blob: 1ee19c91868600a620320c5e37259ea094409ba7 [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{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500108 ext4_lblk_t needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109
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,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500285 ext4_lblk_t i_block,
286 ext4_lblk_t offsets[4], int *boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700287{
Mingming Cao617ba132006-10-11 01:20:53 -0700288 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
289 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
290 const long direct_blocks = EXT4_NDIR_BLOCKS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 indirect_blocks = ptrs,
292 double_blocks = (1 << (ptrs_bits * 2));
293 int n = 0;
294 int final = 0;
295
296 if (i_block < 0) {
Mingming Cao617ba132006-10-11 01:20:53 -0700297 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700298 } else if (i_block < direct_blocks) {
299 offsets[n++] = i_block;
300 final = direct_blocks;
301 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700302 offsets[n++] = EXT4_IND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303 offsets[n++] = i_block;
304 final = ptrs;
305 } else if ((i_block -= indirect_blocks) < double_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700306 offsets[n++] = EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700307 offsets[n++] = i_block >> ptrs_bits;
308 offsets[n++] = i_block & (ptrs - 1);
309 final = ptrs;
310 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
Mingming Cao617ba132006-10-11 01:20:53 -0700311 offsets[n++] = EXT4_TIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700312 offsets[n++] = i_block >> (ptrs_bits * 2);
313 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
314 offsets[n++] = i_block & (ptrs - 1);
315 final = ptrs;
316 } else {
Mingming Cao617ba132006-10-11 01:20:53 -0700317 ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700318 }
319 if (boundary)
320 *boundary = final - 1 - (i_block & (ptrs - 1));
321 return n;
322}
323
324/**
Mingming Cao617ba132006-10-11 01:20:53 -0700325 * ext4_get_branch - read the chain of indirect blocks leading to data
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700326 * @inode: inode in question
327 * @depth: depth of the chain (1 - direct pointer, etc.)
328 * @offsets: offsets of pointers in inode/indirect blocks
329 * @chain: place to store the result
330 * @err: here we store the error value
331 *
332 * Function fills the array of triples <key, p, bh> and returns %NULL
333 * if everything went OK or the pointer to the last filled triple
334 * (incomplete one) otherwise. Upon the return chain[i].key contains
335 * the number of (i+1)-th block in the chain (as it is stored in memory,
336 * i.e. little-endian 32-bit), chain[i].p contains the address of that
337 * number (it points into struct inode for i==0 and into the bh->b_data
338 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
339 * block for i>0 and NULL for i==0. In other words, it holds the block
340 * numbers of the chain, addresses they were taken from (and where we can
341 * verify that chain did not change) and buffer_heads hosting these
342 * numbers.
343 *
344 * Function stops when it stumbles upon zero pointer (absent block)
345 * (pointer to last triple returned, *@err == 0)
346 * or when it gets an IO error reading an indirect block
347 * (ditto, *@err == -EIO)
348 * or when it notices that chain had been changed while it was reading
349 * (ditto, *@err == -EAGAIN)
350 * or when it reads all @depth-1 indirect blocks successfully and finds
351 * the whole chain, all way to the data (returns %NULL, *err == 0).
352 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500353static Indirect *ext4_get_branch(struct inode *inode, int depth,
354 ext4_lblk_t *offsets,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700355 Indirect chain[4], int *err)
356{
357 struct super_block *sb = inode->i_sb;
358 Indirect *p = chain;
359 struct buffer_head *bh;
360
361 *err = 0;
362 /* i_data is not going away, no lock needed */
Mingming Cao617ba132006-10-11 01:20:53 -0700363 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700364 if (!p->key)
365 goto no_block;
366 while (--depth) {
367 bh = sb_bread(sb, le32_to_cpu(p->key));
368 if (!bh)
369 goto failure;
370 /* Reader: pointers */
371 if (!verify_chain(chain, p))
372 goto changed;
373 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
374 /* Reader: end */
375 if (!p->key)
376 goto no_block;
377 }
378 return NULL;
379
380changed:
381 brelse(bh);
382 *err = -EAGAIN;
383 goto no_block;
384failure:
385 *err = -EIO;
386no_block:
387 return p;
388}
389
390/**
Mingming Cao617ba132006-10-11 01:20:53 -0700391 * ext4_find_near - find a place for allocation with sufficient locality
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700392 * @inode: owner
393 * @ind: descriptor of indirect block.
394 *
395 * This function returns the prefered place for block allocation.
396 * It is used when heuristic for sequential allocation fails.
397 * Rules are:
398 * + if there is a block to the left of our position - allocate near it.
399 * + if pointer will live in indirect block - allocate near that block.
400 * + if pointer will live in inode - allocate in the same
401 * cylinder group.
402 *
403 * In the latter case we colour the starting block by the callers PID to
404 * prevent it from clashing with concurrent allocations for a different inode
405 * in the same block group. The PID is used here so that functionally related
406 * files will be close-by on-disk.
407 *
408 * Caller must make sure that @ind is valid and will stay that way.
409 */
Mingming Cao617ba132006-10-11 01:20:53 -0700410static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700411{
Mingming Cao617ba132006-10-11 01:20:53 -0700412 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700413 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
414 __le32 *p;
Mingming Cao617ba132006-10-11 01:20:53 -0700415 ext4_fsblk_t bg_start;
416 ext4_grpblk_t colour;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700417
418 /* Try to find previous block */
419 for (p = ind->p - 1; p >= start; p--) {
420 if (*p)
421 return le32_to_cpu(*p);
422 }
423
424 /* No such thing, so let's try location of indirect block */
425 if (ind->bh)
426 return ind->bh->b_blocknr;
427
428 /*
429 * It is going to be referred to from the inode itself? OK, just put it
430 * into the same cylinder group then.
431 */
Mingming Cao617ba132006-10-11 01:20:53 -0700432 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700433 colour = (current->pid % 16) *
Mingming Cao617ba132006-10-11 01:20:53 -0700434 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700435 return bg_start + colour;
436}
437
438/**
Mingming Cao617ba132006-10-11 01:20:53 -0700439 * ext4_find_goal - find a prefered place for allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700440 * @inode: owner
441 * @block: block we want
442 * @chain: chain of indirect blocks
443 * @partial: pointer to the last triple within a chain
444 * @goal: place to store the result.
445 *
446 * Normally this function find the prefered place for block allocation,
447 * stores it in *@goal and returns zero.
448 */
449
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500450static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451 Indirect chain[4], Indirect *partial)
452{
Mingming Cao617ba132006-10-11 01:20:53 -0700453 struct ext4_block_alloc_info *block_i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700454
Mingming Cao617ba132006-10-11 01:20:53 -0700455 block_i = EXT4_I(inode)->i_block_alloc_info;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700456
457 /*
458 * try the heuristic for sequential allocation,
459 * failing that at least try to get decent locality.
460 */
461 if (block_i && (block == block_i->last_alloc_logical_block + 1)
462 && (block_i->last_alloc_physical_block != 0)) {
463 return block_i->last_alloc_physical_block + 1;
464 }
465
Mingming Cao617ba132006-10-11 01:20:53 -0700466 return ext4_find_near(inode, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700467}
468
469/**
Mingming Cao617ba132006-10-11 01:20:53 -0700470 * ext4_blks_to_allocate: Look up the block map and count the number
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700471 * of direct blocks need to be allocated for the given branch.
472 *
473 * @branch: chain of indirect blocks
474 * @k: number of blocks need for indirect blocks
475 * @blks: number of data blocks to be mapped.
476 * @blocks_to_boundary: the offset in the indirect block
477 *
478 * return the total number of blocks to be allocate, including the
479 * direct and indirect blocks.
480 */
Mingming Cao617ba132006-10-11 01:20:53 -0700481static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482 int blocks_to_boundary)
483{
484 unsigned long count = 0;
485
486 /*
487 * Simple case, [t,d]Indirect block(s) has not allocated yet
488 * then it's clear blocks on that path have not allocated
489 */
490 if (k > 0) {
491 /* right now we don't handle cross boundary allocation */
492 if (blks < blocks_to_boundary + 1)
493 count += blks;
494 else
495 count += blocks_to_boundary + 1;
496 return count;
497 }
498
499 count++;
500 while (count < blks && count <= blocks_to_boundary &&
501 le32_to_cpu(*(branch[0].p + count)) == 0) {
502 count++;
503 }
504 return count;
505}
506
507/**
Mingming Cao617ba132006-10-11 01:20:53 -0700508 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700509 * @indirect_blks: the number of blocks need to allocate for indirect
510 * blocks
511 *
512 * @new_blocks: on return it will store the new block numbers for
513 * the indirect blocks(if needed) and the first direct block,
514 * @blks: on return it will store the total number of allocated
515 * direct blocks
516 */
Mingming Cao617ba132006-10-11 01:20:53 -0700517static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
518 ext4_fsblk_t goal, int indirect_blks, int blks,
519 ext4_fsblk_t new_blocks[4], int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700520{
521 int target, i;
522 unsigned long count = 0;
523 int index = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700524 ext4_fsblk_t current_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700525 int ret = 0;
526
527 /*
528 * Here we try to allocate the requested multiple blocks at once,
529 * on a best-effort basis.
530 * To build a branch, we should allocate blocks for
531 * the indirect blocks(if not allocated yet), and at least
532 * the first direct block of this branch. That's the
533 * minimum number of blocks need to allocate(required)
534 */
535 target = blks + indirect_blks;
536
537 while (1) {
538 count = target;
539 /* allocating blocks for indirect blocks and direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -0700540 current_block = ext4_new_blocks(handle,inode,goal,&count,err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700541 if (*err)
542 goto failed_out;
543
544 target -= count;
545 /* allocate blocks for indirect blocks */
546 while (index < indirect_blks && count) {
547 new_blocks[index++] = current_block++;
548 count--;
549 }
550
551 if (count > 0)
552 break;
553 }
554
555 /* save the new block number for the first direct block */
556 new_blocks[index] = current_block;
557
558 /* total number of blocks allocated for direct blocks */
559 ret = count;
560 *err = 0;
561 return ret;
562failed_out:
563 for (i = 0; i <index; i++)
Mingming Cao617ba132006-10-11 01:20:53 -0700564 ext4_free_blocks(handle, inode, new_blocks[i], 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700565 return ret;
566}
567
568/**
Mingming Cao617ba132006-10-11 01:20:53 -0700569 * ext4_alloc_branch - allocate and set up a chain of blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700570 * @inode: owner
571 * @indirect_blks: number of allocated indirect blocks
572 * @blks: number of allocated direct blocks
573 * @offsets: offsets (in the blocks) to store the pointers to next.
574 * @branch: place to store the chain in.
575 *
576 * This function allocates blocks, zeroes out all but the last one,
577 * links them into chain and (if we are synchronous) writes them to disk.
578 * In other words, it prepares a branch that can be spliced onto the
579 * inode. It stores the information about that chain in the branch[], in
Mingming Cao617ba132006-10-11 01:20:53 -0700580 * the same format as ext4_get_branch() would do. We are calling it after
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581 * we had read the existing part of chain and partial points to the last
582 * triple of that (one with zero ->key). Upon the exit we have the same
Mingming Cao617ba132006-10-11 01:20:53 -0700583 * picture as after the successful ext4_get_block(), except that in one
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700584 * place chain is disconnected - *branch->p is still zero (we did not
585 * set the last link), but branch->key contains the number that should
586 * be placed into *branch->p to fill that gap.
587 *
588 * If allocation fails we free all blocks we've allocated (and forget
589 * their buffer_heads) and return the error value the from failed
Mingming Cao617ba132006-10-11 01:20:53 -0700590 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700591 * as described above and return 0.
592 */
Mingming Cao617ba132006-10-11 01:20:53 -0700593static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
594 int indirect_blks, int *blks, ext4_fsblk_t goal,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500595 ext4_lblk_t *offsets, Indirect *branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596{
597 int blocksize = inode->i_sb->s_blocksize;
598 int i, n = 0;
599 int err = 0;
600 struct buffer_head *bh;
601 int num;
Mingming Cao617ba132006-10-11 01:20:53 -0700602 ext4_fsblk_t new_blocks[4];
603 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700604
Mingming Cao617ba132006-10-11 01:20:53 -0700605 num = ext4_alloc_blocks(handle, inode, goal, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700606 *blks, new_blocks, &err);
607 if (err)
608 return err;
609
610 branch[0].key = cpu_to_le32(new_blocks[0]);
611 /*
612 * metadata blocks and data blocks are allocated.
613 */
614 for (n = 1; n <= indirect_blks; n++) {
615 /*
616 * Get buffer_head for parent block, zero it out
617 * and set the pointer to new one, then send
618 * parent to disk.
619 */
620 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
621 branch[n].bh = bh;
622 lock_buffer(bh);
623 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700624 err = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625 if (err) {
626 unlock_buffer(bh);
627 brelse(bh);
628 goto failed;
629 }
630
631 memset(bh->b_data, 0, blocksize);
632 branch[n].p = (__le32 *) bh->b_data + offsets[n];
633 branch[n].key = cpu_to_le32(new_blocks[n]);
634 *branch[n].p = branch[n].key;
635 if ( n == indirect_blks) {
636 current_block = new_blocks[n];
637 /*
638 * End of chain, update the last new metablock of
639 * the chain to point to the new allocated
640 * data blocks numbers
641 */
642 for (i=1; i < num; i++)
643 *(branch[n].p + i) = cpu_to_le32(++current_block);
644 }
645 BUFFER_TRACE(bh, "marking uptodate");
646 set_buffer_uptodate(bh);
647 unlock_buffer(bh);
648
Mingming Cao617ba132006-10-11 01:20:53 -0700649 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
650 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700651 if (err)
652 goto failed;
653 }
654 *blks = num;
655 return err;
656failed:
657 /* Allocation failed, free what we already allocated */
658 for (i = 1; i <= n ; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700659 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700660 ext4_journal_forget(handle, branch[i].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700661 }
662 for (i = 0; i <indirect_blks; i++)
Mingming Cao617ba132006-10-11 01:20:53 -0700663 ext4_free_blocks(handle, inode, new_blocks[i], 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664
Mingming Cao617ba132006-10-11 01:20:53 -0700665 ext4_free_blocks(handle, inode, new_blocks[i], num);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700666
667 return err;
668}
669
670/**
Mingming Cao617ba132006-10-11 01:20:53 -0700671 * ext4_splice_branch - splice the allocated branch onto inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700672 * @inode: owner
673 * @block: (logical) number of block we are adding
674 * @chain: chain of indirect blocks (with a missing link - see
Mingming Cao617ba132006-10-11 01:20:53 -0700675 * ext4_alloc_branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700676 * @where: location of missing link
677 * @num: number of indirect blocks we are adding
678 * @blks: number of direct blocks we are adding
679 *
680 * This function fills the missing link and does all housekeeping needed in
681 * inode (->i_blocks, etc.). In case of success we end up with the full
682 * chain to new block and return 0.
683 */
Mingming Cao617ba132006-10-11 01:20:53 -0700684static int ext4_splice_branch(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500685 ext4_lblk_t block, Indirect *where, int num, int blks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700686{
687 int i;
688 int err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700689 struct ext4_block_alloc_info *block_i;
690 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700691
Mingming Cao617ba132006-10-11 01:20:53 -0700692 block_i = EXT4_I(inode)->i_block_alloc_info;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700693 /*
694 * If we're splicing into a [td]indirect block (as opposed to the
695 * inode) then we need to get write access to the [td]indirect block
696 * before the splice.
697 */
698 if (where->bh) {
699 BUFFER_TRACE(where->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700700 err = ext4_journal_get_write_access(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700701 if (err)
702 goto err_out;
703 }
704 /* That's it */
705
706 *where->p = where->key;
707
708 /*
709 * Update the host buffer_head or inode to point to more just allocated
710 * direct blocks blocks
711 */
712 if (num == 0 && blks > 1) {
713 current_block = le32_to_cpu(where->key) + 1;
714 for (i = 1; i < blks; i++)
715 *(where->p + i ) = cpu_to_le32(current_block++);
716 }
717
718 /*
719 * update the most recently allocated logical & physical block
720 * in i_block_alloc_info, to assist find the proper goal block for next
721 * allocation
722 */
723 if (block_i) {
724 block_i->last_alloc_logical_block = block + blks - 1;
725 block_i->last_alloc_physical_block =
726 le32_to_cpu(where[num].key) + blks - 1;
727 }
728
729 /* We are done with atomic stuff, now do the rest of housekeeping */
730
Kalpak Shahef7f3832007-07-18 09:15:20 -0400731 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700732 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700733
734 /* had we spliced it onto indirect block? */
735 if (where->bh) {
736 /*
737 * If we spliced it onto an indirect block, we haven't
738 * altered the inode. Note however that if it is being spliced
739 * onto an indirect block at the very end of the file (the
740 * file is growing) then we *will* alter the inode to reflect
741 * the new i_size. But that is not done here - it is done in
Mingming Cao617ba132006-10-11 01:20:53 -0700742 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700743 */
744 jbd_debug(5, "splicing indirect only\n");
Mingming Cao617ba132006-10-11 01:20:53 -0700745 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
746 err = ext4_journal_dirty_metadata(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700747 if (err)
748 goto err_out;
749 } else {
750 /*
751 * OK, we spliced it into the inode itself on a direct block.
752 * Inode was dirtied above.
753 */
754 jbd_debug(5, "splicing direct\n");
755 }
756 return err;
757
758err_out:
759 for (i = 1; i <= num; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700760 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700761 ext4_journal_forget(handle, where[i].bh);
762 ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700763 }
Mingming Cao617ba132006-10-11 01:20:53 -0700764 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765
766 return err;
767}
768
769/*
770 * Allocation strategy is simple: if we have to allocate something, we will
771 * have to go the whole way to leaf. So let's do it before attaching anything
772 * to tree, set linkage between the newborn blocks, write them if sync is
773 * required, recheck the path, free and repeat if check fails, otherwise
774 * set the last missing link (that will protect us from any truncate-generated
775 * removals - all blocks on the path are immune now) and possibly force the
776 * write on the parent block.
777 * That has a nice additional property: no special recovery from the failed
778 * allocations is needed - we simply release blocks and do not touch anything
779 * reachable from inode.
780 *
781 * `handle' can be NULL if create == 0.
782 *
783 * The BKL may not be held on entry here. Be sure to take it early.
784 * return > 0, # of blocks mapped or allocated.
785 * return = 0, if plain lookup failed.
786 * return < 0, error case.
787 */
Mingming Cao617ba132006-10-11 01:20:53 -0700788int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500789 ext4_lblk_t iblock, unsigned long maxblocks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700790 struct buffer_head *bh_result,
791 int create, int extend_disksize)
792{
793 int err = -EIO;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500794 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700795 Indirect chain[4];
796 Indirect *partial;
Mingming Cao617ba132006-10-11 01:20:53 -0700797 ext4_fsblk_t goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700798 int indirect_blks;
799 int blocks_to_boundary = 0;
800 int depth;
Mingming Cao617ba132006-10-11 01:20:53 -0700801 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700802 int count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700803 ext4_fsblk_t first_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804
805
Alex Tomasa86c6182006-10-11 01:21:03 -0700806 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700807 J_ASSERT(handle != NULL || create == 0);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500808 depth = ext4_block_to_path(inode, iblock, offsets,
809 &blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700810
811 if (depth == 0)
812 goto out;
813
Mingming Cao617ba132006-10-11 01:20:53 -0700814 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815
816 /* Simplest case - block found, no allocation needed */
817 if (!partial) {
818 first_block = le32_to_cpu(chain[depth - 1].key);
819 clear_buffer_new(bh_result);
820 count++;
821 /*map more blocks*/
822 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao617ba132006-10-11 01:20:53 -0700823 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700824
825 if (!verify_chain(chain, partial)) {
826 /*
827 * Indirect block might be removed by
828 * truncate while we were reading it.
829 * Handling of that case: forget what we've
830 * got now. Flag the err as EAGAIN, so it
831 * will reread.
832 */
833 err = -EAGAIN;
834 count = 0;
835 break;
836 }
837 blk = le32_to_cpu(*(chain[depth-1].p + count));
838
839 if (blk == first_block + count)
840 count++;
841 else
842 break;
843 }
844 if (err != -EAGAIN)
845 goto got_it;
846 }
847
848 /* Next simple case - plain lookup or failed read of indirect block */
849 if (!create || err == -EIO)
850 goto cleanup;
851
852 mutex_lock(&ei->truncate_mutex);
853
854 /*
855 * If the indirect block is missing while we are reading
Mingming Cao617ba132006-10-11 01:20:53 -0700856 * the chain(ext4_get_branch() returns -EAGAIN err), or
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700857 * if the chain has been changed after we grab the semaphore,
858 * (either because another process truncated this branch, or
859 * another get_block allocated this branch) re-grab the chain to see if
860 * the request block has been allocated or not.
861 *
862 * Since we already block the truncate/other get_block
863 * at this point, we will have the current copy of the chain when we
864 * splice the branch into the tree.
865 */
866 if (err == -EAGAIN || !verify_chain(chain, partial)) {
867 while (partial > chain) {
868 brelse(partial->bh);
869 partial--;
870 }
Mingming Cao617ba132006-10-11 01:20:53 -0700871 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700872 if (!partial) {
873 count++;
874 mutex_unlock(&ei->truncate_mutex);
875 if (err)
876 goto cleanup;
877 clear_buffer_new(bh_result);
878 goto got_it;
879 }
880 }
881
882 /*
883 * Okay, we need to do block allocation. Lazily initialize the block
884 * allocation info here if necessary
885 */
886 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
Mingming Cao617ba132006-10-11 01:20:53 -0700887 ext4_init_block_alloc_info(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888
Mingming Cao617ba132006-10-11 01:20:53 -0700889 goal = ext4_find_goal(inode, iblock, chain, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700890
891 /* the number of blocks need to allocate for [d,t]indirect blocks */
892 indirect_blks = (chain + depth) - partial - 1;
893
894 /*
895 * Next look up the indirect map to count the totoal number of
896 * direct blocks to allocate for this branch.
897 */
Mingming Cao617ba132006-10-11 01:20:53 -0700898 count = ext4_blks_to_allocate(partial, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700899 maxblocks, blocks_to_boundary);
900 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700901 * Block out ext4_truncate while we alter the tree
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700902 */
Mingming Cao617ba132006-10-11 01:20:53 -0700903 err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700904 offsets + (partial - chain), partial);
905
906 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700907 * The ext4_splice_branch call will free and forget any buffers
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700908 * on the new chain if there is a failure, but that risks using
909 * up transaction credits, especially for bitmaps where the
910 * credits cannot be returned. Can we handle this somehow? We
911 * may need to return -EAGAIN upwards in the worst case. --sct
912 */
913 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -0700914 err = ext4_splice_branch(handle, inode, iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700915 partial, indirect_blks, count);
916 /*
917 * i_disksize growing is protected by truncate_mutex. Don't forget to
918 * protect it if you're about to implement concurrent
Mingming Cao617ba132006-10-11 01:20:53 -0700919 * ext4_get_block() -bzzz
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920 */
921 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
922 ei->i_disksize = inode->i_size;
923 mutex_unlock(&ei->truncate_mutex);
924 if (err)
925 goto cleanup;
926
927 set_buffer_new(bh_result);
928got_it:
929 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
930 if (count > blocks_to_boundary)
931 set_buffer_boundary(bh_result);
932 err = count;
933 /* Clean up and exit */
934 partial = chain + depth - 1; /* the whole chain */
935cleanup:
936 while (partial > chain) {
937 BUFFER_TRACE(partial->bh, "call brelse");
938 brelse(partial->bh);
939 partial--;
940 }
941 BUFFER_TRACE(bh_result, "returned");
942out:
943 return err;
944}
945
Mingming Cao617ba132006-10-11 01:20:53 -0700946#define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700947
Mingming Cao617ba132006-10-11 01:20:53 -0700948static int ext4_get_block(struct inode *inode, sector_t iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700949 struct buffer_head *bh_result, int create)
950{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800951 handle_t *handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700952 int ret = 0;
953 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
954
955 if (!create)
956 goto get_block; /* A read */
957
958 if (max_blocks == 1)
959 goto get_block; /* A single block get */
960
961 if (handle->h_transaction->t_state == T_LOCKED) {
962 /*
963 * Huge direct-io writes can hold off commits for long
964 * periods of time. Let this commit run.
965 */
Mingming Cao617ba132006-10-11 01:20:53 -0700966 ext4_journal_stop(handle);
967 handle = ext4_journal_start(inode, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700968 if (IS_ERR(handle))
969 ret = PTR_ERR(handle);
970 goto get_block;
971 }
972
Mingming Cao617ba132006-10-11 01:20:53 -0700973 if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700974 /*
975 * Getting low on buffer credits...
976 */
Mingming Cao617ba132006-10-11 01:20:53 -0700977 ret = ext4_journal_extend(handle, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 if (ret > 0) {
979 /*
980 * Couldn't extend the transaction. Start a new one.
981 */
Mingming Cao617ba132006-10-11 01:20:53 -0700982 ret = ext4_journal_restart(handle, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700983 }
984 }
985
986get_block:
987 if (ret == 0) {
Alex Tomasa86c6182006-10-11 01:21:03 -0700988 ret = ext4_get_blocks_wrap(handle, inode, iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700989 max_blocks, bh_result, create, 0);
990 if (ret > 0) {
991 bh_result->b_size = (ret << inode->i_blkbits);
992 ret = 0;
993 }
994 }
995 return ret;
996}
997
998/*
999 * `handle' can be NULL if create is zero
1000 */
Mingming Cao617ba132006-10-11 01:20:53 -07001001struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001002 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001003{
1004 struct buffer_head dummy;
1005 int fatal = 0, err;
1006
1007 J_ASSERT(handle != NULL || create == 0);
1008
1009 dummy.b_state = 0;
1010 dummy.b_blocknr = -1000;
1011 buffer_trace_init(&dummy.b_history);
Alex Tomasa86c6182006-10-11 01:21:03 -07001012 err = ext4_get_blocks_wrap(handle, inode, block, 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013 &dummy, create, 1);
1014 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001015 * ext4_get_blocks_handle() returns number of blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016 * mapped. 0 in case of a HOLE.
1017 */
1018 if (err > 0) {
1019 if (err > 1)
1020 WARN_ON(1);
1021 err = 0;
1022 }
1023 *errp = err;
1024 if (!err && buffer_mapped(&dummy)) {
1025 struct buffer_head *bh;
1026 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1027 if (!bh) {
1028 *errp = -EIO;
1029 goto err;
1030 }
1031 if (buffer_new(&dummy)) {
1032 J_ASSERT(create != 0);
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -04001033 J_ASSERT(handle != NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001034
1035 /*
1036 * Now that we do not always journal data, we should
1037 * keep in mind whether this should always journal the
1038 * new buffer as metadata. For now, regular file
Mingming Cao617ba132006-10-11 01:20:53 -07001039 * writes use ext4_get_block instead, so it's not a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 * problem.
1041 */
1042 lock_buffer(bh);
1043 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001044 fatal = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001045 if (!fatal && !buffer_uptodate(bh)) {
1046 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1047 set_buffer_uptodate(bh);
1048 }
1049 unlock_buffer(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001050 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1051 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052 if (!fatal)
1053 fatal = err;
1054 } else {
1055 BUFFER_TRACE(bh, "not a new buffer");
1056 }
1057 if (fatal) {
1058 *errp = fatal;
1059 brelse(bh);
1060 bh = NULL;
1061 }
1062 return bh;
1063 }
1064err:
1065 return NULL;
1066}
1067
Mingming Cao617ba132006-10-11 01:20:53 -07001068struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001069 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001070{
1071 struct buffer_head * bh;
1072
Mingming Cao617ba132006-10-11 01:20:53 -07001073 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074 if (!bh)
1075 return bh;
1076 if (buffer_uptodate(bh))
1077 return bh;
1078 ll_rw_block(READ_META, 1, &bh);
1079 wait_on_buffer(bh);
1080 if (buffer_uptodate(bh))
1081 return bh;
1082 put_bh(bh);
1083 *err = -EIO;
1084 return NULL;
1085}
1086
1087static int walk_page_buffers( handle_t *handle,
1088 struct buffer_head *head,
1089 unsigned from,
1090 unsigned to,
1091 int *partial,
1092 int (*fn)( handle_t *handle,
1093 struct buffer_head *bh))
1094{
1095 struct buffer_head *bh;
1096 unsigned block_start, block_end;
1097 unsigned blocksize = head->b_size;
1098 int err, ret = 0;
1099 struct buffer_head *next;
1100
1101 for ( bh = head, block_start = 0;
1102 ret == 0 && (bh != head || !block_start);
1103 block_start = block_end, bh = next)
1104 {
1105 next = bh->b_this_page;
1106 block_end = block_start + blocksize;
1107 if (block_end <= from || block_start >= to) {
1108 if (partial && !buffer_uptodate(bh))
1109 *partial = 1;
1110 continue;
1111 }
1112 err = (*fn)(handle, bh);
1113 if (!ret)
1114 ret = err;
1115 }
1116 return ret;
1117}
1118
1119/*
1120 * To preserve ordering, it is essential that the hole instantiation and
1121 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -07001122 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -07001123 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001124 * prepare_write() is the right place.
1125 *
Mingming Cao617ba132006-10-11 01:20:53 -07001126 * Also, this function can nest inside ext4_writepage() ->
1127 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001128 * has generated enough buffer credits to do the whole page. So we won't
1129 * block on the journal in that case, which is good, because the caller may
1130 * be PF_MEMALLOC.
1131 *
Mingming Cao617ba132006-10-11 01:20:53 -07001132 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001133 * quota file writes. If we were to commit the transaction while thus
1134 * reentered, there can be a deadlock - we would be holding a quota
1135 * lock, and the commit would never complete if another thread had a
1136 * transaction open and was blocking on the quota lock - a ranking
1137 * violation.
1138 *
Mingming Caodab291a2006-10-11 01:21:01 -07001139 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001140 * will _not_ run commit under these circumstances because handle->h_ref
1141 * is elevated. We'll still have enough credits for the tiny quotafile
1142 * write.
1143 */
1144static int do_journal_get_write_access(handle_t *handle,
1145 struct buffer_head *bh)
1146{
1147 if (!buffer_mapped(bh) || buffer_freed(bh))
1148 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001149 return ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001150}
1151
Nick Pigginbfc1af62007-10-16 01:25:05 -07001152static int ext4_write_begin(struct file *file, struct address_space *mapping,
1153 loff_t pos, unsigned len, unsigned flags,
1154 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001155{
Nick Pigginbfc1af62007-10-16 01:25:05 -07001156 struct inode *inode = mapping->host;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001157 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001158 handle_t *handle;
1159 int retries = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001160 struct page *page;
1161 pgoff_t index;
1162 unsigned from, to;
1163
1164 index = pos >> PAGE_CACHE_SHIFT;
1165 from = pos & (PAGE_CACHE_SIZE - 1);
1166 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001167
1168retry:
Nick Pigginbfc1af62007-10-16 01:25:05 -07001169 page = __grab_cache_page(mapping, index);
1170 if (!page)
1171 return -ENOMEM;
1172 *pagep = page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001173
Nick Pigginbfc1af62007-10-16 01:25:05 -07001174 handle = ext4_journal_start(inode, needed_blocks);
1175 if (IS_ERR(handle)) {
1176 unlock_page(page);
1177 page_cache_release(page);
1178 ret = PTR_ERR(handle);
1179 goto out;
1180 }
1181
1182 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1183 ext4_get_block);
1184
1185 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001186 ret = walk_page_buffers(handle, page_buffers(page),
1187 from, to, NULL, do_journal_get_write_access);
1188 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001189
1190 if (ret) {
Andrew Morton7479d2b2007-04-01 23:49:44 -07001191 ext4_journal_stop(handle);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001192 unlock_page(page);
1193 page_cache_release(page);
1194 }
1195
Mingming Cao617ba132006-10-11 01:20:53 -07001196 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001197 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001198out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001199 return ret;
1200}
1201
Mingming Cao617ba132006-10-11 01:20:53 -07001202int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001203{
Mingming Caodab291a2006-10-11 01:21:01 -07001204 int err = jbd2_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001205 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001206 ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001207 bh, handle, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001208 return err;
1209}
1210
Nick Pigginbfc1af62007-10-16 01:25:05 -07001211/* For write_end() in data=journal mode */
1212static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001213{
1214 if (!buffer_mapped(bh) || buffer_freed(bh))
1215 return 0;
1216 set_buffer_uptodate(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001217 return ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001218}
1219
1220/*
Nick Pigginbfc1af62007-10-16 01:25:05 -07001221 * Generic write_end handler for ordered and writeback ext4 journal modes.
1222 * We can't use generic_write_end, because that unlocks the page and we need to
1223 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1224 * after block_write_end.
1225 */
1226static int ext4_generic_write_end(struct file *file,
1227 struct address_space *mapping,
1228 loff_t pos, unsigned len, unsigned copied,
1229 struct page *page, void *fsdata)
1230{
1231 struct inode *inode = file->f_mapping->host;
1232
1233 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1234
1235 if (pos+copied > inode->i_size) {
1236 i_size_write(inode, pos+copied);
1237 mark_inode_dirty(inode);
1238 }
1239
1240 return copied;
1241}
1242
1243/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001244 * We need to pick up the new inode size which generic_commit_write gave us
1245 * `file' can be NULL - eg, when called from page_symlink().
1246 *
Mingming Cao617ba132006-10-11 01:20:53 -07001247 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001248 * buffers are managed internally.
1249 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001250static int ext4_ordered_write_end(struct file *file,
1251 struct address_space *mapping,
1252 loff_t pos, unsigned len, unsigned copied,
1253 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254{
Mingming Cao617ba132006-10-11 01:20:53 -07001255 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001256 struct inode *inode = file->f_mapping->host;
1257 unsigned from, to;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001258 int ret = 0, ret2;
1259
Nick Pigginbfc1af62007-10-16 01:25:05 -07001260 from = pos & (PAGE_CACHE_SIZE - 1);
1261 to = from + len;
1262
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001263 ret = walk_page_buffers(handle, page_buffers(page),
Mingming Cao617ba132006-10-11 01:20:53 -07001264 from, to, NULL, ext4_journal_dirty_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001265
1266 if (ret == 0) {
1267 /*
Nick Pigginbfc1af62007-10-16 01:25:05 -07001268 * generic_write_end() will run mark_inode_dirty() if i_size
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001269 * changes. So let's piggyback the i_disksize mark_inode_dirty
1270 * into that.
1271 */
1272 loff_t new_i_size;
1273
Nick Pigginbfc1af62007-10-16 01:25:05 -07001274 new_i_size = pos + copied;
Mingming Cao617ba132006-10-11 01:20:53 -07001275 if (new_i_size > EXT4_I(inode)->i_disksize)
1276 EXT4_I(inode)->i_disksize = new_i_size;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001277 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1278 page, fsdata);
1279 if (copied < 0)
1280 ret = copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001281 }
Mingming Cao617ba132006-10-11 01:20:53 -07001282 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001283 if (!ret)
1284 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001285 unlock_page(page);
1286 page_cache_release(page);
1287
1288 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001289}
1290
Nick Pigginbfc1af62007-10-16 01:25:05 -07001291static int ext4_writeback_write_end(struct file *file,
1292 struct address_space *mapping,
1293 loff_t pos, unsigned len, unsigned copied,
1294 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001295{
Mingming Cao617ba132006-10-11 01:20:53 -07001296 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001297 struct inode *inode = file->f_mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001298 int ret = 0, ret2;
1299 loff_t new_i_size;
1300
Nick Pigginbfc1af62007-10-16 01:25:05 -07001301 new_i_size = pos + copied;
Mingming Cao617ba132006-10-11 01:20:53 -07001302 if (new_i_size > EXT4_I(inode)->i_disksize)
1303 EXT4_I(inode)->i_disksize = new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001304
Nick Pigginbfc1af62007-10-16 01:25:05 -07001305 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1306 page, fsdata);
1307 if (copied < 0)
1308 ret = copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001309
Mingming Cao617ba132006-10-11 01:20:53 -07001310 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001311 if (!ret)
1312 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001313 unlock_page(page);
1314 page_cache_release(page);
1315
1316 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001317}
1318
Nick Pigginbfc1af62007-10-16 01:25:05 -07001319static int ext4_journalled_write_end(struct file *file,
1320 struct address_space *mapping,
1321 loff_t pos, unsigned len, unsigned copied,
1322 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001323{
Mingming Cao617ba132006-10-11 01:20:53 -07001324 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001325 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001326 int ret = 0, ret2;
1327 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001328 unsigned from, to;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001329
Nick Pigginbfc1af62007-10-16 01:25:05 -07001330 from = pos & (PAGE_CACHE_SIZE - 1);
1331 to = from + len;
1332
1333 if (copied < len) {
1334 if (!PageUptodate(page))
1335 copied = 0;
1336 page_zero_new_buffers(page, from+copied, to);
1337 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001338
1339 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001340 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001341 if (!partial)
1342 SetPageUptodate(page);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001343 if (pos+copied > inode->i_size)
1344 i_size_write(inode, pos+copied);
Mingming Cao617ba132006-10-11 01:20:53 -07001345 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1346 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1347 EXT4_I(inode)->i_disksize = inode->i_size;
1348 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001349 if (!ret)
1350 ret = ret2;
1351 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001352
Mingming Cao617ba132006-10-11 01:20:53 -07001353 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001354 if (!ret)
1355 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001356 unlock_page(page);
1357 page_cache_release(page);
1358
1359 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001360}
1361
1362/*
1363 * bmap() is special. It gets used by applications such as lilo and by
1364 * the swapper to find the on-disk block of a specific piece of data.
1365 *
1366 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07001367 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001368 * filesystem and enables swap, then they may get a nasty shock when the
1369 * data getting swapped to that swapfile suddenly gets overwritten by
1370 * the original zero's written out previously to the journal and
1371 * awaiting writeback in the kernel's buffer cache.
1372 *
1373 * So, if we see any bmap calls here on a modified, data-journaled file,
1374 * take extra steps to flush any blocks which might be in the cache.
1375 */
Mingming Cao617ba132006-10-11 01:20:53 -07001376static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001377{
1378 struct inode *inode = mapping->host;
1379 journal_t *journal;
1380 int err;
1381
Mingming Cao617ba132006-10-11 01:20:53 -07001382 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001383 /*
1384 * This is a REALLY heavyweight approach, but the use of
1385 * bmap on dirty files is expected to be extremely rare:
1386 * only if we run lilo or swapon on a freshly made file
1387 * do we expect this to happen.
1388 *
1389 * (bmap requires CAP_SYS_RAWIO so this does not
1390 * represent an unprivileged user DOS attack --- we'd be
1391 * in trouble if mortal users could trigger this path at
1392 * will.)
1393 *
Mingming Cao617ba132006-10-11 01:20:53 -07001394 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001395 * regular files. If somebody wants to bmap a directory
1396 * or symlink and gets confused because the buffer
1397 * hasn't yet been flushed to disk, they deserve
1398 * everything they get.
1399 */
1400
Mingming Cao617ba132006-10-11 01:20:53 -07001401 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1402 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07001403 jbd2_journal_lock_updates(journal);
1404 err = jbd2_journal_flush(journal);
1405 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001406
1407 if (err)
1408 return 0;
1409 }
1410
Mingming Cao617ba132006-10-11 01:20:53 -07001411 return generic_block_bmap(mapping,block,ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001412}
1413
1414static int bget_one(handle_t *handle, struct buffer_head *bh)
1415{
1416 get_bh(bh);
1417 return 0;
1418}
1419
1420static int bput_one(handle_t *handle, struct buffer_head *bh)
1421{
1422 put_bh(bh);
1423 return 0;
1424}
1425
Mingming Caodab291a2006-10-11 01:21:01 -07001426static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001427{
1428 if (buffer_mapped(bh))
Mingming Cao617ba132006-10-11 01:20:53 -07001429 return ext4_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001430 return 0;
1431}
1432
1433/*
1434 * Note that we always start a transaction even if we're not journalling
1435 * data. This is to preserve ordering: any hole instantiation within
Mingming Cao617ba132006-10-11 01:20:53 -07001436 * __block_write_full_page -> ext4_get_block() should be journalled
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001437 * along with the data so we don't crash and then get metadata which
1438 * refers to old data.
1439 *
1440 * In all journalling modes block_write_full_page() will start the I/O.
1441 *
1442 * Problem:
1443 *
Mingming Cao617ba132006-10-11 01:20:53 -07001444 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1445 * ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001446 *
1447 * Similar for:
1448 *
Mingming Cao617ba132006-10-11 01:20:53 -07001449 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001450 *
Mingming Cao617ba132006-10-11 01:20:53 -07001451 * Same applies to ext4_get_block(). We will deadlock on various things like
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001452 * lock_journal and i_truncate_mutex.
1453 *
1454 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1455 * allocations fail.
1456 *
1457 * 16May01: If we're reentered then journal_current_handle() will be
1458 * non-zero. We simply *return*.
1459 *
1460 * 1 July 2001: @@@ FIXME:
1461 * In journalled data mode, a data buffer may be metadata against the
1462 * current transaction. But the same file is part of a shared mapping
1463 * and someone does a writepage() on it.
1464 *
1465 * We will move the buffer onto the async_data list, but *after* it has
1466 * been dirtied. So there's a small window where we have dirty data on
1467 * BJ_Metadata.
1468 *
1469 * Note that this only applies to the last partial page in the file. The
1470 * bit which block_write_full_page() uses prepare/commit for. (That's
1471 * broken code anyway: it's wrong for msync()).
1472 *
1473 * It's a rare case: affects the final partial page, for journalled data
1474 * where the file is subject to bith write() and writepage() in the same
1475 * transction. To fix it we'll need a custom block_write_full_page().
1476 * We'll probably need that anyway for journalling writepage() output.
1477 *
1478 * We don't honour synchronous mounts for writepage(). That would be
1479 * disastrous. Any write() or metadata operation will sync the fs for
1480 * us.
1481 *
1482 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1483 * we don't need to open a transaction here.
1484 */
Mingming Cao617ba132006-10-11 01:20:53 -07001485static int ext4_ordered_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001486 struct writeback_control *wbc)
1487{
1488 struct inode *inode = page->mapping->host;
1489 struct buffer_head *page_bufs;
1490 handle_t *handle = NULL;
1491 int ret = 0;
1492 int err;
1493
1494 J_ASSERT(PageLocked(page));
1495
1496 /*
1497 * We give up here if we're reentered, because it might be for a
1498 * different filesystem.
1499 */
Mingming Cao617ba132006-10-11 01:20:53 -07001500 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501 goto out_fail;
1502
Mingming Cao617ba132006-10-11 01:20:53 -07001503 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001504
1505 if (IS_ERR(handle)) {
1506 ret = PTR_ERR(handle);
1507 goto out_fail;
1508 }
1509
1510 if (!page_has_buffers(page)) {
1511 create_empty_buffers(page, inode->i_sb->s_blocksize,
1512 (1 << BH_Dirty)|(1 << BH_Uptodate));
1513 }
1514 page_bufs = page_buffers(page);
1515 walk_page_buffers(handle, page_bufs, 0,
1516 PAGE_CACHE_SIZE, NULL, bget_one);
1517
Mingming Cao617ba132006-10-11 01:20:53 -07001518 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001519
1520 /*
1521 * The page can become unlocked at any point now, and
1522 * truncate can then come in and change things. So we
1523 * can't touch *page from now on. But *page_bufs is
1524 * safe due to elevated refcount.
1525 */
1526
1527 /*
1528 * And attach them to the current transaction. But only if
1529 * block_write_full_page() succeeded. Otherwise they are unmapped,
1530 * and generally junk.
1531 */
1532 if (ret == 0) {
1533 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
Mingming Caodab291a2006-10-11 01:21:01 -07001534 NULL, jbd2_journal_dirty_data_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001535 if (!ret)
1536 ret = err;
1537 }
1538 walk_page_buffers(handle, page_bufs, 0,
1539 PAGE_CACHE_SIZE, NULL, bput_one);
Mingming Cao617ba132006-10-11 01:20:53 -07001540 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541 if (!ret)
1542 ret = err;
1543 return ret;
1544
1545out_fail:
1546 redirty_page_for_writepage(wbc, page);
1547 unlock_page(page);
1548 return ret;
1549}
1550
Mingming Cao617ba132006-10-11 01:20:53 -07001551static int ext4_writeback_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001552 struct writeback_control *wbc)
1553{
1554 struct inode *inode = page->mapping->host;
1555 handle_t *handle = NULL;
1556 int ret = 0;
1557 int err;
1558
Mingming Cao617ba132006-10-11 01:20:53 -07001559 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 goto out_fail;
1561
Mingming Cao617ba132006-10-11 01:20:53 -07001562 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001563 if (IS_ERR(handle)) {
1564 ret = PTR_ERR(handle);
1565 goto out_fail;
1566 }
1567
Mingming Cao617ba132006-10-11 01:20:53 -07001568 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1569 ret = nobh_writepage(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001570 else
Mingming Cao617ba132006-10-11 01:20:53 -07001571 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001572
Mingming Cao617ba132006-10-11 01:20:53 -07001573 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001574 if (!ret)
1575 ret = err;
1576 return ret;
1577
1578out_fail:
1579 redirty_page_for_writepage(wbc, page);
1580 unlock_page(page);
1581 return ret;
1582}
1583
Mingming Cao617ba132006-10-11 01:20:53 -07001584static int ext4_journalled_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001585 struct writeback_control *wbc)
1586{
1587 struct inode *inode = page->mapping->host;
1588 handle_t *handle = NULL;
1589 int ret = 0;
1590 int err;
1591
Mingming Cao617ba132006-10-11 01:20:53 -07001592 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001593 goto no_write;
1594
Mingming Cao617ba132006-10-11 01:20:53 -07001595 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001596 if (IS_ERR(handle)) {
1597 ret = PTR_ERR(handle);
1598 goto no_write;
1599 }
1600
1601 if (!page_has_buffers(page) || PageChecked(page)) {
1602 /*
1603 * It's mmapped pagecache. Add buffers and journal it. There
1604 * doesn't seem much point in redirtying the page here.
1605 */
1606 ClearPageChecked(page);
1607 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
Mingming Cao617ba132006-10-11 01:20:53 -07001608 ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001609 if (ret != 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001610 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001611 goto out_unlock;
1612 }
1613 ret = walk_page_buffers(handle, page_buffers(page), 0,
1614 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1615
1616 err = walk_page_buffers(handle, page_buffers(page), 0,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001617 PAGE_CACHE_SIZE, NULL, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001618 if (ret == 0)
1619 ret = err;
Mingming Cao617ba132006-10-11 01:20:53 -07001620 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001621 unlock_page(page);
1622 } else {
1623 /*
1624 * It may be a page full of checkpoint-mode buffers. We don't
1625 * really know unless we go poke around in the buffer_heads.
1626 * But block_write_full_page will do the right thing.
1627 */
Mingming Cao617ba132006-10-11 01:20:53 -07001628 ret = block_write_full_page(page, ext4_get_block, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001629 }
Mingming Cao617ba132006-10-11 01:20:53 -07001630 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001631 if (!ret)
1632 ret = err;
1633out:
1634 return ret;
1635
1636no_write:
1637 redirty_page_for_writepage(wbc, page);
1638out_unlock:
1639 unlock_page(page);
1640 goto out;
1641}
1642
Mingming Cao617ba132006-10-11 01:20:53 -07001643static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001644{
Mingming Cao617ba132006-10-11 01:20:53 -07001645 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001646}
1647
1648static int
Mingming Cao617ba132006-10-11 01:20:53 -07001649ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001650 struct list_head *pages, unsigned nr_pages)
1651{
Mingming Cao617ba132006-10-11 01:20:53 -07001652 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001653}
1654
Mingming Cao617ba132006-10-11 01:20:53 -07001655static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001656{
Mingming Cao617ba132006-10-11 01:20:53 -07001657 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001658
1659 /*
1660 * If it's a full truncate we just forget about the pending dirtying
1661 */
1662 if (offset == 0)
1663 ClearPageChecked(page);
1664
Mingming Caodab291a2006-10-11 01:21:01 -07001665 jbd2_journal_invalidatepage(journal, page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001666}
1667
Mingming Cao617ba132006-10-11 01:20:53 -07001668static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001669{
Mingming Cao617ba132006-10-11 01:20:53 -07001670 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001671
1672 WARN_ON(PageChecked(page));
1673 if (!page_has_buffers(page))
1674 return 0;
Mingming Caodab291a2006-10-11 01:21:01 -07001675 return jbd2_journal_try_to_free_buffers(journal, page, wait);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001676}
1677
1678/*
1679 * If the O_DIRECT write will extend the file then add this inode to the
1680 * orphan list. So recovery will truncate it back to the original size
1681 * if the machine crashes during the write.
1682 *
1683 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1684 * crashes then stale disk data _may_ be exposed inside the file.
1685 */
Mingming Cao617ba132006-10-11 01:20:53 -07001686static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001687 const struct iovec *iov, loff_t offset,
1688 unsigned long nr_segs)
1689{
1690 struct file *file = iocb->ki_filp;
1691 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07001692 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001693 handle_t *handle = NULL;
1694 ssize_t ret;
1695 int orphan = 0;
1696 size_t count = iov_length(iov, nr_segs);
1697
1698 if (rw == WRITE) {
1699 loff_t final_size = offset + count;
1700
Mingming Cao617ba132006-10-11 01:20:53 -07001701 handle = ext4_journal_start(inode, DIO_CREDITS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001702 if (IS_ERR(handle)) {
1703 ret = PTR_ERR(handle);
1704 goto out;
1705 }
1706 if (final_size > inode->i_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001707 ret = ext4_orphan_add(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001708 if (ret)
1709 goto out_stop;
1710 orphan = 1;
1711 ei->i_disksize = inode->i_size;
1712 }
1713 }
1714
1715 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1716 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07001717 ext4_get_block, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001718
1719 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001720 * Reacquire the handle: ext4_get_block() can restart the transaction
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001721 */
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001722 handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001723
1724out_stop:
1725 if (handle) {
1726 int err;
1727
1728 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07001729 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001730 if (orphan && ret > 0) {
1731 loff_t end = offset + ret;
1732 if (end > inode->i_size) {
1733 ei->i_disksize = end;
1734 i_size_write(inode, end);
1735 /*
1736 * We're going to return a positive `ret'
1737 * here due to non-zero-length I/O, so there's
1738 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07001739 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001740 * ignore it.
1741 */
Mingming Cao617ba132006-10-11 01:20:53 -07001742 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001743 }
1744 }
Mingming Cao617ba132006-10-11 01:20:53 -07001745 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001746 if (ret == 0)
1747 ret = err;
1748 }
1749out:
1750 return ret;
1751}
1752
1753/*
Mingming Cao617ba132006-10-11 01:20:53 -07001754 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001755 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1756 * much here because ->set_page_dirty is called under VFS locks. The page is
1757 * not necessarily locked.
1758 *
1759 * We cannot just dirty the page and leave attached buffers clean, because the
1760 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1761 * or jbddirty because all the journalling code will explode.
1762 *
1763 * So what we do is to mark the page "pending dirty" and next time writepage
1764 * is called, propagate that into the buffers appropriately.
1765 */
Mingming Cao617ba132006-10-11 01:20:53 -07001766static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001767{
1768 SetPageChecked(page);
1769 return __set_page_dirty_nobuffers(page);
1770}
1771
Mingming Cao617ba132006-10-11 01:20:53 -07001772static const struct address_space_operations ext4_ordered_aops = {
1773 .readpage = ext4_readpage,
1774 .readpages = ext4_readpages,
1775 .writepage = ext4_ordered_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001776 .sync_page = block_sync_page,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001777 .write_begin = ext4_write_begin,
1778 .write_end = ext4_ordered_write_end,
Mingming Cao617ba132006-10-11 01:20:53 -07001779 .bmap = ext4_bmap,
1780 .invalidatepage = ext4_invalidatepage,
1781 .releasepage = ext4_releasepage,
1782 .direct_IO = ext4_direct_IO,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001783 .migratepage = buffer_migrate_page,
1784};
1785
Mingming Cao617ba132006-10-11 01:20:53 -07001786static const struct address_space_operations ext4_writeback_aops = {
1787 .readpage = ext4_readpage,
1788 .readpages = ext4_readpages,
1789 .writepage = ext4_writeback_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001790 .sync_page = block_sync_page,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001791 .write_begin = ext4_write_begin,
1792 .write_end = ext4_writeback_write_end,
Mingming Cao617ba132006-10-11 01:20:53 -07001793 .bmap = ext4_bmap,
1794 .invalidatepage = ext4_invalidatepage,
1795 .releasepage = ext4_releasepage,
1796 .direct_IO = ext4_direct_IO,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001797 .migratepage = buffer_migrate_page,
1798};
1799
Mingming Cao617ba132006-10-11 01:20:53 -07001800static const struct address_space_operations ext4_journalled_aops = {
1801 .readpage = ext4_readpage,
1802 .readpages = ext4_readpages,
1803 .writepage = ext4_journalled_writepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001804 .sync_page = block_sync_page,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001805 .write_begin = ext4_write_begin,
1806 .write_end = ext4_journalled_write_end,
Mingming Cao617ba132006-10-11 01:20:53 -07001807 .set_page_dirty = ext4_journalled_set_page_dirty,
1808 .bmap = ext4_bmap,
1809 .invalidatepage = ext4_invalidatepage,
1810 .releasepage = ext4_releasepage,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001811};
1812
Mingming Cao617ba132006-10-11 01:20:53 -07001813void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001814{
Mingming Cao617ba132006-10-11 01:20:53 -07001815 if (ext4_should_order_data(inode))
1816 inode->i_mapping->a_ops = &ext4_ordered_aops;
1817 else if (ext4_should_writeback_data(inode))
1818 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001819 else
Mingming Cao617ba132006-10-11 01:20:53 -07001820 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001821}
1822
1823/*
Mingming Cao617ba132006-10-11 01:20:53 -07001824 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001825 * up to the end of the block which corresponds to `from'.
1826 * This required during truncate. We need to physically zero the tail end
1827 * of that block so it doesn't yield old data if the file is later grown.
1828 */
Alex Tomasa86c6182006-10-11 01:21:03 -07001829int ext4_block_truncate_page(handle_t *handle, struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001830 struct address_space *mapping, loff_t from)
1831{
Mingming Cao617ba132006-10-11 01:20:53 -07001832 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001833 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001834 unsigned blocksize, length, pos;
1835 ext4_lblk_t iblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001836 struct inode *inode = mapping->host;
1837 struct buffer_head *bh;
1838 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001839
1840 blocksize = inode->i_sb->s_blocksize;
1841 length = blocksize - (offset & (blocksize - 1));
1842 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1843
1844 /*
1845 * For "nobh" option, we can only work if we don't need to
1846 * read-in the page - otherwise we create buffers to do the IO.
1847 */
1848 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
Mingming Cao617ba132006-10-11 01:20:53 -07001849 ext4_should_writeback_data(inode) && PageUptodate(page)) {
Eric Sandeenfc0e15a2007-07-18 09:20:44 -04001850 zero_user_page(page, offset, length, KM_USER0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001851 set_page_dirty(page);
1852 goto unlock;
1853 }
1854
1855 if (!page_has_buffers(page))
1856 create_empty_buffers(page, blocksize, 0);
1857
1858 /* Find the buffer that contains "offset" */
1859 bh = page_buffers(page);
1860 pos = blocksize;
1861 while (offset >= pos) {
1862 bh = bh->b_this_page;
1863 iblock++;
1864 pos += blocksize;
1865 }
1866
1867 err = 0;
1868 if (buffer_freed(bh)) {
1869 BUFFER_TRACE(bh, "freed: skip");
1870 goto unlock;
1871 }
1872
1873 if (!buffer_mapped(bh)) {
1874 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07001875 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001876 /* unmapped? It's a hole - nothing to do */
1877 if (!buffer_mapped(bh)) {
1878 BUFFER_TRACE(bh, "still unmapped");
1879 goto unlock;
1880 }
1881 }
1882
1883 /* Ok, it's mapped. Make sure it's up-to-date */
1884 if (PageUptodate(page))
1885 set_buffer_uptodate(bh);
1886
1887 if (!buffer_uptodate(bh)) {
1888 err = -EIO;
1889 ll_rw_block(READ, 1, &bh);
1890 wait_on_buffer(bh);
1891 /* Uhhuh. Read error. Complain and punt. */
1892 if (!buffer_uptodate(bh))
1893 goto unlock;
1894 }
1895
Mingming Cao617ba132006-10-11 01:20:53 -07001896 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001897 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07001898 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001899 if (err)
1900 goto unlock;
1901 }
1902
Eric Sandeenfc0e15a2007-07-18 09:20:44 -04001903 zero_user_page(page, offset, length, KM_USER0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001904
1905 BUFFER_TRACE(bh, "zeroed end of block");
1906
1907 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001908 if (ext4_should_journal_data(inode)) {
1909 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001910 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07001911 if (ext4_should_order_data(inode))
1912 err = ext4_journal_dirty_data(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001913 mark_buffer_dirty(bh);
1914 }
1915
1916unlock:
1917 unlock_page(page);
1918 page_cache_release(page);
1919 return err;
1920}
1921
1922/*
1923 * Probably it should be a library function... search for first non-zero word
1924 * or memcmp with zero_page, whatever is better for particular architecture.
1925 * Linus?
1926 */
1927static inline int all_zeroes(__le32 *p, __le32 *q)
1928{
1929 while (p < q)
1930 if (*p++)
1931 return 0;
1932 return 1;
1933}
1934
1935/**
Mingming Cao617ba132006-10-11 01:20:53 -07001936 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001937 * @inode: inode in question
1938 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07001939 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001940 * @chain: place to store the pointers to partial indirect blocks
1941 * @top: place to the (detached) top of branch
1942 *
Mingming Cao617ba132006-10-11 01:20:53 -07001943 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001944 *
1945 * When we do truncate() we may have to clean the ends of several
1946 * indirect blocks but leave the blocks themselves alive. Block is
1947 * partially truncated if some data below the new i_size is refered
1948 * from it (and it is on the path to the first completely truncated
1949 * data block, indeed). We have to free the top of that path along
1950 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07001951 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001952 * finishes, we may safely do the latter, but top of branch may
1953 * require special attention - pageout below the truncation point
1954 * might try to populate it.
1955 *
1956 * We atomically detach the top of branch from the tree, store the
1957 * block number of its root in *@top, pointers to buffer_heads of
1958 * partially truncated blocks - in @chain[].bh and pointers to
1959 * their last elements that should not be removed - in
1960 * @chain[].p. Return value is the pointer to last filled element
1961 * of @chain.
1962 *
1963 * The work left to caller to do the actual freeing of subtrees:
1964 * a) free the subtree starting from *@top
1965 * b) free the subtrees whose roots are stored in
1966 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1967 * c) free the subtrees growing from the inode past the @chain[0].
1968 * (no partially truncated stuff there). */
1969
Mingming Cao617ba132006-10-11 01:20:53 -07001970static Indirect *ext4_find_shared(struct inode *inode, int depth,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001971 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001972{
1973 Indirect *partial, *p;
1974 int k, err;
1975
1976 *top = 0;
1977 /* Make k index the deepest non-null offest + 1 */
1978 for (k = depth; k > 1 && !offsets[k-1]; k--)
1979 ;
Mingming Cao617ba132006-10-11 01:20:53 -07001980 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001981 /* Writer: pointers */
1982 if (!partial)
1983 partial = chain + k-1;
1984 /*
1985 * If the branch acquired continuation since we've looked at it -
1986 * fine, it should all survive and (new) top doesn't belong to us.
1987 */
1988 if (!partial->key && *partial->p)
1989 /* Writer: end */
1990 goto no_top;
1991 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1992 ;
1993 /*
1994 * OK, we've found the last block that must survive. The rest of our
1995 * branch should be detached before unlocking. However, if that rest
1996 * of branch is all ours and does not grow immediately from the inode
1997 * it's easier to cheat and just decrement partial->p.
1998 */
1999 if (p == chain + k - 1 && p > chain) {
2000 p->p--;
2001 } else {
2002 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07002003 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002004#if 0
2005 *p->p = 0;
2006#endif
2007 }
2008 /* Writer: end */
2009
2010 while(partial > p) {
2011 brelse(partial->bh);
2012 partial--;
2013 }
2014no_top:
2015 return partial;
2016}
2017
2018/*
2019 * Zero a number of block pointers in either an inode or an indirect block.
2020 * If we restart the transaction we must again get write access to the
2021 * indirect block for further modification.
2022 *
2023 * We release `count' blocks on disk, but (last - first) may be greater
2024 * than `count' because there can be holes in there.
2025 */
Mingming Cao617ba132006-10-11 01:20:53 -07002026static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
2027 struct buffer_head *bh, ext4_fsblk_t block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002028 unsigned long count, __le32 *first, __le32 *last)
2029{
2030 __le32 *p;
2031 if (try_to_extend_transaction(handle, inode)) {
2032 if (bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002033 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2034 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002035 }
Mingming Cao617ba132006-10-11 01:20:53 -07002036 ext4_mark_inode_dirty(handle, inode);
2037 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002038 if (bh) {
2039 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002040 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002041 }
2042 }
2043
2044 /*
2045 * Any buffers which are on the journal will be in memory. We find
Mingming Caodab291a2006-10-11 01:21:01 -07002046 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002047 * on them. We've already detached each block from the file, so
Mingming Caodab291a2006-10-11 01:21:01 -07002048 * bforget() in jbd2_journal_forget() should be safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002049 *
Mingming Caodab291a2006-10-11 01:21:01 -07002050 * AKPM: turn on bforget in jbd2_journal_forget()!!!
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002051 */
2052 for (p = first; p < last; p++) {
2053 u32 nr = le32_to_cpu(*p);
2054 if (nr) {
2055 struct buffer_head *bh;
2056
2057 *p = 0;
2058 bh = sb_find_get_block(inode->i_sb, nr);
Mingming Cao617ba132006-10-11 01:20:53 -07002059 ext4_forget(handle, 0, inode, bh, nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002060 }
2061 }
2062
Mingming Cao617ba132006-10-11 01:20:53 -07002063 ext4_free_blocks(handle, inode, block_to_free, count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002064}
2065
2066/**
Mingming Cao617ba132006-10-11 01:20:53 -07002067 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002068 * @handle: handle for this transaction
2069 * @inode: inode we are dealing with
2070 * @this_bh: indirect buffer_head which contains *@first and *@last
2071 * @first: array of block numbers
2072 * @last: points immediately past the end of array
2073 *
2074 * We are freeing all blocks refered from that array (numbers are stored as
2075 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2076 *
2077 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2078 * blocks are contiguous then releasing them at one time will only affect one
2079 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2080 * actually use a lot of journal space.
2081 *
2082 * @this_bh will be %NULL if @first and @last point into the inode's direct
2083 * block pointers.
2084 */
Mingming Cao617ba132006-10-11 01:20:53 -07002085static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002086 struct buffer_head *this_bh,
2087 __le32 *first, __le32 *last)
2088{
Mingming Cao617ba132006-10-11 01:20:53 -07002089 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002090 unsigned long count = 0; /* Number of blocks in the run */
2091 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2092 corresponding to
2093 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07002094 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002095 __le32 *p; /* Pointer into inode/ind
2096 for current block */
2097 int err;
2098
2099 if (this_bh) { /* For indirect block */
2100 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002101 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002102 /* Important: if we can't update the indirect pointers
2103 * to the blocks, we can't free them. */
2104 if (err)
2105 return;
2106 }
2107
2108 for (p = first; p < last; p++) {
2109 nr = le32_to_cpu(*p);
2110 if (nr) {
2111 /* accumulate blocks to free if they're contiguous */
2112 if (count == 0) {
2113 block_to_free = nr;
2114 block_to_free_p = p;
2115 count = 1;
2116 } else if (nr == block_to_free + count) {
2117 count++;
2118 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002119 ext4_clear_blocks(handle, inode, this_bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002120 block_to_free,
2121 count, block_to_free_p, p);
2122 block_to_free = nr;
2123 block_to_free_p = p;
2124 count = 1;
2125 }
2126 }
2127 }
2128
2129 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07002130 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002131 count, block_to_free_p, p);
2132
2133 if (this_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002134 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2135 ext4_journal_dirty_metadata(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002136 }
2137}
2138
2139/**
Mingming Cao617ba132006-10-11 01:20:53 -07002140 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002141 * @handle: JBD handle for this transaction
2142 * @inode: inode we are dealing with
2143 * @parent_bh: the buffer_head which contains *@first and *@last
2144 * @first: array of block numbers
2145 * @last: pointer immediately past the end of array
2146 * @depth: depth of the branches to free
2147 *
2148 * We are freeing all blocks refered from these branches (numbers are
2149 * stored as little-endian 32-bit) and updating @inode->i_blocks
2150 * appropriately.
2151 */
Mingming Cao617ba132006-10-11 01:20:53 -07002152static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002153 struct buffer_head *parent_bh,
2154 __le32 *first, __le32 *last, int depth)
2155{
Mingming Cao617ba132006-10-11 01:20:53 -07002156 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002157 __le32 *p;
2158
2159 if (is_handle_aborted(handle))
2160 return;
2161
2162 if (depth--) {
2163 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002164 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002165 p = last;
2166 while (--p >= first) {
2167 nr = le32_to_cpu(*p);
2168 if (!nr)
2169 continue; /* A hole */
2170
2171 /* Go read the buffer for the next level down */
2172 bh = sb_bread(inode->i_sb, nr);
2173
2174 /*
2175 * A read failure? Report error and clear slot
2176 * (should be rare).
2177 */
2178 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002179 ext4_error(inode->i_sb, "ext4_free_branches",
Mingming Cao2ae02102006-10-11 01:21:11 -07002180 "Read failure, inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002181 inode->i_ino, nr);
2182 continue;
2183 }
2184
2185 /* This zaps the entire block. Bottom up. */
2186 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07002187 ext4_free_branches(handle, inode, bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002188 (__le32*)bh->b_data,
2189 (__le32*)bh->b_data + addr_per_block,
2190 depth);
2191
2192 /*
2193 * We've probably journalled the indirect block several
2194 * times during the truncate. But it's no longer
2195 * needed and we now drop it from the transaction via
Mingming Caodab291a2006-10-11 01:21:01 -07002196 * jbd2_journal_revoke().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002197 *
2198 * That's easy if it's exclusively part of this
2199 * transaction. But if it's part of the committing
Mingming Caodab291a2006-10-11 01:21:01 -07002200 * transaction then jbd2_journal_forget() will simply
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002201 * brelse() it. That means that if the underlying
Mingming Cao617ba132006-10-11 01:20:53 -07002202 * block is reallocated in ext4_get_block(),
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002203 * unmap_underlying_metadata() will find this block
2204 * and will try to get rid of it. damn, damn.
2205 *
2206 * If this block has already been committed to the
2207 * journal, a revoke record will be written. And
2208 * revoke records must be emitted *before* clearing
2209 * this block's bit in the bitmaps.
2210 */
Mingming Cao617ba132006-10-11 01:20:53 -07002211 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002212
2213 /*
2214 * Everything below this this pointer has been
2215 * released. Now let this top-of-subtree go.
2216 *
2217 * We want the freeing of this indirect block to be
2218 * atomic in the journal with the updating of the
2219 * bitmap block which owns it. So make some room in
2220 * the journal.
2221 *
2222 * We zero the parent pointer *after* freeing its
2223 * pointee in the bitmaps, so if extend_transaction()
2224 * for some reason fails to put the bitmap changes and
2225 * the release into the same transaction, recovery
2226 * will merely complain about releasing a free block,
2227 * rather than leaking blocks.
2228 */
2229 if (is_handle_aborted(handle))
2230 return;
2231 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002232 ext4_mark_inode_dirty(handle, inode);
2233 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002234 }
2235
Mingming Cao617ba132006-10-11 01:20:53 -07002236 ext4_free_blocks(handle, inode, nr, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002237
2238 if (parent_bh) {
2239 /*
2240 * The block which we have just freed is
2241 * pointed to by an indirect block: journal it
2242 */
2243 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002244 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002245 parent_bh)){
2246 *p = 0;
2247 BUFFER_TRACE(parent_bh,
Mingming Cao617ba132006-10-11 01:20:53 -07002248 "call ext4_journal_dirty_metadata");
2249 ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002250 parent_bh);
2251 }
2252 }
2253 }
2254 } else {
2255 /* We have reached the bottom of the tree. */
2256 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07002257 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002258 }
2259}
2260
2261/*
Mingming Cao617ba132006-10-11 01:20:53 -07002262 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002263 *
Mingming Cao617ba132006-10-11 01:20:53 -07002264 * We block out ext4_get_block() block instantiations across the entire
2265 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002266 * simultaneously on behalf of the same inode.
2267 *
2268 * As we work through the truncate and commmit bits of it to the journal there
2269 * is one core, guiding principle: the file's tree must always be consistent on
2270 * disk. We must be able to restart the truncate after a crash.
2271 *
2272 * The file's tree may be transiently inconsistent in memory (although it
2273 * probably isn't), but whenever we close off and commit a journal transaction,
2274 * the contents of (the filesystem + the journal) must be consistent and
2275 * restartable. It's pretty simple, really: bottom up, right to left (although
2276 * left-to-right works OK too).
2277 *
2278 * Note that at recovery time, journal replay occurs *before* the restart of
2279 * truncate against the orphan inode list.
2280 *
2281 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07002282 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002283 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07002284 * ext4_truncate() to have another go. So there will be instantiated blocks
2285 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002286 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07002287 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002288 */
Mingming Cao617ba132006-10-11 01:20:53 -07002289void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002290{
2291 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07002292 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002293 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07002294 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002295 struct address_space *mapping = inode->i_mapping;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002296 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002297 Indirect chain[4];
2298 Indirect *partial;
2299 __le32 nr = 0;
2300 int n;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002301 ext4_lblk_t last_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002302 unsigned blocksize = inode->i_sb->s_blocksize;
2303 struct page *page;
2304
2305 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2306 S_ISLNK(inode->i_mode)))
2307 return;
Mingming Cao617ba132006-10-11 01:20:53 -07002308 if (ext4_inode_is_fast_symlink(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002309 return;
2310 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2311 return;
2312
2313 /*
2314 * We have to lock the EOF page here, because lock_page() nests
Mingming Caodab291a2006-10-11 01:21:01 -07002315 * outside jbd2_journal_start().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002316 */
2317 if ((inode->i_size & (blocksize - 1)) == 0) {
2318 /* Block boundary? Nothing to do */
2319 page = NULL;
2320 } else {
2321 page = grab_cache_page(mapping,
2322 inode->i_size >> PAGE_CACHE_SHIFT);
2323 if (!page)
2324 return;
2325 }
2326
Alex Tomasa86c6182006-10-11 01:21:03 -07002327 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
2328 return ext4_ext_truncate(inode, page);
2329
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002330 handle = start_transaction(inode);
2331 if (IS_ERR(handle)) {
2332 if (page) {
2333 clear_highpage(page);
2334 flush_dcache_page(page);
2335 unlock_page(page);
2336 page_cache_release(page);
2337 }
2338 return; /* AKPM: return what? */
2339 }
2340
2341 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07002342 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002343
2344 if (page)
Mingming Cao617ba132006-10-11 01:20:53 -07002345 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002346
Mingming Cao617ba132006-10-11 01:20:53 -07002347 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002348 if (n == 0)
2349 goto out_stop; /* error */
2350
2351 /*
2352 * OK. This truncate is going to happen. We add the inode to the
2353 * orphan list, so that if this truncate spans multiple transactions,
2354 * and we crash, we will resume the truncate when the filesystem
2355 * recovers. It also marks the inode dirty, to catch the new size.
2356 *
2357 * Implication: the file must always be in a sane, consistent
2358 * truncatable state while each transaction commits.
2359 */
Mingming Cao617ba132006-10-11 01:20:53 -07002360 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002361 goto out_stop;
2362
2363 /*
2364 * The orphan list entry will now protect us from any crash which
2365 * occurs before the truncate completes, so it is now safe to propagate
2366 * the new, shorter inode size (held for now in i_size) into the
2367 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07002368 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002369 */
2370 ei->i_disksize = inode->i_size;
2371
2372 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002373 * From here we block out all ext4_get_block() callers who want to
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002374 * modify the block allocation tree.
2375 */
2376 mutex_lock(&ei->truncate_mutex);
2377
2378 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07002379 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2380 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002381 goto do_indirects;
2382 }
2383
Mingming Cao617ba132006-10-11 01:20:53 -07002384 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002385 /* Kill the top of shared branch (not detached) */
2386 if (nr) {
2387 if (partial == chain) {
2388 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07002389 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002390 &nr, &nr+1, (chain+n-1) - partial);
2391 *partial->p = 0;
2392 /*
2393 * We mark the inode dirty prior to restart,
2394 * and prior to stop. No need for it here.
2395 */
2396 } else {
2397 /* Shared branch grows from an indirect block */
2398 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002399 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002400 partial->p,
2401 partial->p+1, (chain+n-1) - partial);
2402 }
2403 }
2404 /* Clear the ends of indirect blocks on the shared branch */
2405 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07002406 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002407 (__le32*)partial->bh->b_data+addr_per_block,
2408 (chain+n-1) - partial);
2409 BUFFER_TRACE(partial->bh, "call brelse");
2410 brelse (partial->bh);
2411 partial--;
2412 }
2413do_indirects:
2414 /* Kill the remaining (whole) subtrees */
2415 switch (offsets[0]) {
2416 default:
Mingming Cao617ba132006-10-11 01:20:53 -07002417 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002418 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002419 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2420 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002421 }
Mingming Cao617ba132006-10-11 01:20:53 -07002422 case EXT4_IND_BLOCK:
2423 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002424 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002425 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2426 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002427 }
Mingming Cao617ba132006-10-11 01:20:53 -07002428 case EXT4_DIND_BLOCK:
2429 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002430 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07002431 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2432 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002433 }
Mingming Cao617ba132006-10-11 01:20:53 -07002434 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002435 ;
2436 }
2437
Mingming Cao617ba132006-10-11 01:20:53 -07002438 ext4_discard_reservation(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002439
2440 mutex_unlock(&ei->truncate_mutex);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002441 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002442 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002443
2444 /*
2445 * In a multi-transaction truncate, we only make the final transaction
2446 * synchronous
2447 */
2448 if (IS_SYNC(inode))
2449 handle->h_sync = 1;
2450out_stop:
2451 /*
2452 * If this was a simple ftruncate(), and the file will remain alive
2453 * then we need to clear up the orphan record which we created above.
2454 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07002455 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002456 * orphan info for us.
2457 */
2458 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002459 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002460
Mingming Cao617ba132006-10-11 01:20:53 -07002461 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002462}
2463
Mingming Cao617ba132006-10-11 01:20:53 -07002464static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2465 unsigned long ino, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002466{
Avantika Mathurfd2d4292008-01-28 23:58:27 -05002467 unsigned long desc, group_desc;
2468 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002469 unsigned long offset;
Mingming Cao617ba132006-10-11 01:20:53 -07002470 ext4_fsblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002471 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002472 struct ext4_group_desc * gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002473
Mingming Cao617ba132006-10-11 01:20:53 -07002474 if (!ext4_valid_inum(sb, ino)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002475 /*
2476 * This error is already checked for in namei.c unless we are
2477 * looking at an NFS filehandle, in which case no error
2478 * report is needed
2479 */
2480 return 0;
2481 }
2482
Mingming Cao617ba132006-10-11 01:20:53 -07002483 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2484 if (block_group >= EXT4_SB(sb)->s_groups_count) {
2485 ext4_error(sb,"ext4_get_inode_block","group >= groups count");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002486 return 0;
2487 }
2488 smp_rmb();
Mingming Cao617ba132006-10-11 01:20:53 -07002489 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2490 desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2491 bh = EXT4_SB(sb)->s_group_desc[group_desc];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002492 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002493 ext4_error (sb, "ext4_get_inode_block",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002494 "Descriptor not loaded");
2495 return 0;
2496 }
2497
Alexandre Ratchov0d1ee422006-10-11 01:21:14 -07002498 gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2499 desc * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002500 /*
2501 * Figure out the offset within the block group inode table
2502 */
Mingming Cao617ba132006-10-11 01:20:53 -07002503 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2504 EXT4_INODE_SIZE(sb);
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07002505 block = ext4_inode_table(sb, gdp) +
2506 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002507
2508 iloc->block_group = block_group;
Mingming Cao617ba132006-10-11 01:20:53 -07002509 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002510 return block;
2511}
2512
2513/*
Mingming Cao617ba132006-10-11 01:20:53 -07002514 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002515 * underlying buffer_head on success. If 'in_mem' is true, we have all
2516 * data in memory that is needed to recreate the on-disk version of this
2517 * inode.
2518 */
Mingming Cao617ba132006-10-11 01:20:53 -07002519static int __ext4_get_inode_loc(struct inode *inode,
2520 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002521{
Mingming Cao617ba132006-10-11 01:20:53 -07002522 ext4_fsblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002523 struct buffer_head *bh;
2524
Mingming Cao617ba132006-10-11 01:20:53 -07002525 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002526 if (!block)
2527 return -EIO;
2528
2529 bh = sb_getblk(inode->i_sb, block);
2530 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002531 ext4_error (inode->i_sb, "ext4_get_inode_loc",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002532 "unable to read inode block - "
Mingming Cao2ae02102006-10-11 01:21:11 -07002533 "inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002534 inode->i_ino, block);
2535 return -EIO;
2536 }
2537 if (!buffer_uptodate(bh)) {
2538 lock_buffer(bh);
2539 if (buffer_uptodate(bh)) {
2540 /* someone brought it uptodate while we waited */
2541 unlock_buffer(bh);
2542 goto has_buffer;
2543 }
2544
2545 /*
2546 * If we have all information of the inode in memory and this
2547 * is the only valid inode in the block, we need not read the
2548 * block.
2549 */
2550 if (in_mem) {
2551 struct buffer_head *bitmap_bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002552 struct ext4_group_desc *desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002553 int inodes_per_buffer;
2554 int inode_offset, i;
Avantika Mathurfd2d4292008-01-28 23:58:27 -05002555 ext4_group_t block_group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002556 int start;
2557
2558 block_group = (inode->i_ino - 1) /
Mingming Cao617ba132006-10-11 01:20:53 -07002559 EXT4_INODES_PER_GROUP(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002560 inodes_per_buffer = bh->b_size /
Mingming Cao617ba132006-10-11 01:20:53 -07002561 EXT4_INODE_SIZE(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002562 inode_offset = ((inode->i_ino - 1) %
Mingming Cao617ba132006-10-11 01:20:53 -07002563 EXT4_INODES_PER_GROUP(inode->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002564 start = inode_offset & ~(inodes_per_buffer - 1);
2565
2566 /* Is the inode bitmap in cache? */
Mingming Cao617ba132006-10-11 01:20:53 -07002567 desc = ext4_get_group_desc(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002568 block_group, NULL);
2569 if (!desc)
2570 goto make_io;
2571
2572 bitmap_bh = sb_getblk(inode->i_sb,
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07002573 ext4_inode_bitmap(inode->i_sb, desc));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002574 if (!bitmap_bh)
2575 goto make_io;
2576
2577 /*
2578 * If the inode bitmap isn't in cache then the
2579 * optimisation may end up performing two reads instead
2580 * of one, so skip it.
2581 */
2582 if (!buffer_uptodate(bitmap_bh)) {
2583 brelse(bitmap_bh);
2584 goto make_io;
2585 }
2586 for (i = start; i < start + inodes_per_buffer; i++) {
2587 if (i == inode_offset)
2588 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07002589 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002590 break;
2591 }
2592 brelse(bitmap_bh);
2593 if (i == start + inodes_per_buffer) {
2594 /* all other inodes are free, so skip I/O */
2595 memset(bh->b_data, 0, bh->b_size);
2596 set_buffer_uptodate(bh);
2597 unlock_buffer(bh);
2598 goto has_buffer;
2599 }
2600 }
2601
2602make_io:
2603 /*
2604 * There are other valid inodes in the buffer, this inode
2605 * has in-inode xattrs, or we don't have this inode in memory.
2606 * Read the block from disk.
2607 */
2608 get_bh(bh);
2609 bh->b_end_io = end_buffer_read_sync;
2610 submit_bh(READ_META, bh);
2611 wait_on_buffer(bh);
2612 if (!buffer_uptodate(bh)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002613 ext4_error(inode->i_sb, "ext4_get_inode_loc",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002614 "unable to read inode block - "
Mingming Cao2ae02102006-10-11 01:21:11 -07002615 "inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002616 inode->i_ino, block);
2617 brelse(bh);
2618 return -EIO;
2619 }
2620 }
2621has_buffer:
2622 iloc->bh = bh;
2623 return 0;
2624}
2625
Mingming Cao617ba132006-10-11 01:20:53 -07002626int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002627{
2628 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07002629 return __ext4_get_inode_loc(inode, iloc,
2630 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002631}
2632
Mingming Cao617ba132006-10-11 01:20:53 -07002633void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002634{
Mingming Cao617ba132006-10-11 01:20:53 -07002635 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002636
2637 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07002638 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002639 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07002640 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002641 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07002642 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002643 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07002644 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002645 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07002646 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002647 inode->i_flags |= S_DIRSYNC;
2648}
2649
Jan Karaff9ddf72007-07-18 09:24:20 -04002650/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2651void ext4_get_inode_flags(struct ext4_inode_info *ei)
2652{
2653 unsigned int flags = ei->vfs_inode.i_flags;
2654
2655 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
2656 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
2657 if (flags & S_SYNC)
2658 ei->i_flags |= EXT4_SYNC_FL;
2659 if (flags & S_APPEND)
2660 ei->i_flags |= EXT4_APPEND_FL;
2661 if (flags & S_IMMUTABLE)
2662 ei->i_flags |= EXT4_IMMUTABLE_FL;
2663 if (flags & S_NOATIME)
2664 ei->i_flags |= EXT4_NOATIME_FL;
2665 if (flags & S_DIRSYNC)
2666 ei->i_flags |= EXT4_DIRSYNC_FL;
2667}
2668
Mingming Cao617ba132006-10-11 01:20:53 -07002669void ext4_read_inode(struct inode * inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002670{
Mingming Cao617ba132006-10-11 01:20:53 -07002671 struct ext4_iloc iloc;
2672 struct ext4_inode *raw_inode;
2673 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002674 struct buffer_head *bh;
2675 int block;
2676
Mingming Cao617ba132006-10-11 01:20:53 -07002677#ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2678 ei->i_acl = EXT4_ACL_NOT_CACHED;
2679 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002680#endif
2681 ei->i_block_alloc_info = NULL;
2682
Mingming Cao617ba132006-10-11 01:20:53 -07002683 if (__ext4_get_inode_loc(inode, &iloc, 0))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002684 goto bad_inode;
2685 bh = iloc.bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002686 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002687 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2688 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2689 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2690 if(!(test_opt (inode->i_sb, NO_UID32))) {
2691 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2692 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2693 }
2694 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2695 inode->i_size = le32_to_cpu(raw_inode->i_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002696
2697 ei->i_state = 0;
2698 ei->i_dir_start_lookup = 0;
2699 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2700 /* We now have enough fields to check if the inode was active or not.
2701 * This is needed because nfsd might try to access dead inodes
2702 * the test is that same one that e2fsck uses
2703 * NeilBrown 1999oct15
2704 */
2705 if (inode->i_nlink == 0) {
2706 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07002707 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002708 /* this inode is deleted */
2709 brelse (bh);
2710 goto bad_inode;
2711 }
2712 /* The only unlinked inodes we let through here have
2713 * valid i_mode and are being read by the orphan
2714 * recovery code: that's fine, we're about to complete
2715 * the process of deleting those. */
2716 }
2717 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2718 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002719 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07002720 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2721 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07002722 ei->i_file_acl |=
2723 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002724 if (!S_ISREG(inode->i_mode)) {
2725 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2726 } else {
2727 inode->i_size |=
2728 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2729 }
2730 ei->i_disksize = inode->i_size;
2731 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2732 ei->i_block_group = iloc.block_group;
2733 /*
2734 * NOTE! The in-memory inode i_data array is in little-endian order
2735 * even on big-endian machines: we do NOT byteswap the block numbers!
2736 */
Mingming Cao617ba132006-10-11 01:20:53 -07002737 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002738 ei->i_data[block] = raw_inode->i_block[block];
2739 INIT_LIST_HEAD(&ei->i_orphan);
2740
Mingming Cao617ba132006-10-11 01:20:53 -07002741 if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2742 EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002743 /*
2744 * When mke2fs creates big inodes it does not zero out
Mingming Cao617ba132006-10-11 01:20:53 -07002745 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002746 * so ignore those first few inodes.
2747 */
2748 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07002749 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07002750 EXT4_INODE_SIZE(inode->i_sb)) {
2751 brelse (bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002752 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07002753 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002754 if (ei->i_extra_isize == 0) {
2755 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07002756 ei->i_extra_isize = sizeof(struct ext4_inode) -
2757 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002758 } else {
2759 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07002760 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002761 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07002762 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2763 ei->i_state |= EXT4_STATE_XATTR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002764 }
2765 } else
2766 ei->i_extra_isize = 0;
2767
Kalpak Shahef7f3832007-07-18 09:15:20 -04002768 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
2769 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
2770 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
2771 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
2772
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002773 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002774 inode->i_op = &ext4_file_inode_operations;
2775 inode->i_fop = &ext4_file_operations;
2776 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002777 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002778 inode->i_op = &ext4_dir_inode_operations;
2779 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002780 } else if (S_ISLNK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002781 if (ext4_inode_is_fast_symlink(inode))
2782 inode->i_op = &ext4_fast_symlink_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002783 else {
Mingming Cao617ba132006-10-11 01:20:53 -07002784 inode->i_op = &ext4_symlink_inode_operations;
2785 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002786 }
2787 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002788 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002789 if (raw_inode->i_block[0])
2790 init_special_inode(inode, inode->i_mode,
2791 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2792 else
2793 init_special_inode(inode, inode->i_mode,
2794 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2795 }
2796 brelse (iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002797 ext4_set_inode_flags(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002798 return;
2799
2800bad_inode:
2801 make_bad_inode(inode);
2802 return;
2803}
2804
2805/*
2806 * Post the struct inode info into an on-disk inode location in the
2807 * buffer-cache. This gobbles the caller's reference to the
2808 * buffer_head in the inode location struct.
2809 *
2810 * The caller must have write access to iloc->bh.
2811 */
Mingming Cao617ba132006-10-11 01:20:53 -07002812static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002813 struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -07002814 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002815{
Mingming Cao617ba132006-10-11 01:20:53 -07002816 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2817 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002818 struct buffer_head *bh = iloc->bh;
2819 int err = 0, rc, block;
2820
2821 /* For fields not not tracking in the in-memory inode,
2822 * initialise them to zero for new inodes. */
Mingming Cao617ba132006-10-11 01:20:53 -07002823 if (ei->i_state & EXT4_STATE_NEW)
2824 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002825
Jan Karaff9ddf72007-07-18 09:24:20 -04002826 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002827 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2828 if(!(test_opt(inode->i_sb, NO_UID32))) {
2829 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2830 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2831/*
2832 * Fix up interoperability with old kernels. Otherwise, old inodes get
2833 * re-used with the upper 16 bits of the uid/gid intact
2834 */
2835 if(!ei->i_dtime) {
2836 raw_inode->i_uid_high =
2837 cpu_to_le16(high_16_bits(inode->i_uid));
2838 raw_inode->i_gid_high =
2839 cpu_to_le16(high_16_bits(inode->i_gid));
2840 } else {
2841 raw_inode->i_uid_high = 0;
2842 raw_inode->i_gid_high = 0;
2843 }
2844 } else {
2845 raw_inode->i_uid_low =
2846 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2847 raw_inode->i_gid_low =
2848 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2849 raw_inode->i_uid_high = 0;
2850 raw_inode->i_gid_high = 0;
2851 }
2852 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2853 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002854
2855 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
2856 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
2857 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
2858 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
2859
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002860 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2861 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2862 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07002863 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2864 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07002865 raw_inode->i_file_acl_high =
2866 cpu_to_le16(ei->i_file_acl >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002867 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2868 if (!S_ISREG(inode->i_mode)) {
2869 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2870 } else {
2871 raw_inode->i_size_high =
2872 cpu_to_le32(ei->i_disksize >> 32);
2873 if (ei->i_disksize > 0x7fffffffULL) {
2874 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07002875 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2876 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2877 EXT4_SB(sb)->s_es->s_rev_level ==
2878 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002879 /* If this is the first large file
2880 * created, add a flag to the superblock.
2881 */
Mingming Cao617ba132006-10-11 01:20:53 -07002882 err = ext4_journal_get_write_access(handle,
2883 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002884 if (err)
2885 goto out_brelse;
Mingming Cao617ba132006-10-11 01:20:53 -07002886 ext4_update_dynamic_rev(sb);
2887 EXT4_SET_RO_COMPAT_FEATURE(sb,
2888 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002889 sb->s_dirt = 1;
2890 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07002891 err = ext4_journal_dirty_metadata(handle,
2892 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002893 }
2894 }
2895 }
2896 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2897 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2898 if (old_valid_dev(inode->i_rdev)) {
2899 raw_inode->i_block[0] =
2900 cpu_to_le32(old_encode_dev(inode->i_rdev));
2901 raw_inode->i_block[1] = 0;
2902 } else {
2903 raw_inode->i_block[0] = 0;
2904 raw_inode->i_block[1] =
2905 cpu_to_le32(new_encode_dev(inode->i_rdev));
2906 raw_inode->i_block[2] = 0;
2907 }
Mingming Cao617ba132006-10-11 01:20:53 -07002908 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002909 raw_inode->i_block[block] = ei->i_data[block];
2910
2911 if (ei->i_extra_isize)
2912 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2913
Mingming Cao617ba132006-10-11 01:20:53 -07002914 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2915 rc = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002916 if (!err)
2917 err = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07002918 ei->i_state &= ~EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002919
2920out_brelse:
2921 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002922 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002923 return err;
2924}
2925
2926/*
Mingming Cao617ba132006-10-11 01:20:53 -07002927 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002928 *
2929 * We are called from a few places:
2930 *
2931 * - Within generic_file_write() for O_SYNC files.
2932 * Here, there will be no transaction running. We wait for any running
2933 * trasnaction to commit.
2934 *
2935 * - Within sys_sync(), kupdate and such.
2936 * We wait on commit, if tol to.
2937 *
2938 * - Within prune_icache() (PF_MEMALLOC == true)
2939 * Here we simply return. We can't afford to block kswapd on the
2940 * journal commit.
2941 *
2942 * In all cases it is actually safe for us to return without doing anything,
2943 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07002944 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002945 * knfsd.
2946 *
2947 * Note that we are absolutely dependent upon all inode dirtiers doing the
2948 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2949 * which we are interested.
2950 *
2951 * It would be a bug for them to not do this. The code:
2952 *
2953 * mark_inode_dirty(inode)
2954 * stuff();
2955 * inode->i_size = expr;
2956 *
2957 * is in error because a kswapd-driven write_inode() could occur while
2958 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2959 * will no longer be on the superblock's dirty inode list.
2960 */
Mingming Cao617ba132006-10-11 01:20:53 -07002961int ext4_write_inode(struct inode *inode, int wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002962{
2963 if (current->flags & PF_MEMALLOC)
2964 return 0;
2965
Mingming Cao617ba132006-10-11 01:20:53 -07002966 if (ext4_journal_current_handle()) {
Mingming Caob38bd332007-07-19 01:48:35 -07002967 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002968 dump_stack();
2969 return -EIO;
2970 }
2971
2972 if (!wait)
2973 return 0;
2974
Mingming Cao617ba132006-10-11 01:20:53 -07002975 return ext4_force_commit(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002976}
2977
2978/*
Mingming Cao617ba132006-10-11 01:20:53 -07002979 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002980 *
2981 * Called from notify_change.
2982 *
2983 * We want to trap VFS attempts to truncate the file as soon as
2984 * possible. In particular, we want to make sure that when the VFS
2985 * shrinks i_size, we put the inode on the orphan list and modify
2986 * i_disksize immediately, so that during the subsequent flushing of
2987 * dirty pages and freeing of disk blocks, we can guarantee that any
2988 * commit will leave the blocks being flushed in an unused state on
2989 * disk. (On recovery, the inode will get truncated and the blocks will
2990 * be freed, so we have a strong guarantee that no future commit will
2991 * leave these blocks visible to the user.)
2992 *
2993 * Called with inode->sem down.
2994 */
Mingming Cao617ba132006-10-11 01:20:53 -07002995int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002996{
2997 struct inode *inode = dentry->d_inode;
2998 int error, rc = 0;
2999 const unsigned int ia_valid = attr->ia_valid;
3000
3001 error = inode_change_ok(inode, attr);
3002 if (error)
3003 return error;
3004
3005 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3006 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3007 handle_t *handle;
3008
3009 /* (user+group)*(old+new) structure, inode write (sb,
3010 * inode block, ? - but truncate inode update has it) */
Mingming Cao617ba132006-10-11 01:20:53 -07003011 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
3012 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003013 if (IS_ERR(handle)) {
3014 error = PTR_ERR(handle);
3015 goto err_out;
3016 }
3017 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3018 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07003019 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003020 return error;
3021 }
3022 /* Update corresponding info in inode so that everything is in
3023 * one transaction */
3024 if (attr->ia_valid & ATTR_UID)
3025 inode->i_uid = attr->ia_uid;
3026 if (attr->ia_valid & ATTR_GID)
3027 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07003028 error = ext4_mark_inode_dirty(handle, inode);
3029 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003030 }
3031
3032 if (S_ISREG(inode->i_mode) &&
3033 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3034 handle_t *handle;
3035
Mingming Cao617ba132006-10-11 01:20:53 -07003036 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003037 if (IS_ERR(handle)) {
3038 error = PTR_ERR(handle);
3039 goto err_out;
3040 }
3041
Mingming Cao617ba132006-10-11 01:20:53 -07003042 error = ext4_orphan_add(handle, inode);
3043 EXT4_I(inode)->i_disksize = attr->ia_size;
3044 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003045 if (!error)
3046 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07003047 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003048 }
3049
3050 rc = inode_setattr(inode, attr);
3051
Mingming Cao617ba132006-10-11 01:20:53 -07003052 /* If inode_setattr's call to ext4_truncate failed to get a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003053 * transaction handle at all, we need to clean up the in-core
3054 * orphan list manually. */
3055 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003056 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003057
3058 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07003059 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003060
3061err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07003062 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003063 if (!error)
3064 error = rc;
3065 return error;
3066}
3067
3068
3069/*
3070 * How many blocks doth make a writepage()?
3071 *
3072 * With N blocks per page, it may be:
3073 * N data blocks
3074 * 2 indirect block
3075 * 2 dindirect
3076 * 1 tindirect
3077 * N+5 bitmap blocks (from the above)
3078 * N+5 group descriptor summary blocks
3079 * 1 inode block
3080 * 1 superblock.
Mingming Cao617ba132006-10-11 01:20:53 -07003081 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003082 *
Mingming Cao617ba132006-10-11 01:20:53 -07003083 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003084 *
3085 * With ordered or writeback data it's the same, less the N data blocks.
3086 *
3087 * If the inode's direct blocks can hold an integral number of pages then a
3088 * page cannot straddle two indirect blocks, and we can only touch one indirect
3089 * and dindirect block, and the "5" above becomes "3".
3090 *
3091 * This still overestimates under most circumstances. If we were to pass the
3092 * start and end offsets in here as well we could do block_to_path() on each
3093 * block and work out the exact number of indirects which are touched. Pah.
3094 */
3095
Alex Tomasa86c6182006-10-11 01:21:03 -07003096int ext4_writepage_trans_blocks(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003097{
Mingming Cao617ba132006-10-11 01:20:53 -07003098 int bpp = ext4_journal_blocks_per_page(inode);
3099 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003100 int ret;
3101
Alex Tomasa86c6182006-10-11 01:21:03 -07003102 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3103 return ext4_ext_writepage_trans_blocks(inode, bpp);
3104
Mingming Cao617ba132006-10-11 01:20:53 -07003105 if (ext4_should_journal_data(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003106 ret = 3 * (bpp + indirects) + 2;
3107 else
3108 ret = 2 * (bpp + indirects) + 2;
3109
3110#ifdef CONFIG_QUOTA
3111 /* We know that structure was already allocated during DQUOT_INIT so
3112 * we will be updating only the data blocks + inodes */
Mingming Cao617ba132006-10-11 01:20:53 -07003113 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003114#endif
3115
3116 return ret;
3117}
3118
3119/*
Mingming Cao617ba132006-10-11 01:20:53 -07003120 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003121 * Give this, we know that the caller already has write access to iloc->bh.
3122 */
Mingming Cao617ba132006-10-11 01:20:53 -07003123int ext4_mark_iloc_dirty(handle_t *handle,
3124 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003125{
3126 int err = 0;
3127
3128 /* the do_update_inode consumes one bh->b_count */
3129 get_bh(iloc->bh);
3130
Mingming Caodab291a2006-10-11 01:21:01 -07003131 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Mingming Cao617ba132006-10-11 01:20:53 -07003132 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003133 put_bh(iloc->bh);
3134 return err;
3135}
3136
3137/*
3138 * On success, We end up with an outstanding reference count against
3139 * iloc->bh. This _must_ be cleaned up later.
3140 */
3141
3142int
Mingming Cao617ba132006-10-11 01:20:53 -07003143ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3144 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003145{
3146 int err = 0;
3147 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07003148 err = ext4_get_inode_loc(inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003149 if (!err) {
3150 BUFFER_TRACE(iloc->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07003151 err = ext4_journal_get_write_access(handle, iloc->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003152 if (err) {
3153 brelse(iloc->bh);
3154 iloc->bh = NULL;
3155 }
3156 }
3157 }
Mingming Cao617ba132006-10-11 01:20:53 -07003158 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003159 return err;
3160}
3161
3162/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003163 * Expand an inode by new_extra_isize bytes.
3164 * Returns 0 on success or negative error number on failure.
3165 */
3166int ext4_expand_extra_isize(struct inode *inode, unsigned int new_extra_isize,
3167 struct ext4_iloc iloc, handle_t *handle)
3168{
3169 struct ext4_inode *raw_inode;
3170 struct ext4_xattr_ibody_header *header;
3171 struct ext4_xattr_entry *entry;
3172
3173 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
3174 return 0;
3175
3176 raw_inode = ext4_raw_inode(&iloc);
3177
3178 header = IHDR(inode, raw_inode);
3179 entry = IFIRST(header);
3180
3181 /* No extended attributes present */
3182 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
3183 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
3184 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
3185 new_extra_isize);
3186 EXT4_I(inode)->i_extra_isize = new_extra_isize;
3187 return 0;
3188 }
3189
3190 /* try to expand with EAs present */
3191 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
3192 raw_inode, handle);
3193}
3194
3195/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003196 * What we do here is to mark the in-core inode as clean with respect to inode
3197 * dirtiness (it may still be data-dirty).
3198 * This means that the in-core inode may be reaped by prune_icache
3199 * without having to perform any I/O. This is a very good thing,
3200 * because *any* task may call prune_icache - even ones which
3201 * have a transaction open against a different journal.
3202 *
3203 * Is this cheating? Not really. Sure, we haven't written the
3204 * inode out, but prune_icache isn't a user-visible syncing function.
3205 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3206 * we start and wait on commits.
3207 *
3208 * Is this efficient/effective? Well, we're being nice to the system
3209 * by cleaning up our inodes proactively so they can be reaped
3210 * without I/O. But we are potentially leaving up to five seconds'
3211 * worth of inodes floating about which prune_icache wants us to
3212 * write out. One way to fix that would be to get prune_icache()
3213 * to do a write_super() to free up some memory. It has the desired
3214 * effect.
3215 */
Mingming Cao617ba132006-10-11 01:20:53 -07003216int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003217{
Mingming Cao617ba132006-10-11 01:20:53 -07003218 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003219 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3220 static unsigned int mnt_count;
3221 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003222
3223 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07003224 err = ext4_reserve_inode_write(handle, inode, &iloc);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003225 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
3226 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
3227 /*
3228 * We need extra buffer credits since we may write into EA block
3229 * with this same handle. If journal_extend fails, then it will
3230 * only result in a minor loss of functionality for that inode.
3231 * If this is felt to be critical, then e2fsck should be run to
3232 * force a large enough s_min_extra_isize.
3233 */
3234 if ((jbd2_journal_extend(handle,
3235 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
3236 ret = ext4_expand_extra_isize(inode,
3237 sbi->s_want_extra_isize,
3238 iloc, handle);
3239 if (ret) {
3240 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04003241 if (mnt_count !=
3242 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003243 ext4_warning(inode->i_sb, __FUNCTION__,
3244 "Unable to expand inode %lu. Delete"
3245 " some EAs or run e2fsck.",
3246 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04003247 mnt_count =
3248 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04003249 }
3250 }
3251 }
3252 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003253 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07003254 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003255 return err;
3256}
3257
3258/*
Mingming Cao617ba132006-10-11 01:20:53 -07003259 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003260 *
3261 * We're really interested in the case where a file is being extended.
3262 * i_size has been changed by generic_commit_write() and we thus need
3263 * to include the updated inode in the current transaction.
3264 *
3265 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3266 * are allocated to the file.
3267 *
3268 * If the inode is marked synchronous, we don't honour that here - doing
3269 * so would cause a commit on atime updates, which we don't bother doing.
3270 * We handle synchronous inodes at the highest possible level.
3271 */
Mingming Cao617ba132006-10-11 01:20:53 -07003272void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003273{
Mingming Cao617ba132006-10-11 01:20:53 -07003274 handle_t *current_handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003275 handle_t *handle;
3276
Mingming Cao617ba132006-10-11 01:20:53 -07003277 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003278 if (IS_ERR(handle))
3279 goto out;
3280 if (current_handle &&
3281 current_handle->h_transaction != handle->h_transaction) {
3282 /* This task has a transaction open against a different fs */
3283 printk(KERN_EMERG "%s: transactions do not match!\n",
3284 __FUNCTION__);
3285 } else {
3286 jbd_debug(5, "marking dirty. outer handle=%p\n",
3287 current_handle);
Mingming Cao617ba132006-10-11 01:20:53 -07003288 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003289 }
Mingming Cao617ba132006-10-11 01:20:53 -07003290 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003291out:
3292 return;
3293}
3294
3295#if 0
3296/*
3297 * Bind an inode's backing buffer_head into this transaction, to prevent
3298 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07003299 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003300 * returns no iloc structure, so the caller needs to repeat the iloc
3301 * lookup to mark the inode dirty later.
3302 */
Mingming Cao617ba132006-10-11 01:20:53 -07003303static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003304{
Mingming Cao617ba132006-10-11 01:20:53 -07003305 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003306
3307 int err = 0;
3308 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07003309 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003310 if (!err) {
3311 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07003312 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003313 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07003314 err = ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003315 iloc.bh);
3316 brelse(iloc.bh);
3317 }
3318 }
Mingming Cao617ba132006-10-11 01:20:53 -07003319 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003320 return err;
3321}
3322#endif
3323
Mingming Cao617ba132006-10-11 01:20:53 -07003324int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003325{
3326 journal_t *journal;
3327 handle_t *handle;
3328 int err;
3329
3330 /*
3331 * We have to be very careful here: changing a data block's
3332 * journaling status dynamically is dangerous. If we write a
3333 * data block to the journal, change the status and then delete
3334 * that block, we risk forgetting to revoke the old log record
3335 * from the journal and so a subsequent replay can corrupt data.
3336 * So, first we make sure that the journal is empty and that
3337 * nobody is changing anything.
3338 */
3339
Mingming Cao617ba132006-10-11 01:20:53 -07003340 journal = EXT4_JOURNAL(inode);
Dave Hansend6995942007-07-18 08:33:51 -04003341 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003342 return -EROFS;
3343
Mingming Caodab291a2006-10-11 01:21:01 -07003344 jbd2_journal_lock_updates(journal);
3345 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003346
3347 /*
3348 * OK, there are no updates running now, and all cached data is
3349 * synced to disk. We are now in a completely consistent state
3350 * which doesn't have anything in the journal, and we know that
3351 * no filesystem updates are running, so it is safe to modify
3352 * the inode's in-core data-journaling state flag now.
3353 */
3354
3355 if (val)
Mingming Cao617ba132006-10-11 01:20:53 -07003356 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003357 else
Mingming Cao617ba132006-10-11 01:20:53 -07003358 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3359 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003360
Mingming Caodab291a2006-10-11 01:21:01 -07003361 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003362
3363 /* Finally we can mark the inode as dirty. */
3364
Mingming Cao617ba132006-10-11 01:20:53 -07003365 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003366 if (IS_ERR(handle))
3367 return PTR_ERR(handle);
3368
Mingming Cao617ba132006-10-11 01:20:53 -07003369 err = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003370 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07003371 ext4_journal_stop(handle);
3372 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003373
3374 return err;
3375}