KM_USER0);
1782 memset(kaddr + offset, 0, length);
1783 flush_dcache_page(page);
1784 kunmap_atomic(kaddr, KM_USER0);
1785 set_page_dirty(page);
1786 goto unlock;
1787 }
1788
1789 if (!page_has_buffers(page))
1790 create_empty_buffers(page, blocksize, 0);
1791
1792 /* Find the buffer that contains "offset" */
1793 bh = page_buffers(page);
1794 pos = blocksize;
1795 while (offset >= pos) {
1796 bh = bh->b_this_page;
1797 iblock++;
1798 pos += blocksize;
1799 }
1800
1801 err = 0;
1802 if (buffer_freed(bh)) {
1803 BUFFER_TRACE(bh, "freed: skip");
1804 goto unlock;
1805 }
1806
1807 if (!buffer_mapped(bh)) {
1808 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07001809 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001810 /* unmapped? It's a hole - nothing to do */
1811 if (!buffer_mapped(bh)) {
1812 BUFFER_TRACE(bh, "still unmapped");
1813 goto unlock;
1814 }
1815 }
1816
1817 /* Ok, it's mapped. Make sure it's up-to-date */
1818 if (PageUptodate(page))
1819 set_buffer_uptodate(bh);
1820
1821 if (!buffer_uptodate(bh)) {
1822 err = -EIO;
1823 ll_rw_block(READ, 1, &bh);
1824 wait_on_buffer(bh);
1825 /* Uhhuh. Read error. Complain and punt. */
1826 if (!buffer_uptodate(bh))
1827 goto unlock;
1828 }
1829
Mingming Cao617ba132006-10-11 01:20:53 -07001830 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001831 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07001832 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001833 if (err)
1834 goto unlock;
1835 }
1836
1837 kaddr = kmap_atomic(page, KM_USER0);
1838 memset(kaddr + offset, 0, length);
1839 flush_dcache_page(page);
1840 kunmap_atomic(kaddr, KM_USER0);
1841
1842 BUFFER_TRACE(bh, "zeroed end of block");
1843
1844 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001845 if (ext4_should_journal_data(inode)) {
1846 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001847 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07001848 if (ext4_should_order_data(inode))
1849 err = ext4_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001850 mark_buffer_dirty(bh);
1851 }
1852
1853unlock:
1854 unlock_page(page);
1855 page_cache_release(page);
1856 return err;
1857}
1858
1859/*
1860 * Probably it should be a library function... search for first non-zero word
1861 * or memcmp with zero_page, whatever is better for particular architecture.
1862 * Linus?
1863 */
1864static inline int all_zeroes(__le32 *p, __le32 *q)
1865{
1866 while (p < q)
1867 if (*p++)
1868 return 0;
1869 return 1;
1870}
1871
1872/**
Mingming Cao617ba132006-10-11 01:20:53 -07001873 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001874 * @inode: inode in question
1875 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07001876 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001877 * @chain: place to store the pointers to partial indirect blocks
1878 * @top: place to the (detached) top of branch
1879 *
Mingming Cao617ba132006-10-11 01:20:53 -07001880 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001881 *
1882 * When we do truncate() we may have to clean the ends of several
1883 * indirect blocks but leave the blocks themselves alive. Block is
1884 * partially truncated if some data below the new i_size is refered
1885 * from it (and it is on the path to the first completely truncated
1886 * data block, indeed). We have to free the top of that path along
1887 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07001888 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001889 * finishes, we may safely do the latter, but top of branch may
1890 * require special attention - pageout below the truncation point
1891 * might try to populate it.
1892 *
1893 * We atomically detach the top of branch from the tree, store the
1894 * block number of its root in *@top, pointers to buffer_heads of
1895 * partially truncated blocks - in @chain[].bh and pointers to
1896 * their last elements that should not be removed - in
1897 * @chain[].p. Return value is the pointer to last filled element
1898 * of @chain.
1899 *
1900 * The work left to caller to do the actual freeing of subtrees:
1901 * a) free the subtree starting from *@top
1902 * b) free the subtrees whose roots are stored in
1903 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1904 * c) free the subtrees growing from the inode past the @chain[0].
1905 * (no partially truncated stuff there). */
1906
Mingming Cao617ba132006-10-11 01:20:53 -07001907static Indirect *ext4_find_shared(struct inode *inode, int depth,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001908 int offsets[4], Indirect chain[4], __le32 *top)
1909{
1910 Indirect *partial, *p;
1911 int k, err;
1912
1913 *top = 0;
1914 /* Make k index the deepest non-null offest + 1 */
1915 for (k = depth; k > 1 && !offsets[k-1]; k--)
1916 ;
Mingming Cao617ba132006-10-11 01:20:53 -07001917 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001918 /* Writer: pointers */
1919 if (!partial)
1920 partial = chain + k-1;
1921 /*
1922 * If the branch acquired continuation since we've looked at it -
1923 * fine, it should all survive and (new) top doesn't belong to us.
1924 */
1925 if (!partial->key && *partial->p)
1926 /* Writer: end */
1927 goto no_top;
1928 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1929 ;
1930 /*
1931 * OK, we've found the last block that must survive. The rest of our
1932 * branch should be detached before unlocking. However, if that rest
1933 * of branch is all ours and does not grow immediately from the inode
1934 * it's easier to cheat and just decrement partial->p.
1935 */
1936 if (p == chain + k - 1 && p > chain) {
1937 p->p--;
1938 } else {
1939 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07001940 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001941#if 0
1942 *p->p = 0;
1943#endif
1944 }
1945 /* Writer: end */
1946
1947 while(partial > p) {
1948 brelse(partial->bh);
1949 partial--;
1950 }
1951no_top:
1952 return partial;
1953}
1954
1955/*
1956 * Zero a number of block pointers in either an inode or an indirect block.
1957 * If we restart the transaction we must again get write access to the
1958 * indirect block for further modification.
1959 *
1960 * We release `count' blocks on disk, but (last - first) may be greater
1961 * than `count' because there can be holes in there.
1962 */
Mingming Cao617ba132006-10-11 01:20:53 -07001963static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
1964 struct buffer_head *bh, ext4_fsblk_t block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001965 unsigned long count, __le32 *first, __le32 *last)
1966{
1967 __le32 *p;
1968 if (try_to_extend_transaction(handle, inode)) {
1969 if (bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07001970 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1971 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001972 }
Mingming Cao617ba132006-10-11 01:20:53 -07001973 ext4_mark_inode_dirty(handle, inode);
1974 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001975 if (bh) {
1976 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07001977 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001978 }
1979 }
1980
1981 /*
1982 * Any buffers which are on the journal will be in memory. We find
Mingming Caodab291a2006-10-11 01:21:01 -07001983 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001984 * on them. We've already detached each block from the file, so
Mingming Caodab291a2006-10-11 01:21:01 -07001985 * bforget() in jbd2_journal_forget() should be safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001986 *
Mingming Caodab291a2006-10-11 01:21:01 -07001987 * AKPM: turn on bforget in jbd2_journal_forget()!!!
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001988 */
1989 for (p = first; p < last; p++) {
1990 u32 nr = le32_to_cpu(*p);
1991 if (nr) {
1992 struct buffer_head *bh;
1993
1994 *p = 0;
1995 bh = sb_find_get_block(inode->i_sb, nr);
Mingming Cao617ba132006-10-11 01:20:53 -07001996 ext4_forget(handle, 0, inode, bh, nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001997 }
1998 }
1999
Mingming Cao617ba132006-10-11 01:20:53 -07002000 ext4_free_blocks(handle, inode, block_to_free, count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002001}
2002
2003/**
Mingming Cao617ba132006-10-11 01:20:53 -07002004 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002005 * @handle: handle for this transaction
2006 * @inode: inode we are dealing with
2007 * @this_bh: indirect buffer_head which contains *@first and *@last
2008 * @first: array of block numbers
2009 * @last: points immediately past the end of array
2010 *
2011 * We are freeing all blocks refered from that array (numbers are stored as
2012 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2013 *
2014 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2015 * blocks are contiguous then releasing them at one time will only affect one
2016 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2017 * actually use a lot of journal space.
2018 *
2019 * @this_bh will be %NULL if @first and @last point into the inode's direct
2020 * block pointers.
2021 */
Mingming Cao617ba132006-10-11 01:20:53 -07002022static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002023 struct buffer_head *this_bh,
2024 __le32 *first, __le32 *last)
2025{
Mingming Cao617ba132006-10-11 01:20:53 -07002026 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002027 unsigned long count = 0; /* Number of blocks in the run */
2028 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2029 corresponding to
2030 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07002031 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002032 __le32 *p; /* Pointer into inode/ind
2033 for current block */
2034 int err;
2035
2036 if (this_bh) { /* For indirect block */
2037 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002038 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002039 /* Important: if we can't update the indirect pointers
2040 * to the blocks, we can't free them. */
2041 if (err)
2042 return;
2043 }
2044
2045 for (p = first; p < last; p++) {
2046 nr = le32_to_cpu(*p);
2047 if (nr) {
2048 /* accumulate blocks to free if they're contiguous */
2049 if (count == 0) {
2050 block_to_free = nr;
2051 block_to_free_p = p;
2052 count = 1;
2053 } else if (nr == block_to_free + count) {
2054 count++;
2055 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002056 ext4_clear_blocks(handle, inode, this_bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002057 block_to_free,
2058 count, block_to_free_p, p);
2059 block_to_free = nr;
2060 block_to_free_p = p;
2061 count = 1;
2062 }
2063 }
2064 }
2065
2066 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07002067 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002068 count, block_to_free_p, p);
2069
2070 if (this_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002071 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2072 ext4_journal_dirty_metadata(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002073 }
2074}
2075
2076/**
Mingming Cao617ba132006-10-11 01:20:53 -07002077 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002078 * @handle: JBD handle for this transaction
2079 * @inode: inode we are dealing with
2080 * @parent_bh: the buffer_head which contains *@first and *@last
2081 * @first: array of block numbers
2082 * @last: pointer immediately past the end of array
2083 * @depth: depth of the branches to free
2084 *
2085 * We are freeing all blocks refered from these branches (numbers are
2086 * stored as little-endian 32-bit) and updating @inode->i_blocks
2087 * appropriately.
2088 */
Mingming Cao617ba132006-10-11 01:20:53 -07002089static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002090 struct buffer_head *parent_bh,
2091 __le32 *first, __le32 *last, int depth)
2092{
Mingming Cao617ba132006-10-11 01:20:53 -07002093 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002094 __le32 *p;
2095
2096 if (is_handle_aborted(handle))
2097 return;
2098
2099 if (depth--) {
2100 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002101 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002102 p = last;
2103 while (--p >= first) {
2104 nr = le32_to_cpu(*p);
2105 if (!nr)
2106 continue; /* A hole */
2107
2108 /* Go read the buffer for the next level down */
2109 bh = sb_bread(inode->i_sb, nr);
2110
2111 /*
2112 * A read failure? Report error and clear slot
2113 * (should be rare).
2114 */
2115 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002116 ext4_error(inode->i_sb, "ext4_free_branches",
Mingming Cao2ae02102006-10-11 01:21:11 -07002117 "Read failure, inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002118 inode->i_ino, nr);
2119 continue;
2120 }
2121
2122 /* This zaps the entire block. Bottom up. */
2123 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07002124 ext4_free_branches(handle, inode, bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002125 (__le32*)bh->b_data,
2126 (__le32*)bh->b_data + addr_per_block,
2127 depth);
2128
2129 /*
2130 * We've probably journalled the indirect block several
2131 * times during the truncate. But it's no longer
2132 * needed and we now drop it from the transaction via
Mingming Caodab291a2006-10-11 01:21:01 -07002133 * jbd2_journal_revoke().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002134 *
2135 * That's easy if it's exclusively part of this
2136 * transaction. But if it's part of the committing
Mingming Caodab291a2006-10-11 01:21:01 -07002137 * transaction then jbd2_journal_forget() will simply
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002138 * brelse() it. That means that if the underlying
Mingming Cao617ba132006-10-11 01:20:53 -07002139 * block is reallocated in ext4_get_block(),
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002140 * unmap_underlying_metadata() will find this block
2141 * and will try to get rid of it. damn, damn.
2142 *
2143 * If this block has already been committed to the
2144 * journal, a revoke record will be written. And
2145 * revoke records must be emitted *before* clearing
2146 * this block's bit in the bitmaps.
2147 */
Mingming Cao617ba132006-10-11 01:20:53 -07002148 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002149
2150 /*
2151 * Everything below this this pointer has been
2152 * released. Now let this top-of-subtree go.
2153 *
2154 * We want the freeing of this indirect block to be
2155 * atomic in the journal with the updating of the
2156 * bitmap block which owns it. So make some room in
2157 * the journal.
2158 *
2159 * We zero the parent pointer *after* freeing its
2160 * pointee in the bitmaps, so if extend_transaction()
2161 * for some reason fails to put the bitmap changes and
2162 * the release into the same transaction, recovery
2163 * will merely complain about releasing a free block,
2164 * rather than leaking blocks.
2165 */
2166 if (is_handle_aborted(handle))
2167 return;
2168 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002169 ext4_mark_inode_dirty(handle, inode);
2170 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002171 }
2172
Mingming Cao617ba132006-10-11 01:20:53 -07002173 ext4_free_blocks(handle, inode, nr, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002174
2175 if (parent_bh) {
2176 /*
2177 * The block which we have just freed is
2178 * pointed to by an indirect block: journal it
2179 */
2180 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002181 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002182 parent_bh)){
2183 *p = 0;
2184 BUFFER_TRACE(parent_bh,
Mingming Cao617ba132006-10-11 01:20:53 -07002185 "call ext4_journal_dirty_metadata");
2186 ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002187 parent_bh);
2188 }
2189 }
2190 }
2191 } else {
2192 /* We have reached the bottom of the tree. */
2193 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07002194 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002195 }
2196}
2197
2198/*
Mingming Cao617ba132006-10-11 01:20:53 -07002199 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002200 *
Mingming Cao617ba132006-10-11 01:20:53 -07002201 * We block out ext4_get_block() block instantiations across the entire
2202 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002203 * simultaneously on behalf of the same inode.
2204 *
2205 * As we work through the truncate and commmit bits of it to the journal there
2206 * is one core, guiding principle: the file's tree must always be consistent on
2207 * disk. We must be able to restart the truncate after a crash.
2208 *
2209 * The file's tree may be transiently inconsistent in memory (although it
2210 * probably isn't), but whenever we close off and commit a journal transaction,
2211 * the contents of (the filesystem + the journal) must be consistent and
2212 * restartable. It's pretty simple, really: bottom up, right to left (although
2213 * left-to-right works OK too).
2214 *
2215 * Note that at recovery time, journal replay occurs *before* the restart of
2216 * truncate against the orphan inode list.
2217 *
2218 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07002219 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002220 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07002221 * ext4_truncate() to have another go. So there will be instantiated blocks
2222 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002223 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07002224 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002225 */
Mingming Cao617ba132006-10-11 01:20:53 -07002226void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002227{
2228 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07002229 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002230 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07002231 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002232 struct address_space *mapping = inode->i_mapping;
2233 int offsets[4];
2234 Indirect chain[4];
2235 Indirect *partial;
2236 __le32 nr = 0;
2237 int n;
2238 long last_block;
2239 unsigned blocksize = inode->i_sb->s_blocksize;
2240 struct page *page;
2241
2242 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2243 S_ISLNK(inode->i_mode)))
2244 return;
Mingming Cao617ba132006-10-11 01:20:53 -07002245 if (ext4_inode_is_fast_symlink(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002246 return;
2247 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2248 return;
2249
2250 /*
2251 * We have to lock the EOF page here, because lock_page() nests
Mingming Caodab291a2006-10-11 01:21:01 -07002252 * outside jbd2_journal_start().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002253 */
2254 if ((inode->i_size & (blocksize - 1)) == 0) {
2255 /* Block boundary? Nothing to do */
2256 page = NULL;
2257 } else {
2258 page = grab_cache_page(mapping,
2259 inode->i_size >> PAGE_CACHE_SHIFT);
2260 if (!page)
2261 return;
2262 }
2263
Alex Tomasa86c6182006-10-11 01:21:03 -07002264 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
2265 return ext4_ext_truncate(inode, page);
2266
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002267 handle = start_transaction(inode);
2268 if (IS_ERR(handle)) {
2269 if (page) {
2270 clear_highpage(page);
2271 flush_dcache_page(page);
2272 unlock_page(page);
2273 page_cache_release(page);
2274 }
2275 return; /* AKPM: return what? */
2276 }
2277
2278 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07002279 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002280
2281 if (page)
Mingming Cao617ba132006-10-11 01:20:53 -07002282 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002283
Mingming Cao617ba132006-10-11 01:20:53 -07002284 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002285 if (n == 0)
2286 goto out_stop; /* error */
2287
2288 /*
2289 * OK. This truncate is going to happen. We add the inode to the
2290 * orphan list, so that if this truncate spans multiple transactions,
2291 * and we crash, we will resume the truncate when the filesystem
2292 * recovers. It also marks the inode dirty, to catch the new size.
2293 *
2294 * Implication: the file must always be in a sane, consistent
2295 * truncatable state while each transaction commits.
2296 */
Mingming Cao617ba132006-10-11 01:20:53 -07002297 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002298 goto out_stop;
2299
2300 /*
2301 * The orphan list entry will now protect us from any crash which
2302 * occurs before the truncate completes, so it is now safe to propagate
2303 * the new, shorter inode size (held for now in i_size) into the
2304 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07002305 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002306 */
2307 ei->i_disksize = inode->i_size;
2308
2309 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002310 * From here we block out all ext4_get_block() callers who want to
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002311 * modify the block allocation tree.
2312 */
2313 mutex_lock(&ei->truncate_mutex);
2314
2315 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07002316 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2317 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002318 goto do_indirects;
2319 }
2320
Mingming Cao617ba132006-10-11 01:20:53 -07002321 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002322 /* Kill the top of shared branch (not detached) */
2323 if (nr) {
2324 if (partial == chain) {
2325 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07002326 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002327 &nr, &nr+1, (chain+n-1) - partial);
2328 *partial->p = 0;
2329 /*
2330 * We mark the inode dirty prior to restart,
2331 * and prior to stop. No need for it here.
2332 */
2333 } else {
2334 /* Shared branch grows from an indirect block */
2335 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002336 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002337 partial->p,
2338 partial->p+1, (chain+n-1) - partial);
2339 }
2340 }
2341 /* Clear the ends of indirect blocks on the shared branch */
2342 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07002343 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002344 (__le32*)partial->bh->b_data+addr_per_block,
2345 (chain+n-1) - partial);
2346 BUFFER_TRACE(partial->bh, "call brelse");
2347 brelse (partial->bh);
2348 partial--;
2349 }
2350do_indirects:
2351 /* Kill the remaining (whole) subtrees */
2352 switch (offsets[0]) {
2353 default:
Mingming Cao617ba132006-10-11 01:20:53 -07002354 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002355 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002356 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2357 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002358 }
Mingming Cao617ba132006-10-11 01:20:53 -07002359 case EXT4_IND_BLOCK:
2360 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002361 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002362 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2363 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002364 }
Mingming Cao617ba132006-10-11 01:20:53 -07002365 case EXT4_DIND_BLOCK:
2366 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002367 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002368 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2369 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002370 }
Mingming Cao617ba132006-10-11 01:20:53 -07002371 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002372 ;
2373 }
2374
Mingming Cao617ba132006-10-11 01:20:53 -07002375 ext4_discard_reservation(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002376
2377 mutex_unlock(&ei->truncate_mutex);
2378 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
Mingming Cao617ba132006-10-11 01:20:53 -07002379 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002380
2381 /*
2382 * In a multi-transaction truncate, we only make the final transaction
2383 * synchronous
2384 */
2385 if (IS_SYNC(inode))
2386 handle->h_sync = 1;
2387out_stop:
2388 /*
2389 * If this was a simple ftruncate(), and the file will remain alive
2390 * then we need to clear up the orphan record which we created above.
2391 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07002392 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002393 * orphan info for us.
2394 */
2395 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002396 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002397
Mingming Cao617ba132006-10-11 01:20:53 -07002398 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002399}
2400
Mingming Cao617ba132006-10-11 01:20:53 -07002401static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2402 unsigned long ino, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002403{
2404 unsigned long desc, group_desc, block_group;
2405 unsigned long offset;
Mingming Cao617ba132006-10-11 01:20:53 -07002406 ext4_fsblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002407 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002408 struct ext4_group_desc * gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002409
Mingming Cao617ba132006-10-11 01:20:53 -07002410 if (!ext4_valid_inum(sb, ino)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002411 /*
2412 * This error is already checked for in namei.c unless we are
2413 * looking at an NFS filehandle, in which case no error
2414 * report is needed
2415 */
2416 return 0;
2417 }
2418
Mingming Cao617ba132006-10-11 01:20:53 -07002419 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2420 if (block_group >= EXT4_SB(sb)->s_groups_count) {
2421 ext4_error(sb,"ext4_get_inode_block","group >= groups count");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002422 return 0;
2423 }
2424 smp_rmb();
Mingming Cao617ba132006-10-11 01:20:53 -07002425 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2426 desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2427 bh = EXT4_SB(sb)->s_group_desc[group_desc];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002428 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002429 ext4_error (sb, "ext4_get_inode_block",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002430 "Descriptor not loaded");
2431 return 0;
2432 }
2433
Alexandre Ratchov0d1ee422006-10-11 01:21:14 -07002434 gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2435 desc * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002436 /*
2437 * Figure out the offset within the block group inode table
2438 */
Mingming Cao617ba132006-10-11 01:20:53 -07002439 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2440 EXT4_INODE_SIZE(sb);
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07002441 block = ext4_inode_table(sb, gdp) +
2442 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002443
2444 iloc->block_group = block_group;
Mingming Cao617ba132006-10-11 01:20:53 -07002445 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002446 return block;
2447}
2448
2449/*
Mingming Cao617ba132006-10-11 01:20:53 -07002450 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002451 * underlying buffer_head on success. If 'in_mem' is true, we have all
2452 * data in memory that is needed to recreate the on-disk version of this
2453 * inode.
2454 */
Mingming Cao617ba132006-10-11 01:20:53 -07002455static int __ext4_get_inode_loc(struct inode *inode,
2456 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002457{
Mingming Cao617ba132006-10-11 01:20:53 -07002458 ext4_fsblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002459 struct buffer_head *bh;
2460
Mingming Cao617ba132006-10-11 01:20:53 -07002461 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002462 if (!block)
2463 return -EIO;
2464
2465 bh = sb_getblk(inode->i_sb, block);
2466 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002467 ext4_error (inode->i_sb, "ext4_get_inode_loc",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002468 "unable to read inode block - "
Mingming Cao2ae02102006-10-11 01:21:11 -07002469 "inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002470 inode->i_ino, block);
2471 return -EIO;
2472 }
2473 if (!buffer_uptodate(bh)) {
2474 lock_buffer(bh);
2475 if (buffer_uptodate(bh)) {
2476 /* someone brought it uptodate while we waited */
2477 unlock_buffer(bh);
2478 goto has_buffer;
2479 }
2480
2481 /*
2482 * If we have all information of the inode in memory and this
2483 * is the only valid inode in the block, we need not read the
2484 * block.
2485 */
2486 if (in_mem) {
2487 struct buffer_head *bitmap_bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002488 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002489 int inodes_per_buffer;
2490 int inode_offset, i;
2491 int block_group;
2492 int start;
2493
2494 block_group = (inode->i_ino - 1) /
Mingming Cao617ba132006-10-11 01:20:53 -07002495 EXT4_INODES_PER_GROUP(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002496 inodes_per_buffer = bh->b_size /
Mingming Cao617ba132006-10-11 01:20:53 -07002497 EXT4_INODE_SIZE(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002498 inode_offset = ((inode->i_ino - 1) %
Mingming Cao617ba132006-10-11 01:20:53 -07002499 EXT4_INODES_PER_GROUP(inode->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002500 start = inode_offset & ~(inodes_per_buffer - 1);
2501
2502 /* Is the inode bitmap in cache? */
Mingming Cao617ba132006-10-11 01:20:53 -07002503 desc = ext4_get_group_desc(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002504 block_group, NULL);
2505 if (!desc)
2506 goto make_io;
2507
2508 bitmap_bh = sb_getblk(inode->i_sb,
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07002509 ext4_inode_bitmap(inode->i_sb, desc));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002510 if (!bitmap_bh)
2511 goto make_io;
2512
2513 /*
2514 * If the inode bitmap isn't in cache then the
2515 * optimisation may end up performing two reads instead
2516 * of one, so skip it.
2517 */
2518 if (!buffer_uptodate(bitmap_bh)) {
2519 brelse(bitmap_bh);
2520 goto make_io;
2521 }
2522 for (i = start; i < start + inodes_per_buffer; i++) {
2523 if (i == inode_offset)
2524 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07002525 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002526 break;
2527 }
2528 brelse(bitmap_bh);
2529 if (i == start + inodes_per_buffer) {
2530 /* all other inodes are free, so skip I/O */
2531 memset(bh->b_data, 0, bh->b_size);
2532 set_buffer_uptodate(bh);
2533 unlock_buffer(bh);
2534 goto has_buffer;
2535 }
2536 }
2537
2538make_io:
2539 /*
2540 * There are other valid inodes in the buffer, this inode
2541 * has in-inode xattrs, or we don't have this inode in memory.
2542 * Read the block from disk.
2543 */
2544 get_bh(bh);
2545 bh->b_end_io = end_buffer_read_sync;
2546 submit_bh(READ_META, bh);
2547 wait_on_buffer(bh);
2548 if (!buffer_uptodate(bh)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002549 ext4_error(inode->i_sb, "ext4_get_inode_loc",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002550 "unable to read inode block - "
Mingming Cao2ae02102006-10-11 01:21:11 -07002551 "inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002552 inode->i_ino, block);
2553 brelse(bh);
2554 return -EIO;
2555 }
2556 }
2557has_buffer:
2558 iloc->bh = bh;
2559 return 0;
2560}
2561
Mingming Cao617ba132006-10-11 01:20:53 -07002562int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002563{
2564 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07002565 return __ext4_get_inode_loc(inode, iloc,
2566 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002567}
2568
Mingming Cao617ba132006-10-11 01:20:53 -07002569void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002570{
Mingming Cao617ba132006-10-11 01:20:53 -07002571 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002572
2573 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07002574 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002575 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07002576 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002577 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07002578 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002579 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07002580 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002581 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07002582 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002583 inode->i_flags |= S_DIRSYNC;
2584}
2585
Jan Karaff9ddf72007-07-18 09:24:20 -04002586/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2587void ext4_get_inode_flags(struct ext4_inode_info *ei)
2588{
2589 unsigned int flags = ei->vfs_inode.i_flags;
2590
2591 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
2592 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
2593 if (flags & S_SYNC)
2594 ei->i_flags |= EXT4_SYNC_FL;
2595 if (flags & S_APPEND)
2596 ei->i_flags |= EXT4_APPEND_FL;
2597 if (flags & S_IMMUTABLE)
2598 ei->i_flags |= EXT4_IMMUTABLE_FL;
2599 if (flags & S_NOATIME)
2600 ei->i_flags |= EXT4_NOATIME_FL;
2601 if (flags & S_DIRSYNC)
2602 ei->i_flags |= EXT4_DIRSYNC_FL;
2603}
2604
Mingming Cao617ba132006-10-11 01:20:53 -07002605void ext4_read_inode(struct inode * inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002606{
Mingming Cao617ba132006-10-11 01:20:53 -07002607 struct ext4_iloc iloc;
2608 struct ext4_inode *raw_inode;
2609 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002610 struct buffer_head *bh;
2611 int block;
2612
Mingming Cao617ba132006-10-11 01:20:53 -07002613#ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2614 ei->i_acl = EXT4_ACL_NOT_CACHED;
2615 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002616#endif
2617 ei->i_block_alloc_info = NULL;
2618
Mingming Cao617ba132006-10-11 01:20:53 -07002619 if (__ext4_get_inode_loc(inode, &iloc, 0))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002620 goto bad_inode;
2621 bh = iloc.bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002622 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002623 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2624 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2625 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2626 if(!(test_opt (inode->i_sb, NO_UID32))) {
2627 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2628 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2629 }
2630 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2631 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07002632 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2633 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2634 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002635 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2636
2637 ei->i_state = 0;
2638 ei->i_dir_start_lookup = 0;
2639 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2640 /* We now have enough fields to check if the inode was active or not.
2641 * This is needed because nfsd might try to access dead inodes
2642 * the test is that same one that e2fsck uses
2643 * NeilBrown 1999oct15
2644 */
2645 if (inode->i_nlink == 0) {
2646 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07002647 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002648 /* this inode is deleted */
2649 brelse (bh);
2650 goto bad_inode;
2651 }
2652 /* The only unlinked inodes we let through here have
2653 * valid i_mode and are being read by the orphan
2654 * recovery code: that's fine, we're about to complete
2655 * the process of deleting those. */
2656 }
2657 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2658 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Mingming Cao617ba132006-10-11 01:20:53 -07002659#ifdef EXT4_FRAGMENTS
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002660 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2661 ei->i_frag_no = raw_inode->i_frag;
2662 ei->i_frag_size = raw_inode->i_fsize;
2663#endif
2664 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07002665 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2666 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07002667 ei->i_file_acl |=
2668 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002669 if (!S_ISREG(inode->i_mode)) {
2670 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2671 } else {
2672 inode->i_size |=
2673 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2674 }
2675 ei->i_disksize = inode->i_size;
2676 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2677 ei->i_block_group = iloc.block_group;
2678 /*
2679 * NOTE! The in-memory inode i_data array is in little-endian order
2680 * even on big-endian machines: we do NOT byteswap the block numbers!
2681 */
Mingming Cao617ba132006-10-11 01:20:53 -07002682 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002683 ei->i_data[block] = raw_inode->i_block[block];
2684 INIT_LIST_HEAD(&ei->i_orphan);
2685
Mingming Cao617ba132006-10-11 01:20:53 -07002686 if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2687 EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002688 /*
2689 * When mke2fs creates big inodes it does not zero out
Mingming Cao617ba132006-10-11 01:20:53 -07002690 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002691 * so ignore those first few inodes.
2692 */
2693 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07002694 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07002695 EXT4_INODE_SIZE(inode->i_sb)) {
2696 brelse (bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002697 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07002698 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002699 if (ei->i_extra_isize == 0) {
2700 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07002701 ei->i_extra_isize = sizeof(struct ext4_inode) -
2702 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002703 } else {
2704 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07002705 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002706 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07002707 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2708 ei->i_state |= EXT4_STATE_XATTR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002709 }
2710 } else
2711 ei->i_extra_isize = 0;
2712
2713 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002714 inode->i_op = &ext4_file_inode_operations;
2715 inode->i_fop = &ext4_file_operations;
2716 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002717 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002718 inode->i_op = &ext4_dir_inode_operations;
2719 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002720 } else if (S_ISLNK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002721 if (ext4_inode_is_fast_symlink(inode))
2722 inode->i_op = &ext4_fast_symlink_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002723 else {
Mingming Cao617ba132006-10-11 01:20:53 -07002724 inode->i_op = &ext4_symlink_inode_operations;
2725 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002726 }
2727 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002728 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002729 if (raw_inode->i_block[0])
2730 init_special_inode(inode, inode->i_mode,
2731 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2732 else
2733 init_special_inode(inode, inode->i_mode,
2734 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2735 }
2736 brelse (iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002737 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002738 return;
2739
2740bad_inode:
2741 make_bad_inode(inode);
2742 return;
2743}
2744
2745/*
2746 * Post the struct inode info into an on-disk inode location in the
2747 * buffer-cache. This gobbles the caller's reference to the
2748 * buffer_head in the inode location struct.
2749 *
2750 * The caller must have write access to iloc->bh.
2751 */
Mingming Cao617ba132006-10-11 01:20:53 -07002752static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002753 struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -07002754 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002755{
Mingming Cao617ba132006-10-11 01:20:53 -07002756 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2757 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002758 struct buffer_head *bh = iloc->bh;
2759 int err = 0, rc, block;
2760
2761 /* For fields not not tracking in the in-memory inode,
2762 * initialise them to zero for new inodes. */
Mingming Cao617ba132006-10-11 01:20:53 -07002763 if (ei->i_state & EXT4_STATE_NEW)
2764 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002765
Jan Karaff9ddf72007-07-18 09:24:20 -04002766 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002767 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2768 if(!(test_opt(inode->i_sb, NO_UID32))) {
2769 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2770 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2771/*
2772 * Fix up interoperability with old kernels. Otherwise, old inodes get
2773 * re-used with the upper 16 bits of the uid/gid intact
2774 */
2775 if(!ei->i_dtime) {
2776 raw_inode->i_uid_high =
2777 cpu_to_le16(high_16_bits(inode->i_uid));
2778 raw_inode->i_gid_high =
2779 cpu_to_le16(high_16_bits(inode->i_gid));
2780 } else {
2781 raw_inode->i_uid_high = 0;
2782 raw_inode->i_gid_high = 0;
2783 }
2784 } else {
2785 raw_inode->i_uid_low =
2786 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2787 raw_inode->i_gid_low =
2788 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2789 raw_inode->i_uid_high = 0;
2790 raw_inode->i_gid_high = 0;
2791 }
2792 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2793 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2794 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2795 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2796 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2797 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2798 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2799 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
Mingming Cao617ba132006-10-11 01:20:53 -07002800#ifdef EXT4_FRAGMENTS
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002801 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2802 raw_inode->i_frag = ei->i_frag_no;
2803 raw_inode->i_fsize = ei->i_frag_size;
2804#endif
Mingming Cao9b8f1f02006-10-11 01:21:13 -07002805 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2806 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07002807 raw_inode->i_file_acl_high =
2808 cpu_to_le16(ei->i_file_acl >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002809 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2810 if (!S_ISREG(inode->i_mode)) {
2811 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2812 } else {
2813 raw_inode->i_size_high =
2814 cpu_to_le32(ei->i_disksize >> 32);
2815 if (ei->i_disksize > 0x7fffffffULL) {
2816 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07002817 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2818 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2819 EXT4_SB(sb)->s_es->s_rev_level ==
2820 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002821 /* If this is the first large file
2822 * created, add a flag to the superblock.
2823 */
Mingming Cao617ba132006-10-11 01:20:53 -07002824 err = ext4_journal_get_write_access(handle,
2825 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002826 if (err)
2827 goto out_brelse;
Mingming Cao617ba132006-10-11 01:20:53 -07002828 ext4_update_dynamic_rev(sb);
2829 EXT4_SET_RO_COMPAT_FEATURE(sb,
2830 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002831 sb->s_dirt = 1;
2832 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07002833 err = ext4_journal_dirty_metadata(handle,
2834 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002835 }
2836 }
2837 }
2838 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2839 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2840 if (old_valid_dev(inode->i_rdev)) {
2841 raw_inode->i_block[0] =
2842 cpu_to_le32(old_encode_dev(inode->i_rdev));
2843 raw_inode->i_block[1] = 0;
2844 } else {
2845 raw_inode->i_block[0] = 0;
2846 raw_inode->i_block[1] =
2847 cpu_to_le32(new_encode_dev(inode->i_rdev));
2848 raw_inode->i_block[2] = 0;
2849 }
Mingming Cao617ba132006-10-11 01:20:53 -07002850 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002851 raw_inode->i_block[block] = ei->i_data[block];
2852
2853 if (ei->i_extra_isize)
2854 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2855
Mingming Cao617ba132006-10-11 01:20:53 -07002856 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2857 rc = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002858 if (!err)
2859 err = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07002860 ei->i_state &= ~EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002861
2862out_brelse:
2863 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002864 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002865 return err;
2866}
2867
2868/*
Mingming Cao617ba132006-10-11 01:20:53 -07002869 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002870 *
2871 * We are called from a few places:
2872 *
2873 * - Within generic_file_write() for O_SYNC files.
2874 * Here, there will be no transaction running. We wait for any running
2875 * trasnaction to commit.
2876 *
2877 * - Within sys_sync(), kupdate and such.
2878 * We wait on commit, if tol to.
2879 *
2880 * - Within prune_icache() (PF_MEMALLOC == true)
2881 * Here we simply return. We can't afford to block kswapd on the
2882 * journal commit.
2883 *
2884 * In all cases it is actually safe for us to return without doing anything,
2885 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07002886 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002887 * knfsd.
2888 *
2889 * Note that we are absolutely dependent upon all inode dirtiers doing the
2890 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2891 * which we are interested.
2892 *
2893 * It would be a bug for them to not do this. The code:
2894 *
2895 * mark_inode_dirty(inode)
2896 * stuff();
2897 * inode->i_size = expr;
2898 *
2899 * is in error because a kswapd-driven write_inode() could occur while
2900 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2901 * will no longer be on the superblock's dirty inode list.
2902 */
Mingming Cao617ba132006-10-11 01:20:53 -07002903int ext4_write_inode(struct inode *inode, int wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002904{
2905 if (current->flags & PF_MEMALLOC)
2906 return 0;
2907
Mingming Cao617ba132006-10-11 01:20:53 -07002908 if (ext4_journal_current_handle()) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002909 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2910 dump_stack();
2911 return -EIO;
2912 }
2913
2914 if (!wait)
2915 return 0;
2916
Mingming Cao617ba132006-10-11 01:20:53 -07002917 return ext4_force_commit(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002918}
2919
2920/*
Mingming Cao617ba132006-10-11 01:20:53 -07002921 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002922 *
2923 * Called from notify_change.
2924 *
2925 * We want to trap VFS attempts to truncate the file as soon as
2926 * possible. In particular, we want to make sure that when the VFS
2927 * shrinks i_size, we put the inode on the orphan list and modify
2928 * i_disksize immediately, so that during the subsequent flushing of
2929 * dirty pages and freeing of disk blocks, we can guarantee that any
2930 * commit will leave the blocks being flushed in an unused state on
2931 * disk. (On recovery, the inode will get truncated and the blocks will
2932 * be freed, so we have a strong guarantee that no future commit will
2933 * leave these blocks visible to the user.)
2934 *
2935 * Called with inode->sem down.
2936 */
Mingming Cao617ba132006-10-11 01:20:53 -07002937int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002938{
2939 struct inode *inode = dentry->d_inode;
2940 int error, rc = 0;
2941 const unsigned int ia_valid = attr->ia_valid;
2942
2943 error = inode_change_ok(inode, attr);
2944 if (error)
2945 return error;
2946
2947 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2948 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2949 handle_t *handle;
2950
2951 /* (user+group)*(old+new) structure, inode write (sb,
2952 * inode block, ? - but truncate inode update has it) */
Mingming Cao617ba132006-10-11 01:20:53 -07002953 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
2954 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002955 if (IS_ERR(handle)) {
2956 error = PTR_ERR(handle);
2957 goto err_out;
2958 }
2959 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2960 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07002961 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002962 return error;
2963 }
2964 /* Update corresponding info in inode so that everything is in
2965 * one transaction */
2966 if (attr->ia_valid & ATTR_UID)
2967 inode->i_uid = attr->ia_uid;
2968 if (attr->ia_valid & ATTR_GID)
2969 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07002970 error = ext4_mark_inode_dirty(handle, inode);
2971 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002972 }
2973
2974 if (S_ISREG(inode->i_mode) &&
2975 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2976 handle_t *handle;
2977
Mingming Cao617ba132006-10-11 01:20:53 -07002978 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002979 if (IS_ERR(handle)) {
2980 error = PTR_ERR(handle);
2981 goto err_out;
2982 }
2983
Mingming Cao617ba132006-10-11 01:20:53 -07002984 error = ext4_orphan_add(handle, inode);
2985 EXT4_I(inode)->i_disksize = attr->ia_size;
2986 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002987 if (!error)
2988 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07002989 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002990 }
2991
2992 rc = inode_setattr(inode, attr);
2993
Mingming Cao617ba132006-10-11 01:20:53 -07002994 /* If inode_setattr's call to ext4_truncate failed to get a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002995 * transaction handle at all, we need to clean up the in-core
2996 * orphan list manually. */
2997 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002998 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002999
3000 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07003001 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003002
3003err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07003004 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003005 if (!error)
3006 error = rc;
3007 return error;
3008}
3009
3010
3011/*
3012 * How many blocks doth make a writepage()?
3013 *
3014 * With N blocks per page, it may be:
3015 * N data blocks
3016 * 2 indirect block
3017 * 2 dindirect
3018 * 1 tindirect
3019 * N+5 bitmap blocks (from the above)
3020 * N+5 group descriptor summary blocks
3021 * 1 inode block
3022 * 1 superblock.
Mingming Cao617ba132006-10-11 01:20:53 -07003023 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003024 *
Mingming Cao617ba132006-10-11 01:20:53 -07003025 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003026 *
3027 * With ordered or writeback data it's the same, less the N data blocks.
3028 *
3029 * If the inode's direct blocks can hold an integral number of pages then a
3030 * page cannot straddle two indirect blocks, and we can only touch one indirect
3031 * and dindirect block, and the "5" above becomes "3".
3032 *
3033 * This still overestimates under most circumstances. If we were to pass the
3034 * start and end offsets in here as well we could do block_to_path() on each
3035 * block and work out the exact number of indirects which are touched. Pah.
3036 */
3037
Alex Tomasa86c6182006-10-11 01:21:03 -07003038int ext4_writepage_trans_blocks(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003039{
Mingming Cao617ba132006-10-11 01:20:53 -07003040 int bpp = ext4_journal_blocks_per_page(inode);
3041 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003042 int ret;
3043
Alex Tomasa86c6182006-10-11 01:21:03 -07003044 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3045 return ext4_ext_writepage_trans_blocks(inode, bpp);
3046
Mingming Cao617ba132006-10-11 01:20:53 -07003047 if (ext4_should_journal_data(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003048 ret = 3 * (bpp + indirects) + 2;
3049 else
3050 ret = 2 * (bpp + indirects) + 2;
3051
3052#ifdef CONFIG_QUOTA
3053 /* We know that structure was already allocated during DQUOT_INIT so
3054 * we will be updating only the data blocks + inodes */
Mingming Cao617ba132006-10-11 01:20:53 -07003055 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003056#endif
3057
3058 return ret;
3059}
3060
3061/*
Mingming Cao617ba132006-10-11 01:20:53 -07003062 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003063 * Give this, we know that the caller already has write access to iloc->bh.
3064 */
Mingming Cao617ba132006-10-11 01:20:53 -07003065int ext4_mark_iloc_dirty(handle_t *handle,
3066 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003067{
3068 int err = 0;
3069
3070 /* the do_update_inode consumes one bh->b_count */
3071 get_bh(iloc->bh);
3072
Mingming Caodab291a2006-10-11 01:21:01 -07003073 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Mingming Cao617ba132006-10-11 01:20:53 -07003074 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003075 put_bh(iloc->bh);
3076 return err;
3077}
3078
3079/*
3080 * On success, We end up with an outstanding reference count against
3081 * iloc->bh. This _must_ be cleaned up later.
3082 */
3083
3084int
Mingming Cao617ba132006-10-11 01:20:53 -07003085ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3086 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003087{
3088 int err = 0;
3089 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07003090 err = ext4_get_inode_loc(inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003091 if (!err) {
3092 BUFFER_TRACE(iloc->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07003093 err = ext4_journal_get_write_access(handle, iloc->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003094 if (err) {
3095 brelse(iloc->bh);
3096 iloc->bh = NULL;
3097 }
3098 }
3099 }
Mingming Cao617ba132006-10-11 01:20:53 -07003100 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003101 return err;
3102}
3103
3104/*
3105 * What we do here is to mark the in-core inode as clean with respect to inode
3106 * dirtiness (it may still be data-dirty).
3107 * This means that the in-core inode may be reaped by prune_icache
3108 * without having to perform any I/O. This is a very good thing,
3109 * because *any* task may call prune_icache - even ones which
3110 * have a transaction open against a different journal.
3111 *
3112 * Is this cheating? Not really. Sure, we haven't written the
3113 * inode out, but prune_icache isn't a user-visible syncing function.
3114 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3115 * we start and wait on commits.
3116 *
3117 * Is this efficient/effective? Well, we're being nice to the system
3118 * by cleaning up our inodes proactively so they can be reaped
3119 * without I/O. But we are potentially leaving up to five seconds'
3120 * worth of inodes floating about which prune_icache wants us to
3121 * write out. One way to fix that would be to get prune_icache()
3122 * to do a write_super() to free up some memory. It has the desired
3123 * effect.
3124 */
Mingming Cao617ba132006-10-11 01:20:53 -07003125int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003126{
Mingming Cao617ba132006-10-11 01:20:53 -07003127 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003128 int err;
3129
3130 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07003131 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003132 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07003133 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003134 return err;
3135}
3136
3137/*
Mingming Cao617ba132006-10-11 01:20:53 -07003138 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003139 *
3140 * We're really interested in the case where a file is being extended.
3141 * i_size has been changed by generic_commit_write() and we thus need
3142 * to include the updated inode in the current transaction.
3143 *
3144 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3145 * are allocated to the file.
3146 *
3147 * If the inode is marked synchronous, we don't honour that here - doing
3148 * so would cause a commit on atime updates, which we don't bother doing.
3149 * We handle synchronous inodes at the highest possible level.
3150 */
Mingming Cao617ba132006-10-11 01:20:53 -07003151void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003152{
Mingming Cao617ba132006-10-11 01:20:53 -07003153 handle_t *current_handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003154 handle_t *handle;
3155
Mingming Cao617ba132006-10-11 01:20:53 -07003156 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003157 if (IS_ERR(handle))
3158 goto out;
3159 if (current_handle &&
3160 current_handle->h_transaction != handle->h_transaction) {
3161 /* This task has a transaction open against a different fs */
3162 printk(KERN_EMERG "%s: transactions do not match!\n",
3163 __FUNCTION__);
3164 } else {
3165 jbd_debug(5, "marking dirty. outer handle=%p\n",
3166 current_handle);
Mingming Cao617ba132006-10-11 01:20:53 -07003167 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003168 }
Mingming Cao617ba132006-10-11 01:20:53 -07003169 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003170out:
3171 return;
3172}
3173
3174#if 0
3175/*
3176 * Bind an inode's backing buffer_head into this transaction, to prevent
3177 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07003178 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003179 * returns no iloc structure, so the caller needs to repeat the iloc
3180 * lookup to mark the inode dirty later.
3181 */
Mingming Cao617ba132006-10-11 01:20:53 -07003182static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003183{
Mingming Cao617ba132006-10-11 01:20:53 -07003184 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003185
3186 int err = 0;
3187 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07003188 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003189 if (!err) {
3190 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07003191 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003192 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07003193 err = ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003194 iloc.bh);
3195 brelse(iloc.bh);
3196 }
3197 }
Mingming Cao617ba132006-10-11 01:20:53 -07003198 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003199 return err;
3200}
3201#endif
3202
Mingming Cao617ba132006-10-11 01:20:53 -07003203int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003204{
3205 journal_t *journal;
3206 handle_t *handle;
3207 int err;
3208
3209 /*
3210 * We have to be very careful here: changing a data block's
3211 * journaling status dynamically is dangerous. If we write a
3212 * data block to the journal, change the status and then delete
3213 * that block, we risk forgetting to revoke the old log record
3214 * from the journal and so a subsequent replay can corrupt data.
3215 * So, first we make sure that the journal is empty and that
3216 * nobody is changing anything.
3217 */
3218
Mingming Cao617ba132006-10-11 01:20:53 -07003219 journal = EXT4_JOURNAL(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003220 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3221 return -EROFS;
3222
Mingming Caodab291a2006-10-11 01:21:01 -07003223 jbd2_journal_lock_updates(journal);
3224 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003225
3226 /*
3227 * OK, there are no updates running now, and all cached data is
3228 * synced to disk. We are now in a completely consistent state
3229 * which doesn't have anything in the journal, and we know that
3230 * no filesystem updates are running, so it is safe to modify
3231 * the inode's in-core data-journaling state flag now.
3232 */
3233
3234 if (val)
Mingming Cao617ba132006-10-11 01:20:53 -07003235 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003236 else
Mingming Cao617ba132006-10-11 01:20:53 -07003237 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3238 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003239
Mingming Caodab291a2006-10-11 01:21:01 -07003240 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003241
3242 /* Finally we can mark the inode as dirty. */
3243
Mingming Cao617ba132006-10-11 01:20:53 -07003244 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003245 if (IS_ERR(handle))
3246 return PTR_ERR(handle);
3247
Mingming Cao617ba132006-10-11 01:20:53 -07003248 err = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003249 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07003250 ext4_journal_stop(handle);
3251 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003252
3253 return err;
3254}