blob: 5489703d95738ffb075ed7cb275aab94eb4d74e6 [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
Kalpak Shahef7f3832007-07-18 09:15:20 -0400729 inode->i_ctime = ext4_current_time(inode);
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);
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -04001030 J_ASSERT(handle != NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001031
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
Nick Pigginbfc1af62007-10-16 01:25:05 -07001149static int ext4_write_begin(struct file *file, struct address_space *mapping,
1150 loff_t pos, unsigned len, unsigned flags,
1151 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001152{
Nick Pigginbfc1af62007-10-16 01:25:05 -07001153 struct inode *inode = mapping->host;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001154 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001155 handle_t *handle;
1156 int retries = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001157 struct page *page;
1158 pgoff_t index;
1159 unsigned from, to;
1160
1161 index = pos >> PAGE_CACHE_SHIFT;
1162 from = pos & (PAGE_CACHE_SIZE - 1);
1163 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164
1165retry:
Nick Pigginbfc1af62007-10-16 01:25:05 -07001166 page = __grab_cache_page(mapping, index);
1167 if (!page)
1168 return -ENOMEM;
1169 *pagep = page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001170
Nick Pigginbfc1af62007-10-16 01:25:05 -07001171 handle = ext4_journal_start(inode, needed_blocks);
1172 if (IS_ERR(handle)) {
1173 unlock_page(page);
1174 page_cache_release(page);
1175 ret = PTR_ERR(handle);
1176 goto out;
1177 }
1178
1179 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1180 ext4_get_block);
1181
1182 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001183 ret = walk_page_buffers(handle, page_buffers(page),
1184 from, to, NULL, do_journal_get_write_access);
1185 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001186
1187 if (ret) {
Andrew Morton7479d2b2007-04-01 23:49:44 -07001188 ext4_journal_stop(handle);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001189 unlock_page(page);
1190 page_cache_release(page);
1191 }
1192
Mingming Cao617ba132006-10-11 01:20:53 -07001193 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001194 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001195out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001196 return ret;
1197}
1198
Mingming Cao617ba132006-10-11 01:20:53 -07001199int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001200{
Mingming Caodab291a2006-10-11 01:21:01 -07001201 int err = jbd2_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001202 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001203 ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001204 bh, handle, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001205 return err;
1206}
1207
Nick Pigginbfc1af62007-10-16 01:25:05 -07001208/* For write_end() in data=journal mode */
1209static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001210{
1211 if (!buffer_mapped(bh) || buffer_freed(bh))
1212 return 0;
1213 set_buffer_uptodate(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001214 return ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001215}
1216
1217/*
Nick Pigginbfc1af62007-10-16 01:25:05 -07001218 * Generic write_end handler for ordered and writeback ext4 journal modes.
1219 * We can't use generic_write_end, because that unlocks the page and we need to
1220 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1221 * after block_write_end.
1222 */
1223static int ext4_generic_write_end(struct file *file,
1224 struct address_space *mapping,
1225 loff_t pos, unsigned len, unsigned copied,
1226 struct page *page, void *fsdata)
1227{
1228 struct inode *inode = file->f_mapping->host;
1229
1230 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1231
1232 if (pos+copied > inode->i_size) {
1233 i_size_write(inode, pos+copied);
1234 mark_inode_dirty(inode);
1235 }
1236
1237 return copied;
1238}
1239
1240/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001241 * We need to pick up the new inode size which generic_commit_write gave us
1242 * `file' can be NULL - eg, when called from page_symlink().
1243 *
Mingming Cao617ba132006-10-11 01:20:53 -07001244 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001245 * buffers are managed internally.
1246 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001247static int ext4_ordered_write_end(struct file *file,
1248 struct address_space *mapping,
1249 loff_t pos, unsigned len, unsigned copied,
1250 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001251{
Mingming Cao617ba132006-10-11 01:20:53 -07001252 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001253 struct inode *inode = file->f_mapping->host;
1254 unsigned from, to;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001255 int ret = 0, ret2;
1256
Nick Pigginbfc1af62007-10-16 01:25:05 -07001257 from = pos & (PAGE_CACHE_SIZE - 1);
1258 to = from + len;
1259
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001260 ret = walk_page_buffers(handle, page_buffers(page),
Mingming Cao617ba132006-10-11 01:20:53 -07001261 from, to, NULL, ext4_journal_dirty_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001262
1263 if (ret == 0) {
1264 /*
Nick Pigginbfc1af62007-10-16 01:25:05 -07001265 * generic_write_end() will run mark_inode_dirty() if i_size
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001266 * changes. So let's piggyback the i_disksize mark_inode_dirty
1267 * into that.
1268 */
1269 loff_t new_i_size;
1270
Nick Pigginbfc1af62007-10-16 01:25:05 -07001271 new_i_size = pos + copied;
Mingming Cao617ba132006-10-11 01:20:53 -07001272 if (new_i_size > EXT4_I(inode)->i_disksize)
1273 EXT4_I(inode)->i_disksize = new_i_size;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001274 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1275 page, fsdata);
1276 if (copied < 0)
1277 ret = copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001278 }
Mingming Cao617ba132006-10-11 01:20:53 -07001279 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001280 if (!ret)
1281 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001282 unlock_page(page);
1283 page_cache_release(page);
1284
1285 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001286}
1287
Nick Pigginbfc1af62007-10-16 01:25:05 -07001288static int ext4_writeback_write_end(struct file *file,
1289 struct address_space *mapping,
1290 loff_t pos, unsigned len, unsigned copied,
1291 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001292{
Mingming Cao617ba132006-10-11 01:20:53 -07001293 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001294 struct inode *inode = file->f_mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001295 int ret = 0, ret2;
1296 loff_t new_i_size;
1297
Nick Pigginbfc1af62007-10-16 01:25:05 -07001298 new_i_size = pos + copied;
Mingming Cao617ba132006-10-11 01:20:53 -07001299 if (new_i_size > EXT4_I(inode)->i_disksize)
1300 EXT4_I(inode)->i_disksize = new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001301
Nick Pigginbfc1af62007-10-16 01:25:05 -07001302 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1303 page, fsdata);
1304 if (copied < 0)
1305 ret = copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001306
Mingming Cao617ba132006-10-11 01:20:53 -07001307 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001308 if (!ret)
1309 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001310 unlock_page(page);
1311 page_cache_release(page);
1312
1313 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001314}
1315
Nick Pigginbfc1af62007-10-16 01:25:05 -07001316static int ext4_journalled_write_end(struct file *file,
1317 struct address_space *mapping,
1318 loff_t pos, unsigned len, unsigned copied,
1319 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001320{
Mingming Cao617ba132006-10-11 01:20:53 -07001321 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001322 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001323 int ret = 0, ret2;
1324 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001325 unsigned from, to;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001326
Nick Pigginbfc1af62007-10-16 01:25:05 -07001327 from = pos & (PAGE_CACHE_SIZE - 1);
1328 to = from + len;
1329
1330 if (copied < len) {
1331 if (!PageUptodate(page))
1332 copied = 0;
1333 page_zero_new_buffers(page, from+copied, to);
1334 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001335
1336 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001337 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001338 if (!partial)
1339 SetPageUptodate(page);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001340 if (pos+copied > inode->i_size)
1341 i_size_write(inode, pos+copied);
Mingming Cao617ba132006-10-11 01:20:53 -07001342 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1343 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1344 EXT4_I(inode)->i_disksize = inode->i_size;
1345 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001346 if (!ret)
1347 ret = ret2;
1348 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001349
Mingming Cao617ba132006-10-11 01:20:53 -07001350 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001351 if (!ret)
1352 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001353 unlock_page(page);
1354 page_cache_release(page);
1355
1356 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001357}
1358
1359/*
1360 * bmap() is special. It gets used by applications such as lilo and by
1361 * the swapper to find the on-disk block of a specific piece of data.
1362 *
1363 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07001364 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001365 * filesystem and enables swap, then they may get a nasty shock when the
1366 * data getting swapped to that swapfile suddenly gets overwritten by
1367 * the original zero's written out previously to the journal and
1368 * awaiting writeback in the kernel's buffer cache.
1369 *
1370 * So, if we see any bmap calls here on a modified, data-journaled file,
1371 * take extra steps to flush any blocks which might be in the cache.
1372 */
Mingming Cao617ba132006-10-11 01:20:53 -07001373static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001374{
1375 struct inode *inode = mapping->host;
1376 journal_t *journal;
1377 int err;
1378
Mingming Cao617ba132006-10-11 01:20:53 -07001379 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001380 /*
1381 * This is a REALLY heavyweight approach, but the use of
1382 * bmap on dirty files is expected to be extremely rare:
1383 * only if we run lilo or swapon on a freshly made file
1384 * do we expect this to happen.
1385 *
1386 * (bmap requires CAP_SYS_RAWIO so this does not
1387 * represent an unprivileged user DOS attack --- we'd be
1388 * in trouble if mortal users could trigger this path at
1389 * will.)
1390 *
Mingming Cao617ba132006-10-11 01:20:53 -07001391 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001392 * regular files. If somebody wants to bmap a directory
1393 * or symlink and gets confused because the buffer
1394 * hasn't yet been flushed to disk, they deserve
1395 * everything they get.
1396 */
1397
Mingming Cao617ba132006-10-11 01:20:53 -07001398 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1399 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07001400 jbd2_journal_lock_updates(journal);
1401 err = jbd2_journal_flush(journal);
1402 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001403
1404 if (err)
1405 return 0;
1406 }
1407
Mingming Cao617ba132006-10-11 01:20:53 -07001408 return generic_block_bmap(mapping,block,ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001409}
1410
1411static int bget_one(handle_t *handle, struct buffer_head *bh)
1412{
1413 get_bh(bh);
1414 return 0;
1415}
1416
1417static int bput_one(handle_t *handle, struct buffer_head *bh)
1418{
1419 put_bh(bh);
1420 return 0;
1421}
1422
Mingming Caodab291a2006-10-11 01:21:01 -07001423static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001424{
1425 if (buffer_mapped(bh))
Mingming Cao617ba132006-10-11 01:20:53 -07001426 return ext4_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001427 return 0;
1428}
1429
1430/*
1431 * Note that we always start a transaction even if we're not journalling
1432 * data. This is to preserve ordering: any hole instantiation within
Mingming Cao617ba132006-10-11 01:20:53 -07001433 * __block_write_full_page -> ext4_get_block() should be journalled
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001434 * along with the data so we don't crash and then get metadata which
1435 * refers to old data.
1436 *
1437 * In all journalling modes block_write_full_page() will start the I/O.
1438 *
1439 * Problem:
1440 *
Mingming Cao617ba132006-10-11 01:20:53 -07001441 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1442 * ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001443 *
1444 * Similar for:
1445 *
Mingming Cao617ba132006-10-11 01:20:53 -07001446 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001447 *
Mingming Cao617ba132006-10-11 01:20:53 -07001448 * Same applies to ext4_get_block(). We will deadlock on various things like
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001449 * lock_journal and i_truncate_mutex.
1450 *
1451 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1452 * allocations fail.
1453 *
1454 * 16May01: If we're reentered then journal_current_handle() will be
1455 * non-zero. We simply *return*.
1456 *
1457 * 1 July 2001: @@@ FIXME:
1458 * In journalled data mode, a data buffer may be metadata against the
1459 * current transaction. But the same file is part of a shared mapping
1460 * and someone does a writepage() on it.
1461 *
1462 * We will move the buffer onto the async_data list, but *after* it has
1463 * been dirtied. So there's a small window where we have dirty data on
1464 * BJ_Metadata.
1465 *
1466 * Note that this only applies to the last partial page in the file. The
1467 * bit which block_write_full_page() uses prepare/commit for. (That's
1468 * broken code anyway: it's wrong for msync()).
1469 *
1470 * It's a rare case: affects the final partial page, for journalled data
1471 * where the file is subject to bith write() and writepage() in the same
1472 * transction. To fix it we'll need a custom block_write_full_page().
1473 * We'll probably need that anyway for journalling writepage() output.
1474 *
1475 * We don't honour synchronous mounts for writepage(). That would be
1476 * disastrous. Any write() or metadata operation will sync the fs for
1477 * us.
1478 *
1479 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1480 * we don't need to open a transaction here.
1481 */
Mingming Cao617ba132006-10-11 01:20:53 -07001482static int ext4_ordered_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 struct buffer_head *page_bufs;
1487 handle_t *handle = NULL;
1488 int ret = 0;
1489 int err;
1490
1491 J_ASSERT(PageLocked(page));
1492
1493 /*
1494 * We give up here if we're reentered, because it might be for a
1495 * different filesystem.
1496 */
Mingming Cao617ba132006-10-11 01:20:53 -07001497 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001498 goto out_fail;
1499
Mingming Cao617ba132006-10-11 01:20:53 -07001500 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501
1502 if (IS_ERR(handle)) {
1503 ret = PTR_ERR(handle);
1504 goto out_fail;
1505 }
1506
1507 if (!page_has_buffers(page)) {
1508 create_empty_buffers(page, inode->i_sb->s_blocksize,
1509 (1 << BH_Dirty)|(1 << BH_Uptodate));
1510 }
1511 page_bufs = page_buffers(page);
1512 walk_page_buffers(handle, page_bufs, 0,
1513 PAGE_CACHE_SIZE, NULL, bget_one);
1514
Mingming Cao617ba132006-10-11 01:20:53 -07001515 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001516
1517 /*
1518 * The page can become unlocked at any point now, and
1519 * truncate can then come in and change things. So we
1520 * can't touch *page from now on. But *page_bufs is
1521 * safe due to elevated refcount.
1522 */
1523
1524 /*
1525 * And attach them to the current transaction. But only if
1526 * block_write_full_page() succeeded. Otherwise they are unmapped,
1527 * and generally junk.
1528 */
1529 if (ret == 0) {
1530 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
Mingming Caodab291a2006-10-11 01:21:01 -07001531 NULL, jbd2_journal_dirty_data_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001532 if (!ret)
1533 ret = err;
1534 }
1535 walk_page_buffers(handle, page_bufs, 0,
1536 PAGE_CACHE_SIZE, NULL, bput_one);
Mingming Cao617ba132006-10-11 01:20:53 -07001537 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001538 if (!ret)
1539 ret = err;
1540 return ret;
1541
1542out_fail:
1543 redirty_page_for_writepage(wbc, page);
1544 unlock_page(page);
1545 return ret;
1546}
1547
Mingming Cao617ba132006-10-11 01:20:53 -07001548static int ext4_writeback_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001549 struct writeback_control *wbc)
1550{
1551 struct inode *inode = page->mapping->host;
1552 handle_t *handle = NULL;
1553 int ret = 0;
1554 int err;
1555
Mingming Cao617ba132006-10-11 01:20:53 -07001556 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001557 goto out_fail;
1558
Mingming Cao617ba132006-10-11 01:20:53 -07001559 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 if (IS_ERR(handle)) {
1561 ret = PTR_ERR(handle);
1562 goto out_fail;
1563 }
1564
Mingming Cao617ba132006-10-11 01:20:53 -07001565 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1566 ret = nobh_writepage(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001567 else
Mingming Cao617ba132006-10-11 01:20:53 -07001568 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001569
Mingming Cao617ba132006-10-11 01:20:53 -07001570 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001571 if (!ret)
1572 ret = err;
1573 return ret;
1574
1575out_fail:
1576 redirty_page_for_writepage(wbc, page);
1577 unlock_page(page);
1578 return ret;
1579}
1580
Mingming Cao617ba132006-10-11 01:20:53 -07001581static int ext4_journalled_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001582 struct writeback_control *wbc)
1583{
1584 struct inode *inode = page->mapping->host;
1585 handle_t *handle = NULL;
1586 int ret = 0;
1587 int err;
1588
Mingming Cao617ba132006-10-11 01:20:53 -07001589 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001590 goto no_write;
1591
Mingming Cao617ba132006-10-11 01:20:53 -07001592 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001593 if (IS_ERR(handle)) {
1594 ret = PTR_ERR(handle);
1595 goto no_write;
1596 }
1597
1598 if (!page_has_buffers(page) || PageChecked(page)) {
1599 /*
1600 * It's mmapped pagecache. Add buffers and journal it. There
1601 * doesn't seem much point in redirtying the page here.
1602 */
1603 ClearPageChecked(page);
1604 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
Mingming Cao617ba132006-10-11 01:20:53 -07001605 ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001606 if (ret != 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001607 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001608 goto out_unlock;
1609 }
1610 ret = walk_page_buffers(handle, page_buffers(page), 0,
1611 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1612
1613 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001614 PAGE_CACHE_SIZE, NULL, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615 if (ret == 0)
1616 ret = err;
Mingming Cao617ba132006-10-11 01:20:53 -07001617 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001618 unlock_page(page);
1619 } else {
1620 /*
1621 * It may be a page full of checkpoint-mode buffers. We don't
1622 * really know unless we go poke around in the buffer_heads.
1623 * But block_write_full_page will do the right thing.
1624 */
Mingming Cao617ba132006-10-11 01:20:53 -07001625 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001626 }
Mingming Cao617ba132006-10-11 01:20:53 -07001627 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628 if (!ret)
1629 ret = err;
1630out:
1631 return ret;
1632
1633no_write:
1634 redirty_page_for_writepage(wbc, page);
1635out_unlock:
1636 unlock_page(page);
1637 goto out;
1638}
1639
Mingming Cao617ba132006-10-11 01:20:53 -07001640static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001641{
Mingming Cao617ba132006-10-11 01:20:53 -07001642 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001643}
1644
1645static int
Mingming Cao617ba132006-10-11 01:20:53 -07001646ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001647 struct list_head *pages, unsigned nr_pages)
1648{
Mingming Cao617ba132006-10-11 01:20:53 -07001649 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001650}
1651
Mingming Cao617ba132006-10-11 01:20:53 -07001652static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001653{
Mingming Cao617ba132006-10-11 01:20:53 -07001654 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001655
1656 /*
1657 * If it's a full truncate we just forget about the pending dirtying
1658 */
1659 if (offset == 0)
1660 ClearPageChecked(page);
1661
Mingming Caodab291a2006-10-11 01:21:01 -07001662 jbd2_journal_invalidatepage(journal, page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001663}
1664
Mingming Cao617ba132006-10-11 01:20:53 -07001665static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001666{
Mingming Cao617ba132006-10-11 01:20:53 -07001667 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001668
1669 WARN_ON(PageChecked(page));
1670 if (!page_has_buffers(page))
1671 return 0;
Mingming Caodab291a2006-10-11 01:21:01 -07001672 return jbd2_journal_try_to_free_buffers(journal, page, wait);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001673}
1674
1675/*
1676 * If the O_DIRECT write will extend the file then add this inode to the
1677 * orphan list. So recovery will truncate it back to the original size
1678 * if the machine crashes during the write.
1679 *
1680 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1681 * crashes then stale disk data _may_ be exposed inside the file.
1682 */
Mingming Cao617ba132006-10-11 01:20:53 -07001683static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001684 const struct iovec *iov, loff_t offset,
1685 unsigned long nr_segs)
1686{
1687 struct file *file = iocb->ki_filp;
1688 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07001689 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001690 handle_t *handle = NULL;
1691 ssize_t ret;
1692 int orphan = 0;
1693 size_t count = iov_length(iov, nr_segs);
1694
1695 if (rw == WRITE) {
1696 loff_t final_size = offset + count;
1697
Mingming Cao617ba132006-10-11 01:20:53 -07001698 handle = ext4_journal_start(inode, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001699 if (IS_ERR(handle)) {
1700 ret = PTR_ERR(handle);
1701 goto out;
1702 }
1703 if (final_size > inode->i_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001704 ret = ext4_orphan_add(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001705 if (ret)
1706 goto out_stop;
1707 orphan = 1;
1708 ei->i_disksize = inode->i_size;
1709 }
1710 }
1711
1712 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1713 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07001714 ext4_get_block, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001715
1716 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001717 * Reacquire the handle: ext4_get_block() can restart the transaction
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001718 */
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001719 handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001720
1721out_stop:
1722 if (handle) {
1723 int err;
1724
1725 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07001726 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001727 if (orphan && ret > 0) {
1728 loff_t end = offset + ret;
1729 if (end > inode->i_size) {
1730 ei->i_disksize = end;
1731 i_size_write(inode, end);
1732 /*
1733 * We're going to return a positive `ret'
1734 * here due to non-zero-length I/O, so there's
1735 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07001736 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001737 * ignore it.
1738 */
Mingming Cao617ba132006-10-11 01:20:53 -07001739 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001740 }
1741 }
Mingming Cao617ba132006-10-11 01:20:53 -07001742 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001743 if (ret == 0)
1744 ret = err;
1745 }
1746out:
1747 return ret;
1748}
1749
1750/*
Mingming Cao617ba132006-10-11 01:20:53 -07001751 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1753 * much here because ->set_page_dirty is called under VFS locks. The page is
1754 * not necessarily locked.
1755 *
1756 * We cannot just dirty the page and leave attached buffers clean, because the
1757 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1758 * or jbddirty because all the journalling code will explode.
1759 *
1760 * So what we do is to mark the page "pending dirty" and next time writepage
1761 * is called, propagate that into the buffers appropriately.
1762 */
Mingming Cao617ba132006-10-11 01:20:53 -07001763static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764{
1765 SetPageChecked(page);
1766 return __set_page_dirty_nobuffers(page);
1767}
1768
Mingming Cao617ba132006-10-11 01:20:53 -07001769static const struct address_space_operations ext4_ordered_aops = {
1770 .readpage = ext4_readpage,
1771 .readpages = ext4_readpages,
1772 .writepage = ext4_ordered_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001773 .sync_page = block_sync_page,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001774 .write_begin = ext4_write_begin,
1775 .write_end = ext4_ordered_write_end,
Mingming Cao617ba132006-10-11 01:20:53 -07001776 .bmap = ext4_bmap,
1777 .invalidatepage = ext4_invalidatepage,
1778 .releasepage = ext4_releasepage,
1779 .direct_IO = ext4_direct_IO,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001780 .migratepage = buffer_migrate_page,
1781};
1782
Mingming Cao617ba132006-10-11 01:20:53 -07001783static const struct address_space_operations ext4_writeback_aops = {
1784 .readpage = ext4_readpage,
1785 .readpages = ext4_readpages,
1786 .writepage = ext4_writeback_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001787 .sync_page = block_sync_page,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001788 .write_begin = ext4_write_begin,
1789 .write_end = ext4_writeback_write_end,
Mingming Cao617ba132006-10-11 01:20:53 -07001790 .bmap = ext4_bmap,
1791 .invalidatepage = ext4_invalidatepage,
1792 .releasepage = ext4_releasepage,
1793 .direct_IO = ext4_direct_IO,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001794 .migratepage = buffer_migrate_page,
1795};
1796
Mingming Cao617ba132006-10-11 01:20:53 -07001797static const struct address_space_operations ext4_journalled_aops = {
1798 .readpage = ext4_readpage,
1799 .readpages = ext4_readpages,
1800 .writepage = ext4_journalled_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001801 .sync_page = block_sync_page,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001802 .write_begin = ext4_write_begin,
1803 .write_end = ext4_journalled_write_end,
Mingming Cao617ba132006-10-11 01:20:53 -07001804 .set_page_dirty = ext4_journalled_set_page_dirty,
1805 .bmap = ext4_bmap,
1806 .invalidatepage = ext4_invalidatepage,
1807 .releasepage = ext4_releasepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001808};
1809
Mingming Cao617ba132006-10-11 01:20:53 -07001810void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001811{
Mingming Cao617ba132006-10-11 01:20:53 -07001812 if (ext4_should_order_data(inode))
1813 inode->i_mapping->a_ops = &ext4_ordered_aops;
1814 else if (ext4_should_writeback_data(inode))
1815 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001816 else
Mingming Cao617ba132006-10-11 01:20:53 -07001817 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001818}
1819
1820/*
Mingming Cao617ba132006-10-11 01:20:53 -07001821 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001822 * up to the end of the block which corresponds to `from'.
1823 * This required during truncate. We need to physically zero the tail end
1824 * of that block so it doesn't yield old data if the file is later grown.
1825 */
Alex Tomasa86c6182006-10-11 01:21:03 -07001826int ext4_block_truncate_page(handle_t *handle, struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827 struct address_space *mapping, loff_t from)
1828{
Mingming Cao617ba132006-10-11 01:20:53 -07001829 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001830 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1831 unsigned blocksize, iblock, length, pos;
1832 struct inode *inode = mapping->host;
1833 struct buffer_head *bh;
1834 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001835
1836 blocksize = inode->i_sb->s_blocksize;
1837 length = blocksize - (offset & (blocksize - 1));
1838 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1839
1840 /*
1841 * For "nobh" option, we can only work if we don't need to
1842 * read-in the page - otherwise we create buffers to do the IO.
1843 */
1844 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
Mingming Cao617ba132006-10-11 01:20:53 -07001845 ext4_should_writeback_data(inode) && PageUptodate(page)) {
Eric Sandeenfc0e15a2007-07-18 09:20:44 -04001846 zero_user_page(page, offset, length, KM_USER0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001847 set_page_dirty(page);
1848 goto unlock;
1849 }
1850
1851 if (!page_has_buffers(page))
1852 create_empty_buffers(page, blocksize, 0);
1853
1854 /* Find the buffer that contains "offset" */
1855 bh = page_buffers(page);
1856 pos = blocksize;
1857 while (offset >= pos) {
1858 bh = bh->b_this_page;
1859 iblock++;
1860 pos += blocksize;
1861 }
1862
1863 err = 0;
1864 if (buffer_freed(bh)) {
1865 BUFFER_TRACE(bh, "freed: skip");
1866 goto unlock;
1867 }
1868
1869 if (!buffer_mapped(bh)) {
1870 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07001871 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001872 /* unmapped? It's a hole - nothing to do */
1873 if (!buffer_mapped(bh)) {
1874 BUFFER_TRACE(bh, "still unmapped");
1875 goto unlock;
1876 }
1877 }
1878
1879 /* Ok, it's mapped. Make sure it's up-to-date */
1880 if (PageUptodate(page))
1881 set_buffer_uptodate(bh);
1882
1883 if (!buffer_uptodate(bh)) {
1884 err = -EIO;
1885 ll_rw_block(READ, 1, &bh);
1886 wait_on_buffer(bh);
1887 /* Uhhuh. Read error. Complain and punt. */
1888 if (!buffer_uptodate(bh))
1889 goto unlock;
1890 }
1891
Mingming Cao617ba132006-10-11 01:20:53 -07001892 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001893 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07001894 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001895 if (err)
1896 goto unlock;
1897 }
1898
Eric Sandeenfc0e15a2007-07-18 09:20:44 -04001899 zero_user_page(page, offset, length, KM_USER0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001900
1901 BUFFER_TRACE(bh, "zeroed end of block");
1902
1903 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001904 if (ext4_should_journal_data(inode)) {
1905 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001906 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07001907 if (ext4_should_order_data(inode))
1908 err = ext4_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001909 mark_buffer_dirty(bh);
1910 }
1911
1912unlock:
1913 unlock_page(page);
1914 page_cache_release(page);
1915 return err;
1916}
1917
1918/*
1919 * Probably it should be a library function... search for first non-zero word
1920 * or memcmp with zero_page, whatever is better for particular architecture.
1921 * Linus?
1922 */
1923static inline int all_zeroes(__le32 *p, __le32 *q)
1924{
1925 while (p < q)
1926 if (*p++)
1927 return 0;
1928 return 1;
1929}
1930
1931/**
Mingming Cao617ba132006-10-11 01:20:53 -07001932 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001933 * @inode: inode in question
1934 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07001935 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001936 * @chain: place to store the pointers to partial indirect blocks
1937 * @top: place to the (detached) top of branch
1938 *
Mingming Cao617ba132006-10-11 01:20:53 -07001939 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001940 *
1941 * When we do truncate() we may have to clean the ends of several
1942 * indirect blocks but leave the blocks themselves alive. Block is
1943 * partially truncated if some data below the new i_size is refered
1944 * from it (and it is on the path to the first completely truncated
1945 * data block, indeed). We have to free the top of that path along
1946 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07001947 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001948 * finishes, we may safely do the latter, but top of branch may
1949 * require special attention - pageout below the truncation point
1950 * might try to populate it.
1951 *
1952 * We atomically detach the top of branch from the tree, store the
1953 * block number of its root in *@top, pointers to buffer_heads of
1954 * partially truncated blocks - in @chain[].bh and pointers to
1955 * their last elements that should not be removed - in
1956 * @chain[].p. Return value is the pointer to last filled element
1957 * of @chain.
1958 *
1959 * The work left to caller to do the actual freeing of subtrees:
1960 * a) free the subtree starting from *@top
1961 * b) free the subtrees whose roots are stored in
1962 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1963 * c) free the subtrees growing from the inode past the @chain[0].
1964 * (no partially truncated stuff there). */
1965
Mingming Cao617ba132006-10-11 01:20:53 -07001966static Indirect *ext4_find_shared(struct inode *inode, int depth,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001967 int offsets[4], Indirect chain[4], __le32 *top)
1968{
1969 Indirect *partial, *p;
1970 int k, err;
1971
1972 *top = 0;
1973 /* Make k index the deepest non-null offest + 1 */
1974 for (k = depth; k > 1 && !offsets[k-1]; k--)
1975 ;
Mingming Cao617ba132006-10-11 01:20:53 -07001976 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001977 /* Writer: pointers */
1978 if (!partial)
1979 partial = chain + k-1;
1980 /*
1981 * If the branch acquired continuation since we've looked at it -
1982 * fine, it should all survive and (new) top doesn't belong to us.
1983 */
1984 if (!partial->key && *partial->p)
1985 /* Writer: end */
1986 goto no_top;
1987 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1988 ;
1989 /*
1990 * OK, we've found the last block that must survive. The rest of our
1991 * branch should be detached before unlocking. However, if that rest
1992 * of branch is all ours and does not grow immediately from the inode
1993 * it's easier to cheat and just decrement partial->p.
1994 */
1995 if (p == chain + k - 1 && p > chain) {
1996 p->p--;
1997 } else {
1998 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07001999 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002000#if 0
2001 *p->p = 0;
2002#endif
2003 }
2004 /* Writer: end */
2005
2006 while(partial > p) {
2007 brelse(partial->bh);
2008 partial--;
2009 }
2010no_top:
2011 return partial;
2012}
2013
2014/*
2015 * Zero a number of block pointers in either an inode or an indirect block.
2016 * If we restart the transaction we must again get write access to the
2017 * indirect block for further modification.
2018 *
2019 * We release `count' blocks on disk, but (last - first) may be greater
2020 * than `count' because there can be holes in there.
2021 */
Mingming Cao617ba132006-10-11 01:20:53 -07002022static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
2023 struct buffer_head *bh, ext4_fsblk_t block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002024 unsigned long count, __le32 *first, __le32 *last)
2025{
2026 __le32 *p;
2027 if (try_to_extend_transaction(handle, inode)) {
2028 if (bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002029 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2030 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002031 }
Mingming Cao617ba132006-10-11 01:20:53 -07002032 ext4_mark_inode_dirty(handle, inode);
2033 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002034 if (bh) {
2035 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002036 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002037 }
2038 }
2039
2040 /*
2041 * Any buffers which are on the journal will be in memory. We find
Mingming Caodab291a2006-10-11 01:21:01 -07002042 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002043 * on them. We've already detached each block from the file, so
Mingming Caodab291a2006-10-11 01:21:01 -07002044 * bforget() in jbd2_journal_forget() should be safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002045 *
Mingming Caodab291a2006-10-11 01:21:01 -07002046 * AKPM: turn on bforget in jbd2_journal_forget()!!!
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002047 */
2048 for (p = first; p < last; p++) {
2049 u32 nr = le32_to_cpu(*p);
2050 if (nr) {
2051 struct buffer_head *bh;
2052
2053 *p = 0;
2054 bh = sb_find_get_block(inode->i_sb, nr);
Mingming Cao617ba132006-10-11 01:20:53 -07002055 ext4_forget(handle, 0, inode, bh, nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002056 }
2057 }
2058
Mingming Cao617ba132006-10-11 01:20:53 -07002059 ext4_free_blocks(handle, inode, block_to_free, count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002060}
2061
2062/**
Mingming Cao617ba132006-10-11 01:20:53 -07002063 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002064 * @handle: handle for this transaction
2065 * @inode: inode we are dealing with
2066 * @this_bh: indirect buffer_head which contains *@first and *@last
2067 * @first: array of block numbers
2068 * @last: points immediately past the end of array
2069 *
2070 * We are freeing all blocks refered from that array (numbers are stored as
2071 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2072 *
2073 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2074 * blocks are contiguous then releasing them at one time will only affect one
2075 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2076 * actually use a lot of journal space.
2077 *
2078 * @this_bh will be %NULL if @first and @last point into the inode's direct
2079 * block pointers.
2080 */
Mingming Cao617ba132006-10-11 01:20:53 -07002081static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002082 struct buffer_head *this_bh,
2083 __le32 *first, __le32 *last)
2084{
Mingming Cao617ba132006-10-11 01:20:53 -07002085 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002086 unsigned long count = 0; /* Number of blocks in the run */
2087 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2088 corresponding to
2089 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07002090 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002091 __le32 *p; /* Pointer into inode/ind
2092 for current block */
2093 int err;
2094
2095 if (this_bh) { /* For indirect block */
2096 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002097 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002098 /* Important: if we can't update the indirect pointers
2099 * to the blocks, we can't free them. */
2100 if (err)
2101 return;
2102 }
2103
2104 for (p = first; p < last; p++) {
2105 nr = le32_to_cpu(*p);
2106 if (nr) {
2107 /* accumulate blocks to free if they're contiguous */
2108 if (count == 0) {
2109 block_to_free = nr;
2110 block_to_free_p = p;
2111 count = 1;
2112 } else if (nr == block_to_free + count) {
2113 count++;
2114 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002115 ext4_clear_blocks(handle, inode, this_bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002116 block_to_free,
2117 count, block_to_free_p, p);
2118 block_to_free = nr;
2119 block_to_free_p = p;
2120 count = 1;
2121 }
2122 }
2123 }
2124
2125 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07002126 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002127 count, block_to_free_p, p);
2128
2129 if (this_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002130 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2131 ext4_journal_dirty_metadata(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002132 }
2133}
2134
2135/**
Mingming Cao617ba132006-10-11 01:20:53 -07002136 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002137 * @handle: JBD handle for this transaction
2138 * @inode: inode we are dealing with
2139 * @parent_bh: the buffer_head which contains *@first and *@last
2140 * @first: array of block numbers
2141 * @last: pointer immediately past the end of array
2142 * @depth: depth of the branches to free
2143 *
2144 * We are freeing all blocks refered from these branches (numbers are
2145 * stored as little-endian 32-bit) and updating @inode->i_blocks
2146 * appropriately.
2147 */
Mingming Cao617ba132006-10-11 01:20:53 -07002148static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002149 struct buffer_head *parent_bh,
2150 __le32 *first, __le32 *last, int depth)
2151{
Mingming Cao617ba132006-10-11 01:20:53 -07002152 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002153 __le32 *p;
2154
2155 if (is_handle_aborted(handle))
2156 return;
2157
2158 if (depth--) {
2159 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002160 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002161 p = last;
2162 while (--p >= first) {
2163 nr = le32_to_cpu(*p);
2164 if (!nr)
2165 continue; /* A hole */
2166
2167 /* Go read the buffer for the next level down */
2168 bh = sb_bread(inode->i_sb, nr);
2169
2170 /*
2171 * A read failure? Report error and clear slot
2172 * (should be rare).
2173 */
2174 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002175 ext4_error(inode->i_sb, "ext4_free_branches",
Mingming Cao2ae02102006-10-11 01:21:11 -07002176 "Read failure, inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002177 inode->i_ino, nr);
2178 continue;
2179 }
2180
2181 /* This zaps the entire block. Bottom up. */
2182 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07002183 ext4_free_branches(handle, inode, bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002184 (__le32*)bh->b_data,
2185 (__le32*)bh->b_data + addr_per_block,
2186 depth);
2187
2188 /*
2189 * We've probably journalled the indirect block several
2190 * times during the truncate. But it's no longer
2191 * needed and we now drop it from the transaction via
Mingming Caodab291a2006-10-11 01:21:01 -07002192 * jbd2_journal_revoke().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002193 *
2194 * That's easy if it's exclusively part of this
2195 * transaction. But if it's part of the committing
Mingming Caodab291a2006-10-11 01:21:01 -07002196 * transaction then jbd2_journal_forget() will simply
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002197 * brelse() it. That means that if the underlying
Mingming Cao617ba132006-10-11 01:20:53 -07002198 * block is reallocated in ext4_get_block(),
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002199 * unmap_underlying_metadata() will find this block
2200 * and will try to get rid of it. damn, damn.
2201 *
2202 * If this block has already been committed to the
2203 * journal, a revoke record will be written. And
2204 * revoke records must be emitted *before* clearing
2205 * this block's bit in the bitmaps.
2206 */
Mingming Cao617ba132006-10-11 01:20:53 -07002207 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002208
2209 /*
2210 * Everything below this this pointer has been
2211 * released. Now let this top-of-subtree go.
2212 *
2213 * We want the freeing of this indirect block to be
2214 * atomic in the journal with the updating of the
2215 * bitmap block which owns it. So make some room in
2216 * the journal.
2217 *
2218 * We zero the parent pointer *after* freeing its
2219 * pointee in the bitmaps, so if extend_transaction()
2220 * for some reason fails to put the bitmap changes and
2221 * the release into the same transaction, recovery
2222 * will merely complain about releasing a free block,
2223 * rather than leaking blocks.
2224 */
2225 if (is_handle_aborted(handle))
2226 return;
2227 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002228 ext4_mark_inode_dirty(handle, inode);
2229 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002230 }
2231
Mingming Cao617ba132006-10-11 01:20:53 -07002232 ext4_free_blocks(handle, inode, nr, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002233
2234 if (parent_bh) {
2235 /*
2236 * The block which we have just freed is
2237 * pointed to by an indirect block: journal it
2238 */
2239 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002240 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002241 parent_bh)){
2242 *p = 0;
2243 BUFFER_TRACE(parent_bh,
Mingming Cao617ba132006-10-11 01:20:53 -07002244 "call ext4_journal_dirty_metadata");
2245 ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002246 parent_bh);
2247 }
2248 }
2249 }
2250 } else {
2251 /* We have reached the bottom of the tree. */
2252 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07002253 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002254 }
2255}
2256
2257/*
Mingming Cao617ba132006-10-11 01:20:53 -07002258 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002259 *
Mingming Cao617ba132006-10-11 01:20:53 -07002260 * We block out ext4_get_block() block instantiations across the entire
2261 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002262 * simultaneously on behalf of the same inode.
2263 *
2264 * As we work through the truncate and commmit bits of it to the journal there
2265 * is one core, guiding principle: the file's tree must always be consistent on
2266 * disk. We must be able to restart the truncate after a crash.
2267 *
2268 * The file's tree may be transiently inconsistent in memory (although it
2269 * probably isn't), but whenever we close off and commit a journal transaction,
2270 * the contents of (the filesystem + the journal) must be consistent and
2271 * restartable. It's pretty simple, really: bottom up, right to left (although
2272 * left-to-right works OK too).
2273 *
2274 * Note that at recovery time, journal replay occurs *before* the restart of
2275 * truncate against the orphan inode list.
2276 *
2277 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07002278 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002279 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07002280 * ext4_truncate() to have another go. So there will be instantiated blocks
2281 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002282 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07002283 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002284 */
Mingming Cao617ba132006-10-11 01:20:53 -07002285void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002286{
2287 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07002288 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002289 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07002290 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002291 struct address_space *mapping = inode->i_mapping;
2292 int offsets[4];
2293 Indirect chain[4];
2294 Indirect *partial;
2295 __le32 nr = 0;
2296 int n;
2297 long last_block;
2298 unsigned blocksize = inode->i_sb->s_blocksize;
2299 struct page *page;
2300
2301 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2302 S_ISLNK(inode->i_mode)))
2303 return;
Mingming Cao617ba132006-10-11 01:20:53 -07002304 if (ext4_inode_is_fast_symlink(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002305 return;
2306 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2307 return;
2308
2309 /*
2310 * We have to lock the EOF page here, because lock_page() nests
Mingming Caodab291a2006-10-11 01:21:01 -07002311 * outside jbd2_journal_start().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002312 */
2313 if ((inode->i_size & (blocksize - 1)) == 0) {
2314 /* Block boundary? Nothing to do */
2315 page = NULL;
2316 } else {
2317 page = grab_cache_page(mapping,
2318 inode->i_size >> PAGE_CACHE_SHIFT);
2319 if (!page)
2320 return;
2321 }
2322
Alex Tomasa86c6182006-10-11 01:21:03 -07002323 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
2324 return ext4_ext_truncate(inode, page);
2325
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002326 handle = start_transaction(inode);
2327 if (IS_ERR(handle)) {
2328 if (page) {
2329 clear_highpage(page);
2330 flush_dcache_page(page);
2331 unlock_page(page);
2332 page_cache_release(page);
2333 }
2334 return; /* AKPM: return what? */
2335 }
2336
2337 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07002338 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002339
2340 if (page)
Mingming Cao617ba132006-10-11 01:20:53 -07002341 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002342
Mingming Cao617ba132006-10-11 01:20:53 -07002343 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002344 if (n == 0)
2345 goto out_stop; /* error */
2346
2347 /*
2348 * OK. This truncate is going to happen. We add the inode to the
2349 * orphan list, so that if this truncate spans multiple transactions,
2350 * and we crash, we will resume the truncate when the filesystem
2351 * recovers. It also marks the inode dirty, to catch the new size.
2352 *
2353 * Implication: the file must always be in a sane, consistent
2354 * truncatable state while each transaction commits.
2355 */
Mingming Cao617ba132006-10-11 01:20:53 -07002356 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002357 goto out_stop;
2358
2359 /*
2360 * The orphan list entry will now protect us from any crash which
2361 * occurs before the truncate completes, so it is now safe to propagate
2362 * the new, shorter inode size (held for now in i_size) into the
2363 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07002364 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002365 */
2366 ei->i_disksize = inode->i_size;
2367
2368 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002369 * From here we block out all ext4_get_block() callers who want to
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002370 * modify the block allocation tree.
2371 */
2372 mutex_lock(&ei->truncate_mutex);
2373
2374 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07002375 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2376 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002377 goto do_indirects;
2378 }
2379
Mingming Cao617ba132006-10-11 01:20:53 -07002380 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002381 /* Kill the top of shared branch (not detached) */
2382 if (nr) {
2383 if (partial == chain) {
2384 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07002385 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002386 &nr, &nr+1, (chain+n-1) - partial);
2387 *partial->p = 0;
2388 /*
2389 * We mark the inode dirty prior to restart,
2390 * and prior to stop. No need for it here.
2391 */
2392 } else {
2393 /* Shared branch grows from an indirect block */
2394 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002395 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002396 partial->p,
2397 partial->p+1, (chain+n-1) - partial);
2398 }
2399 }
2400 /* Clear the ends of indirect blocks on the shared branch */
2401 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07002402 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002403 (__le32*)partial->bh->b_data+addr_per_block,
2404 (chain+n-1) - partial);
2405 BUFFER_TRACE(partial->bh, "call brelse");
2406 brelse (partial->bh);
2407 partial--;
2408 }
2409do_indirects:
2410 /* Kill the remaining (whole) subtrees */
2411 switch (offsets[0]) {
2412 default:
Mingming Cao617ba132006-10-11 01:20:53 -07002413 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002414 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002415 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2416 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002417 }
Mingming Cao617ba132006-10-11 01:20:53 -07002418 case EXT4_IND_BLOCK:
2419 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002420 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002421 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2422 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002423 }
Mingming Cao617ba132006-10-11 01:20:53 -07002424 case EXT4_DIND_BLOCK:
2425 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002426 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002427 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2428 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002429 }
Mingming Cao617ba132006-10-11 01:20:53 -07002430 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002431 ;
2432 }
2433
Mingming Cao617ba132006-10-11 01:20:53 -07002434 ext4_discard_reservation(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002435
2436 mutex_unlock(&ei->truncate_mutex);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002437 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002438 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002439
2440 /*
2441 * In a multi-transaction truncate, we only make the final transaction
2442 * synchronous
2443 */
2444 if (IS_SYNC(inode))
2445 handle->h_sync = 1;
2446out_stop:
2447 /*
2448 * If this was a simple ftruncate(), and the file will remain alive
2449 * then we need to clear up the orphan record which we created above.
2450 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07002451 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002452 * orphan info for us.
2453 */
2454 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002455 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002456
Mingming Cao617ba132006-10-11 01:20:53 -07002457 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002458}
2459
Mingming Cao617ba132006-10-11 01:20:53 -07002460static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2461 unsigned long ino, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002462{
2463 unsigned long desc, group_desc, block_group;
2464 unsigned long offset;
Mingming Cao617ba132006-10-11 01:20:53 -07002465 ext4_fsblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002466 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002467 struct ext4_group_desc * gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002468
Mingming Cao617ba132006-10-11 01:20:53 -07002469 if (!ext4_valid_inum(sb, ino)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002470 /*
2471 * This error is already checked for in namei.c unless we are
2472 * looking at an NFS filehandle, in which case no error
2473 * report is needed
2474 */
2475 return 0;
2476 }
2477
Mingming Cao617ba132006-10-11 01:20:53 -07002478 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2479 if (block_group >= EXT4_SB(sb)->s_groups_count) {
2480 ext4_error(sb,"ext4_get_inode_block","group >= groups count");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002481 return 0;
2482 }
2483 smp_rmb();
Mingming Cao617ba132006-10-11 01:20:53 -07002484 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2485 desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2486 bh = EXT4_SB(sb)->s_group_desc[group_desc];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002487 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002488 ext4_error (sb, "ext4_get_inode_block",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002489 "Descriptor not loaded");
2490 return 0;
2491 }
2492
Alexandre Ratchov0d1ee422006-10-11 01:21:14 -07002493 gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2494 desc * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002495 /*
2496 * Figure out the offset within the block group inode table
2497 */
Mingming Cao617ba132006-10-11 01:20:53 -07002498 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2499 EXT4_INODE_SIZE(sb);
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07002500 block = ext4_inode_table(sb, gdp) +
2501 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002502
2503 iloc->block_group = block_group;
Mingming Cao617ba132006-10-11 01:20:53 -07002504 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002505 return block;
2506}
2507
2508/*
Mingming Cao617ba132006-10-11 01:20:53 -07002509 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002510 * underlying buffer_head on success. If 'in_mem' is true, we have all
2511 * data in memory that is needed to recreate the on-disk version of this
2512 * inode.
2513 */
Mingming Cao617ba132006-10-11 01:20:53 -07002514static int __ext4_get_inode_loc(struct inode *inode,
2515 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002516{
Mingming Cao617ba132006-10-11 01:20:53 -07002517 ext4_fsblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002518 struct buffer_head *bh;
2519
Mingming Cao617ba132006-10-11 01:20:53 -07002520 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002521 if (!block)
2522 return -EIO;
2523
2524 bh = sb_getblk(inode->i_sb, block);
2525 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002526 ext4_error (inode->i_sb, "ext4_get_inode_loc",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002527 "unable to read inode block - "
Mingming Cao2ae02102006-10-11 01:21:11 -07002528 "inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002529 inode->i_ino, block);
2530 return -EIO;
2531 }
2532 if (!buffer_uptodate(bh)) {
2533 lock_buffer(bh);
2534 if (buffer_uptodate(bh)) {
2535 /* someone brought it uptodate while we waited */
2536 unlock_buffer(bh);
2537 goto has_buffer;
2538 }
2539
2540 /*
2541 * If we have all information of the inode in memory and this
2542 * is the only valid inode in the block, we need not read the
2543 * block.
2544 */
2545 if (in_mem) {
2546 struct buffer_head *bitmap_bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002547 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002548 int inodes_per_buffer;
2549 int inode_offset, i;
2550 int block_group;
2551 int start;
2552
2553 block_group = (inode->i_ino - 1) /
Mingming Cao617ba132006-10-11 01:20:53 -07002554 EXT4_INODES_PER_GROUP(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002555 inodes_per_buffer = bh->b_size /
Mingming Cao617ba132006-10-11 01:20:53 -07002556 EXT4_INODE_SIZE(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002557 inode_offset = ((inode->i_ino - 1) %
Mingming Cao617ba132006-10-11 01:20:53 -07002558 EXT4_INODES_PER_GROUP(inode->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002559 start = inode_offset & ~(inodes_per_buffer - 1);
2560
2561 /* Is the inode bitmap in cache? */
Mingming Cao617ba132006-10-11 01:20:53 -07002562 desc = ext4_get_group_desc(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002563 block_group, NULL);
2564 if (!desc)
2565 goto make_io;
2566
2567 bitmap_bh = sb_getblk(inode->i_sb,
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07002568 ext4_inode_bitmap(inode->i_sb, desc));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002569 if (!bitmap_bh)
2570 goto make_io;
2571
2572 /*
2573 * If the inode bitmap isn't in cache then the
2574 * optimisation may end up performing two reads instead
2575 * of one, so skip it.
2576 */
2577 if (!buffer_uptodate(bitmap_bh)) {
2578 brelse(bitmap_bh);
2579 goto make_io;
2580 }
2581 for (i = start; i < start + inodes_per_buffer; i++) {
2582 if (i == inode_offset)
2583 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07002584 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002585 break;
2586 }
2587 brelse(bitmap_bh);
2588 if (i == start + inodes_per_buffer) {
2589 /* all other inodes are free, so skip I/O */
2590 memset(bh->b_data, 0, bh->b_size);
2591 set_buffer_uptodate(bh);
2592 unlock_buffer(bh);
2593 goto has_buffer;
2594 }
2595 }
2596
2597make_io:
2598 /*
2599 * There are other valid inodes in the buffer, this inode
2600 * has in-inode xattrs, or we don't have this inode in memory.
2601 * Read the block from disk.
2602 */
2603 get_bh(bh);
2604 bh->b_end_io = end_buffer_read_sync;
2605 submit_bh(READ_META, bh);
2606 wait_on_buffer(bh);
2607 if (!buffer_uptodate(bh)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002608 ext4_error(inode->i_sb, "ext4_get_inode_loc",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002609 "unable to read inode block - "
Mingming Cao2ae02102006-10-11 01:21:11 -07002610 "inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002611 inode->i_ino, block);
2612 brelse(bh);
2613 return -EIO;
2614 }
2615 }
2616has_buffer:
2617 iloc->bh = bh;
2618 return 0;
2619}
2620
Mingming Cao617ba132006-10-11 01:20:53 -07002621int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002622{
2623 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07002624 return __ext4_get_inode_loc(inode, iloc,
2625 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002626}
2627
Mingming Cao617ba132006-10-11 01:20:53 -07002628void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002629{
Mingming Cao617ba132006-10-11 01:20:53 -07002630 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002631
2632 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07002633 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002634 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07002635 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002636 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07002637 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002638 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07002639 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002640 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07002641 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002642 inode->i_flags |= S_DIRSYNC;
2643}
2644
Jan Karaff9ddf72007-07-18 09:24:20 -04002645/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2646void ext4_get_inode_flags(struct ext4_inode_info *ei)
2647{
2648 unsigned int flags = ei->vfs_inode.i_flags;
2649
2650 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
2651 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
2652 if (flags & S_SYNC)
2653 ei->i_flags |= EXT4_SYNC_FL;
2654 if (flags & S_APPEND)
2655 ei->i_flags |= EXT4_APPEND_FL;
2656 if (flags & S_IMMUTABLE)
2657 ei->i_flags |= EXT4_IMMUTABLE_FL;
2658 if (flags & S_NOATIME)
2659 ei->i_flags |= EXT4_NOATIME_FL;
2660 if (flags & S_DIRSYNC)
2661 ei->i_flags |= EXT4_DIRSYNC_FL;
2662}
2663
Mingming Cao617ba132006-10-11 01:20:53 -07002664void ext4_read_inode(struct inode * inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002665{
Mingming Cao617ba132006-10-11 01:20:53 -07002666 struct ext4_iloc iloc;
2667 struct ext4_inode *raw_inode;
2668 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002669 struct buffer_head *bh;
2670 int block;
2671
Mingming Cao617ba132006-10-11 01:20:53 -07002672#ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2673 ei->i_acl = EXT4_ACL_NOT_CACHED;
2674 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002675#endif
2676 ei->i_block_alloc_info = NULL;
2677
Mingming Cao617ba132006-10-11 01:20:53 -07002678 if (__ext4_get_inode_loc(inode, &iloc, 0))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002679 goto bad_inode;
2680 bh = iloc.bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002681 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002682 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2683 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2684 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2685 if(!(test_opt (inode->i_sb, NO_UID32))) {
2686 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2687 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2688 }
2689 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2690 inode->i_size = le32_to_cpu(raw_inode->i_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002691
2692 ei->i_state = 0;
2693 ei->i_dir_start_lookup = 0;
2694 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2695 /* We now have enough fields to check if the inode was active or not.
2696 * This is needed because nfsd might try to access dead inodes
2697 * the test is that same one that e2fsck uses
2698 * NeilBrown 1999oct15
2699 */
2700 if (inode->i_nlink == 0) {
2701 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07002702 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002703 /* this inode is deleted */
2704 brelse (bh);
2705 goto bad_inode;
2706 }
2707 /* The only unlinked inodes we let through here have
2708 * valid i_mode and are being read by the orphan
2709 * recovery code: that's fine, we're about to complete
2710 * the process of deleting those. */
2711 }
2712 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2713 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002714 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07002715 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2716 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07002717 ei->i_file_acl |=
2718 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002719 if (!S_ISREG(inode->i_mode)) {
2720 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2721 } else {
2722 inode->i_size |=
2723 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2724 }
2725 ei->i_disksize = inode->i_size;
2726 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2727 ei->i_block_group = iloc.block_group;
2728 /*
2729 * NOTE! The in-memory inode i_data array is in little-endian order
2730 * even on big-endian machines: we do NOT byteswap the block numbers!
2731 */
Mingming Cao617ba132006-10-11 01:20:53 -07002732 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002733 ei->i_data[block] = raw_inode->i_block[block];
2734 INIT_LIST_HEAD(&ei->i_orphan);
2735
Mingming Cao617ba132006-10-11 01:20:53 -07002736 if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2737 EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002738 /*
2739 * When mke2fs creates big inodes it does not zero out
Mingming Cao617ba132006-10-11 01:20:53 -07002740 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002741 * so ignore those first few inodes.
2742 */
2743 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07002744 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07002745 EXT4_INODE_SIZE(inode->i_sb)) {
2746 brelse (bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002747 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07002748 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002749 if (ei->i_extra_isize == 0) {
2750 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07002751 ei->i_extra_isize = sizeof(struct ext4_inode) -
2752 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002753 } else {
2754 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07002755 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002756 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07002757 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2758 ei->i_state |= EXT4_STATE_XATTR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002759 }
2760 } else
2761 ei->i_extra_isize = 0;
2762
Kalpak Shahef7f3832007-07-18 09:15:20 -04002763 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
2764 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
2765 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
2766 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
2767
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002768 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002769 inode->i_op = &ext4_file_inode_operations;
2770 inode->i_fop = &ext4_file_operations;
2771 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002772 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002773 inode->i_op = &ext4_dir_inode_operations;
2774 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002775 } else if (S_ISLNK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002776 if (ext4_inode_is_fast_symlink(inode))
2777 inode->i_op = &ext4_fast_symlink_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002778 else {
Mingming Cao617ba132006-10-11 01:20:53 -07002779 inode->i_op = &ext4_symlink_inode_operations;
2780 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002781 }
2782 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002783 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002784 if (raw_inode->i_block[0])
2785 init_special_inode(inode, inode->i_mode,
2786 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2787 else
2788 init_special_inode(inode, inode->i_mode,
2789 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2790 }
2791 brelse (iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002792 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002793 return;
2794
2795bad_inode:
2796 make_bad_inode(inode);
2797 return;
2798}
2799
2800/*
2801 * Post the struct inode info into an on-disk inode location in the
2802 * buffer-cache. This gobbles the caller's reference to the
2803 * buffer_head in the inode location struct.
2804 *
2805 * The caller must have write access to iloc->bh.
2806 */
Mingming Cao617ba132006-10-11 01:20:53 -07002807static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002808 struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -07002809 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002810{
Mingming Cao617ba132006-10-11 01:20:53 -07002811 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2812 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002813 struct buffer_head *bh = iloc->bh;
2814 int err = 0, rc, block;
2815
2816 /* For fields not not tracking in the in-memory inode,
2817 * initialise them to zero for new inodes. */
Mingming Cao617ba132006-10-11 01:20:53 -07002818 if (ei->i_state & EXT4_STATE_NEW)
2819 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002820
Jan Karaff9ddf72007-07-18 09:24:20 -04002821 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002822 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2823 if(!(test_opt(inode->i_sb, NO_UID32))) {
2824 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2825 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2826/*
2827 * Fix up interoperability with old kernels. Otherwise, old inodes get
2828 * re-used with the upper 16 bits of the uid/gid intact
2829 */
2830 if(!ei->i_dtime) {
2831 raw_inode->i_uid_high =
2832 cpu_to_le16(high_16_bits(inode->i_uid));
2833 raw_inode->i_gid_high =
2834 cpu_to_le16(high_16_bits(inode->i_gid));
2835 } else {
2836 raw_inode->i_uid_high = 0;
2837 raw_inode->i_gid_high = 0;
2838 }
2839 } else {
2840 raw_inode->i_uid_low =
2841 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2842 raw_inode->i_gid_low =
2843 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2844 raw_inode->i_uid_high = 0;
2845 raw_inode->i_gid_high = 0;
2846 }
2847 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2848 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002849
2850 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
2851 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
2852 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
2853 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
2854
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002855 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2856 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2857 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07002858 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2859 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07002860 raw_inode->i_file_acl_high =
2861 cpu_to_le16(ei->i_file_acl >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002862 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2863 if (!S_ISREG(inode->i_mode)) {
2864 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2865 } else {
2866 raw_inode->i_size_high =
2867 cpu_to_le32(ei->i_disksize >> 32);
2868 if (ei->i_disksize > 0x7fffffffULL) {
2869 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07002870 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2871 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2872 EXT4_SB(sb)->s_es->s_rev_level ==
2873 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002874 /* If this is the first large file
2875 * created, add a flag to the superblock.
2876 */
Mingming Cao617ba132006-10-11 01:20:53 -07002877 err = ext4_journal_get_write_access(handle,
2878 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002879 if (err)
2880 goto out_brelse;
Mingming Cao617ba132006-10-11 01:20:53 -07002881 ext4_update_dynamic_rev(sb);
2882 EXT4_SET_RO_COMPAT_FEATURE(sb,
2883 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002884 sb->s_dirt = 1;
2885 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07002886 err = ext4_journal_dirty_metadata(handle,
2887 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002888 }
2889 }
2890 }
2891 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2892 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2893 if (old_valid_dev(inode->i_rdev)) {
2894 raw_inode->i_block[0] =
2895 cpu_to_le32(old_encode_dev(inode->i_rdev));
2896 raw_inode->i_block[1] = 0;
2897 } else {
2898 raw_inode->i_block[0] = 0;
2899 raw_inode->i_block[1] =
2900 cpu_to_le32(new_encode_dev(inode->i_rdev));
2901 raw_inode->i_block[2] = 0;
2902 }
Mingming Cao617ba132006-10-11 01:20:53 -07002903 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002904 raw_inode->i_block[block] = ei->i_data[block];
2905
2906 if (ei->i_extra_isize)
2907 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2908
Mingming Cao617ba132006-10-11 01:20:53 -07002909 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2910 rc = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002911 if (!err)
2912 err = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07002913 ei->i_state &= ~EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002914
2915out_brelse:
2916 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002917 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002918 return err;
2919}
2920
2921/*
Mingming Cao617ba132006-10-11 01:20:53 -07002922 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002923 *
2924 * We are called from a few places:
2925 *
2926 * - Within generic_file_write() for O_SYNC files.
2927 * Here, there will be no transaction running. We wait for any running
2928 * trasnaction to commit.
2929 *
2930 * - Within sys_sync(), kupdate and such.
2931 * We wait on commit, if tol to.
2932 *
2933 * - Within prune_icache() (PF_MEMALLOC == true)
2934 * Here we simply return. We can't afford to block kswapd on the
2935 * journal commit.
2936 *
2937 * In all cases it is actually safe for us to return without doing anything,
2938 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07002939 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002940 * knfsd.
2941 *
2942 * Note that we are absolutely dependent upon all inode dirtiers doing the
2943 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2944 * which we are interested.
2945 *
2946 * It would be a bug for them to not do this. The code:
2947 *
2948 * mark_inode_dirty(inode)
2949 * stuff();
2950 * inode->i_size = expr;
2951 *
2952 * is in error because a kswapd-driven write_inode() could occur while
2953 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2954 * will no longer be on the superblock's dirty inode list.
2955 */
Mingming Cao617ba132006-10-11 01:20:53 -07002956int ext4_write_inode(struct inode *inode, int wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002957{
2958 if (current->flags & PF_MEMALLOC)
2959 return 0;
2960
Mingming Cao617ba132006-10-11 01:20:53 -07002961 if (ext4_journal_current_handle()) {
Mingming Caob38bd332007-07-19 01:48:35 -07002962 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002963 dump_stack();
2964 return -EIO;
2965 }
2966
2967 if (!wait)
2968 return 0;
2969
Mingming Cao617ba132006-10-11 01:20:53 -07002970 return ext4_force_commit(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002971}
2972
2973/*
Mingming Cao617ba132006-10-11 01:20:53 -07002974 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002975 *
2976 * Called from notify_change.
2977 *
2978 * We want to trap VFS attempts to truncate the file as soon as
2979 * possible. In particular, we want to make sure that when the VFS
2980 * shrinks i_size, we put the inode on the orphan list and modify
2981 * i_disksize immediately, so that during the subsequent flushing of
2982 * dirty pages and freeing of disk blocks, we can guarantee that any
2983 * commit will leave the blocks being flushed in an unused state on
2984 * disk. (On recovery, the inode will get truncated and the blocks will
2985 * be freed, so we have a strong guarantee that no future commit will
2986 * leave these blocks visible to the user.)
2987 *
2988 * Called with inode->sem down.
2989 */
Mingming Cao617ba132006-10-11 01:20:53 -07002990int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002991{
2992 struct inode *inode = dentry->d_inode;
2993 int error, rc = 0;
2994 const unsigned int ia_valid = attr->ia_valid;
2995
2996 error = inode_change_ok(inode, attr);
2997 if (error)
2998 return error;
2999
3000 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3001 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3002 handle_t *handle;
3003
3004 /* (user+group)*(old+new) structure, inode write (sb,
3005 * inode block, ? - but truncate inode update has it) */
Mingming Cao617ba132006-10-11 01:20:53 -07003006 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
3007 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003008 if (IS_ERR(handle)) {
3009 error = PTR_ERR(handle);
3010 goto err_out;
3011 }
3012 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3013 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07003014 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003015 return error;
3016 }
3017 /* Update corresponding info in inode so that everything is in
3018 * one transaction */
3019 if (attr->ia_valid & ATTR_UID)
3020 inode->i_uid = attr->ia_uid;
3021 if (attr->ia_valid & ATTR_GID)
3022 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07003023 error = ext4_mark_inode_dirty(handle, inode);
3024 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003025 }
3026
3027 if (S_ISREG(inode->i_mode) &&
3028 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3029 handle_t *handle;
3030
Mingming Cao617ba132006-10-11 01:20:53 -07003031 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003032 if (IS_ERR(handle)) {
3033 error = PTR_ERR(handle);
3034 goto err_out;
3035 }
3036
Mingming Cao617ba132006-10-11 01:20:53 -07003037 error = ext4_orphan_add(handle, inode);
3038 EXT4_I(inode)->i_disksize = attr->ia_size;
3039 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003040 if (!error)
3041 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07003042 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003043 }
3044
3045 rc = inode_setattr(inode, attr);
3046
Mingming Cao617ba132006-10-11 01:20:53 -07003047 /* If inode_setattr's call to ext4_truncate failed to get a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003048 * transaction handle at all, we need to clean up the in-core
3049 * orphan list manually. */
3050 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003051 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003052
3053 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07003054 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003055
3056err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07003057 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003058 if (!error)
3059 error = rc;
3060 return error;
3061}
3062
3063
3064/*
3065 * How many blocks doth make a writepage()?
3066 *
3067 * With N blocks per page, it may be:
3068 * N data blocks
3069 * 2 indirect block
3070 * 2 dindirect
3071 * 1 tindirect
3072 * N+5 bitmap blocks (from the above)
3073 * N+5 group descriptor summary blocks
3074 * 1 inode block
3075 * 1 superblock.
Mingming Cao617ba132006-10-11 01:20:53 -07003076 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003077 *
Mingming Cao617ba132006-10-11 01:20:53 -07003078 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003079 *
3080 * With ordered or writeback data it's the same, less the N data blocks.
3081 *
3082 * If the inode's direct blocks can hold an integral number of pages then a
3083 * page cannot straddle two indirect blocks, and we can only touch one indirect
3084 * and dindirect block, and the "5" above becomes "3".
3085 *
3086 * This still overestimates under most circumstances. If we were to pass the
3087 * start and end offsets in here as well we could do block_to_path() on each
3088 * block and work out the exact number of indirects which are touched. Pah.
3089 */
3090
Alex Tomasa86c6182006-10-11 01:21:03 -07003091int ext4_writepage_trans_blocks(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003092{
Mingming Cao617ba132006-10-11 01:20:53 -07003093 int bpp = ext4_journal_blocks_per_page(inode);
3094 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003095 int ret;
3096
Alex Tomasa86c6182006-10-11 01:21:03 -07003097 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3098 return ext4_ext_writepage_trans_blocks(inode, bpp);
3099
Mingming Cao617ba132006-10-11 01:20:53 -07003100 if (ext4_should_journal_data(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003101 ret = 3 * (bpp + indirects) + 2;
3102 else
3103 ret = 2 * (bpp + indirects) + 2;
3104
3105#ifdef CONFIG_QUOTA
3106 /* We know that structure was already allocated during DQUOT_INIT so
3107 * we will be updating only the data blocks + inodes */
Mingming Cao617ba132006-10-11 01:20:53 -07003108 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003109#endif
3110
3111 return ret;
3112}
3113
3114/*
Mingming Cao617ba132006-10-11 01:20:53 -07003115 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003116 * Give this, we know that the caller already has write access to iloc->bh.
3117 */
Mingming Cao617ba132006-10-11 01:20:53 -07003118int ext4_mark_iloc_dirty(handle_t *handle,
3119 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003120{
3121 int err = 0;
3122
3123 /* the do_update_inode consumes one bh->b_count */
3124 get_bh(iloc->bh);
3125
Mingming Caodab291a2006-10-11 01:21:01 -07003126 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Mingming Cao617ba132006-10-11 01:20:53 -07003127 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003128 put_bh(iloc->bh);
3129 return err;
3130}
3131
3132/*
3133 * On success, We end up with an outstanding reference count against
3134 * iloc->bh. This _must_ be cleaned up later.
3135 */
3136
3137int
Mingming Cao617ba132006-10-11 01:20:53 -07003138ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3139 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003140{
3141 int err = 0;
3142 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07003143 err = ext4_get_inode_loc(inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003144 if (!err) {
3145 BUFFER_TRACE(iloc->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07003146 err = ext4_journal_get_write_access(handle, iloc->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003147 if (err) {
3148 brelse(iloc->bh);
3149 iloc->bh = NULL;
3150 }
3151 }
3152 }
Mingming Cao617ba132006-10-11 01:20:53 -07003153 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003154 return err;
3155}
3156
3157/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003158 * Expand an inode by new_extra_isize bytes.
3159 * Returns 0 on success or negative error number on failure.
3160 */
3161int ext4_expand_extra_isize(struct inode *inode, unsigned int new_extra_isize,
3162 struct ext4_iloc iloc, handle_t *handle)
3163{
3164 struct ext4_inode *raw_inode;
3165 struct ext4_xattr_ibody_header *header;
3166 struct ext4_xattr_entry *entry;
3167
3168 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
3169 return 0;
3170
3171 raw_inode = ext4_raw_inode(&iloc);
3172
3173 header = IHDR(inode, raw_inode);
3174 entry = IFIRST(header);
3175
3176 /* No extended attributes present */
3177 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
3178 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
3179 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
3180 new_extra_isize);
3181 EXT4_I(inode)->i_extra_isize = new_extra_isize;
3182 return 0;
3183 }
3184
3185 /* try to expand with EAs present */
3186 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
3187 raw_inode, handle);
3188}
3189
3190/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003191 * What we do here is to mark the in-core inode as clean with respect to inode
3192 * dirtiness (it may still be data-dirty).
3193 * This means that the in-core inode may be reaped by prune_icache
3194 * without having to perform any I/O. This is a very good thing,
3195 * because *any* task may call prune_icache - even ones which
3196 * have a transaction open against a different journal.
3197 *
3198 * Is this cheating? Not really. Sure, we haven't written the
3199 * inode out, but prune_icache isn't a user-visible syncing function.
3200 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3201 * we start and wait on commits.
3202 *
3203 * Is this efficient/effective? Well, we're being nice to the system
3204 * by cleaning up our inodes proactively so they can be reaped
3205 * without I/O. But we are potentially leaving up to five seconds'
3206 * worth of inodes floating about which prune_icache wants us to
3207 * write out. One way to fix that would be to get prune_icache()
3208 * to do a write_super() to free up some memory. It has the desired
3209 * effect.
3210 */
Mingming Cao617ba132006-10-11 01:20:53 -07003211int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003212{
Mingming Cao617ba132006-10-11 01:20:53 -07003213 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003214 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3215 static unsigned int mnt_count;
3216 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003217
3218 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07003219 err = ext4_reserve_inode_write(handle, inode, &iloc);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003220 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
3221 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
3222 /*
3223 * We need extra buffer credits since we may write into EA block
3224 * with this same handle. If journal_extend fails, then it will
3225 * only result in a minor loss of functionality for that inode.
3226 * If this is felt to be critical, then e2fsck should be run to
3227 * force a large enough s_min_extra_isize.
3228 */
3229 if ((jbd2_journal_extend(handle,
3230 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
3231 ret = ext4_expand_extra_isize(inode,
3232 sbi->s_want_extra_isize,
3233 iloc, handle);
3234 if (ret) {
3235 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04003236 if (mnt_count !=
3237 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003238 ext4_warning(inode->i_sb, __FUNCTION__,
3239 "Unable to expand inode %lu. Delete"
3240 " some EAs or run e2fsck.",
3241 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04003242 mnt_count =
3243 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003244 }
3245 }
3246 }
3247 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003248 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07003249 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003250 return err;
3251}
3252
3253/*
Mingming Cao617ba132006-10-11 01:20:53 -07003254 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003255 *
3256 * We're really interested in the case where a file is being extended.
3257 * i_size has been changed by generic_commit_write() and we thus need
3258 * to include the updated inode in the current transaction.
3259 *
3260 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3261 * are allocated to the file.
3262 *
3263 * If the inode is marked synchronous, we don't honour that here - doing
3264 * so would cause a commit on atime updates, which we don't bother doing.
3265 * We handle synchronous inodes at the highest possible level.
3266 */
Mingming Cao617ba132006-10-11 01:20:53 -07003267void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003268{
Mingming Cao617ba132006-10-11 01:20:53 -07003269 handle_t *current_handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003270 handle_t *handle;
3271
Mingming Cao617ba132006-10-11 01:20:53 -07003272 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003273 if (IS_ERR(handle))
3274 goto out;
3275 if (current_handle &&
3276 current_handle->h_transaction != handle->h_transaction) {
3277 /* This task has a transaction open against a different fs */
3278 printk(KERN_EMERG "%s: transactions do not match!\n",
3279 __FUNCTION__);
3280 } else {
3281 jbd_debug(5, "marking dirty. outer handle=%p\n",
3282 current_handle);
Mingming Cao617ba132006-10-11 01:20:53 -07003283 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003284 }
Mingming Cao617ba132006-10-11 01:20:53 -07003285 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003286out:
3287 return;
3288}
3289
3290#if 0
3291/*
3292 * Bind an inode's backing buffer_head into this transaction, to prevent
3293 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07003294 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003295 * returns no iloc structure, so the caller needs to repeat the iloc
3296 * lookup to mark the inode dirty later.
3297 */
Mingming Cao617ba132006-10-11 01:20:53 -07003298static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003299{
Mingming Cao617ba132006-10-11 01:20:53 -07003300 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003301
3302 int err = 0;
3303 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07003304 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003305 if (!err) {
3306 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07003307 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003308 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07003309 err = ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003310 iloc.bh);
3311 brelse(iloc.bh);
3312 }
3313 }
Mingming Cao617ba132006-10-11 01:20:53 -07003314 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003315 return err;
3316}
3317#endif
3318
Mingming Cao617ba132006-10-11 01:20:53 -07003319int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003320{
3321 journal_t *journal;
3322 handle_t *handle;
3323 int err;
3324
3325 /*
3326 * We have to be very careful here: changing a data block's
3327 * journaling status dynamically is dangerous. If we write a
3328 * data block to the journal, change the status and then delete
3329 * that block, we risk forgetting to revoke the old log record
3330 * from the journal and so a subsequent replay can corrupt data.
3331 * So, first we make sure that the journal is empty and that
3332 * nobody is changing anything.
3333 */
3334
Mingming Cao617ba132006-10-11 01:20:53 -07003335 journal = EXT4_JOURNAL(inode);
Dave Hansend6995942007-07-18 08:33:51 -04003336 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003337 return -EROFS;
3338
Mingming Caodab291a2006-10-11 01:21:01 -07003339 jbd2_journal_lock_updates(journal);
3340 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003341
3342 /*
3343 * OK, there are no updates running now, and all cached data is
3344 * synced to disk. We are now in a completely consistent state
3345 * which doesn't have anything in the journal, and we know that
3346 * no filesystem updates are running, so it is safe to modify
3347 * the inode's in-core data-journaling state flag now.
3348 */
3349
3350 if (val)
Mingming Cao617ba132006-10-11 01:20:53 -07003351 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003352 else
Mingming Cao617ba132006-10-11 01:20:53 -07003353 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3354 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003355
Mingming Caodab291a2006-10-11 01:21:01 -07003356 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003357
3358 /* Finally we can mark the inode as dirty. */
3359
Mingming Cao617ba132006-10-11 01:20:53 -07003360 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003361 if (IS_ERR(handle))
3362 return PTR_ERR(handle);
3363
Mingming Cao617ba132006-10-11 01:20:53 -07003364 err = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003365 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07003366 ext4_journal_stop(handle);
3367 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003368
3369 return err;
3370}